Detailed Description
Exemplary embodiments of the present disclosure are described below in conjunction with the accompanying drawings, which include various details of the embodiments of the present disclosure to facilitate understanding, and should be considered as merely exemplary. Accordingly, one of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.
In the technical scheme of the disclosure, the acquisition, storage, application and the like of the related personal information of the user all conform to the regulations of related laws and regulations, necessary security measures are taken, and the public order harmony is not violated.
The embodiment of the disclosure provides an information interaction method, device, electronic equipment and storage medium based on a large language model. The information interaction method based on the large language model comprises the steps of responding to the acquired demand description information, processing the demand description information by utilizing the large language model to obtain a task execution strategy matched with the demand description information, wherein the task execution strategy comprises a plurality of demand task attributes, at least one of the plurality of demand task attributes is matched with demand intention represented by the demand description information, determining a strategy execution result related to the task execution strategy according to the demand task attributes in the task execution strategy, generating feedback information according to the strategy execution result, and displaying the feedback information.
According to the embodiment of the disclosure, the requirement description information is processed by using the large language model to obtain the task execution strategy, so that part of requirement task attributes in the task execution strategy are matched with the requirement intents represented by the requirement description information, the task execution strategy containing a plurality of requirement task attributes can further satisfy potential requirement intents which are not expressed by the requirement description information on the basis of satisfying the requirement intents represented by the requirement description information, a strategy execution result is determined according to the requirement task attributes in the task execution strategy, feedback information is further generated, the potential requirement intents of a target object can be sufficiently satisfied by the feedback information, the input operation process of inputting description texts representing the potential requirement intents in the next information interaction process of the target object is reduced, and the information acquisition efficiency of the target object is improved.
FIG. 1 schematically illustrates an exemplary system architecture to which large language model based information interaction methods and apparatus may be applied, according to embodiments of the present disclosure.
It should be noted that fig. 1 is only an example of a system architecture to which embodiments of the present disclosure may be applied to assist those skilled in the art in understanding the technical content of the present disclosure, but does not mean that embodiments of the present disclosure may not be used in other devices, systems, environments, or scenarios. For example, in another embodiment, an exemplary system architecture to which the information interaction method and apparatus based on a large language model may be applied may include a terminal device, but the terminal device may implement the information interaction method and apparatus based on a large language model provided by the embodiments of the present disclosure without interacting with a server.
As shown in fig. 1, a system architecture 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used as a medium to provide communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired and/or wireless communication links, and the like.
The user may interact with the server 105 via the network 104 using the terminal devices 101, 102, 103 to receive or send messages or the like. Various communication client applications may be installed on the terminal devices 101, 102, 103, such as a knowledge reading class application, a web browser application, a search class application, an instant messaging tool, a mailbox client and/or social platform software, etc. (as examples only).
The terminal devices 101, 102, 103 may be a variety of electronic devices having a display screen and supporting web browsing, including but not limited to smartphones, tablets, laptop and desktop computers, and the like.
The server 105 may be a server providing various services, such as a background management server (by way of example only) providing support for content browsed by the user using the terminal devices 101, 102, 103. The background management server may analyze and process the received data such as the user request, and feed back the processing result (e.g., the web page, information, or data obtained or generated according to the user request) to the terminal device.
The server 105 may also be a cloud server, also called a cloud computing server or a cloud host, and is a host product in a cloud computing service system, so as to solve the defects of large management difficulty and weak service expansibility in the traditional physical hosts and VPS service ('' Virtual PRIVATE SERVER '' or 'VPS' for short). The server may also be a server of a distributed system or a server that incorporates a blockchain.
It should be noted that, the information interaction method based on the large language model provided in the embodiments of the present disclosure may be generally executed by the server 105. Accordingly, the information interaction device based on the large language model provided by the embodiments of the present disclosure may be generally disposed in the server 105. The information interaction method based on the large language model provided by the embodiments of the present disclosure may also be performed by a server or a server cluster that is different from the server 105 and is capable of communicating with the terminal devices 101, 102, 103 and/or the server 105. Accordingly, the information interaction device based on the large language model provided by the embodiments of the present disclosure may also be provided in a server or a server cluster different from the server 105 and capable of communicating with the terminal devices 101, 102, 103 and/or the server 105.
Or the information interaction method based on the large language model provided by the embodiment of the present disclosure may be performed by the terminal device 101, 102 or 103. Accordingly, the information interaction device based on the large language model provided by the embodiment of the present disclosure may also be provided in the terminal device 101, 102 or 103.
It should be understood that the number of terminal devices, networks and servers in fig. 1 is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation.
Fig. 2 schematically illustrates a flowchart of a method of information interaction based on a large language model according to an embodiment of the present disclosure.
As shown in FIG. 2, the information interaction method based on the large language model comprises operations S210-S230.
In response to obtaining the requirement description information, processing the requirement description information by using a large language model to obtain a task execution strategy matched with the requirement description information in operation S210, wherein the task execution strategy comprises a plurality of requirement task attributes, and at least one of the plurality of requirement task attributes is matched with the requirement intention represented by the requirement description information.
In operation S220, a policy execution result related to the task execution policy is determined according to the required task attribute in the task execution policy.
In operation S230, feedback information is generated according to the policy execution result and displayed.
According to an embodiment of the present disclosure, the requirement description information may be natural language information for characterizing an information acquisition requirement of the target object according to an embodiment of the present disclosure. For example, the demand description information may be "how the xx marketing scheme performs in the near future," which may characterize the target object's acquisition demand intention for revenue information of the xx marketing scheme.
According to an embodiment of the present disclosure, the requirement description information may be acquired based on an input operation of the target object with respect to an information input box in the interactive interface. But not limited to this, the requirement description information may be obtained in other manners, for example, the requirement description information represented by the sound information may be obtained by identifying the requirement description information through a voice recognition algorithm based on the sound information that is collected by the sound collecting device and expressed by the target object, and the embodiment of the disclosure does not limit the specific manner of obtaining the requirement description information.
According to embodiments of the present disclosure, a large language model (LLM: large Language Model) may include a deep learning model trained using a large amount of text data, which may include, for example, a neural network model built based on a Transformer model or the like, a large language model to understand meaning of language text, and generating natural language text. The large language model may handle natural language tasks. Since large language models typically contain billions of parameters, large scale parameters can help large language models learn complex patterns in natural language data, and thus perform well in natural language processing (NLP: natural Language Processing) tasks.
According to embodiments of the present disclosure, a task execution policy may characterize one or more tasks that need to be performed in order to obtain demand-related information, which may be matched to the demand intent characterized by the demand description information, and may also be matched to the potential demand intent characterized by the demand description information. Among the plurality of demand task attributes of the task execution strategy, one part of the demand task attributes are matched with the demand intention represented by the demand description information, and the other part of the demand task attributes are matched with the potential demand intention represented by the demand description information. For example, the plurality of demand task attributes may include a demand task attribute a, a demand task attribute B, and a demand task attribute C, the demand task attribute a matching the demand intent characterized by the demand description information, the demand task attribute B and the demand task attribute C matching the potential demand intent characterized by the demand description information. The number of demand task attributes in the task execution policy may be greater than the number of demand intents characterized by the demand description information. It should be noted that the potential demand intention may include a demand intention that is deeper than the demand intention on the basis of the demand intention characterized by the demand description information.
According to the embodiment of the disclosure, the large language model can be pre-trained according to business application data, so that the large language model can predict the demand task attribute matched with the demand intention and the potential demand intention (or potential demand intention) according to the demand description text expressed by the natural language. Or the corresponding prompt information can be obtained or edited based on the requirement description information, and the prompt information is processed through a large language model to generate the requirement task attribute matched with the potential requirement intention.
According to embodiments of the present disclosure, the demand task attribute may include task configuration parameters, task types, and the like of a task for acquiring demand-related information matching the demand intent and the potential demand intent, the demand task may be generated based on the demand task attribute, and part or all of the demand-related information may be acquired by executing the demand task.
For example, the requirement task attribute may include attribute parameters such as a table name, a data table generation time, a data name and the like of the data table to be acquired, so that a requirement task may be generated according to the requirement task attribute, the requirement task may include a query statement for querying data, and the data matched with the requirement task attribute in the data table may be acquired by executing the query statement, so as to obtain a requirement task execution result.
According to the embodiment of the disclosure, according to the requirement task attribute in the task execution strategy, determining a strategy execution result related to the task execution strategy, generating a requirement task through the requirement task attribute, obtaining the task execution result through executing the requirement task, wherein the task execution result can contain information matched with the requirement intention and can also contain information matched with the potential requirement intention. So that policy execution results can be obtained from a plurality of task execution results.
According to an embodiment of the present disclosure, generating feedback information according to the policy execution result may include taking the policy execution result as feedback information. But not limited to, or may also include populating a preset configuration template based on policy enforcement results to obtain feedback information. The embodiment of the present disclosure does not limit a specific manner of determining feedback information, as long as the policy execution result can be characterized.
According to embodiments of the present disclosure, processing demand description information using a large language model to obtain a task execution policy that matches the demand description information may include processing the demand description information using the large language model to obtain a multi-level demand intent, and determining the task execution policy based on the multi-level demand intent.
According to an embodiment of the present disclosure, the multi-level demand intention may include a plurality of demand intents whose demand level is continuously expanded, and the demand intention of the current level may be intention information after the intention expansion of the demand intention of the previous level. The large language model can be pre-trained by selecting multiple rounds of interaction information as samples, so that the pre-trained large language model can predict multi-level demand intention matched with the demand description information.
According to embodiments of the present disclosure, determining a task execution policy based on a multi-level demand intent may include determining demand task attributes corresponding to demand intents of respective levels of the multi-level demand intent, and deriving the task execution policy based on the plurality of demand task attributes.
According to the embodiment of the disclosure, the multi-level requirement intention recognition is performed on the requirement description information through the large language model, and the requirement of the extended deeper level can be mined on the requirement description information, so that the operation steps of inputting the potential requirement of the deep level by the target object through multiple rounds of interaction are reduced.
According to the embodiment of the disclosure, processing the demand description information by using the large language model, obtaining the multi-level demand intention may include processing the demand description information and the demand intention of the previous level by using the large language model for the current level, obtaining the demand intention of the current level, wherein the demand intention of the first level in the multi-level demand intention is obtained by processing the demand description information by using the large language model.
According to the embodiment of the disclosure, the detected demand intention of the last level can be processed by iteratively utilizing the large language model based on the strong natural language semantic understanding capability and text prediction capability of the large language model, so that the demand intention of the current level is predicted, and the continuous expansion and mining of the demand intention can be realized, and the obtained multi-level demand intention comprises the potential demand intention represented by the demand description information.
According to the embodiment of the disclosure, processing the demand description information by using the large language model, and obtaining the multi-level demand intention may further include processing the demand description information by using the large language model and obtaining the multi-level demand intention based on the associated query information queried by the demand description information.
According to embodiments of the present disclosure, a part or all of the requirement description information may be used as a query keyword (or query), and the associated query information may be searched using a search tool or a query statement. The associated query information may contain one or more of contextual information, expertise, and the like, related to the demand description information, thereby helping the large language model to understand the semantics of the demand description information expression and predict multi-level demand intents including potential demand intents.
According to the embodiment of the disclosure, the knowledge base related to business service can be constructed so as to conveniently inquire the related inquiry information from the knowledge base according to the requirement description information, so that the correlation between the related inquiry information and the requirement description information is improved.
In accordance with an embodiment of the present disclosure, before determining the task execution policy according to the multi-level demand intents, the information interaction method may further include exposing at least one demand intention of the multi-level demand intents;
According to embodiments of the present disclosure, determining a task execution policy based on the multi-level demand intents may include determining a task execution policy based on a demand intention associated with a selection operation among the multi-level demand intents.
According to the embodiment of the disclosure, the matching degree of the multi-level demand intention and the real intention of the target object can be improved by displaying the demand intention for the target object to confirm, so that the obtained feedback information can truly meet the demand of the target object.
According to the embodiment of the disclosure, determining a strategy execution result related to a task execution strategy according to a demand task attribute in the task execution strategy can comprise determining a processed completion demand task attribute from a plurality of demand task attributes of the current task execution strategy, processing the completion demand task attribute and a task execution result related to the completion demand task attribute by using a large language model to obtain a to-be-processed demand task attribute, updating the current task execution strategy according to the to-be-processed demand task attribute to obtain a new task execution strategy, and determining a strategy execution result according to the new task execution strategy.
According to an embodiment of the present disclosure, the current task execution policy may include a task execution policy obtained after being updated, or may include a task execution policy that is not updated. Completion of the required task attribute may refer to the required task having been generated from the completion task attribute and that the required task has been performed.
According to the embodiment of the disclosure, the processing of the task completion requirement attribute by using the large language model may include inputting a part or all of the task completion requirement attribute which is already executed and completed in the current task execution strategy and a task execution result corresponding to the part or all of the task completion requirement attribute into the large language model, so that the large language model can understand the execution condition of the current task execution strategy, and further judge whether the current task execution strategy needs to be updated, so that the new task execution strategy can be further matched with the requirement intention and the potential requirement intention represented by the requirement description information.
According to the embodiment of the disclosure, the new task execution strategy can be used as the current task execution strategy, and the current task execution strategy is updated in a continuous iterative optimization mode through the current task execution attribute of the completion requirement and the task execution result corresponding to the current task execution attribute of the completion requirement, so that the strategy execution result corresponding to the task execution strategy obtained by the last update can meet the multi-level requirement intention of the target object, the interaction time of the target object for acquiring the information is saved, and meanwhile, the communication space and the occupation of computing resources generated by the information interaction are saved.
According to the embodiment of the disclosure, updating the current task execution policy according to the pending demand task attribute may include updating an outstanding demand task attribute that has not been processed in the current task execution policy according to the pending demand task attribute.
According to the embodiment of the disclosure, the outstanding demand task attribute which is not processed in the current task execution strategy can be updated through the pending demand task attribute, so that the pending demand task attribute in the new task execution strategy obtained after updating can be adapted to the potential demand intention of the target object, and a task execution result can be obtained according to the updated pending demand task attribute. Therefore, the self-thinking of the task execution strategy can be realized according to the large language model, and further the outstanding demand task attribute which is not processed is corrected in time, so that the current task execution strategy is adjusted in time, the calculation resource waste caused by executing the demand task attribute with larger difference with the demand intention is avoided, and the influence of feedback information generated according to the uncorrelated task execution result on the use experience of the target object is avoided.
According to the embodiment of the disclosure, updating the current task execution policy according to the pending demand task attribute may further include updating the completion demand task attribute in the current task execution policy according to the pending demand task attribute.
According to the embodiment of the disclosure, by updating the task attribute of the completion demand in the current task execution strategy, the task can be detected based on the language understanding capability of the large language model, and the correlation between the task of the completion demand attribute and the demand intention or the potential demand intention which is processed currently is smaller, so that the task attribute to be processed can be output to replace the task attribute of the completion demand, further, the predicted error of demand judgment is corrected through self-reactive thinking, or the current task execution strategy is timely adjusted according to the task execution result related to the task attribute to be completed, so that the new task execution strategy can be more suitable for the demand intention and the potential demand intention of the target object, and the influence of feedback information generated according to the irrelevant task execution result on the use experience of the target object is avoided.
In one embodiment of the present disclosure, the attribute of a partially or fully completed demand task in the current task execution policy may also be updated according to the demand task to be processed, and the attribute of a partially or fully incomplete demand task in the current task execution policy may also be updated.
According to embodiments of the present disclosure, the current task execution policy may include execution dependencies between current demand task attributes. The step of updating the current task execution strategy according to the attribute of the to-be-processed demand task can further comprise the step of updating an execution dependency relationship contained in the current task execution strategy, so that the new task execution strategy can execute the demand task according to the new execution dependency relationship, and the suitability of a finally obtained strategy execution result, the demand intention of the target object and the potential demand intention is improved.
According to embodiments of the present disclosure, task execution policies may be characterized based on task execution topologies, nodes in which may characterize demand task attributes, and edge relationships between demand task attributes may characterize execution dependencies.
Fig. 3A schematically illustrates a schematic diagram of a task execution strategy according to an embodiment of the present disclosure.
Fig. 3B schematically illustrates a schematic diagram of a task execution strategy according to another embodiment of the present disclosure.
As shown in fig. 3A, the current task execution policy 300A may include a plurality of demand task attributes, namely, an a demand task attribute 311, a B demand task attribute 321, a C demand task attribute 322, and a D demand task attribute 331. The task execution policy 300A may be characterized by a task execution topology, and the a-requirement task attribute 311, the B-requirement task attribute 321, the C-requirement task attribute 322, and the D-requirement task attribute 331 are nodes in the task execution topology, respectively, and execution dependency relationships among the a-requirement task attribute 311, the B-requirement task attribute 321, the C-requirement task attribute 322, and the D-requirement task attribute 331 are characterized by edge relationships. The task execution policy 300A may represent logical relationships that require task attributes to be processed through the task execution topology. Wherein, the a-demand task attribute 311 and the B-demand task attribute 321 may be completion demand task attributes that have already been processed for completion, and the C-demand task attribute 322 and the D-demand task attribute 331 may be incomplete demand task attributes that have not yet been processed.
In one embodiment of the present disclosure, as shown in fig. 3A and 3B, a large language model may be used to process the a-requirement task attribute 311 and the B-requirement task attribute 321, and the task execution results associated with each of the a-requirement task attribute 311 and the B-requirement task attribute 321, to obtain the task attribute to be processed and update indication information of the task attribute to be processed. The update indication information of the pending task attribute may indicate updating the B-demanded task attribute 321 in the current task execution policy 300A. Thus, the current task execution policy 300A is updated according to the attributes of the task to be processed, and a new task execution policy 300B can be obtained. The new task execution policy 300B may include a B 'demand task attribute 321'.
Fig. 4A schematically illustrates a schematic diagram of a task execution strategy according to yet another embodiment of the present disclosure.
Fig. 4B schematically illustrates a schematic diagram of a task execution strategy according to still another embodiment of the present disclosure.
As shown in fig. 4A, the current task execution policy 400A may include a plurality of required task attributes, namely, an A1 required task attribute 411, a B1 required task attribute 421, a C1 required task attribute 422, and a D1 required task attribute 431. The task execution policy 400A may be characterized by a task execution topology, and the A1 demand task attribute 411, the B1 demand task attribute 421, the C1 demand task attribute 422, and the D1 demand task attribute 431 are nodes in the task execution topology, and the execution dependency between the A1 demand task attribute 411, the B1 demand task attribute 421, the C1 demand task attribute 422, and the D1 demand task attribute 431 is characterized by an edge relationship. Task execution policy 400A may represent logical relationships that require task attributes to be processed through the task execution topology. Wherein, the A1 demand task attribute 411, the B1 demand task attribute 421 and the C1 demand task attribute 422 may be completion demand task attributes that have already been processed for completion, and the D1 demand task attribute 431 may be incomplete demand task attributes that have not yet been processed.
In one embodiment of the present disclosure, as shown in fig. 4A and 4B, the A1 demand task attribute 411, the B1 demand task attribute 421, and the C1 demand task attribute 422, and the task execution results associated with each of the A1 demand task attribute 411, the B1 demand task attribute 421, and the C1 demand task attribute 422 may be processed using a large language model, to obtain the task attribute to be processed and update indication information of the task attribute to be processed. The update indication information of the task attribute to be processed may indicate to update the B1 demand task attribute 421, the C1 demand task attribute 422, and the D1 demand task attribute 431 in the current task execution policy 400A, and the update indication information of the task attribute to be processed may also indicate to update the execution dependency relationship in the current task execution policy 400A. Thus, the current task execution policy 400A is updated according to the attributes of the task to be processed, and a new task execution policy 400B can be obtained. The new task execution policy 400B may include an A1 demand task attribute 411, a B1 'demand task attribute 421', a C1 'demand task attribute 422', and a D1 'demand task attribute 431', and the execution dependency relationship between the demand task attributes of the new task execution policy 400B is updated according to the update prompt information output by the large language model, so that the obtained new task execution policy 400B can perform timely thinking-back and adjustment in combination with the task execution result of the completed demand task attribute, so that the new task execution policy 400B can meet the demand intention of the target object.
According to embodiments of the present disclosure, the new task execution policy includes execution dependencies between new demand task attributes.
According to the embodiment of the disclosure, determining the policy execution result according to the new task execution policy may include determining a first target demand task according to a first target demand task attribute and a first dependent task execution result associated with the first dependent demand task attribute, wherein the new task execution policy includes the first target demand task attribute and the first dependent demand task attribute with an execution dependency relationship therebetween, invoking a first target service resource associated with the first target demand task attribute to execute the first target demand task to obtain a task execution result associated with the first target demand task attribute, and the policy execution result includes a task execution result associated with the first target demand task attribute.
According to an embodiment of the present disclosure, the first target demand task attribute may include a demand task attribute that needs to be processed, the first dependent demand task attribute may include a demand task attribute of a previous level associated with the first target demand task attribute, or may further include a demand task attribute of one level apart or multiple levels apart associated with the first target demand task attribute. The first dependent demand task attribute may be determined by one skilled in the art based on the actual demand as long as the first target demand task attribute is satisfied with a direct or indirect execution dependency relationship.
According to an embodiment of the present disclosure, determining the first target demand task according to the first target demand task attribute and the first dependent task execution result associated with the first dependent demand task attribute may include generating the first target demand task according to a task allocation parameter included in the first target demand task attribute and the first dependent task execution result that has been generated.
According to an embodiment of the present disclosure, the first target demand task attribute may include a target service resource identifier adapted to characterize an association between the first target demand task attribute and the first target service resource. The first target service resource is called to execute the first target demand task through the target service resource identifier, so that the target object can be prevented from executing the first target demand task through sending a resource calling instruction or through resource selection operation, automatic calling of the service resource is realized, and the interactive operation duration of the target object is saved.
According to an embodiment of the disclosure, the first target demand task attribute is characterized based on a hint template. The method comprises the steps of determining a first target demand task according to a first target demand task attribute and a first dependent task execution result related to the first dependent demand task attribute, updating the first target demand task attribute according to the first dependent task execution result related to the first dependent demand task attribute to obtain first target task prompt information, and processing the first target task prompt information by utilizing a large language model to obtain the first target demand task.
According to embodiments of the present disclosure, the hint template (or template) may be a task configuration parameter for helping the large language model understand the first target demand task attribute, the hint template may be based on a hint tag sequence that controls the large language model to accurately predict, and the hint tag sequence may include any type of hint tag such as characters, fields, words, and the like. And filling the first target demand task attribute according to the first dependent task execution result, so that the first target task prompt information can be obtained. Therefore, the first target task prompt information is processed by using the large language model, and the obtained first target demand task can be more accurately matched with the demand intention and the potential demand intention. Therefore, after the first target demand task is executed, the task execution result related to the first target demand task attribute can be more accurately matched with the demand of the target object.
According to embodiments of the present disclosure, the task execution policy further includes execution dependencies between the plurality of demand task attributes. Determining a policy execution result related to the task execution policy according to the requirement task attribute in the task execution policy may include generating a second target requirement task according to a second target requirement task attribute and a task execution result related to a second dependent requirement task attribute, where the task execution policy includes the second dependent requirement task attribute and the second target requirement task attribute, an execution dependency relationship exists between the second dependent requirement task attribute and the second target requirement task attribute, and invoking a second target service resource related to the second target requirement task attribute to execute the second target requirement task to obtain a task execution result related to the second target requirement task attribute, and the policy execution result includes a task execution result related to the second target requirement task attribute.
According to the embodiment of the disclosure, under the condition that the task execution strategy is not updated, according to the execution dependency relationship among the plurality of requirement task attributes indicated by the task execution strategy, the second target requirement task attributes in the task execution strategy are sequentially selected, the second target task is generated according to the second dependency requirement task attributes with the execution dependency relationship with the second target requirement task attributes and the task execution results associated with the second dependency requirement task attributes, so that automatic generation of the target task to be executed can be realized, and the operation steps of filling the task configuration parameters to generate the task to be executed are avoided. And the second target demand task is executed by calling the second target service resource associated with the second target demand task attribute, so that the accurate automatic resource calling can be realized, and the problem of long resource calling operation time caused by manual operation selection is avoided.
According to an embodiment of the present disclosure, the second target demand task attribute comprises a target service resource identification adapted to characterize an association between the second target demand task attribute and the second target service resource.
According to the embodiment of the disclosure, the second target demand task attribute is characterized based on a prompt template, wherein generating the second target demand task according to the second target demand task attribute and the task execution result associated with the second dependent demand task attribute comprises updating the second target demand task attribute according to the task execution result associated with the second dependent demand task attribute to obtain second target task prompt information, and processing the second target task prompt information by utilizing a large language model to obtain the second target demand task.
According to embodiments of the present disclosure, the prompt template (or template) may be a task configuration parameter for helping the large language model understand the second target demand task attribute, the prompt template may be based on a prompt tag sequence that controls the large language model to accurately predict, and the prompt tag sequence may include any type of prompt tag such as characters, fields, words, and the like. And filling the second target demand task attribute according to the second dependent task execution result, so that second target task prompt information can be obtained. Therefore, the large language model is utilized to process the second target task prompt information, and the obtained second target demand task can be more accurately matched with the demand intention and the potential demand intention. Therefore, after the second target demand task is executed, the task execution result related to the attribute of the second target demand task can be more accurately matched with the demand of the target object.
According to the embodiment of the disclosure, the information interaction method based on the large language model can further comprise the step of displaying target task execution results, wherein the target task execution results comprise task execution results associated with the first target requirement task attribute.
According to the embodiment of the disclosure, the task execution result associated with the first target demand task attribute is displayed, so that the target object can browse the currently acquired demand related information in time, and timeliness of information display is improved.
According to the embodiment of the disclosure, the target object can execute any type of operations such as selection operation, editing operation and the like according to the displayed target task execution result, so that the target object can be involved in the execution process of the current task execution strategy in time, and the task execution strategy can be adjusted adaptively according to the current requirement of the target object, so that the target object can acquire the information matched with the requirement of the target object.
According to embodiments of the present disclosure, the target task execution results may further include task execution results associated with the second target demand task attribute.
According to the embodiment of the disclosure, the target object can browse the task execution result associated with the second target requirement task attribute in time, and can execute any type of operations such as selecting operation, editing operation and the like according to the task execution result associated with the second target requirement task attribute, so that the target object can be conveniently and timely involved in the execution process of the current task execution strategy, and the task execution strategy can be adaptively adjusted according to the current requirement of the target object, so that the target object can conveniently acquire the information matched with the requirement of the target object.
Fig. 5 schematically illustrates an application scenario diagram of a large language model-based information interaction method according to an embodiment of the present disclosure.
As shown in fig. 5, in the application scenario 500, the target object may input the requirement description information 501 into the large language model M510 based on the input requirement description information 501 "how the account is integrally performed", and output the task execution policy 510. The 1 st target demand task, e.g., data acquisition task 521, may be generated based on the 1 st target demand task attribute 511 "acquire basic impression data for marketing scheme 1111xxxx using report query cast" in task execution policy 510. The 1 st target task execution result (e.g., may be "i have acquired the base impression data XX") may be obtained by invoking a target service resource, such as the data collection plug-in set 531, associated with the 1 st target demand task attribute 511 to execute the data collection task 521.
As shown in FIG. 5, the 2 nd target demand task, e.g., data analysis task 522, may be generated from the 2 nd target demand task attribute 512 "using report analysis casting, analyzing report data, knowing the current advertising effectiveness, comparative analysis", and the 1 st target task execution result in the task execution policy 510. The target task execution result (which may be, for example, "I have completed analysis XX of reimbursement data") may be obtained by invoking a target service resource, such as data analysis plug-in set 532, associated with target demand task attribute 512 to execute data analysis task 522.
As shown in fig. 5, the 3 rd target demand task, e.g., the effect diagnostic analysis task 523, may be generated from the 3 rd target demand task attribute 513 "use effect analysis plug-in, perform impression effect diagnostic analysis" in the task execution policy 510, and the 2 nd target task execution result. By invoking a target service resource associated with the 3 rd target demand task attribute 513, such as the effects diagnosis analysis plug-in set 533, to execute the data analysis task 523, a 3 rd target task execution result (which may be, for example, "i have completed effects diagnosis, found that CTR (Click-Through-Rate) is problematic") may be obtained.
As shown in FIG. 5, the 4 th target demand task, e.g., the generative rewrite and recommendation task 524, may be generated from the 4 th target demand task attribute 514 "use optimize cast, analyze how to perform account optimization" in the task execution policy 510, and the 3 rd target task execution results. The 4 th target task execution result (which may be, for example, "use optimized cast, the way to analyze how to promote click through rate may be xxx") may be obtained by invoking a target service resource associated with the 4 th target demand task attribute 514, such as the generative optimization plug-in set 534, to execute the generative rewrite and recommendation task 524.
As shown in fig. 5, according to the 5 th target demand task attribute 515 in the task execution policy 510, "summarize analysis results and end tasks using end commands", and the 1 st target task execution result to the 4 th target task execution result may generate the 5 th target demand task, execute the 5 th target demand task as the 5 th target service resource based on the large language model M510, may summarize the 1 st target task execution result to the 4 th target task execution result, and obtain the 5 th target task execution result, and may use the 5 th target task execution result as the policy execution result.
According to the embodiment of the disclosure, the task execution result can be characterized based on natural language description, so that the execution of the multi-demand task can be completed in an internal dialogue interaction mode in the execution process of the task execution strategy, the comprehensiveness and the generative capability of the large language model can be conveniently called, and the readability of the task execution result is improved. Therefore, only the target object is required to send out an open expression requirement description of how the query information is "how the account is integrally represented", so that the end-to-end one-stop data analysis process of basic report forms, report form analysis, problem diagnosis and action optimization can be successfully completed, and the quality of business service is improved.
According to the embodiment of the disclosure, the service resource can realize integration of the multi-type plug-in resource, so that the multi-type plug-in can be conveniently called in real time through the requirement task attribute, and the task execution strategy can be adjusted through continuous iterative optimization of the task execution strategy until the requirement intention of the target object is met.
The information interaction method provided according to the embodiments of the present disclosure may be implemented in a manner as shown in table 1, for example, may be implemented in an online scheme or an offline scheme.
TABLE 1
According to the information interaction method based on the large language model, which is provided by the embodiment of the disclosure, an advertising marketing intelligent system integrating perception, thinking and action can be constructed. The advertising marketing intelligent system comprises an inference planning subsystem, a memory subsystem, a perception subsystem, a action subsystem and the like. The advertising marketing intelligent system can realize the global capability coordination of advertising marketing business by taking a large language model as a kernel. Wherein the function of each subsystem is as follows.
And the planning subsystem triggers deep reasoning thinking through a task execution strategy paradigm based on a large language model, enhances based on a marketing knowledge base, and carries out reinforcement learning based on prompt information so as to realize task planning of a given marketing business target. Specifically, the method is responsible for task planning, subtask disassembly, and correction of thinking back in the execution process, and aims at guiding collaborative tasks. The planning subsystem may have a variety of switchable thinking paradigms. For example, the task execution strategy can be continuously and iteratively optimized in a mode of (a) planning, task execution, task analysis, task execution result disfigurement and task execution strategy updating for a plurality of times. The jeopardy process can process the task execution result output by the previous turn based on the large language model to judge whether the task nodes of the remaining turns can meet the requirements of users, so that the flow nodes of the remaining turns can be timely adjusted, for example, the task nodes can be added, modified and deleted, and further task planning updating is realized. The mode (b) can execute the task execution strategy for multi-round subtask disassembly, task priority ordering and high-priority task execution paradigm. The mode (c) can execute topology planning for the tasks of the one-time global task execution strategy and execute the tasks according to the topology sequence. The ways (b) and (c) b and c can be used for improving the reply efficiency for the information input by the user without a dislike flow.
A memory subsystem stores information perceived from the environment and uses the recorded data to facilitate future reply accuracy actions. The memory modules in the memory subsystem can help the large language model accumulate experience, self-evolve, and accomplish tasks in a more consistent, rational, and efficient manner. The memory subsystem comprises a short-term memory bank and a working memory bank, wherein the working memory bank is used for storing the context information (such as intermediate reasoning process and result) of the task, and the storage and the retrieval can be realized based on the vector database. The long-term memory bank can store mass knowledge in the marketing business field. For example, the method can comprise personalized information, customer behavior habit and the like which are self-submitted by customers, pre-constructed commercial knowledge graphs, historical analysis knowledge and the like, and authoritative theories and case knowledge in various fields and the like.
The perception subsystem can expand the perception space from plain text to multi-mode fields including text, vision, hearing and the like, so that the perception subsystem can acquire and utilize more information from a network environment more effectively. Such as advertising documents, advertising pictures, advertising videos, rich media information of customer landing pages, etc., can be perceived by the agents of the advertising marketing intelligence system to provide valuable business information. The relevant landing page can be accessed and called according to the input information of the user, and the information corresponding to the requirement is correspondingly acquired, or action tasks such as putting advertisements can be performed.
At the heart of the action subsystem is the use of tools. The tool can expand the action space of the agent constructed based on the large language model. With the aid of tools, agents built based on large language models can utilize various external resources during the reasoning and planning phases. Tools used by the row subsystem include, but are not limited to, the following tools (or plug-ins, service resources)
And the text tool is a LLM tool based on an externally hung natural language model or secondary packaging, so that the quality of the text in fluency, correlation, diversity, specialty and controllability is improved.
Knowledge retrieval tools-search based tools can improve the range and quality of knowledge available to an agent by means of external databases, knowledge maps and web pages.
The business domain tool can enhance the expertise of the intelligent agent in the corresponding domain and execute the specific business action.
Natural language to query statement tool-LLM based controllers can generate query statements to query databases or convert user queries to search requests and use search engines to obtain desired results. And interpreting the codes or generating corresponding query sentences according to the input information, and executing the query sentences to perform data retrieval.
API (interface) tools-APIs of various capabilities, given structured descriptive information, can be considered as available tools for agents.
And the code interpreter is in butt joint with the LaTeX compiler by utilizing the Python interpreter, so that the performance of the code interpreter in complex tasks and mathematical calculation tasks is improved.
And (3) a multi-mode tool, namely, comprehensively completing complex tasks by using a plurality of AI models, such as a decision model or a generation model of image/video/audio/multi-mode and the like.
LLM-based Agent can create tools by generating executable programs or integrating existing tools into more powerful tools, and can learn self-debugging, continuously building own tool libraries.
And the feedback tool is used for abstracting feedback information into a tool, and agents can actively or passively search for feedback information intervention so as to better fit task targets, advance task progress and deliver results with high quality. The method can be displayed at the front end based on background interaction, and can provide an accessed interface for a user or maintainer, so that the task is propelled.
Perceived action tool-is considered a key paradigm that combines model intelligence with the physical world. Tools are used that actively sense, understand and interact with the physical environment, such as viewing, manipulating, navigating, etc. The device and the equipment in the physical world can be connected, and the device and the equipment can be controlled according to the feedback information. By utilizing the capabilities of various tool plug-ins, the data, knowledge and capabilities of the advertising marketing business domain can be organically integrated, autonomous decision making can be realized, and a proper plug-in combination is selected to finish the marketing business target task.
FIG. 6 schematically illustrates a functional framework diagram of an advertising marketing intelligence system suitable for performing a large language model based information interaction method in accordance with an embodiment of the present disclosure.
As shown in fig. 6, the agent 600 may provide feedback information for the open description of the natural language of the target object during each business service of the advertisement marketing link and the advertisement delivery link.
For example, feedback information including industry analysis, bid analysis, marketing ideas, etc. may be provided in the pre-sale link at the advertising marketing link. The selling knots provide feedback information such as emotion analysis, intent analysis, recommended language program, etc. And providing feedback information such as marketing bars, automatic labels, next marketing plans and the like in the after-sale link.
For another example, marketing suggestion feedback information of each stage before, during and after the advertisement delivery can be provided in the advertisement delivery link, including marketing suggestion feedback information of advertisement target and strategy customization, keyword research and selection, audience and other delivery settings, bidding and budget, advertisement creative production, detection and optimization, data reporting and analysis and the like. The advertising marketing intelligent system can realize the comprehensive service capability of artificial intelligence multi-mode, global knowledge, industry field experience, advertising service capability (comprising functions of analysis, insight, delivery, diagnosis, optimization and the like) +professional small model and large language model based on the intelligent agent, and realize the autonomous solution of the user demand of the advertising marketing service domain.
Fig. 7 schematically illustrates a framework diagram suitable for performing a large language model based information interaction method according to an embodiment of the present disclosure.
As shown in fig. 7, the intelligent system framework 700 may be adapted to implement the large language model based information interaction method provided according to the embodiments of the present disclosure. The intelligent system framework 700 can include a software development tool wrapper, a service layer, a model layer, and a storage layer. The software development kit (Software Development Kit, SDK) layer may include a business encapsulation software development kit module and an artificial intelligence software development kit module. The software development tool cladding can be a core carrier oriented to the atomization artificial intelligence capability and the general scene business capability, so that a target object can conveniently input the requirement description information in time, and feedback information is obtained. The model layer can comprise deep learning model support of various types such as a large language model, an open source model and the like, and can also realize functions such as model training customization, large model tuning and the like through a model tuning module. The storage layer is responsible for data persistence of the whole system and comprises a vector subsystem, a cache subsystem, a memory subsystem, a meta-information subsystem, a service corpus and the like. The intelligent system framework 700 can provide powerful support for an agent to implement the information interaction method provided by the embodiments of the present disclosure.
Fig. 8 schematically illustrates an architecture diagram suitable for implementing an advertising marketing system in accordance with an embodiment of the present disclosure.
As shown in FIG. 8, the advertising marketing system 800 may be adapted to perform the large language model based information interaction method provided by the embodiments of the present disclosure. Advertising marketing system 800 may include an interaction layer at the top layer, an agent at the middle layer, and a plug-in layer. The interaction layer may implement information interaction with the target object based on a graphical rendering component library and UI (interface design) interaction protocol. The agent may implement other functions such as the generated content function through interactions with the large language model and the business model. The intelligent agent can also interact with modules such as a rights management control module, an operation and maintenance monitoring module, an intelligent agent policy evolution module, a small flow control module, a data ETL (Extract-Transform-Load, also called a data warehouse) system module and the like, so that functions such as plug-in resource calling and data storage are realized. The intelligent agent can also perform data interaction with a marketing domain library, a marketing domain knowledge base, an analysis platform, a business report center, a multi-mode intelligent module and a public domain capability knowledge base based on the plug-ins such as a retrieval enhancement plug-in, a query statement plug-in, a business plug-in cluster, a public domain plug-in and other plug-ins.
According to embodiments of the present disclosure, an advertising marketing system may integrate diverse plug-ins based on a mid-platform design, with a greater degree of intellectualization being the richer the plug-in capability. The modular deployment of the agent may take a centralized mode. The intelligent agent realizes the switching and the generation of various task execution strategies through the switching of various types of large language models. And various plug-in resources are called based on the generated task execution strategy, so that dynamic arrangement and static arrangement flexible switching of the plug-in resources are realized. Multiple plug-ins supporting task execution policies simultaneously generate 1 composite plug-in to be flexibly invoked. In the intelligent interaction process of the user, the intermediate thinking process, such as the intermediate task execution result, can be displayed, so that the decision process of the user participation is realized. Meanwhile, the data flywheel function can be realized. The data flywheel functions can comprise a brain (a large language model) and a cerebellum (a MoE, mixture of Experts, a mixed expert model), and the decision layer is made to intelligently and rapidly iterate by referring to the thought of 'knowledge distillation' through business effect feedback information.
Fig. 9 schematically illustrates an architecture diagram suitable for implementing an agent in accordance with an embodiment of the present disclosure.
As shown in fig. 9, an agent 900 may include an artificial intelligence business services module, a basic capability + business capability module, a plan & think back module, a memory capability module, a memory middle module, an action module, an agent module, an artificial intelligence model capability module, an underlying business capability module. The memory capability module may include a long-term memory unit and a working memory unit. The plan & thinking-back module may include a feedback unit, bored optimization unit, task execution policy unit, and task decomposition unit. The basic capability and business capability module can comprise various plug-in resources such as a search plug-in, an interface plug-in, a permission authentication plug-in and the like. The bottom layer business capability module can comprise a function unit such as an optimization center, an effect platform, a three-party platform interface and the like, and is used for realizing the functions such as information display, data calling and the like of feedback information. The artificial intelligence model capability module may include a text processing unit, an image processing unit, a video processing unit, a multimodal processing model, and the like, constructed based on artificial intelligence algorithms. The artificial intelligence model capability module and the underlying business capability module can enable each functional unit to be automatically scheduled and assembled by the agent module or the action module.
Fig. 10 schematically illustrates a schematic diagram suitable for performing agent evolution in accordance with an embodiment of the present disclosure.
As shown in fig. 10, the agent may be developed jointly based on developers, product personnel, operators, testers, for example, developer diversified plug-in resources, such as artificial intelligence plug-ins, business service plug-in capabilities, and the like. The product personnel may provide marketing-related knowledge. Operators may provide industry methods. Industry experience is input to the agent. A tester may provide a business test case. The agent can provide advertising marketing business service information to the user, can iteratively optimize the technical scheme based on the interaction information of the user, and provides the technical scheme to the data flywheel module, the group intelligent module and the field experience module so as to facilitate the iterative optimization of the function of the agent.
Fig. 11 schematically illustrates a schematic diagram suitable for performing an agent evolution according to another embodiment of the present disclosure.
As shown in fig. 11, the agent may provide a task execution policy to the service module, and may obtain a policy execution result based on the task execution policy, and the service module may, for example, transmit a personalized scheme uploaded by the target object, and transmit a platform arrangement scheme generated based on the agent to a personalized scheme library and a marketing scheme library to enrich knowledge information for optimizing the model, and may also obtain public expertise through a public domain knowledge library. The post-training module may obtain knowledge information from any one or more of the personalized solution store, the marketing solution store, and the public domain knowledge store through the knowledge collection module. The post training module can strengthen marketing knowledge through the interaction information provided by the intelligent agent, can optimize the large language model based on the collected information, and can carry out knowledge distillation on the models of the mixed expert model 1, the mixed expert model 2.
Therefore, the intelligent agent can be changed from a traditional program development mode to a teaching intelligent agent mode, so that the intelligent agent is more and more powerful. Through the thinking-back function of the intelligent agent, the intelligent agent can be continuously 'educated' by combining information such as plug-in resources, knowledge resources, examples and the like provided by each related person, and the logic thinking paradigm of the intelligent agent is continuously optimized to expand the knowledge range, so that the intelligent of the intelligent agent is continuously evolved, and the new business value is expanded.
FIG. 12 schematically illustrates a flow chart of a method of large language model based information interaction in accordance with another embodiment of the present disclosure.
As shown in FIG. 12, the large language model-based information interaction method includes operations S1210-S1230.
In operation S1210, the demand description information is processed using at least one agent of the plurality of agents in response to the demand description information input by the target object, resulting in feedback information.
In operation S1220, at least one information interaction is realized based on the feedback information using at least two of the plurality of agents, and interactive feedback information is generated.
In operation S1230, interactive feedback information is presented.
According to the embodiments of the present disclosure, the agent may include a device adapted to implement an information processing function based on a deep learning algorithm, for example, may include a software agent such as a virtual device constructed based on a deep learning algorithm, or may further include a hardware agent such as a chip having an information processing function, and the embodiments of the present disclosure do not limit a specific structure or type of the agent as long as the information processing function can be performed.
According to the embodiment of the disclosure, the intelligent agent is suitable for responding to the acquired demand description information, processing the demand description information by utilizing a large language model to obtain a task execution strategy matched with the demand description information, wherein the task execution strategy comprises a plurality of demand task attributes, at least one of which is matched with the demand intention represented by the demand description information;
According to the embodiment of the disclosure, information interaction is performed through at least two of the plurality of agents, and further, interaction feedback information is generated, so that the requirement intention of a target object can be realized through an interaction mode among the plurality of agents, and execution of a requirement task is advanced, so that the task execution can be advanced by simulating artificial interaction capability and thinking capability under the condition of no manual intervention, the generated interaction feedback information can facilitate the execution process and reasoning process of a display task, and the readability and the understandability of acquired data are improved, and further, the user experience is improved.
According to the embodiment of the disclosure, determining a strategy execution result related to a task execution strategy according to a demand task attribute in the task execution strategy comprises determining a processed completion demand task attribute from a plurality of demand task attributes of the current task execution strategy, processing the completion demand task attribute by using a large language model and obtaining a to-be-processed demand task attribute according to the task execution result related to the completion demand task attribute, updating the current task execution strategy according to the to-be-processed demand task attribute to obtain a new task execution strategy, and determining a strategy execution result according to the new task execution strategy.
According to the embodiments of the present disclosure, at least one of the plurality of agents may be used to implement the information interaction method based on the large language model provided by the embodiments of the present disclosure, which is not described herein in detail.
Fig. 13 schematically illustrates an application scenario diagram of a large language model-based information interaction method according to another embodiment of the present disclosure.
An interactive interface 1300 may be included in the application scenario as shown in fig. 13. The target object 1310 may input the requirement description 1311 "how the scheme performs", and the agent 1320 may obtain the 1 st feedback 1321 "the scheme is analyzed as xxx by three months according to the requirement description 1311. The agent 1330 may perform the information interaction method provided in the embodiments of the present disclosure according to the 1 st feedback information 1321 as the requirement description information, to obtain the 1 st interaction feedback information 1331, "data analysis is represented by a data table. The agent 1320 may execute the information interaction method provided by the embodiment of the present disclosure with the 1 st interaction feedback information 1331 as the requirement description information to obtain the 2 nd interaction feedback information 1322.
Fig. 14 schematically illustrates a block diagram of a large language model based information interaction device according to an embodiment of the present disclosure.
As shown in fig. 14, the large language model-based information interaction device 1400 may include a task execution policy obtaining module 1410, a policy execution result determining module 1420, and a feedback module 1430.
The task execution policy obtaining module 1410 is configured to process the requirement description information by using a large language model in response to obtaining the requirement description information, and obtain a task execution policy matched with the requirement description information, where the task execution policy includes a plurality of requirement task attributes, and at least one of the plurality of requirement task attributes is matched with a requirement intention represented by the requirement description information.
The policy execution result determining module 1420 is configured to determine a policy execution result related to the task execution policy according to the required task attribute in the task execution policy.
And the feedback module 1430 is configured to generate feedback information according to the policy execution result and display the feedback information.
The strategy execution result determining module comprises a first determining sub-module, a to-be-processed demand task attribute obtaining sub-module, an updating sub-module and a strategy execution result determining sub-module.
The first determining submodule is used for determining the completed required task attribute of the processed completed required task from a plurality of required task attributes of the current task execution strategy.
And the to-be-processed demand task attribute obtaining sub-module is used for processing the completion demand task attribute and the task execution result related to the completion demand task attribute by utilizing the large language model to obtain the to-be-processed demand task attribute.
And the updating sub-module is used for updating the current task execution strategy according to the task attribute of the to-be-processed requirement to obtain a new task execution strategy.
And the strategy execution result determining sub-module is used for determining a strategy execution result according to the new task execution strategy.
According to an embodiment of the present disclosure, the update sub-module comprises a first update unit.
And the first updating unit is used for updating the attribute of the task to be completed in the current task execution strategy according to the attribute of the task to be processed.
According to an embodiment of the present disclosure, the update sub-module comprises a second update unit.
And the second updating unit is used for updating the outstanding demand task attribute which is not processed in the current task execution strategy according to the to-be-processed demand task attribute.
According to embodiments of the present disclosure, the new task execution policy includes execution dependencies between new demand task attributes.
According to the embodiment of the disclosure, the strategy execution result determining submodule comprises a first target demand task determining unit and a first task execution result obtaining unit.
The first target demand task determining unit is configured to determine a first target demand task according to a first target demand task attribute and a first dependent task execution result associated with the first dependent demand task attribute, where the new task execution policy includes the first target demand task attribute and the first dependent demand task attribute, and an execution dependency relationship is provided between the first target demand task attribute and the first dependent demand task attribute.
The first task execution result obtaining unit is used for calling a first target service resource associated with the first target requirement task attribute to execute the first target requirement task to obtain a task execution result associated with the first target requirement task attribute, and the strategy execution result comprises the task execution result associated with the first target requirement task attribute.
According to an embodiment of the disclosure, the first target demand task attribute is characterized based on a hint template.
According to the embodiment of the disclosure, the first target demand task determining unit comprises a first updating subunit and a first target demand task obtaining unit.
And the first updating subunit is used for updating the first target demand task attribute according to the first dependent task execution result associated with the first dependent demand task attribute to obtain first target task prompt information.
The first target demand task obtaining unit is used for processing the first target task prompt information by utilizing the large language model to obtain a first target demand task.
According to embodiments of the present disclosure, the task execution policy further includes execution dependencies between the plurality of demand task attributes.
The strategy execution result determining module comprises a second target demand task obtaining sub-module and a second task execution result obtaining sub-module.
The second target demand task obtaining sub-module is used for generating a second target demand task according to a second target demand task attribute and a task execution result associated with the second dependent demand task attribute, wherein the task execution strategy comprises a second dependent demand task attribute and a second target demand task attribute, and an execution dependency relationship is arranged between the second dependent demand task attribute and the second target demand task attribute.
And the second task execution result obtaining sub-module is used for calling a second target service resource associated with the second target demand task attribute to execute the second target demand task to obtain a task execution result associated with the second target demand task attribute, and the strategy execution result comprises the task execution result associated with the second target demand task attribute.
According to an embodiment of the present disclosure, the second target demand task attribute is characterized based on a hint template.
According to the embodiment of the disclosure, the second target demand task obtaining submodule comprises a second target task prompt information obtaining unit and a second target demand task obtaining unit.
And the second target task prompt information obtaining unit is used for updating the second target requirement task attribute according to the task execution result associated with the second dependency requirement task attribute to obtain second target task prompt information.
And the second target demand task obtaining unit is used for processing the second target task prompt information by utilizing the large language model to obtain a second target demand task.
According to an embodiment of the present disclosure, the first target demand task attribute or the second target demand task attribute includes a target service resource identifier adapted to characterize an association between the first target demand task attribute and the first target service resource or adapted to characterize an association between the second target demand task attribute and the second target service resource.
According to an embodiment of the disclosure, the information interaction device based on the large language model further comprises a first display module.
The first display module is used for displaying target task execution results, wherein the target task execution results comprise task execution results associated with first target demand task attributes or task execution results associated with second target demand task attributes.
According to an embodiment of the disclosure, the task execution policy acquisition module includes a multi-level demand intent acquisition sub-module and a task execution policy determination sub-module.
The multi-level demand intention obtaining submodule is used for processing the demand description information by using the large language model to obtain the multi-level demand intention.
And the task execution strategy determination submodule is used for determining the task execution strategy according to the multi-level demand intention.
According to an embodiment of the present disclosure, the multi-level demand intention obtaining submodule includes a first demand intention obtaining unit.
The first demand intention obtaining unit is used for processing the demand description information and the demand intention of the previous level by utilizing the large language model aiming at the current level to obtain the demand intention of the current level, wherein the demand intention of the first level in the multi-level demand intention is obtained by utilizing the large language model to process the demand description information.
According to an embodiment of the present disclosure, the multi-level demand intention obtaining submodule includes a second demand intention obtaining unit.
The second demand intention obtaining unit is used for processing the demand description information by using the large language model and obtaining the multi-level demand intention based on the associated query information queried by the demand description information.
According to the embodiment of the disclosure, the information interaction device based on the large language model further comprises a second display module and a demand intention determining module.
And the second display module is used for displaying at least one demand intention in the multi-level demand intention.
A demand intent determination module for determining a demand intent associated with a selection operation in response to the selection operation for at least one demand intent.
According to an embodiment of the present disclosure, the task execution policy determination submodule includes a task execution policy determination unit.
And the task execution strategy determining unit is used for determining a task execution strategy according to the requirement intention associated with the selection operation in the multi-level requirement intention.
Fig. 15 schematically illustrates a block diagram of a large language model based information interaction device according to another embodiment of the present disclosure.
As shown in fig. 15, the information interaction device 1500 based on the large language model includes a feedback information obtaining module 1510, an interaction feedback information generating module 1520, and an interaction feedback information exhibiting module 1530.
The feedback information obtaining module 1510 is configured to process the demand description information by using at least one agent of the plurality of agents in response to the demand description information input by the target object to obtain feedback information, where the agent is adapted to process the demand description information by using the large language model in response to obtaining the demand description information to obtain a task execution policy matched with the demand description information, where the task execution policy includes a plurality of demand task attributes, at least one of the plurality of demand task attributes is matched with a demand intention represented by the demand description information, determine a policy execution result related to the task execution policy according to the demand task attributes in the task execution policy, and generate the feedback information and display the feedback information according to the policy execution result.
The interactive feedback information generating module 1520 is configured to utilize at least two of the plurality of agents to implement at least one information interaction based on the feedback information, and generate interactive feedback information.
The interactive feedback information display module 1530 is configured to display interactive feedback information.
According to embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.
According to an embodiment of the present disclosure, an electronic device includes at least one processor and a memory communicatively coupled to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a method provided in accordance with an embodiment of the present disclosure.
According to an embodiment of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform a method provided according to an embodiment of the present disclosure.
According to an embodiment of the present disclosure, a computer program product comprising a computer program which, when executed by a processor, implements a method provided according to an embodiment of the present disclosure.
Fig. 16 schematically illustrates a block diagram of an electronic device adapted to implement a large language model based information interaction method, according to an embodiment of the disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the disclosure described and/or claimed herein.
As shown in fig. 16, the apparatus 1600 includes a computing unit 1601 that can perform various appropriate actions and processes according to a computer program stored in a Read Only Memory (ROM) 1602 or a computer program loaded from a storage unit 1608 into a Random Access Memory (RAM) 1603. In RAM 1603, various programs and data required for operation of device 1600 may also be stored. The computing unit 1601, ROM 1602, and RAM 1603 are connected to each other by a bus 1604. An input/output (I/O) interface 1605 is also connected to the bus 1604.
Various components in the device 1600 are connected to I/O interfaces 1605, including an input unit 1606, such as a keyboard, mouse, etc., an output unit 1607, such as various types of displays, speakers, etc., a storage unit 1608, such as a magnetic disk, optical disk, etc., and a communication unit 1609, such as a network card, modem, wireless communication transceiver, etc. Communication unit 1609 allows device 1600 to exchange information/data with other devices via a computer network, such as the internet, and/or various telecommunications networks.
The computing unit 1601 may be a variety of general purpose and/or special purpose processing components with processing and computing capabilities. Some examples of computing unit 1601 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various specialized Artificial Intelligence (AI) computing chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1601 performs the respective methods and processes described above, for example, an information interaction method based on a large language model. For example, in some embodiments, the large language model based information interaction method may be implemented as a computer software program tangibly embodied on a machine-readable medium, e.g., the storage unit 1608. In some embodiments, some or all of the computer programs may be loaded and/or installed onto device 1600 via ROM 1602 and/or communication unit 1609. When a computer program is loaded into the RAM 1 603 and executed by the computing unit 1601, one or more steps of the large language model based information interaction method described above may be performed. Alternatively, in other embodiments, the computing unit 1601 may be configured to perform the large language model based information interaction method by any other suitable means (e.g. by means of firmware).
Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuit systems, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), systems On Chip (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include being implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be a special or general purpose programmable processor, operable to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.
Program code for carrying out methods of the present disclosure may be written in any combination of one or more programming languages. These program code may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus such that the program code, when executed by the processor or controller, causes the functions/operations specified in the flowchart and/or block diagram to be implemented. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.
In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user, for example, feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback), and input from the user may be received in any form, including acoustic input, speech input, or tactile input.
The systems and techniques described here can be implemented in a computing system that includes a background component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such background, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a Local Area Network (LAN), a Wide Area Network (WAN), and the Internet.
The computer system may include a client and a server. The client and server are typically remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server may be a cloud server, a server of a distributed system, or a server incorporating a blockchain.
It should be appreciated that various forms of the flows shown above may be used to reorder, add, or delete steps. For example, the steps recited in the present disclosure may be performed in parallel or sequentially or in a different order, provided that the desired results of the technical solutions of the present disclosure are achieved, and are not limited herein.
The above detailed description should not be taken as limiting the scope of the present disclosure. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present disclosure are intended to be included within the scope of the present disclosure.