CN114281969A - Reply sentence recommendation method and device, electronic equipment and storage medium - Google Patents
Reply sentence recommendation method and device, electronic equipment and storage mediumDownload PDFInfo
- Publication number
- CN114281969A CN114281969ACN202111567908.7ACN202111567908ACN114281969ACN 114281969 ACN114281969 ACN 114281969ACN 202111567908 ACN202111567908 ACN 202111567908ACN 114281969 ACN114281969 ACN 114281969A
- Authority
- CN
- China
- Prior art keywords
- intention
- user
- chain
- agent
- corpus
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034methodMethods0.000titleclaimsabstractdescription59
- 230000007704transitionEffects0.000claimsabstractdescription25
- 238000012546transferMethods0.000claimsdescription45
- 230000015654memoryEffects0.000claimsdescription29
- 230000006870functionEffects0.000claimsdescription18
- 238000012545processingMethods0.000claimsdescription17
- 239000011159matrix materialSubstances0.000claimsdescription15
- 238000004891communicationMethods0.000claimsdescription14
- 238000004590computer programMethods0.000claimsdescription13
- 238000012163sequencing techniqueMethods0.000claimsdescription9
- 230000011218segmentationEffects0.000claimsdescription7
- 238000000605extractionMethods0.000claimsdescription6
- 230000008901benefitEffects0.000abstractdescription5
- 239000003795chemical substances by applicationSubstances0.000description99
- 230000000875corresponding effectEffects0.000description52
- 238000005516engineering processMethods0.000description8
- 238000010586diagramMethods0.000description7
- 230000003287optical effectEffects0.000description7
- 238000013473artificial intelligenceMethods0.000description6
- 230000008878couplingEffects0.000description3
- 238000010168coupling processMethods0.000description3
- 238000005859coupling reactionMethods0.000description3
- 239000000463materialSubstances0.000description3
- 238000005065miningMethods0.000description3
- 230000008569processEffects0.000description3
- 238000013528artificial neural networkMethods0.000description2
- 230000003068static effectEffects0.000description2
- 230000017105transpositionEffects0.000description2
- 230000009286beneficial effectEffects0.000description1
- 230000001276controlling effectEffects0.000description1
- 230000002596correlated effectEffects0.000description1
- 238000013135deep learningMethods0.000description1
- 238000011161developmentMethods0.000description1
- 230000018109developmental processEffects0.000description1
- 238000009396hybridizationMethods0.000description1
- 230000003993interactionEffects0.000description1
- 239000004973liquid crystal related substanceSubstances0.000description1
- 238000010801machine learningMethods0.000description1
- 238000003058natural language processingMethods0.000description1
Images
Landscapes
- Machine Translation (AREA)
Abstract
Translated fromChinese
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a method and a device for recommending reply sentences, electronic equipment and a storage medium.
Background
Currently, in the conventional dialogue model, a user is usually given a single intention tag based on the current reply of the user, and a corresponding dialogue is searched according to the intention tag for replying. Although the algorithm level is mature, the method can only analyze a single round of conversation, and in a real-world scene, the conversation is basically a multi-round conversation scene based on context, and at the moment, it is not enough to only consider the current round of speaking content of the user.
The existing solution is to mashup the intentions of the contexts together by arranging and combining the intentions learned from a single round of conversation. However, this method can only combine the intentions of the contexts, and cannot deeply mine the intention connection between the contexts, so that the obtained conclusion is not accurate enough.
Disclosure of Invention
In order to solve the above problems in the prior art, embodiments of the application provide a method and an apparatus for recommending a reply sentence, an electronic device, and a storage medium, which can deeply mine an intention link between contexts, improve accuracy of intention identification, accurately match a reply sentence, and improve user experience.
In a first aspect, an embodiment of the present application provides a reply sentence recommendation method, including:
searching historical conversation data according to the voice information of the user at the current moment, and generating a user intention chain and an agent intention chain, wherein the user intention chain is used for identifying the intention trend of the user in the current round of conversation, and the agent intention chain is used for identifying the intention trend of an agent in the current round of conversation;
determining the user maximum income intention corresponding to the user intention chain and the seat maximum income intention corresponding to the seat intention chain according to the application field to which the voice information at the current moment belongs;
establishing a first functional relationship between the maximum profit intention of the user and the user intention chain, and determining the user transfer probability of the user intention chain according to the first functional relationship;
establishing a second functional relationship among the user transfer probability, the agent maximum income intention and the agent intention chain, and grading at least one preset answer dialog according to the second functional relationship to obtain at least one score, wherein the at least one score is in one-to-one correspondence with the at least one answer dialog;
and generating a reply sentence according to the reply grammar corresponding to the highest score in the at least one score, and pushing the reply sentence to the agent.
In a second aspect, an embodiment of the present application provides a reply sentence recommendation apparatus, including:
the query module is used for searching historical conversation data according to the voice information of the user at the current moment and generating a user intention chain and an agent intention chain, wherein the user intention chain is used for identifying the intention trend of the user in the current round of conversation, and the agent intention chain is used for identifying the intention trend of an agent in the current round of conversation, which has a conversation with the user;
the analysis module is used for determining the user maximum income intention corresponding to the user intention chain and the seat maximum income intention corresponding to the seat intention chain according to the application field to which the voice information at the current moment belongs;
the processing module is used for establishing a first functional relationship between the maximum income intention of the user and the user intention chain, determining the user transfer probability of the user intention chain according to the first functional relationship, establishing a second functional relationship among the user transfer probability, the agent maximum income intention and the agent intention chain, and grading at least one preset answer dialog according to the second functional relationship to obtain at least one score, wherein the at least one score is in one-to-one correspondence with the at least one answer dialog;
and the recommendation module is used for generating a reply sentence according to the reply speech corresponding to the highest score in the at least one score and pushing the reply sentence to the agent.
In a third aspect, an embodiment of the present application provides an electronic device, including: a processor coupled to the memory, the memory for storing a computer program, the processor for executing the computer program stored in the memory to cause the electronic device to perform the method of the first aspect.
In a fourth aspect, embodiments of the present application provide a computer-readable storage medium having a computer program stored thereon, the computer program causing a computer to perform the method according to the first aspect.
In a fifth aspect, embodiments of the present application provide a computer program product comprising a non-transitory computer readable storage medium storing a computer program, the computer operable to cause the computer to perform a method according to the first aspect.
The implementation of the embodiment of the application has the following beneficial effects:
in the embodiment of the application, historical conversation data of the current round of conversation is acquired through the voice information of the user at the current moment, and a user intention chain and an agent intention chain are established. Then, based on the application field to which the voice information at the current moment belongs, the maximum profit intentions of the user and the seat in the field are determined, namely the final communication target. Therefore, the transition probability of the user intention chain is confirmed by establishing a first functional relationship between the maximum profit intention of the user and the user intention chain, and then the transition probability of the user intention chain is used as a parameter of a transition matrix of the agent intention chain to disturb the functional relationship between the agent intention chain and the maximum profit intention of the agent, so as to obtain a second functional relationship. Finally, the scores of the candidate dialogs are confirmed through the second functional relation, so that a reply sentence is generated through the dialogs with the highest scores, and the reply is carried out on the user. Based on the method, the seat intention chain is disturbed through the user intention chain, so that the user intention chain influences and games the seat intention chain mutually, and finally convergence is carried out according to respective corresponding maximized income, so that deep mining of intention connection between contexts is realized, inference logic is introduced, accuracy of intention identification is improved, answer sentences are accurately matched, and the interpretability is strong.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.
Fig. 1 is a schematic diagram of a hardware structure of a reply sentence recommendation apparatus according to an embodiment of the present application;
fig. 2 is a schematic flowchart of a method for recommending a reply sentence according to an embodiment of the present application;
fig. 3 is a schematic flowchart of a method for searching historical dialog data according to voice information of a user at a current time to generate a user intention chain and an agent intention chain according to an embodiment of the present application;
fig. 4 is a schematic flow chart of a method for extracting an intention of each of n agent sentences to obtain n agent intention features according to an embodiment of the present application;
fig. 5 is a block diagram illustrating functional modules of a reply sentence recommendation apparatus according to an embodiment of the present application;
fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is apparent that the described embodiments are some embodiments of the present application, but not all embodiments of the present application. All other embodiments obtained by a person of ordinary skill in the art without any inventive work based on the embodiments in the present application are within the scope of protection of the present application.
The terms "first," "second," "third," and "fourth," etc. in the description and claims of this application and in the accompanying drawings are used for distinguishing between different objects and not for describing a particular order. Furthermore, the terms "include" and "have," as well as any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, article, or apparatus that comprises a list of steps or elements is not limited to only those steps or elements listed, but may alternatively include other steps or elements not listed, or inherent to such process, method, article, or apparatus.
Reference herein to "an embodiment" means that a particular feature, result, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. It is explicitly and implicitly understood by one skilled in the art that the embodiments described herein can be combined with other embodiments.
First, referring to fig. 1, fig. 1 is a schematic diagram of a hardware structure of a reply sentence recommendation device according to an embodiment of the present application. The replysentence recommendation apparatus 100 includes at least oneprocessor 101, acommunication line 102, amemory 103, and at least onecommunication interface 104.
In this embodiment, theprocessor 101 may be a general processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more ics for controlling the execution of programs according to the present disclosure.
Thecommunication link 102, which may include a path, carries information between the aforementioned components.
Thecommunication interface 104 may be any transceiver or other device (e.g., an antenna, etc.) for communicating with other devices or communication networks, such as an ethernet, RAN, Wireless Local Area Network (WLAN), etc.
Thememory 103 may be, but is not limited to, a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a Random Access Memory (RAM) or other type of dynamic storage device that can store information and instructions, an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disk storage, optical disk storage (including compact disc, laser disc, optical disc, digital versatile disc, blu-ray disc, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer.
In this embodiment, thememory 103 may be independent and connected to theprocessor 101 through thecommunication line 102. Thememory 103 may also be integrated with theprocessor 101. Thememory 103 provided in the embodiments of the present application may generally have a nonvolatile property. Thememory 103 is used for storing computer-executable instructions for executing the scheme of the application, and is controlled by theprocessor 101 to execute. Theprocessor 101 is configured to execute computer-executable instructions stored in thememory 103, thereby implementing the methods provided in the embodiments of the present application described below.
In alternative embodiments, computer-executable instructions may also be referred to as application code, which is not specifically limited in this application.
In alternative embodiments,processor 101 may include one or more CPUs, such as CPU0 and CPU1 of FIG. 1.
In an alternative embodiment, the replysentence recommendation apparatus 100 may include a plurality of processors, such as theprocessor 101 and theprocessor 107 in fig. 1. Each of these processors may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. A processor herein may refer to one or more devices, circuits, and/or processing cores for processing data (e.g., computer program instructions).
In an alternative embodiment, if the replysentence recommendation apparatus 100 is a server, for example, it may be an independent server, or may be a cloud server that provides basic cloud computing services such as cloud service, cloud database, cloud computing, cloud function, cloud storage, web service, cloud communication, middleware service, domain name service, security service, Content Delivery Network (CDN), big data and artificial intelligence platform, and the like. The replysentence recommendation apparatus 100 may further include anoutput device 105 and aninput device 106. Theoutput device 105 is in communication with theprocessor 101 and may display information in a variety of ways. For example, theoutput device 105 may be a Liquid Crystal Display (LCD), a Light Emitting Diode (LED) display device, a Cathode Ray Tube (CRT) display device, a projector (projector), or the like. Theinput device 106 is in communication with theprocessor 101 and may receive user input in a variety of ways. For example, theinput device 106 may be a mouse, a keyboard, a touch screen device, or a sensing device, among others.
The above-described replysentence recommendation apparatus 100 may be a general-purpose device or a special-purpose device. The present embodiment does not limit the type of the replysentence recommendation apparatus 100.
Next, it should be noted that the embodiments disclosed in the present application may acquire and process related data based on artificial intelligence technology. Among them, Artificial Intelligence (AI) is a theory, method, technique and application system that simulates, extends and expands human Intelligence using a digital computer or a machine controlled by a digital computer, senses the environment, acquires knowledge and uses the knowledge to obtain the best result.
The artificial intelligence infrastructure generally includes technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a robot technology, a biological recognition technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and the like.
Finally, the reply sentence recommendation method in the application can be applied to scenes of e-commerce sales, off-line entity sales, service promotion, seat telephone outbound, social platform promotion and the like. In the application, the answer sentence recommendation method is mainly described by taking an agent telephone outbound call scene as an example, and answer sentence recommendation methods in other scenes are similar to the implementation manner in the agent telephone outbound call scene and are not described herein.
Hereinafter, a reply sentence recommendation method disclosed in the present application will be explained:
referring to fig. 2, fig. 2 is a schematic flowchart of a method for recommending a reply sentence according to an embodiment of the present application. The reply sentence recommendation method comprises the following steps:
201: and searching historical dialogue data according to the voice information of the user at the current moment to generate a user intention chain and an agent intention chain.
In this embodiment, the user intention chain is used to identify the intention trend of the user in the current round of conversation, and the agent intention chain is used to identify the intention trend of the agent having a conversation with the user in the current round of conversation. Illustratively, by means of a preset intention considering range, extracting the dialog data which is in the range before the current time, and sequentially extracting the intention of each dialog data, so that the intentions are sorted according to the time when the dialog occurs, the corresponding intention chain can be obtained.
Based on this, the present embodiment provides a method for searching historical dialog data according to the speech information of the user at the current time, and generating a user intention chain and an agent intention chain, as shown in fig. 3, the method includes:
301: and identifying the voice information of the user to obtain a first sentence.
In this embodiment, the voice information may be analyzed to obtain the acoustic feature of the voice information, and then the dialect category of the voice information may be determined according to the acoustic feature. And then, acquiring an audio transposition formula corresponding to the dialect type, and converting the voice information into standard voice through the audio transposition formula. Therefore, the standard voice is subjected to feature extraction to obtain corresponding audio features, and then matching is carried out in a preset neural network according to the audio features to obtain a pinyin text matched with the audio features. Specifically, the pinyin text may be composed of at least one first pinyin-meta text, and the first pinyin-meta text refers to any one of an initial or a final.
In this embodiment, after obtaining the pinyin text, matching may be performed in the neural network according to each first pinyin element text in at least one first pinyin element text in the pinyin text, so as to obtain at least one first character corresponding to the at least one first pinyin element text one to one. Then, the at least one first character is arranged according to the arrangement sequence of the at least one first pinyin element text in the pinyin text according to the corresponding relation between the at least one first character and the at least one first pinyin element text, and a first sentence can be obtained.
302: and extracting the first n pairs of question-answer pairs of the first sentence according to historical dialogue data to obtain the n pairs of question-answer pairs.
In the present embodiment, n is an integer greater than or equal to 1, and specifically, the specific value of n may be determined by a preset range of intent considerations, which may be set to different values according to the application field. For example, for a domain that is strongly related to a dialog, the value of n may be set smaller. In the field with weak conversation correlation, because the correlation between each pair of conversations is weak, the number of analysis rounds of the conversation can be enlarged to obtain an accurate analysis result, and therefore the n value can be set to be larger.
303: and splitting the question-answer pairs of the n pairs to obtain n user sentences and n seat sentences.
In the present embodiment, the n user sentences correspond to the n pairs of questions and answers one to one, and the n agent sentences correspond to the n pairs of questions and answers one to one.
In an alternative embodiment, the historical dialogue data may also be a queue of statements that stores two interrelated queues, where one queue is used to store user dialogue data for users and the other pair is used to store agent dialogue data for agents. Meanwhile, each user dialogue data in the user dialogue data and each agent dialogue data in the agent dialogue data comprise a dialogue identifier, and the user dialogue data and the agent dialogue data with the same dialogue identifier form a question-answer pair. Namely, the dialog mark is the same in the user dialog data and the agent dialog data, and the agent dialog data is a sentence for replying the user dialog data. Therefore, the question-answer logicality in the historical dialogue data can be guaranteed, and the dialogue data of the user and the agent are stored separately, so that the search is facilitated.
304: and respectively extracting the intention of each of the n agent sentences to obtain n agent intention characteristics.
In the present embodiment, there is provided a method for extracting an intention for each of n agent sentences to obtain n agent intention features, as shown in fig. 4, the method including:
401: and carrying out word segmentation processing on each seat sentence to obtain a key phrase.
In this embodiment, a set of separators may be preset, which includes some separators commonly used in chinese, for example: punctuation marks, special marks, diagrams, conjunctions, stop words, etc. And then matching each agent statement with the separator set, replacing the separator existing in each agent conversation with a space, and segmenting each agent statement to obtain at least one candidate field. Then, the forward maximum matching is carried out on each candidate field in the at least one candidate field and the general word segmentation dictionary respectively. And when the candidate field is successfully matched with the words in the dictionary, extracting the successfully matched words in the candidate field, and adding the words serving as a keyword into the keyword group.
402: and calculating the similarity between the key phrase and each corpus in at least one corpus in a preset corpus to obtain at least one corpus similarity corresponding to at least one corpus one by one.
In the present embodiment, each corpus of at least one corpus pre-stored in the corpus corresponds to one intention feature, but each intention feature may correspond to a plurality of corpora. The linguistic data are generated by analyzing historical outbound data to obtain different expression modes corresponding to each intention characteristic and combining keywords corresponding to each intention characteristic. In short, the corpus may be understood as a common expression pattern of its corresponding intended features.
Meanwhile, in the present embodiment, each intention feature corresponding to at least one corpus is determined according to a specific scenario in the application field corresponding to the corpus, for example, for a scenario of an agent outgoing call in the bank field, the intention features that may be included in the corpus are: the interest rate, interest, amount, repayment time, red-separating proportion and other strongly related intention characteristics.
In addition, each corpus also corresponds to a corpus key phrase obtained by performing word segmentation processing on each corpus, and the specific implementation manner may refer to the word segmentation processing on each agent sentence instep 401, which is not described herein again.
Based on this, in the present embodiment, the similarity between the keyword group and each corpus can be expressed by formula (i):
wherein x represents the x-th corpus in the corpus, yxRepresenting the number of key words in the corpus key phrase corresponding to the x-th corpus, y representing the number of key words in the key phrase, aiCharacteristic value representing the ith keyword in the keyword group, biAnd the coefficient of the keyword corresponding to the ith keyword is shown, wherein i is an integer which is greater than or equal to 1.
Further, the feature value of the ith keyword can be represented by a formula (II):
wherein L isiInverse document frequency, D, representing the ith keywordiWord frequency difference, H, representing the ith keywordiWord length, G, representing the ith keywordiAnd k is an adjustment coefficient and is an integer greater than or equal to 1, and k can be equal to 1 in general.
Further, the coefficient b of the keyword corresponding to the ith keywordiThe value of (b) can be determined by determining whether the ith keyword exists in the corpus keyword group corresponding to the xth corpus. Specifically, the keyword coefficient b is obtained when the ith keyword exists in the corpus keyword group corresponding to thexth corpus i1 is ═ 1; when the ith keyword does not exist in the corpus keyword group corresponding to the xth corpus, the keyword coefficient bi＝0。
403: and determining at least one language material candidate in at least one language material according to at least one language material similarity.
In this embodiment, the corpus similarity corresponding to each corpus candidate in the at least one corpus candidate is greater than the first threshold.
404: and determining a target corpus in at least one candidate corpus according to the seat intention characteristics of the previous sentence of each seat sentence, and taking the intention characteristics corresponding to the target corpus as the seat intention characteristics of each seat sentence.
In this embodiment, the intention feature corresponding to each corpus candidate in at least one corpus candidate is matched with the corresponding agent sentence. Based on the above, the association degree between the intention feature corresponding to each candidate corpus and the seat intention feature of the previous sentence of the seat sentence can be calculated, the intention feature with the highest association degree with the current dialogue in the intention features corresponding to each candidate corpus is determined, and then the intention feature is taken as the finally determined seat intention feature. Therefore, the development direction of the current sentence is reasonably deduced by combining the intentions, so that the obtained intention characteristics are more fit with the current conversation scene, and the accuracy of the subsequent analysis result is further improved.
In an optional implementation manner, the target sentence may also be determined by obtaining the user intention characteristics of the user sentences in the question and answer pair to which each agent sentence belongs, and calculating the association degree between the intention characteristics corresponding to each candidate corpus and the user intention characteristics. Context can also be made tight and user intent and agent intent can be made to interact, enabling deep mining of intent relationships between contexts.
305: and sequencing the n seat intention characteristics according to the sequence of the occurrence time of each seat statement to obtain a seat intention chain.
Specifically, when n is 5, the obtained first 5 seat sentences are sorted according to the occurrence time sequence as follows: [statement 1, statement 2, statement 3, statement 4, statement 5 ]. Through willful extraction, the following can be obtained:willingness feature 1 forstatement 1, willingness feature 2 for statement 2, willingness feature 3 for statement 3, willingness feature 4 for statement 4, and willingness feature 5 for statement 5. Therefore, the seat intention chain can be obtained by sequencing according to the occurrence time sequence of the seat sentences: [ wishfeature 1, wish feature 2, wish feature 3, wish feature 4, wish feature 5 ].
306: and respectively carrying out intention extraction on the first sentence and each user sentence in the n user sentences to obtain n +1 user intention characteristics.
In this embodiment, the method for extracting the intention of each of the first sentence and the n user sentences is similar to the method for extracting the intention of each of the n agent sentences instep 304, and is not described herein again.
307: and sequencing the n +1 user intention characteristics according to the sequence of the first statement and the occurrence time of each user statement to obtain a user intention chain.
In this embodiment, the method for sorting the n +1 user intention features is similar to the method for sorting the n agent intention features instep 305, and is not described herein again.
202: and determining the user maximum income intention corresponding to the user intention chain and the seat maximum income intention corresponding to the seat intention chain according to the application field to which the voice information at the current moment belongs.
In this embodiment, the final purpose of each of the user and the agent can be determined according to the application field to which the voice information at the current time belongs, and the final purpose can be used as the maximum profit intention of each of the user and the agent. Specifically, for the scenario of an agent outgoing call in the banking domain, the final purpose of the agent is usually: successful promotion deals, the customer end purpose is: lower the price or improve the benefits, and is cost-effective. Both parties to a conversation converge in the conversation, each towards their respective final destination. Therefore, in this scenario, the user maximum profit intention corresponding to the user intention chain may be set as: low to high value; the maximum income intention of the seat corresponding to the seat intention chain is set as: and (6) carrying out hybridization.
203: and establishing a first functional relation between the maximum profit intention of the user and the user intention chain, and determining the user transfer probability of the user intention chain according to the first functional relation.
In this embodiment, the user intention chain may be converted into a first hidden markov chain, and a first transition probability of each link in the first hidden markov chain may be determined. And then, establishing a first functional relation by taking the maximum profit intention of the user as an objective function of the first transfer probability of each link.
Specifically, the first functional relationship may be constructed by introducing the maximum profit intention of the user into the game tree, that is, taking the maximum profit of the user as an objective function of the user and a user intention chain as a hidden markov chain in the game tree. In a hidden markov chain, each state is gradually conducted through the transition probability, so that the next transition in the hidden markov chain can be determined only by determining the transition probability.
For example, the maximum profit-and-interest of the user may be set as shown in formula (c):
max(M)＝Z(s)＝max(r(s)).........③
where max (m) represents the maximum benefit intent of the user, and r(s) is an objective function with respect to z(s), whereby the transfer matrix z(s) may be set to p(s) based on historical data, i.e.: and z(s) ═ p(s), wherein p(s) is the user transition probability.
204: and establishing a second functional relationship among the user transfer probability, the agent maximum income intention and the agent intention chain, and grading at least one preset answer dialog according to the second functional relationship to obtain at least one score in one-to-one correspondence with the at least one answer dialog.
In this embodiment, the agent intention chain may be converted into a second hidden markov chain, and a second transition probability of each link in the second hidden markov chain may be determined. And establishing a third functional relation by taking the maximum seat income intention as an objective function of the second transition probability of each link. And then, determining parameters of the seat transfer matrix according to the second transfer probability of each link and the user transfer probability. And finally, rewriting the third function according to the parameters of the agent transfer matrix to obtain a second function relation.
Specifically, the user transfer probability may be used as a parameter of a transfer matrix between the agent maximum profit intention and the agent intention chain, that is, the transfer matrix z (k) between the agent maximum profit intention and the agent intention chain may be expressed by the formula (r):
Z(k)＝tp(k)+ep(s).........④
where t, e are weight parameters and p(s) is the transition probability determined by the user intent chain.
Therefore, the functional relation between the seat maximum profit intention and the seat intention chain is rewritten through the user transfer probability p(s), so that the second functional relation can be obtained, the user intention and the seat intention are correlated, the hidden correlation characteristics are mined, and the subsequent intention prediction is more accurate.
In an alternative embodiment, the first functional relationship may also be directly used as a parameter of the transfer matrix between the agent maximum benefit intention and the agent intention chain, that is, the transfer matrix z (k) between the agent maximum benefit intention and the agent intention chain may be expressed by the formula (c):
Z(k)＝tp(k)+emax(M).........⑤
where max (m) is the first functional relationship obtained instep 203.
205: and generating a reply sentence according to the reply grammar corresponding to the highest score in the at least one score, and pushing the reply sentence to the agent.
In summary, in the reply sentence recommendation method provided by the present invention, the historical dialogue data of the current round of dialogue is obtained through the voice information of the user at the current time, and a user intention chain and an agent intention chain are established. Then, based on the application field to which the voice information at the current moment belongs, the maximum profit intentions of the user and the seat in the field are determined, namely the final communication target. Therefore, the transition probability of the user intention chain is confirmed by establishing a first functional relationship between the maximum profit intention of the user and the user intention chain, and then the transition probability of the user intention chain is used as a parameter of a transition matrix of the agent intention chain to disturb the functional relationship between the agent intention chain and the maximum profit intention of the agent, so as to obtain a second functional relationship. Finally, the scores of the candidate dialogs are confirmed through the second functional relation, so that a reply sentence is generated through the dialogs with the highest scores, and the reply is carried out on the user. Based on the method, the seat intention chain is disturbed through the user intention chain, so that the user intention chain influences and games the seat intention chain mutually, and finally convergence is carried out according to respective corresponding maximized income, so that deep mining of intention connection between contexts is realized, inference logic is introduced, accuracy of intention identification is improved, answer sentences are accurately matched, and the interpretability is strong.
Referring to fig. 5, fig. 5 is a block diagram illustrating functional modules of a device for recommending reply sentences according to an embodiment of the present application. As shown in fig. 5, the replysentence recommendation apparatus 500 includes:
thequery module 501 is configured to search historical dialog data according to the voice information of the user at the current time, and generate a user intention chain and an agent intention chain, where the user intention chain is used to identify an intention trend of the user in the current round of dialog, and the agent intention chain is used to identify an intention trend of an agent performing a dialog with the user in the current round of dialog;
theanalysis module 502 is configured to determine, according to the application field to which the voice information at the current time belongs, a user maximum profit intention corresponding to the user intention chain and an agent maximum profit intention corresponding to the agent intention chain;
theprocessing module 503 is configured to establish a first functional relationship between the user maximum profit intention and the user intention chain, determine a user transfer probability of the user intention chain according to the first functional relationship, establish a second functional relationship between the user transfer probability, the agent maximum profit intention and the agent intention chain, and score at least one preset answer dialog according to the second functional relationship to obtain at least one score, where the at least one score is in one-to-one correspondence with the at least one answer dialog;
and the recommendingmodule 504 is configured to generate a reply sentence according to the reply grammar corresponding to the highest score in the at least one score, and push the reply sentence to the agent.
In the embodiment of the present invention, in terms of searching historical dialog data according to the speech information of the user at the current time, and generating a user intention chain and an agent intention chain, thequery module 501 is specifically configured to:
recognizing voice information of a user to obtain a first sentence;
extracting the first n pairs of question-answer pairs of the first statement according to historical dialogue data to obtain n pairs of question-answer pairs, wherein n is an integer greater than or equal to 1;
splitting the question-answer pairs to obtain n user sentences and n seat sentences, wherein the n user sentences correspond to the question-answers pairs one by one, and the n seat sentences correspond to the question-answers pairs one by one;
respectively extracting intentions of each of the n agent sentences to obtain n agent intention characteristics;
sequencing the n seat intention characteristics according to the sequence of the occurrence time of each seat statement to obtain a seat intention chain;
respectively extracting intentions of the first sentence and each user sentence in the n user sentences to obtain n +1 user intention characteristics;
and sequencing the n +1 user intention characteristics according to the sequence of the first statement and the occurrence time of each user statement to obtain a user intention chain.
In an embodiment of the present invention, in terms of respectively performing intent extraction on each of n agent sentences to obtain n agent intent features, thequery module 501 is specifically configured to:
carrying out word segmentation processing on each seat sentence to obtain a key phrase;
calculating the similarity between the key phrase and each corpus in at least one corpus in a preset corpus to obtain at least one corpus similarity, wherein the at least one corpus similarity corresponds to the at least one corpus one by one;
determining at least one corpus candidate in at least one corpus according to at least one corpus similarity, wherein the corpus similarity corresponding to each corpus candidate in the at least one corpus candidate is greater than a first threshold;
and determining a target corpus in at least one candidate corpus according to the seat intention characteristics of the previous sentence of each seat sentence, and taking the intention characteristics corresponding to the target corpus as the seat intention characteristics of each seat sentence.
In the embodiment of the present invention, the similarity between the keyword group and each corpus can be represented by the formula (c):
wherein x represents the x-th corpus in the corpus, yxRepresenting the number of key words in the corpus key phrase corresponding to the x-th corpus, y representing the number of key words in the key phrase, aiCharacteristic value representing the ith keyword in the keyword group, biAnd the coefficient of the keyword corresponding to the ith keyword is shown, wherein i is an integer which is greater than or equal to 1.
In an embodiment of the present invention, the feature value of the ith keyword may be represented by formula (c):
wherein L isiInverse document frequency, D, representing the ith keywordiWord frequency difference, H, representing the ith keywordiWord length, G, representing the ith keywordiAnd k is an adjustment coefficient and is an integer greater than or equal to 1.
In an embodiment of the present invention, in terms of establishing the first functional relationship between the user maximum profit intention and the user intention chain, theprocessing module 503 is specifically configured to:
converting the user intention chain into a first hidden Markov chain, and determining a first transfer probability of each link in the first hidden Markov chain;
and establishing a first functional relation by taking the maximum profit intention of the user as an objective function of the first transfer probability of each link.
In an embodiment of the present invention, in establishing the second functional relationship between the user transfer probability, the agent maximum profit intent, and the agent intent chain, theprocessing module 503 is specifically configured to:
converting the seat intention chain into a second hidden Markov chain, and determining a second transition probability of each link in the second hidden Markov chain;
establishing a third functional relation by taking the agent maximum income intention as a target function of the second transition probability of each link;
determining parameters of the seat transfer matrix according to the second transfer probability of each link and the user transfer probability;
and rewriting the third function according to the parameters of the seat transfer matrix to obtain a second function relation.
Referring to fig. 6, fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure. As shown in fig. 6, theelectronic device 600 includes atransceiver 601, aprocessor 602, and amemory 603. Connected to each other by abus 604. Thememory 603 is used to store computer programs and data, and can transfer data stored in thememory 603 to theprocessor 602.
Theprocessor 602 is configured to read the computer program in thememory 603 to perform the following operations:
searching historical conversation data according to the voice information of the user at the current moment, and generating a user intention chain and an agent intention chain, wherein the user intention chain is used for identifying the intention trend of the user in the current round of conversation, and the agent intention chain is used for identifying the intention trend of an agent in the current round of conversation;
determining the user maximum income intention corresponding to the user intention chain and the seat maximum income intention corresponding to the seat intention chain according to the application field to which the voice information at the current moment belongs;
establishing a first functional relationship between the maximum profit intention of the user and the user intention chain, and determining the user transfer probability of the user intention chain according to the first functional relationship;
establishing a second functional relationship among the user transfer probability, the agent maximum income intention and the agent intention chain, and grading at least one preset answer dialog according to the second functional relationship to obtain at least one score, wherein the at least one score is in one-to-one correspondence with the at least one answer dialog;
and generating a reply sentence according to the reply grammar corresponding to the highest score in the at least one score, and pushing the reply sentence to the agent.
In an embodiment of the present invention, in searching historical dialog data according to the speech information of the user at the current time, and generating a user intention chain and an agent intention chain, theprocessor 602 is specifically configured to perform the following operations:
recognizing voice information of a user to obtain a first sentence;
extracting the first n pairs of question-answer pairs of the first statement according to historical dialogue data to obtain n pairs of question-answer pairs, wherein n is an integer greater than or equal to 1;
splitting the question-answer pairs to obtain n user sentences and n seat sentences, wherein the n user sentences correspond to the question-answers pairs one by one, and the n seat sentences correspond to the question-answers pairs one by one;
respectively extracting intentions of each of the n agent sentences to obtain n agent intention characteristics;
sequencing the n seat intention characteristics according to the sequence of the occurrence time of each seat statement to obtain a seat intention chain;
respectively extracting intentions of the first sentence and each user sentence in the n user sentences to obtain n +1 user intention characteristics;
and sequencing the n +1 user intention characteristics according to the sequence of the first statement and the occurrence time of each user statement to obtain a user intention chain.
In an embodiment of the present invention, in terms of performing intent extraction on each of n agent sentences to obtain n agent intent features, theprocessor 602 is specifically configured to perform the following operations:
carrying out word segmentation processing on each seat sentence to obtain a key phrase;
calculating the similarity between the key phrase and each corpus in at least one corpus in a preset corpus to obtain at least one corpus similarity, wherein the at least one corpus similarity corresponds to the at least one corpus one by one;
determining at least one corpus candidate in at least one corpus according to at least one corpus similarity, wherein the corpus similarity corresponding to each corpus candidate in the at least one corpus candidate is greater than a first threshold;
and determining a target corpus in at least one candidate corpus according to the seat intention characteristics of the previous sentence of each seat sentence, and taking the intention characteristics corresponding to the target corpus as the seat intention characteristics of each seat sentence.
In the embodiment of the present invention, the similarity between the key word group and each corpus can be expressed by the formula (r):
wherein x represents the x-th corpus in the corpus, yxRepresenting the number of key words in the corpus key phrase corresponding to the x-th corpus, y representing the number of key words in the key phrase, aiCharacteristic value representing the ith keyword in the keyword group, biAnd the coefficient of the keyword corresponding to the ith keyword is shown, wherein i is an integer which is greater than or equal to 1.
In an embodiment of the present invention, the feature value of the ith keyword may be represented by formula ninthly:
wherein L isiInverse document frequency, D, representing the ith keywordiWord frequency difference, H, representing the ith keywordiWord length, G, representing the ith keywordiAnd k is an adjustment coefficient and is an integer greater than or equal to 1.
In an embodiment of the present invention, in terms of establishing the first functional relationship between the user's maximum profit intent and the chain of user intentions, theprocessor 602 is specifically configured to:
converting the user intention chain into a first hidden Markov chain, and determining a first transfer probability of each link in the first hidden Markov chain;
and establishing a first functional relation by taking the maximum profit intention of the user as an objective function of the first transfer probability of each link.
In an embodiment of the present invention, in establishing the second functional relationship between the user transition probability, the agent maximum profit intent and the agent intent chain, theprocessor 602 is specifically configured to:
converting the seat intention chain into a second hidden Markov chain, and determining a second transition probability of each link in the second hidden Markov chain;
establishing a third functional relation by taking the agent maximum income intention as a target function of the second transition probability of each link;
determining parameters of the seat transfer matrix according to the second transfer probability of each link and the user transfer probability;
and rewriting the third function according to the parameters of the seat transfer matrix to obtain a second function relation.
It should be understood that the reply sentence recommendation device in the present application may include a smart Phone (e.g., an Android Phone, an iOS Phone, a Windows Phone, etc.), a tablet computer, a palm computer, a notebook computer, a Mobile Internet device MID (MID), a robot, a wearable device, etc. The above reply sentence recommending apparatus is merely an example, and is not exhaustive, and includes but is not limited to the above reply sentence recommending apparatus. In practical applications, the above reply sentence recommendation apparatus may further include: intelligent vehicle-mounted terminal, computer equipment and the like.
Through the above description of the embodiments, those skilled in the art will clearly understand that the present invention can be implemented by combining software and a hardware platform. With this understanding in mind, all or part of the technical solutions of the present invention that contribute to the background can be embodied in the form of a software product, which can be stored in a storage medium, such as a ROM/RAM, a magnetic disk, an optical disk, etc., and includes instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods according to the embodiments or some parts of the embodiments.
Accordingly, the present application embodiment also provides a computer-readable storage medium storing a computer program, which is executed by a processor to implement part or all of the steps of any one of the reply sentence recommendation methods as set forth in the above method embodiments. For example, the storage medium may include a hard disk, a floppy disk, an optical disk, a magnetic tape, a magnetic disk, a flash memory, and the like.
Embodiments of the present application also provide a computer program product comprising a non-transitory computer readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps of any of the reply sentence recommendation methods as set out in the above method embodiments.
It should be noted that, for simplicity of description, the above-mentioned method embodiments are described as a series of acts or combination of acts, but those skilled in the art will recognize that the present application is not limited by the order of acts described, as some steps may occur in other orders or concurrently depending on the application. Further, those skilled in the art should also appreciate that the embodiments described in the specification are all alternative embodiments and that the acts and modules referred to are not necessarily required by the application.
In the above embodiments, the description of each embodiment has its own emphasis, and for parts not described in detail in a certain embodiment, reference may be made to the description of other embodiments.
In the several embodiments provided in the present application, it should be understood that the disclosed apparatus may be implemented in other manners. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is merely a logical division, and other divisions may be realized in practice, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection of some interfaces, devices or units, and may be an electric or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.
In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit. The integrated unit may be implemented in the form of hardware, or may be implemented in the form of a software program module.
The integrated units, if implemented in the form of software program modules and sold or used as stand-alone products, may be stored in a computer readable memory. Based on such understanding, the technical solution of the present application may be substantially implemented or a part of or all or part of the technical solution contributing to the prior art may be embodied in the form of a software product stored in a memory, and including several instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method described in the embodiments of the present application. And the aforementioned memory comprises: a U-disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a removable hard disk, a magnetic or optical disk, and other various media capable of storing program codes.
Those skilled in the art will appreciate that all or part of the steps in the methods of the above embodiments may be implemented by associated hardware instructed by a program, which may be stored in a computer-readable memory, and the memory may include: flash Memory disks, Read-Only memories (ROMs), Random Access Memories (RAMs), magnetic or optical disks, and the like.
The foregoing detailed description of the embodiments of the present application has been presented to illustrate the principles and implementations of the present application, and the above description of the embodiments is only provided to help understand the methods and their core ideas of the present application; meanwhile, for a person skilled in the art, according to the idea of the present application, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present application.
Claims (10)
1. A reply sentence recommendation method, the method comprising:
searching historical conversation data according to the voice information of the user at the current moment, and generating a user intention chain and an agent intention chain, wherein the user intention chain is used for identifying the intention trend of the user in the current round of conversation, and the agent intention chain is used for identifying the intention trend of an agent in the current round of conversation, which has a conversation with the user;
determining the user maximum income intention corresponding to the user intention chain and the seat maximum income intention corresponding to the seat intention chain according to the application field to which the voice information at the current moment belongs;
establishing a first functional relationship between the user maximum income intention and the user intention chain, and determining the user transfer probability of the user intention chain according to the first functional relationship;
establishing a second functional relationship among the user transfer probability, the seat maximum profit intention and the seat intention chain, and grading at least one preset answer dialog according to the second functional relationship to obtain at least one score, wherein the at least one score is in one-to-one correspondence with the at least one answer dialog;
and generating a reply sentence according to the reply dialect corresponding to the highest score in the at least one score, and pushing the reply sentence to the agent.
2. The method according to claim 1, wherein the searching historical dialogue data according to the voice information of the user at the current moment to generate a user intention chain and an agent intention chain comprises:
recognizing voice information of a user to obtain a first sentence;
extracting the first n pairs of question-answer pairs of the first statement according to historical dialogue data to obtain n pairs of question-answer pairs, wherein n is an integer greater than or equal to 1;
splitting the n pairs of question-answer pairs to obtain n user sentences and n seat sentences, wherein the n user sentences are in one-to-one correspondence with the n pairs of question-answer pairs, and the n seat sentences are in one-to-one correspondence with the n pairs of question-answer pairs;
respectively extracting intentions of each of the n agent sentences to obtain n agent intention features;
sequencing the n seat intention features according to the sequence of the occurrence time of each seat statement to obtain the seat intention chain;
respectively extracting intentions of the first sentence and each user sentence in the n user sentences to obtain n +1 user intention characteristics;
and sequencing the n +1 user intention characteristics according to the sequence of the first statement and the occurrence time of each user statement to obtain the user intention chain.
3. The method according to claim 2, wherein the performing intent extraction on each of the n agent sentences to obtain n agent intent features comprises:
carrying out word segmentation processing on each seat sentence to obtain a key phrase;
calculating the similarity between the key phrase and each corpus in at least one corpus in a preset corpus to obtain at least one corpus similarity, wherein the at least one corpus similarity corresponds to the at least one corpus one by one;
determining at least one corpus candidate in the at least one corpus according to at least one corpus similarity, wherein the corpus similarity corresponding to each corpus candidate in the at least one corpus candidate is greater than a first threshold;
and determining a target corpus in the at least one candidate corpus according to the seat intention characteristics of the previous sentence of each seat sentence, and taking the intention characteristics corresponding to the target corpus as the seat intention characteristics of each seat sentence.
4. The method according to claim 3, wherein the similarity between the keyword group and each corpus satisfies the following formula:
wherein x represents the x-th corpus in the corpus, and yxRepresenting the number of key words in the corpus key phrase corresponding to the x-th corpus, y representing the number of key words in the key phrase, aiRepresenting the characteristic value of the ith keyword in the keyword set, biAnd representing the keyword coefficient corresponding to the ith keyword, wherein i is an integer greater than or equal to 1.
5. The method of claim 4, wherein the feature value of the ith keyword satisfies the following formula:
wherein L isiRepresents the firstinverse document frequency of i keywords, DiWord frequency difference, H, representing the ith keywordiWord length, G, representing the ith keywordiAnd k is an adjustment coefficient and is an integer greater than or equal to 1.
6. The method of claim 1, wherein establishing a first functional relationship between the user's maximum profit intent and the chain of user intentions comprises:
converting the user intention chain into a first hidden Markov chain, and determining a first transition probability of each link in the first hidden Markov chain;
and establishing the first functional relationship by taking the maximum profit intention of the user as an objective function of the first transfer probability of each link.
7. The method of claim 6, wherein establishing a second functional relationship between the user transition probability, the agent maximum gain intent, and the chain of agent intentions comprises:
converting the seat intention chain into a second hidden Markov chain, and determining a second transition probability of each link in the second hidden Markov chain;
establishing a third functional relation by taking the agent maximum profit intention as an objective function of the second transition probability of each link;
determining parameters of an agent transfer matrix according to the second transfer probability of each link and the user transfer probability;
and rewriting the third function according to the parameters of the agent transfer matrix to obtain the second functional relation.
8. An answer sentence recommendation apparatus, characterized in that the apparatus comprises:
the query module is used for searching historical conversation data according to the voice information of the user at the current moment and generating a user intention chain and an agent intention chain, wherein the user intention chain is used for identifying the intention trend of the user in the current round of conversation, and the agent intention chain is used for identifying the intention trend of an agent in the current round of conversation, which has a conversation with the user;
the analysis module is used for determining the user maximum income intention corresponding to the user intention chain and the seat maximum income intention corresponding to the seat intention chain according to the application field to which the voice information at the current moment belongs;
the processing module is used for establishing a first functional relationship between the user maximum income intention and the user intention chain, determining a user transfer probability of the user intention chain according to the first functional relationship, establishing a second functional relationship among the user transfer probability, the seat maximum income intention and the seat intention chain, and grading at least one preset answer dialog according to the second functional relationship to obtain at least one score, wherein the at least one score is in one-to-one correspondence with the at least one answer dialog;
and the recommending module is used for generating a reply sentence according to the reply grammar corresponding to the highest score in the at least one score and pushing the reply sentence to the agent.
9. An electronic device comprising a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the processor, the one or more programs including instructions for performing the steps in the method of any of claims 1-7.
10. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program which is executed by a processor to implement the method according to any one of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111567908.7ACN114281969A (en) | 2021-12-20 | 2021-12-20 | Reply sentence recommendation method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111567908.7ACN114281969A (en) | 2021-12-20 | 2021-12-20 | Reply sentence recommendation method and device, electronic equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114281969Atrue CN114281969A (en) | 2022-04-05 |
Family
ID=80873374
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111567908.7APendingCN114281969A (en) | 2021-12-20 | 2021-12-20 | Reply sentence recommendation method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114281969A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118690856A (en)* | 2024-08-26 | 2024-09-24 | 深圳市华金硕科技有限公司 | Intelligent question-answering method and system for after-sales service of components based on deep learning |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109829038A (en)* | 2018-12-11 | 2019-05-31 | 平安科技（深圳）有限公司 | Question and answer feedback method, device, equipment and storage medium based on deep learning |
| CN111723189A (en)* | 2020-06-23 | 2020-09-29 | 贝壳技术有限公司 | Interactive question and answer prompting method and device, storage medium and electronic equipment |
- 2021
- 2021-12-20CNCN202111567908.7Apatent/CN114281969A/enactivePending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109829038A (en)* | 2018-12-11 | 2019-05-31 | 平安科技（深圳）有限公司 | Question and answer feedback method, device, equipment and storage medium based on deep learning |
| CN111723189A (en)* | 2020-06-23 | 2020-09-29 | 贝壳技术有限公司 | Interactive question and answer prompting method and device, storage medium and electronic equipment |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118690856A (en)* | 2024-08-26 | 2024-09-24 | 深圳市华金硕科技有限公司 | Intelligent question-answering method and system for after-sales service of components based on deep learning |
| CN118690856B (en)* | 2024-08-26 | 2024-10-22 | 深圳市华金硕科技有限公司 | Intelligent question-answering method and system for after-sales service of components based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111914568B (en) | Method, device and equipment for generating text sentence and readable storage medium | |
| EP4113357A1 (en) | Method and apparatus for recognizing entity, electronic device and storage medium | |
| CN110334179B (en) | Question-answer processing method, device, computer equipment and storage medium | |
| CN115309877B (en) | Dialogue generation method, dialogue model training method and device | |
| CN114416943B (en) | Training method and device for dialogue model, electronic equipment and storage medium | |
| CN113505293B (en) | Information pushing method and device, electronic equipment and storage medium | |
| CN109753661A (en) | A machine reading comprehension method, device, device and storage medium | |
| CN112349294A (en) | Voice processing method and device, computer readable medium and electronic equipment | |
| CN114528851A (en) | Reply statement determination method and device, electronic equipment and storage medium | |
| CN110717021A (en) | Input text and related device for obtaining artificial intelligence interview | |
| CN108345612A (en) | A kind of question processing method and device, a kind of device for issue handling | |
| CN115272540A (en) | Processing method and device based on virtual customer service image, equipment and medium | |
| CN112926308A (en) | Method, apparatus, device, storage medium and program product for matching text | |
| CN111680136A (en) | Method and device for matching spoken language and semantics | |
| CN114048757A (en) | A sign language synthesis method, device, computer equipment and storage medium | |
| CN114218356B (en) | Semantic recognition method, device, equipment and storage medium based on artificial intelligence | |
| WO2020199590A1 (en) | Mood detection analysis method and related device | |
| CN114969195A (en) | Dialogue Content Mining Method and Dialogue Content Evaluation Model Generation Method | |
| CN112860995B (en) | Interaction method, device, client, server and storage medium | |
| CN114281969A (en) | Reply sentence recommendation method and device, electronic equipment and storage medium | |
| CN119149681A (en) | Information processing method, information processing apparatus, computer device, storage medium, and program product | |
| CN110377706B (en) | Search sentence mining method and device based on deep learning | |
| US20220351085A1 (en) | Method and apparatus for presenting candidate character string, and method and apparatus for training discriminative model | |
| CN116741141A (en) | Speech synthesis method, speech synthesis device, electronic device, and storage medium | |
| CN114519094A (en) | Method and device for conversational recommendation based on random state and electronic equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | Application publication date:20220405 | |
| WD01 | Invention patent application deemed withdrawn after publication |