Disclosure of Invention
The embodiment of the invention aims to provide a video live broadcast resource recommendation method, a video live broadcast resource recommendation system and a video live broadcast resource recommendation device, which can provide personalized recommendation for users and reduce recommendation time delay. The specific technical scheme is as follows:
in a first aspect of the present invention, there is first provided a live video resource recommendation method, including:
receiving a video live broadcast resource request sent by a target client;
inquiring each video live broadcast resource corresponding to the target client in the corresponding relation between the pre-recorded client and the video live broadcast resource to be used as a first video live broadcast resource; wherein, the correspondence is: based on the user characteristics of the user logging in the client, the similarity between the user characteristics and the resource characteristics of the live video resources is determined;
recommending the video live broadcast resource to the target client based on the arrangement sequence of the first video live broadcast resource in the corresponding relation.
Optionally, if the target client does not exist in the corresponding relationship, recommending the video live broadcast resource to the target client based on a preset video live broadcast resource.
Optionally, after pushing the live video resource to the target client based on the arrangement order of the live video resource in the correspondence, the method further includes:
when reaching a preset moment, determining a video live broadcast resource browsed by a target user of the target client after last recommendation, and taking the video live broadcast resource as a second video live broadcast resource;
determining a video live asset associated with the second video live asset as a third video live asset;
based on the similarity between the resource characteristics of the third video live broadcast resource and the user characteristics of the target user, sequencing the third video live broadcast resource to obtain a sequencing result;
and updating the video live broadcast resources corresponding to the target clients in the corresponding relation according to the sorting result.
Optionally, the method is applied to a resource recommendation server in a resource recommendation system, and the resource recommendation system further comprises a resource sequencing server;
The step of sorting the third live video resources based on the resource characteristics of the third live video resources to obtain a sorting result includes:
and sending the resource characteristics of the third live video resources to the resource ordering server so that the resource ordering server orders the third live video resources based on the similarity between the resource characteristics of the third live video resources and the user characteristics of the target user to obtain an ordering result, and sending the ordering result to the resource recommending server.
Optionally, the sorting the third live video resource based on the similarity between the resource feature of the third live video resource and the user feature of the target user to obtain a sorting result includes:
inputting the resource characteristics of the third video live broadcast resource and the user characteristics of the target user into a pre-trained browsing prediction network model aiming at each third video live broadcast resource to obtain the probability of the target user browsing the third video live broadcast resource, wherein the probability is used as the similarity between the resource characteristics of the third video live broadcast resource and the user characteristics of the target user; the browsing prediction network model is obtained by training based on a preset training sample; the preset training sample comprises user characteristics of a sample user and resource characteristics of video live broadcast resources browsed by the sample user;
And sequencing the third live video resources according to the sequence of the corresponding similarity from large to small to obtain a sequencing result.
Optionally, after determining the last recommendation, logging in the live video resource browsed by the user of the target client as the second live video resource, the method further includes:
and aiming at each second video live broadcast resource, taking the resource characteristics of the second video live broadcast resource and the user characteristics of the target user as training data, and adjusting the model parameters of the browsing prediction network model to update the browsing prediction network model.
In a second aspect of the present invention, a live video resource recommendation method is provided, where the method is applied to a resource ordering server in a resource recommendation system, and the resource recommendation system further includes a resource recommendation server, and the method includes:
receiving resource characteristics of a third live video resource sent by the resource recommendation server; wherein the third live video asset is associated with a second live video asset; the second live video resources are live video resources which are determined by the resource recommendation server at a preset moment and are browsed by a target user logged in a target client after the last recommendation;
Based on the similarity between the resource characteristics of the third video live broadcast resource and the user characteristics of the target user, sequencing the third video live broadcast resource to obtain a sequencing result;
and sending the sequencing result to the resource recommendation server so that the resource recommendation server records the corresponding relation between the target client and the sequencing result, and recommending the video live broadcast resource to the target client based on the corresponding relation when receiving the video live broadcast resource request sent by the target client.
Optionally, the sorting the third live video resource based on the similarity between the resource feature of the third live video resource and the user feature of the target user to obtain a sorting result includes:
inputting the resource characteristics of the third video live broadcast resource and the user characteristics of the target user into a pre-trained browsing prediction network model aiming at each third video live broadcast resource to obtain the probability of the target user browsing the third video live broadcast resource, wherein the probability is used as the similarity between the resource characteristics of the third video live broadcast resource and the user characteristics of the target user; the browsing prediction network model is obtained by training based on a preset training sample; the preset training sample comprises user characteristics of a sample user and resource characteristics of video live broadcast resources browsed by the sample user;
And sequencing the third live video resources according to the sequence of the corresponding similarity from large to small to obtain a sequencing result.
In a third aspect of the present invention, there is also provided a live video resource recommendation system, the system including:
the live video resource recommendation system comprises a target client and a resource recommendation server, wherein:
the client is used for sending a live video resource request to the resource recommendation server;
the resource recommendation server is used for receiving a video live broadcast resource request sent by a client, and inquiring each video live broadcast resource corresponding to the target client in a corresponding relation between the pre-recorded client and the video live broadcast resource to be used as a first video live broadcast resource; wherein, the correspondence is: based on the user characteristics of the user logging in the client, the similarity between the user characteristics and the resource characteristics of the live video resources is determined; recommending the video live broadcast resource to the target client based on the arrangement sequence of the first video live broadcast resource in the corresponding relation.
Optionally, the resource recommendation server is further configured to recommend a live video resource to the target client based on a preset live video resource if the target client does not exist in the corresponding relationship.
Optionally, the resource recommendation server is further configured to, after pushing the video live resources to the target client based on the arrangement order of the first video live resources in the corresponding relationship, determine a resource feature of a third video live resource associated with a second video live resource browsed by a target user logging in the target client after the last recommendation;
based on the similarity between the resource characteristics of the third video live broadcast resource and the user characteristics of the target user, sequencing the third video live broadcast resource to obtain a sequencing result;
and updating the video live broadcast resources corresponding to the target clients in the corresponding relation according to the sorting result.
Optionally, the live video resource recommendation system further comprises a resource ordering server;
the resource recommendation server is further configured to send a resource feature of the third live video resource to the resource ordering server;
and the resource ordering server is used for ordering the third live video resources based on the similarity between the resource characteristics of the third live video resources and the user characteristics of the target user to obtain an ordering result, and sending the ordering result to the resource recommending server.
Optionally, the resource ordering server is configured to input, for each third live video resource, a resource feature of the third live video resource and a user feature of the target user to a pre-trained browsing prediction network model, to obtain a probability that the target user browses the third live video resource, where the probability is used as a similarity between the resource feature of the third live video resource and the user feature of the target user; the browsing prediction network model is obtained by training based on a preset training sample; the preset training sample comprises user characteristics of a sample user and resource characteristics of video live broadcast resources browsed by the sample user;
and sequencing the third live video resources according to the sequence of the corresponding similarity from large to small to obtain a sequencing result.
Optionally, the live video resource recommendation system further comprises a resource acquisition server;
the resource acquisition server is used for determining a second live video resource browsed by the user of the target client after the last recommendation according to the browsing behavior data of the target user sent by the target client;
Determining a live video resource associated with the second live video resource as a third live video resource;
storing the resource characteristics of the third live video resource into a first preset storage space;
the resource recommendation server is configured to obtain a resource feature of the third live video resource from the first preset storage space.
Optionally, the resource ordering server is further configured to obtain the browsing prediction network model from a second preset storage space;
the resource obtaining server is further configured to update, for each second live video resource, a model parameter of a browsing prediction network model in the second preset storage space with a resource feature of the second live video resource and a user feature of the target user as training data.
In a fourth aspect of the present invention, there is also provided a live video resource recommendation apparatus, the apparatus including:
the information receiving module is used for receiving a video live broadcast resource request sent by the target client;
the first video live broadcast resource query module is used for querying each video live broadcast resource corresponding to the target client side in the corresponding relation between the pre-recorded client side and the video live broadcast resource as a first video live broadcast resource; wherein, the correspondence is: based on the user characteristics of the user logging in the client, the similarity between the user characteristics and the resource characteristics of the live video resources is determined;
And the video live broadcast resource recommending module is used for recommending the video live broadcast resource to the target client based on the arrangement sequence of the first video live broadcast resource in the corresponding relation.
Optionally, the live video resource recommendation module is further configured to recommend live video resources to the target client based on preset live video resources if the target client does not exist in the corresponding relationship.
Optionally, the apparatus further includes:
the second live video resource determining module is used for determining live video resources browsed by a user logged in the target client after the last recommendation when the preset moment is reached, and taking the live video resources as the live video resources;
a third live video resource determining module, configured to determine a live video resource associated with the second live video resource as a third live video resource;
the ordering module is used for ordering the third video live broadcast resources based on the similarity between the resource characteristics of the third video live broadcast resources and the user characteristics of the target users to obtain an ordering result;
and the video live broadcast resource updating module is used for updating the video live broadcast resources corresponding to the target client in the corresponding relation according to the sorting result.
Optionally, the device is applied to a resource recommendation server in a resource recommendation system, and the resource recommendation system further comprises a resource sequencing server;
the sorting module is further configured to send the resource characteristics of the third live video resource to the resource sorting server, so that the resource sorting server sorts the third live video resource based on the similarity between the resource characteristics of the third live video resource and the user characteristics of the target user, to obtain a sorting result, and to send the sorting result to the resource recommendation server.
Optionally, the sorting module is specifically configured to input, for each third live video resource, a resource feature of the third live video resource and a user feature of the target user into a pre-trained browsing prediction network model, so as to obtain a probability that the target user browses the third live video resource, where the probability is used as a similarity between the resource feature of the third live video resource and the user feature of the target user; the browsing prediction network model is obtained by training based on a preset training sample; the preset training sample comprises user characteristics of a sample user and resource characteristics of video live broadcast resources browsed by the sample user; and sequencing the third live video resources according to the sequence of the corresponding similarity from large to small to obtain a sequencing result.
Optionally, the apparatus further includes:
and the browse prediction network model updating module is used for adjusting the model parameters of the browse prediction network model by taking the resource characteristics of the second video live broadcast resources and the user characteristics of the target user as training data aiming at each second video live broadcast resource so as to update the browse prediction network model.
In a fifth aspect of the present invention, there is also provided a live video resource recommendation device, where the device is applied to a resource ordering server in a resource recommendation system, the resource recommendation system further includes a resource recommendation server, and the device includes:
the resource feature receiving module is used for receiving the resource feature of the third live video resource sent by the resource recommending server; wherein the third live video asset is associated with a second live video asset; the second live video resources are live video resources which are determined by the resource recommendation server at a preset moment and are browsed by a target user logged in a target client after the last recommendation;
the ordering module is used for ordering the third video live broadcast resources based on the similarity between the resource characteristics of the third video live broadcast resources and the user characteristics of the target users to obtain an ordering result;
And the sequencing result sending module is used for sending the sequencing result to the resource recommending server so that the resource recommending server records the corresponding relation between the target client and the sequencing result, and recommending the video live broadcast resource to the target client based on the corresponding relation when receiving the video live broadcast resource request sent by the target client.
Optionally, the ranking module is specifically configured to input, for each third live video resource, a resource feature of the third live video resource and a user feature of the target user into a pre-trained browsing prediction network model, so as to obtain a probability that the target user browses the third live video resource, where the probability is used as a similarity between the resource feature of the third live video resource and the user feature of the target user; the browsing prediction network model is obtained by training based on a preset training sample; the preset training sample comprises user characteristics of a sample user and resource characteristics of video live broadcast resources browsed by the sample user; and sequencing the third live video resources according to the sequence of the corresponding similarity from large to small to obtain a sequencing result.
In yet another aspect of the present invention, there is also provided an electronic device including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory perform communication with each other through the communication bus;
a memory for storing a computer program;
and the processor is used for realizing any one of the live video resource recommendation methods when executing the programs stored in the memory.
In still another aspect of the present invention, there is further provided a computer readable storage medium, where a computer program is stored, where the computer program when executed by a processor implements any one of the live video resource recommendation methods described above.
In yet another aspect of the implementation of the present invention, there is also provided a computer program product containing instructions that, when run on a computer, cause the computer to perform any of the live video resource recommendation methods described above.
By adopting the video live broadcast resource recommendation method provided by the embodiment of the invention, a video live broadcast resource request sent by a target client is received; inquiring each video live broadcast resource corresponding to the target client in the corresponding relation between the pre-recorded client and the video live broadcast resource to be used as a first video live broadcast resource; wherein, the corresponding relation is: based on the user characteristics of the user logging in the client, the similarity between the user characteristics and the resource characteristics of the live video resources is determined; recommending the video live broadcast resources to the target client based on the arrangement sequence of the first video live broadcast resources in the corresponding relation.
The video live broadcast resource recommended to the target client is determined based on the similarity between the user characteristics of the user logging in the client and the resource characteristics of the video live broadcast resource, so that the method provided by the embodiment of the application can realize personalized recommendation for the user. In addition, the method of the embodiment of the invention can pre-determine the video live broadcast resources corresponding to the client, correspondingly, when receiving the video live broadcast resource request, directly recommend the video live broadcast resources to the client, and can reduce the time delay of recommendation.
Of course, not all of the above-described advantages need be achieved simultaneously in practicing any one of the products or methods of the present application.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In the related art, aiming at a live scene, a current hot video live broadcast resource can be recommended to a user. However, recommendation is performed to the user only based on the hot video live broadcast resources, personalized requirements of the user cannot be met, and user experience is poor.
In order to solve the above problems, the present invention provides a live video resource recommendation method, which can be applied to a server (hereinafter referred to as a target server).
Referring to fig. 1, fig. 1 is a flowchart of a live video resource recommendation method provided by an embodiment of the present invention, including:
s101: and receiving a video live broadcast resource request sent by the target client.
S102: and querying each video live broadcast resource corresponding to the target client in the corresponding relation between the pre-recorded client and the video live broadcast resource as a first video live broadcast resource.
Wherein, the corresponding relation is: and determining the similarity between the user characteristics of the user logged in the client and the resource characteristics of the live video resource.
S103: recommending the video live broadcast resources to the target client based on the arrangement sequence of the first video live broadcast resources in the corresponding relation.
The video live broadcast resource recommended to the target client is determined based on the similarity between the user characteristics of the user logging in the client and the resource characteristics of the video live broadcast resource, so that the method provided by the embodiment of the application can realize personalized recommendation for the user. In addition, the method of the embodiment of the invention can pre-determine the video live broadcast resources corresponding to the client, correspondingly, when receiving the video live broadcast resource request, directly recommend the video live broadcast resources to the client, and can reduce the time delay of recommendation.
The live video resources mentioned in the embodiments of the present invention correspond to a host, that is, one live video resource may represent one host.
In step S101, the target client may be an application installed in an intelligent device such as a mobile phone or a computer, for browsing live video resources.
When the target user is detected to be logged in, the target client sends a live video resource request to the target server; or when the target user refreshes the live video recommended page of the target client, the target client sends a live video resource request to the recommended server; or when the target user is watching live broadcast, the target client side sends a video live broadcast resource request to the recommendation server at regular time.
In step S102, the target server may determine, in advance, a correspondence between the client and the live video resource, where the correspondence is in multiple forms. For example, the live video resource recommendation system allocates an independent storage space for each client in the database, and records live video resources corresponding to the client in the corresponding relation in the storage space; or the recommendation system allocates a shared storage space in the database, and stores the corresponding relation between each client and the live video resource in the shared storage space.
In addition, the target server may also set a storage period for the above correspondence, which may be empirically set, such as setting the storage period to 6 hours, 24 hours, or the like. When the offline time of the target user from the target client reaches the storage time, the corresponding relation between the target user and the corresponding video live broadcast resource can be deleted.
In one implementation manner, the corresponding relationship may be a corresponding relationship between an identifier of a client and an identifier of a live video resource, the identifier of the live video resource may be a live video room address of a corresponding anchor, and the identifier of the client may be an account number of a user logging in the client.
The user characteristics of the user may include: personal information of the user and/or browsing behavior information. Wherein the personal information may include at least one of: age, gender, physical location of the user, type of video live asset of interest. The browsing behavior information may include at least one of: the time length of the user watching the recommended video live resources and comment content of the user aiming at the watched video live resources.
The resource characteristics of the live video resource may include a corresponding anchor characteristic, an offline characteristic, and a real-time characteristic of the anchor. Wherein the anchor feature may include at least one of: age, sex, type of anchor. The offline feature may include at least one of: behavior of the anchor history live broadcast and index parameters of the anchor history live broadcast. The act of hosting the historical live broadcast may include at least one of: live singing, live dancing and live game. The live room history index parameters of the anchor may include at least one of: average duration of the historical live broadcast, average online number of the historical live broadcast, and average praise of the historical live broadcast. The real-time characteristics may include at least one of: the current state of the anchor and the current index parameters of the live broadcasting room of the anchor. The current status of the anchor may include at least one of: singing, dancing, playing. The current index parameters of the live room of the anchor may include at least one of: the current live time length, the current number of online people and the current praise number.
In step S103, in one manner, the target server may directly recommend the first live video resource to the target client.
In another manner, the target server may further filter the first live video resource, for example, remove live video resources that have been previously recommended to the target client, and/or live video resources corresponding to the anchor that is not currently online, and recommend the filtered live video resources to the target client.
In one embodiment, referring to fig. 2, after S101, the method may further include the steps of:
s104: if the target client does not exist in the corresponding relation, recommending the video live broadcast resource to the target client based on the preset video live broadcast resource.
When a target user logs in a target client for the first time to watch live broadcast, the target server does not record the corresponding relation between the target client and the video live broadcast resource yet. Or when the offline time of the target user exceeds the storage duration of the corresponding relationship, the target server deletes the corresponding relationship between the target client and the live video resource, so that the corresponding relationship does not have the target client. At this time, the video live broadcast resource cannot be recommended to the target client based on the correspondence.
The preset video live broadcast resource can be the video live broadcast resource with the largest number of current online people, or can be the video live broadcast resource with the largest number of current praise, or can be the video live broadcast resource with the largest number of current received gifts.
In addition, the target server can update the video live broadcast resources corresponding to the target client in the corresponding relation based on the browsing behavior of the user.
In one embodiment, referring to fig. 3, after S103, the method may further include the steps of:
s105: and when the preset time is reached, determining the video live broadcast resource browsed by the user of the login target client after the last recommendation, and taking the video live broadcast resource as a second video live broadcast resource.
S106: and determining the video live resources associated with the second video live resources as third video live resources.
S107: and sequencing the third video live broadcast resources based on the similarity between the resource characteristics of the third video live broadcast resources and the user characteristics of the target user to obtain a sequencing result.
S108: and updating the video live broadcast resources corresponding to the target clients in the corresponding relation according to the sequencing result.
In the embodiment of the invention, the target server can update the corresponding relationship according to a preset period. That is, when the time corresponding to the preset period is reached (i.e., when the preset time is reached), the target server may determine browsing behavior data of the target user after the last recommendation, and determine the third live video resource based on the browsing behavior, so as to update the live video resource corresponding to the target client according to the third live video resource.
It may be appreciated that if the video live broadcast resource is recommended to the target client based on the arrangement order of the first video live broadcast resource in the correspondence, the determining the second video live broadcast resource includes: and in the first live video resources, logging in live video resources browsed by a target user of the target client. If the video live broadcast resource is recommended to the target client based on the preset video live broadcast resource last time, determining a second video live broadcast resource comprises: and in the preset video live broadcast resources, logging in the video live broadcast resources browsed by the target user of the target client.
In step S105, in one implementation, after the target server recommends the live video resource to the target client, the target client may obtain browsing behavior data of the target user for the recommended live video resource. For example, the browsing behavior data may include: and (5) identifying the video live broadcast resources which the target user has browsed.
Then, the target client may add the browsing behavior data to a preset message queue, and correspondingly, the target server may acquire the browsing behavior data from the preset message queue to determine the third live video resource. The preset message queue may be a kafka message queue.
In step S106, the target server may determine, from the other live video resources, a live video resource having at least one resource feature identical to that of the second live video resource, as a third live video resource. Or, the target server may also determine, as the third live video resource, live video resources corresponding to other live video resources of interest to the anchor corresponding to the second live video resource. Or, the target server may also determine, as the third live video resource, live video resources corresponding to other anchors that have focused on the anchor corresponding to the second live video resource.
In step S107, the target server may calculate a similarity between each third live video resource and the target user, and rank the third live video resources according to the similarity.
In one implementation manner, for each third live video resource, a feature vector corresponding to a resource feature of the third live video resource and a feature vector corresponding to a user feature of a target user may be determined, and then, a similarity between the two feature vectors may be calculated as a similarity between the target user and the third live video resource.
In another implementation, the similarity between the target user and each third live video resource may also be calculated based on a pre-trained network model.
In step S108, the target server may determine the video live resources corresponding to the target client in the corresponding relationship, and update the corresponding relationship as the third video live resources after sequencing. Or, the video live broadcast resources corresponding to the target client in the corresponding relation can be determined, and the video live broadcast resources are the preset number of video live broadcast resources in the third video live broadcast resources after sequencing, so that the corresponding relation is updated.
In one embodiment, the method may be applied to a resource recommendation server (i.e., the target server described above) in a resource recommendation system, which may further include a resource ranking server.
Thus, the target server may send the resource characteristics of the third live video resource to the resource ordering server, and the resource ordering server may order the third live video resource.
The method for ordering the third live video resources by the resource ordering server may refer to the method for ordering the third live video resources by the target server in step S107.
In one implementation, the resource recommendation server may be a server corresponding to a near-line layer of the recommendation system, for example, the resource recommendation server may be a cloud server deployed with a container in which the near-line layer is deployed. In addition, the resource ordering server may be a physical server.
Based on the architecture, the third live video resources are ordered through the physical server, so that the service load of the near-line layer can be reduced, the speed of the near-line layer responding to the live video resource request of the user is improved, and the time delay is reduced.
In addition, the near-line layer is deployed in a container of the cloud end, so that flexible deployment of the near-line layer can be realized. When a new function needs to be added to the near line layer, the new function can be rapidly on line.
In one implementation, the resource ranking server may register for services in the zookeeper, become a provider of ranking services, and write its URL (Uniform Resource Locator ) address, so that the resource recommendation server may subscribe to the ranking services in the zookeeper. The resource recommendation server sends a micro-service request containing the identification of the third live video resource to the resource ordering server through the URL address of the resource ordering server. Further, the resource ordering server may order the third live video resource. Correspondingly, the resource recommending server can subscribe the service registered in the zookeeper by the resource sequencing server in the zookeeper, become a consumer of the sequencing service, write in the URL address of the consumer, so that after the resource sequencing server obtains the sequencing result of the third live video resource, the sequencing result of the third live video resource is returned to the resource recommending server through the URL address of the resource recommending server.
In one embodiment, referring to fig. 4, on the basis of fig. 3, step S107 may include:
s1071: and inputting the resource characteristics of the third video live broadcast resources and the user characteristics of the target user into a pre-trained browsing prediction network model aiming at each third video live broadcast resource to obtain the probability of browsing the third video live broadcast resources by the target user, wherein the probability is used as the similarity between the resource characteristics of the third video live broadcast resources and the user characteristics of the target user.
The browsing prediction network model is obtained by training based on a preset training sample. The preset training sample comprises user characteristics of a sample user and resource characteristics of video live broadcast resources browsed by the sample user.
S1072: and sequencing the third live video resources according to the sequence of the corresponding similarity from large to small to obtain a sequencing result.
Wherein the browse predictive network model may be set by a technician based on experience and business requirements. For example, the browse prediction network model may be an LR (Logistic Regression ) model, a GBDT (Gradient Boost Decision Tree, iterative decision tree) model, a GBDT-FM (Gradient Boost Decision Tree-Factorization Machines, iterative decision tree-factorization) model, or a Deep (Deep, depth ordering) model.
In one embodiment, referring to fig. 5, after step S105, the method may further include the steps of:
s109: and aiming at each second video live broadcast resource, taking the resource characteristics of the second video live broadcast resource and the user characteristics of the target user as training data, and adjusting the model parameters of the browsing prediction network model to update the browsing prediction network model.
In the embodiment of the invention, the target server can update the browsing prediction network model so as to improve the accuracy of the browsing prediction network model.
In one implementation manner, the target server may input the resource characteristics of the second live video resource and the user characteristics of the target user into a browsing prediction network model to obtain a prediction probability of the target user for the second live video resource, and calculate a loss function value between the prediction probability and a tag of the target user for the second live video resource. Further, model parameters of the browsing prediction network model are adjusted based on the loss function value. The tag of the target user for the second live video asset may indicate whether the target user browses the second live video asset.
In another mode, the target server combines the resource features of the second live video resource two by two to obtain the high-order combined feature of the second live video resource, and/or combines the user features of the target user two by two to obtain the high-order combined feature of the target user.
The target server takes the resource characteristics of the second video live broadcast resource, the high-order combination characteristics of the second video live broadcast resource, the user characteristics of the target user and the high-order combination characteristics of the target user as training data, inputs a browse prediction network model, obtains the prediction probability of the target user for the second video live broadcast resource, and calculates a loss function value between the prediction probability and the label of the target user for the second video live broadcast resource. Further, model parameters of the browsing prediction network model are adjusted based on the loss function value. The tag of the target user for the second live video asset may indicate whether the target user browses the second live video asset.
It can be appreciated that the above process of updating the browsing prediction network model, that is, the process of training the browsing prediction network model again, is based on the resource characteristics of the second live video resource and the user characteristics of the target user.
Therefore, the process of training the browsing prediction network model based on the preset training sample can refer to the process of updating the browsing prediction network model based on the resource characteristics of the second live video resource and the user characteristics of the target user.
Based on the same inventive concept, the embodiment of the invention provides a live video resource recommendation method, which can also be applied to a resource ordering server in a resource recommendation system, wherein the resource recommendation system also comprises a resource recommendation server. Referring to fig. 6, the method may include the steps of:
s601: and the receiving resource recommendation server transmits the resource characteristics of the third live video resource.
Wherein the third live video asset is associated with the second live video asset; the second live video resources are live video resources which are determined by the resource recommendation server at a preset moment and are browsed by a target user logged in the target client after the last recommendation.
S602: and sequencing the third video live broadcast resources based on the similarity between the resource characteristics of the third video live broadcast resources and the user characteristics of the target user to obtain a sequencing result.
S603: and sending the sequencing result to the resource recommendation server so that the resource recommendation server updates the corresponding relation, and recommending the video live broadcast resource to the target client based on the updated corresponding relation when receiving the video live broadcast resource request sent by the target client.
The video live broadcast resources recommended to the target client by the resource sequencing server are determined based on the similarity between the user characteristics of the user logging in the client and the resource characteristics of the video live broadcast resources, so that the method provided by the embodiment of the application can provide personalized recommendation for the user. In addition, the method of the embodiment of the invention can pre-determine the video live broadcast resources corresponding to the client, correspondingly, when receiving the video live broadcast resource request, directly recommend the video live broadcast resources to the client, and can reduce the time delay of recommendation.
S602 may refer to the process of ordering the third live video resource by the target server in step S107.
In one embodiment, referring to fig. 7, on the basis of fig. 6, step S602 may include:
s6021: and inputting the resource characteristics of the third video live broadcast resources and the user characteristics of the target user into a pre-trained browsing prediction network model aiming at each third video live broadcast resource to obtain the probability of browsing the third video live broadcast resources by the target user, wherein the probability is used as the similarity between the resource characteristics of the third video live broadcast resources and the user characteristics of the target user.
The browsing prediction network model is obtained by training based on a preset training sample; the preset training sample comprises user characteristics of a sample user and resource characteristics of video live broadcast resources browsed by the sample user.
S6022: and sequencing the third live video resources according to the sequence of the corresponding similarity from large to small to obtain a sequencing result.
Steps S6021-S6022 may be referred to the description of steps S1071-S1072 above.
Based on the same inventive concept, the embodiment of the present invention further provides a live video resource recommendation system, referring to fig. 8, fig. 8 is a structural diagram of the live video resource recommendation system provided in the embodiment of the present application, including:
and the client 801 is used for sending the live video resource request to the resource recommendation server.
The resource recommendation server 802 is configured to receive a live video resource request sent by a client, and query each live video resource corresponding to a target client as a first live video resource in a pre-recorded corresponding relationship between the client and the live video resource.
Wherein, the corresponding relation is: based on the user characteristics of the user logging in the client, the similarity between the user characteristics and the resource characteristics of the live video resources is determined; recommending the video live broadcast resources to the target client based on the arrangement sequence of the first video live broadcast resources in the corresponding relation.
In one embodiment, the resource recommendation server 802 is further configured to recommend the live video resource to the target client based on the preset live video resource if the target client does not exist in the correspondence.
In one embodiment, the resource recommendation server 802 is further configured to determine, after pushing the live video resource to the target client based on the arrangement order of the live video resource in the correspondence, a resource feature of a third live video resource associated with the second live video resource that has been browsed by the target user logged in the target client after the last recommendation; based on the similarity between the resource characteristics of the third video live broadcast resource and the user characteristics of the target user, sequencing the third video live broadcast resource to obtain a sequencing result; and updating the video live broadcast resources corresponding to the target clients in the corresponding relation according to the sequencing result.
In one embodiment, the resource recommendation server 802 is further configured to send the resource characteristics of the third live video resource to the resource ordering server.
And the resource ordering server 803 is configured to order the third live video resource based on the similarity between the resource feature of the third live video resource and the user feature of the target user, obtain an ordering result, and send the ordering result to the resource recommendation server.
In one embodiment, the resource ordering server 803 is specifically configured to input, for each third live video resource, a resource feature of the third live video resource and a user feature of a target user into a pre-trained browsing prediction network model, so as to obtain a probability that the target user browses the third live video resource, where the probability is used as a similarity between the resource feature of the third live video resource and the user feature of the target user; the browsing prediction network model is obtained by training based on a preset training sample; the preset training samples comprise user characteristics of sample users and resource characteristics of video live broadcast resources browsed by the sample users; and sequencing the third live video resources according to the sequence of the corresponding similarity from large to small to obtain a sequencing result.
In one embodiment, the resource obtaining server 804 is configured to determine, according to browsing behavior data of a target user sent by a target client, a second live video resource that has been browsed by the user of the target client in the first live video resource; determining a live video asset associated with the second live video asset as a third live video asset; and storing the resource characteristics of the third video live broadcast resource into a first preset storage space.
The resource recommendation server 802 is further configured to obtain a resource characteristic of the third live video resource from the first preset storage space.
In the embodiment of the present invention, the resource obtaining server 804 receives the browsing behavior data sent by the target client, analyzes the browsing behavior data, determines the second live video resource browsed by the user of the target client after the last recommendation, determines the live video resource associated with the second live video resource through the resource characteristics of the second live video resource, and stores the resource characteristics of the third live video resource into the first preset storage space.
In one implementation, the target client may add the browsing behavior data to a preset message queue, and correspondingly, the target server may obtain the browsing behavior data from the preset message queue to determine the third live video resource. The preset message queue may be a kafka message queue.
In one embodiment, the resource ordering server 803 is further configured to obtain a browsing prediction network model from the second preset storage space.
The resource obtaining server 804 is further configured to update, for each second live video resource, the model parameters of the browsing prediction network model in the second preset storage space with the resource characteristics of the second live video resource and the user characteristics of the target user as training data.
In an embodiment of the present invention, the resource obtaining server 804 is configured to update the browse prediction network model.
In one implementation, the target client may send browsing behavior data to the resource acquisition server 804, which may also include user characteristics of the user.
The resource obtaining server 804 may update the browsing prediction network model stored in the second preset storage space based on the identifier of the live video resource in the browsing behavior data and the user characteristics of the user, so that the resource ordering server 803 may obtain the updated browsing prediction network model from the second preset storage space.
Referring to fig. 9, fig. 9 is a schematic diagram of live video resource recommendation according to an embodiment of the present invention. In fig. 9, the APP backend corresponds to the target client, the cloud container deploys a near-line layer, and a server corresponding to the near-line layer representing the recommendation system is deployed in a container of the cloud server, where the server corresponding to the near-line layer corresponds to the resource recommendation server. The physical machine deploys a predictor (predictive model) ranking module, which indicates that the resource ranking module is deployed on the physical machine, and the ranking module corresponds to the resource ranking server.
Couchbase personalized cache index, representing the storage index of the personalized cache stored in Couchbase. The personalized cache corresponds to the video live broadcast resources in the corresponding relation. And the storage index of the personalized cache is used for searching the video live broadcast resources in the corresponding relation in the Couchbase. And the non-personalized cache index represents a storage index of a non-personalized cache stored in the Couchbase, the non-personalized cache corresponds to the pre-stored video live broadcast resource, and the storage index of the non-personalized cache is used for searching the pre-set video live broadcast resource in the Couchbase.
The APP back end sends a query request, namely, a live video resource request, to the near line layer. After the near line layer receives the resource request, the index is queried. That is, the corresponding relation is queried, the live video resources are preset, and the query result, that is, the queried live video resources are returned.
The APP back end can acquire user behavior data, and the user behavior data can comprise live video resources browsed by a target user logging in the target client.
And the APP back end generates a user behavior pingback message according to the user behavior data and adds the user behavior pingback message to a Kafka message queue. The near line layer may construct a personalized cache index according to the Kafka message queue.
The algorithm engineering babel (translator) task is completed by the resource acquisition server. The algorithm engineering is applied to the resource acquisition server, and can determine the second live video resource browsed by the target user according to the user behavior ping back message in the Kafka message queue.
The Hive table is used for recording all currently existing live video resources, and all live video resources can comprise the identification of all live video resources and the anchor features of corresponding anchor of all live video resources. The Bl live table (Boolean honeycomb table) is used for recording each currently existing video live broadcast resource, and corresponds to the offline characteristic and the real-time characteristic of the anchor.
And searching and determining a main broadcasting recall source inverted row in the Hive table according to the main broadcasting characteristics of the main broadcasting corresponding to the second video live broadcasting resource through algorithm engineering, wherein the main broadcasting recall source inverted row corresponds to the third video live broadcasting resource. The algorithm engineering uses a babel tool, and can translate the inverted format of the anchor recall source into a key-value (key value pair) format to be stored in the Couchbase, wherein the key can be one anchor feature in anchor features of the anchor corresponding to the second live video resource, and the value can be the associated live video resource queried by utilizing the resource feature.
The algorithm engineering searches and determines the offline characteristics of each anchor in the anchor recall source reverse row in the Bl live table. And the algorithm engineering uses a babel tool to translate the format of the front row of the searched offline features into a key-value format and store the key-value format into the Couchbase, wherein the key can be a host in the inverted row of the anchor recall source, and the value can be the offline feature corresponding to the host. The offline features are arranged by adopting forward index, namely, the offline features are arranged according to each anchor. The algorithm engineering uses a Flink tool to output Batch in a Bl Hive table, determines the real-time characteristics of each anchor in the anchor recall source reverse row, and stores the real-time characteristics in the Couchbase.
The algorithm engineering searches the anchor features of the anchor corresponding to the second video live broadcast resource in the live table, searches the offline features and the real-time features of the anchor corresponding to the second video live broadcast resource in the live table, and uses the anchor features, the offline features and the real-time features of the anchor corresponding to the second video live broadcast resource as the resource features of the second video live broadcast resource. The algorithm engineering searches user characteristics of a target user logging in the APP in the Hive table, and translates the queried resource characteristics and user characteristics into key-value by using a table tool. The key-value of the resource feature can be one video live resource in the second video live resource, and the value can be the corresponding resource feature of the video live resource. The key-value of the user feature can be a client identifier, and the value can be the user feature of the user corresponding to the client. The algorithm engineering can also update the model file stored in Couchbase by using the translated resource characteristics and the corresponding user characteristics as sample data, namely updating the browsing prediction network model. The model files may include model files of GBDT, GBDT-FM models.
Micro api call: GBDT/GBDT-FM, representing the anchor features, offline features and real-time features of anchor recall sources loaded from the database by the near-line layer, encapsulated into a micro API call (micro service request) and sent to the ranking module. The ordering module may periodically (e.g., every 5 minutes) scan the Couchbase to obtain the model file generated by the algorithm engineering described above, and load it into the cache. When the sorting module receives a micro-service request from a near-line layer, processing target user characteristics, anchor characteristics, offline characteristics and real-time characteristic input of an anchor recall source in the micro-service request based on a browsing prediction network model represented by the model file so as to sort the anchor recall source to obtain a sorting result, and returning the sorting result to the near-line layer. After receiving the ordering result, the near line layer deletes the personalized cache index of the target user in the database, takes the ordering result as a new personalized cache construction index and stores the new personalized cache construction index into the database.
The near-line layer can also obtain live video resources of the popular anchor from MySQL (relational database management system), and store the live video resources of the popular anchor into the Couchbase database as non-personalized cache.
The video live broadcast resource recommendation method provided by the embodiment of the invention is used for performing the pressure test, wherein in the test process, the number of video live broadcast resources is 3000, 6-path recalls are adopted, namely 6-class resource characteristics are determined according to the second video live broadcast resources, and video live broadcast resources containing at least one class of resource characteristics in the 6-class resource characteristics are determined to be used as third video live broadcast resources. In addition, the number of users tested was 200 ten thousand, and QPS (Query Per Second) was: 5000 to 8000, and the test results shown in Table (1) were obtained.
Watch (1)
| User ratio | Recommended time delay |
| 50％ | 35ms |
| 90％ | 37ms |
| 95％ | 41ms |
| 99％ | 49ms |
In table (1), in all users, the recommended time delay of 50% of the users is not more than 35 ms, the recommended time delay of 90% of the users is not more than 37 ms, the recommended time delay of 95% of the users is not more than 41 ms, and the recommended time delay of 99% of the users is not more than 49 ms, so that the recommended time delay is maintained at a small value.
Based on the same inventive concept, the embodiment of the present invention further provides a live video resource recommendation device, referring to fig. 10, and fig. 10 is a structural diagram of the live video resource recommendation device provided in the embodiment of the present application, including:
An information receiving module 1001, configured to receive a live video resource request sent by a target client;
the first live video resource query module 1002 is configured to query, in a corresponding relationship between a pre-recorded client and live video resources, each live video resource corresponding to a target client as a first live video resource; wherein, the corresponding relation is: based on the user characteristics of the user logging in the client, the similarity between the user characteristics and the resource characteristics of the live video resources is determined;
and the live video resource recommending module 1003 is configured to recommend live video resources to the target client based on the arrangement order of the first live video resources in the correspondence.
In one embodiment, the live video resource recommendation module 1003 is further configured to recommend live video resources to the target client based on the preset live video resources if the target client does not exist in the correspondence.
And the second live video resource determining module is used for determining live video resources browsed by a user logging in the target client side in the first live video resources as the second live video resources when the preset moment is reached.
And the third live video resource determining module is used for determining the live video resource associated with the second live video resource as the third live video resource.
And the ordering module is used for ordering the third video live broadcast resources based on the similarity between the resource characteristics of the third video live broadcast resources and the user characteristics of the target user to obtain an ordering result.
And the video live broadcast resource updating module is used for updating the video live broadcast resources corresponding to the target clients in the corresponding relation according to the sequencing result.
In one embodiment, the ranking module is further configured to send the resource characteristics of the third live video resource to the resource ranking server, so that the resource ranking server ranks the third live video resource based on the similarity between the user characteristics of the target user and the resource characteristics of the third live video resource, to obtain a ranking result, and to send the ranking result to the resource recommendation server.
In one embodiment, the ranking module is specifically configured to input, for each third live video resource, a resource feature of the third live video resource and a user feature of a target user into a pre-trained browsing prediction network model, to obtain a probability that the target user browses the third live video resource, where the probability is used as a similarity between the resource feature of the third live video resource and the user feature of the target user; the browsing prediction network model is obtained by training based on a preset training sample; the preset training samples comprise user characteristics of sample users and resource characteristics of video live broadcast resources browsed by the sample users; and sequencing the third live video resources according to the sequence of the corresponding similarity from large to small to obtain a sequencing result.
And the browse prediction network model updating module is used for aiming at each second video live broadcast resource, taking the resource characteristics of the second video live broadcast resource and the user characteristics of the target user as training data, and adjusting the model parameters of the browse prediction network model so as to update the browse prediction network model.
Based on the same inventive concept, the embodiment of the invention also provides a video live broadcast resource recommendation device. Referring to fig. 11, fig. 11 is a block diagram of a live video resource recommendation device provided in an embodiment of the present application, including:
the resource feature receiving module 1101 is configured to receive a resource feature of the third live video resource sent by the resource recommendation server; wherein the third live video asset is associated with the second live video asset; the second live video resources are live video resources which are determined at a preset moment by a resource recommendation server and are browsed by a target user of a target client in first live video resources corresponding to the target client in a pre-recorded corresponding relation;
the ordering module 1102 is configured to order the third live video resource based on a similarity between the resource feature of the third live video resource and the user feature of the target user, so as to obtain an ordering result;
The sequencing result sending module 1103 is configured to send a sequencing result to the resource recommendation server, so that the resource recommendation server records a corresponding relationship between the target client and the sequencing result, and when receiving a live video resource request sent by the target client, recommends live video resources to the target client based on the corresponding relationship. The ranking module 1102 is specifically configured to input, for each third live video resource, a resource feature of the third live video resource and a user feature of a target user into a pre-trained browsing prediction network model, so as to obtain a probability that the target user browses the third live video resource, where the probability is used as a similarity between the resource feature of the third live video resource and the user feature of the target user; the browsing prediction network model is obtained by training based on a preset training sample; the preset training samples comprise user characteristics of sample users and resource characteristics of video live broadcast resources browsed by the sample users;
and sequencing the third live video resources according to the sequence of the corresponding similarity from large to small to obtain a sequencing result.
The embodiment of the invention also provides an electronic device, as shown in fig. 12, which comprises a processor 1201, a communication interface 1202, a memory 1203 and a communication bus 1204, wherein the processor 1201, the communication interface 1202 and the memory 1203 complete the communication with each other through the communication bus 1204,
A memory 1203 for storing a computer program;
the processor 1201, when executing the program stored in the memory 1203, performs the following steps:
receiving a video live broadcast resource request sent by a target client;
inquiring each video live broadcast resource corresponding to the target client in the corresponding relation between the pre-recorded client and the video live broadcast resource to be used as a first video live broadcast resource; wherein, the correspondence is: based on the user characteristics of the user logging in the client, the similarity between the user characteristics and the resource characteristics of the live video resources is determined;
recommending the video live broadcast resource to the target client based on the arrangement sequence of the first video live broadcast resource in the corresponding relation.
The embodiment of the present invention further provides an electronic device, as shown in fig. 13, including a processor 1301, a communication interface 1302, a memory 1303 and a communication bus 1304, where the processor 1301, the communication interface 1302, and the memory 1303 complete communication with each other through the communication bus 1304,
a memory 1303 for storing a computer program;
processor 1301, when executing the program stored in memory 1303, implements the following steps:
Receiving resource characteristics of a third live video resource sent by the resource recommendation server; wherein the third live video asset is associated with a second live video asset; the second live video resource is a live video resource which is determined by the resource recommendation server at a preset moment, and after the last recommendation, a target user of the target client logs in the live video resource which is browsed by the target user;
based on the similarity between the resource characteristics of the third video live broadcast resource and the user characteristics of the target user, sequencing the third video live broadcast resource to obtain a sequencing result;
and sending the sequencing result to the resource recommendation server so that the resource recommendation server records the corresponding relation between the target client and the sequencing result, and recommending the video live broadcast resource to the target client based on the corresponding relation when receiving the video live broadcast resource request sent by the target client.
The communication bus mentioned by the above electronic device may be a peripheral component interconnect standard (Peripheral Component Interconnect, abbreviated as PCI) bus or an extended industry standard architecture (Extended Industry Standard Architecture, abbreviated as EISA) bus, or the like. The communication bus may be classified as an address bus, a data bus, a control bus, or the like. For ease of illustration, the figures are shown with only one bold line, but not with only one bus or one type of bus.
The communication interface is used for communication between the electronic device and other devices.
The memory may include random access memory (Random Access Memory, RAM) or non-volatile memory (non-volatile memory), such as at least one disk memory. Optionally, the memory may also be at least one memory device located remotely from the aforementioned processor.
The processor may be a general-purpose processor, including a central processing unit (Central Processing Unit, CPU for short), a network processor (Network Processor, NP for short), etc.; but also digital signal processors (Digital Signal Processor, DSP for short), application specific integrated circuits (Application Specific Integrated Circuit, ASIC for short), field-programmable gate arrays (Field-Programmable Gate Array, FPGA for short) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.
In yet another embodiment of the present invention, a computer readable storage medium is provided, where a computer program is stored, where the computer program is executed by a processor to implement the live video resource recommendation method according to any one of the foregoing embodiments.
In yet another embodiment of the present invention, a computer program product containing instructions that, when executed on a computer, cause the computer to perform the live video asset recommendation method of any of the above embodiments is also provided.
In the above embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loaded and executed on a computer, produces a flow or function in accordance with embodiments of the present invention, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a computer network, or other programmable apparatus. The computer instructions may be stored in or transmitted from one computer-readable storage medium to another, for example, by wired (e.g., coaxial cable, optical fiber, digital Subscriber Line (DSL)), or wireless (e.g., infrared, wireless, microwave, etc.). The computer readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that contains an integration of one or more available media. The usable medium may be a magnetic medium (e.g., floppy Disk, hard Disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid State Disk (SSD)), etc.
It is noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
In this specification, each embodiment is described in a related manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments. In particular, for an apparatus, a system, an electronic device, a computer readable storage medium, and a computer program product, the description is relatively simple, as it is substantially similar to the method embodiments, and relevant places are referred to in the section of the description of the method embodiments.
The foregoing description is only of the preferred embodiments of the present invention and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.