AI-driven live content intelligent recommendation method, device and computer equipment

Abstract

The embodiment of the present invention provides an AI-driven intelligent recommendation method, device and computer equipment for live content, which belongs to the field of data recommendation technology. The method includes: calculating the target interaction value of the target user and the associated user according to the target user behavior of the target user and the associated user behavior of the associated user; obtaining similar users of the target user from the associated users according to the target interaction value, and obtaining similar user behaviors of similar users from the associated user behavior; using a multi-target classification model to classify different features of the live content that the target user and similar users pay attention to according to the target user behavior and similar user behavior to obtain the neighboring users of the target user; recommending live content to the target user according to the neighboring user behavior of the neighboring users to obtain an initial recommendation result; using an emotion recognition model to identify the emotion of the target user according to the target user behavior to obtain the target emotion; adjusting the initial recommendation result according to the target emotion to obtain the target recommendation result.

Description

Technical Field

[0001] The present invention relates to the field of data recommendation technology, and in particular to an AI-driven live content intelligent recommendation method, device and computer equipment. Background Art

[0002] With the rapid development and popularization of Internet technology, the live broadcast industry has shown explosive growth in recent years. Live broadcast platforms have not only become a stage for content creators to showcase their talents, but also an important channel for users to obtain information and entertainment. In this context, the application of intelligent recommendation systems in the live broadcast industry is becoming more and more extensive, becoming one of the key technologies to improve user experience and increase user stickiness. However, the intelligent recommendation technology currently adopted by most live broadcast platforms mainly relies on the similarity between users' historical viewing content for recommendation. Although this similarity-based recommendation algorithm can predict users' interests to a certain extent, this method lacks the mining of users' deep-seated needs, and its recommendation results are often difficult to truly mine users' deep-seated needs, resulting in a certain deviation between the recommendation results and users' actual needs. Summary of the invention

[0003] The main purpose of the embodiments of the present invention is to provide an AI-driven live content intelligent recommendation method, device and computer equipment, aiming to solve the problem that the related technology lacks the exploration of users' deep-seated needs when recommending content, and its recommendation results are often difficult to truly explore the users' deep-seated needs, thus resulting in a certain deviation between the recommendation results and the actual needs of the users.

[0004] In a first aspect, an embodiment of the present invention provides an AI-driven live content intelligent recommendation method, comprising:

[0005] Obtaining a target user behavior corresponding to the target user when watching the live broadcast, and obtaining associated users corresponding to the target user and associated user behaviors corresponding to the associated users when watching the live broadcast;

[0006] Calculate the target interaction values ​​corresponding to the target user and the associated user according to the target user behavior and the associated user behavior;

[0007] Obtain similar users corresponding to the target user from the associated users according to the target interaction value, and obtain similar user behaviors corresponding to the similar users from the associated user behaviors;

[0008] Using a multi-objective classification model to classify different features of the live content that the target user and the similar users pay attention to according to the target user behavior and the similar user behavior, to obtain neighboring users corresponding to the target user;

[0009] Recommending live content to the target user according to the neighboring user behaviors corresponding to the neighboring users to obtain an initial recommendation result;

[0010] Using an emotion recognition model to identify the emotion of the target user according to the target user behavior to obtain a corresponding target emotion;

[0011] The initial recommendation result is adjusted according to the target emotion to obtain a target recommendation result.

[0012] In a second aspect, an embodiment of the present invention provides an AI-driven live content intelligent recommendation device, comprising:

[0013] A data acquisition module, used to obtain the target user behavior corresponding to the target user when watching the live broadcast, and obtain the associated users corresponding to the target user and the associated user behavior corresponding to the associated users when watching the live broadcast;

[0014] An interaction calculation module, used to calculate the target interaction values ​​corresponding to the target user and the associated user according to the target user behavior and the associated user behavior;

[0015] A user screening module, configured to obtain similar users corresponding to the target user from the associated users according to the target interaction value, and obtain similar user behaviors corresponding to the similar users from the associated user behaviors;

[0016] A user determination module is used to classify different features of the target user and the similar users' attention to live content by using a multi-target classification model according to the target user behavior and the similar user behavior, and obtain neighboring users corresponding to the target user;

[0017] An initial recommendation module, used to recommend live content to the target user according to the neighboring user behaviors corresponding to the neighboring users, to obtain an initial recommendation result;

[0018] An emotion recognition module, used to use an emotion recognition model to recognize the emotion of the target user according to the target user's behavior to obtain a corresponding target emotion;

[0019] The recommendation adjustment module is used to adjust the initial recommendation result according to the target emotion to obtain the target recommendation result.

[0020] In a third aspect, an embodiment of the present invention further provides a computer device, comprising a processor, a memory, a computer program stored in the memory and executable by the processor, and a data bus for realizing connection and communication between the processor and the memory, wherein when the computer program is executed by the processor, the steps of any one of the AI-driven live content intelligent recommendation methods provided in the specification of the present invention are implemented.

[0021] The embodiment of the present invention provides an AI-driven intelligent recommendation method, device and computer equipment for live broadcast content, the method comprising: obtaining a target user behavior corresponding to a target user when watching a live broadcast, and obtaining associated users corresponding to the target user and associated user behaviors corresponding to the associated users when watching the live broadcast; calculating target interaction values ​​corresponding to the target user and the associated users according to the target user behavior and the associated user behavior; obtaining similar users corresponding to the target user from the associated users according to the target interaction value, and obtaining similar user behaviors corresponding to the similar users from the associated user behavior; using a multi-target classification model to classify different features of the live broadcast content that the target user and the similar users pay attention to according to the target user behavior and the similar user behavior, and obtaining neighboring users corresponding to the target user; recommending live broadcast content to the target user according to the neighboring user behaviors corresponding to the neighboring users, and obtaining an initial recommendation result; using an emotion recognition model to identify the emotion of the target user according to the target user behavior to obtain a corresponding target emotion; adjusting the initial recommendation result according to the target emotion to obtain a target recommendation result. By analyzing the target interaction value between the target user's target user behavior and the associated user behavior of the associated users, this method can more accurately understand the interests and preferences of the target user, and then obtain similar users corresponding to the target user, so as to identify neighboring users with similar interests to the target user by feature classification of user behavior through a multi-target classification model, and then dig out the deep user needs of the target user, thereby improving the accuracy and relevance of the recommendation. Then, the emotion recognition model is used to analyze the emotions of the target user in real time, and the recommended content can be dynamically adjusted according to the emotional state of the target user to ensure that the recommended content matches the current emotion of the target user, thereby improving the user experience. It also solves the problem that the related technology lacks the mining of the user's deep needs when performing content recommendation, and its recommendation results are often difficult to truly dig out the user's deep needs, resulting in a certain deviation between the recommendation results and the actual needs of the user. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0023] Figure 1 A flowchart of an AI-driven live content intelligent recommendation method provided by an embodiment of the present invention;

[0024] Figure 2 A schematic diagram of the module structure of an AI-driven live content intelligent recommendation device provided by an embodiment of the present invention;

[0025] Figure 3 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0028] It should be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0029] The embodiments of the present invention provide an AI-driven live content intelligent recommendation method, device, and computer device. The AI-driven live content intelligent recommendation method can be applied to a computer device, which can be an electronic device such as a tablet computer, a laptop computer, a desktop computer, a personal digital assistant, and a wearable device. The computer device can be a server or a server cluster.

[0030] Some embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0031] Please refer to Figure 1 , Figure 1 A flowchart of an AI-driven live content intelligent recommendation method provided in an embodiment of the present invention.

[0032] like Figure 1 As shown, the AI-driven live content intelligent recommendation method includes steps S101 to S107.

[0033] Step S101: obtaining a target user behavior corresponding to a target user when watching a live broadcast, and obtaining associated users corresponding to the target user and associated user behaviors corresponding to the associated users when watching a live broadcast.

[0034] For example, the target user is the user of the live content to be recommended, the associated user is the user who watches the same live content or similar live content as the target user, and the associated user is the user whose behavior overlaps with the target user. Thus, the target user behavior corresponding to the target user when watching the live broadcast is obtained, the target user behavior includes but is not limited to the historical behavior and real-time behavior of the target user when watching the live broadcast, and the associated user includes but is not limited to the historical behavior and real-time behavior of the associated user when watching the live broadcast.

[0035] Step S102: Calculate target interaction values ​​corresponding to the target user and the associated users according to the target user behavior and the associated user behavior.

[0036] For example, the target user behavior includes, but is not limited to, the corresponding viewing time, interaction frequency (such as comments, likes, and shares), the type of content viewed, viewing duration, and skip rate when the target user watches the live broadcast. The associated user behavior includes, but is not limited to, the corresponding viewing time, interaction frequency (such as comments, likes, and shares), the type of content viewed, viewing duration, and skip rate when the associated user watches the live broadcast.

[0037] Exemplarily, when the types of content viewed in the target user behavior and the associated user behavior are the same, the similarity between the target user's corresponding viewing time, interaction frequency (such as comments, likes, sharing), type of content viewed, viewing duration, and skip rate and the associated user's corresponding viewing time, interaction frequency (such as comments, likes, sharing), type of content viewed, viewing duration, and skip rate is calculated, thereby fusing the similarities of the above different parameters to obtain the target interaction values ​​corresponding to the target user and the associated user.

[0038] In some embodiments, the calculating the target interaction value corresponding to the target user and the associated user according to the target user behavior and the associated user behavior includes: determining whether there is communication between the target user and the associated user according to the target user behavior and the associated user behavior, and obtaining a first communication value; determining the same user behavior between the target user and the associated user according to the target user behavior and the associated user behavior, and determining a second communication value between the target user and the associated user according to the same user behavior; fusing the first communication value and the second communication value to obtain a first interaction value between the target user and the associated user; obtaining a third communication value of the target user operating on the relevant data of the associated user from the target user behavior; obtaining a fourth communication value of the associated user operating on the relevant data of the target user from the associated user behavior; obtaining other user behaviors corresponding to other users and the other users, and determining a fifth communication value between the target user and the associated user according to the other user behaviors; fusing the third communication value, the fourth communication value, and the fifth communication value to obtain a second interaction value between the target user and the associated user; and determining the target interaction value corresponding to the target user and the associated user according to the first interaction value and the second interaction value.

[0039] Exemplarily, whether the target user and the associated user follow each other is obtained from the target user behavior and the associated user behavior, and a first communication value is obtained. When the target user and the associated user follow each other, the first communication value is 1, otherwise the first communication value is 0.

[0040] Exemplarily, an intersection calculation is performed based on the target user behavior and the associated user behavior to obtain the same user behavior between the target user and the associated users, and a union calculation is performed based on the target user behavior and the associated user behavior to obtain all user behaviors between the target user and the associated users, thereby dividing the number of behaviors corresponding to the same user behavior by the number of behaviors corresponding to all user behaviors to obtain the second communication value between the target user and the associated users.

[0041] Exemplarily, the first interaction value and the second interaction value are weighted and fused according to historical experience or expert experience to obtain the first interaction value between the target user and the associated user. The first interaction value is used to characterize the interaction intensity between the target user and the associated user at the behavioral level.

[0042] Exemplarily, the relevant data of the associated user includes the recommended operations, published operations, non-recommended operations, etc. corresponding to the associated user, so as to obtain the target user's operations corresponding to the relevant data of the associated user from the target user's behavior, and then obtain the third communication value.

[0043] For example, when the operation of the associated user is a recommended operation, and the corresponding operation of the target user for the recommended operation obtained from the target user behavior is marking as a favorite or is also a recommended operation, the third communication value is increased by 1; when the operation of the associated user is a recommended operation, and the corresponding operation of the target user for the recommended operation obtained from the target user behavior is a non-recommended operation, the third communication value is reduced by 1, and so on. That is, the third communication value is the target user's recognition of the relevant operation of the associated user.

[0044] Exemplarily, the target user's relevant data includes the target user's corresponding recommended operations, published operations, non-recommended operations, etc., so that the operations of the associated users corresponding to the target user's relevant data are obtained from the associated user behaviors, thereby obtaining the fourth communication value.

[0045] For example, when the target user's operation is not a recommended operation, and the corresponding operation of the associated user for the not recommended operation obtained from the associated user behavior is marked as a favorite or is also a recommended operation, the fourth communication value is subtracted by 1 or other numerical values; when the target user's operation is a recommended operation, and the corresponding operation of the associated user for the recommended operation obtained from the associated user behavior is a recommended operation, the fourth communication value is added by 1, and so on. That is, the fourth communication value is the associated user's recognition of the target user's related operation.

[0046] Exemplarily, other user behaviors corresponding to other users and other users are obtained, thereby obtaining corresponding operations of other users in other user behaviors, and obtaining first feedback of the target user on the corresponding operations in other user behaviors from the target user behaviors, and obtaining second feedback of the associated user on the corresponding operations in other user behaviors from the associated user behaviors, and then calculating the similarity between the first feedback and the second feedback to determine the fifth communication value between the target user and the associated user according to the similarity. That is, the fifth communication value is the similarity between the target user's recognition of the related operations of other users and the associated user's recognition of the related operations of other users to characterize the recognition between the target user and the associated user.

[0047] Exemplarily, weights corresponding to the third communication value, the fourth communication value and the fifth communication value are set respectively according to expert experience or historical experience, so that the third communication value, the fourth communication value and the fifth communication value are weightedly fused according to different weights to obtain the second interaction value between the target user and the associated user.

[0048] Exemplarily, the first interaction value and the second interaction value are summed to obtain target interaction values ​​corresponding to the target user and the associated users, wherein the target interaction value is used to indicate the final recognition between the target user and the associated users.

[0049] In some embodiments, determining the second communication value between the target user and the associated user based on the same user behavior includes: splitting the same user behavior to obtain a first user behavior in which the target user operates according to the corresponding behavior of the associated user and a second user behavior in which the associated user operates according to the corresponding behavior of the target user; obtaining a first behavior corresponding to the target user and a first weight corresponding to the first behavior from the first user behavior; obtaining a first quantity corresponding to the first behavior from the first user behavior and a first total quantity corresponding to the target user under the first behavior from the target user behavior; determining a first influence value of the associated user on the target user based on the first weight, the first quantity, and the first total quantity; obtaining a second behavior corresponding to the associated user and a second weight corresponding to the second behavior from the second user behavior; obtaining a second quantity corresponding to the second behavior from the second user behavior and a second total quantity corresponding to the associated user under the second behavior from the associated user behavior; determining a second influence value of the target user on the associated user based on the second weight, the second quantity, and the second total quantity; and determining the second communication value between the target user and the associated user by fusing the first influence value and the second influence value.

[0050] Exemplarily, the same user behavior is the behavior in which the target user and the associated user have the same operation in the same live video or the same type of video. Thus, the same user behavior is split to obtain the first user behavior in which the target user operates according to the corresponding behavior of the associated user and the second user behavior in which the associated user operates according to the corresponding behavior of the target user, that is, the first user behavior is the behavior of the target user under the influence of the associated user on the target user; the second user behavior is the behavior of the associated user under the influence of the target user on the associated user.

[0051] For example, the same user behavior is that the target user and the associated user watched the target live video for a long time, then if the target user watched the target live video for a long time due to the recommendation of the associated user to watch the target live video for a long time, then the target user's long-term viewing operation on the target live video is determined as the first user behavior of the target user according to the corresponding behavior of the associated user; similarly, if the associated user watched the target live video for a long time due to the recommendation of the target user to watch the target live video for a long time, then the associated user's long-term viewing operation on the target live video is determined as the second user behavior of the associated user according to the corresponding behavior of the target user.

[0052] Exemplarily, the first behavior implemented by the target user on the related operation of the associated user is obtained from the first user behavior. The first behavior can be the same as the related operation of the associated user or different from the related operation of the associated user. The same as the related operation of the associated user is used to represent that the target user approves the related operation of the associated user, and different from the related operation of the associated user is used to represent that the target user does not approve the related operation of the associated user. Then, the first weight corresponding to the first behavior is set based on experience. Thus, the first quantity corresponding to the first behavior is obtained from the first user behavior and the first total quantity corresponding to the target user under the first behavior is obtained from the target user behavior, and then the first quantity and the first total quantity are ratioed to obtain a first ratio, and then the first ratio and the first weight are weighted summed to obtain a first influence value of the target user on the associated user. The first influence value is used to represent the influence degree of the associated user on the target user or the recognition degree of the target user on the associated user.

[0053] Exemplarily, the second behavior implemented by the associated user on the relevant operation of the target user is obtained from the second user behavior. The second behavior can be the same as the relevant operation of the target user or different from the relevant operation of the target user. The same as the relevant operation of the target user is used to represent that the associated user approves the relevant operation of the target user, and different from the relevant operation of the target user is used to represent that the associated user does not approve the relevant operation of the target user. Then, the second weight corresponding to the second behavior is set based on experience. Thus, the second quantity corresponding to the second behavior is obtained from the second user behavior and the second total quantity corresponding to the associated user under the second behavior is obtained from the associated user behavior, and then the second quantity and the second total quantity are ratioed to obtain a second ratio, and then the second ratio and the second weight are weighted summed to obtain a second influence value of the associated user on the target user. The second influence value is used to represent the degree of influence of the target user on the associated user or the degree of approval of the associated user on the target user.

[0054] Exemplarily, the second influence value is added by 1 and then divided by 2 to obtain a third influence value, and then the first influence value and the third influence value are multiplied to obtain a second communication value between the target user and the associated user. The second communication value is used to characterize the degree of recognition between the target user and the associated user.

[0055] In some embodiments, determining the fifth communication value between the target user and the associated user based on the other user behaviors includes: obtaining a third user behavior corresponding to the target user's operation on the other user's information and obtaining a fourth user behavior corresponding to the associated user's operation on the other user's information from the other user behaviors; performing an intersection calculation on the third user behavior and the fourth user behavior to obtain a first behavior quantity and performing a union calculation on the third user behavior and the fourth user behavior to obtain a second behavior quantity; performing data statistics on the third user behavior to obtain a third behavior quantity and performing data statistics on the fourth user behavior to obtain a fourth behavior quantity; performing a behavior overlap calculation based on the first behavior quantity, the second behavior quantity, the third behavior quantity, and the fourth behavior quantity to obtain the fifth communication value; wherein the fifth communication value is obtained according to the following formula:

[0056] ;

[0057] in, represents the fifth AC value, represents the number of the first behavior, Indicates the number of the second behavior, represents the number of the third line, Indicates that the fourth line is a quantity.

[0058] Exemplarily, a third user behavior corresponding to the target user operating on the other user's information is obtained from other user behaviors, and a fourth user behavior corresponding to the associated user operating on the other user's information is obtained. That is, the third user behavior is the behavior of the target user under the influence of other users on the target user; the fourth user behavior is the behavior of the associated user under the influence of other users on the associated user.

[0059] For example, other user behaviors include other users watching the target live video for a long time. If the target user also watches the target live video for a long time due to the recommendation of other users to watch the target live video for a long time, then the target user's long-term viewing operation on the target live video is determined as a third user behavior of the target user performing an operation according to the corresponding behaviors of other users; similarly, if the associated user also watches the target live video for a long time due to the recommendation of other users to watch the target live video for a long time, then the associated user's long-term viewing operation on the target live video is determined as a fourth user behavior of the associated user performing an operation according to the corresponding behaviors of other users.

[0060] Exemplarily, the intersection calculation is performed on the third user behavior and the fourth user behavior to obtain the number of first behaviors corresponding to the same behavior between the behavior of the target user under the influence of other users on the target user and the behavior of the associated user under the influence of other users on the associated user.

[0061] Exemplarily, the third user behavior and the fourth user behavior are union calculated to obtain the number of second behaviors corresponding to all behaviors between the behavior of the target user under the influence of other users on the target user and the behavior of the associated user under the influence of other users on the associated user.

[0062] Exemplarily, data statistics are performed on the third user behavior to obtain the third behavior quantity of all behaviors of the target user under the influence of other users on the target user, and data statistics are performed on the fourth user behavior to obtain the fourth behavior quantity of all behaviors of the associated users under the influence of other users on the associated users.

[0063] Exemplarily, the behavior overlap degree is calculated for the first behavior number, the second behavior number, the third behavior number, and the fourth behavior number according to the following formula to obtain the fifth AC value:

[0064] ;

[0065] in, represents the fifth AC value, Indicates the number of the first line, Indicates the number of the second line, Indicates the number of the third line, Indicates that the fourth line is a quantity.

[0066] For example, according to Calculate the behavior entropy value corresponding to the target user, according to Calculate the behavior entropy value corresponding to the associated user, The mutual influence information between the target user and the associated user is represented, and the degree of behavioral overlap between the target user and the associated user is calculated according to the above formula to obtain the fifth communication value. That is, the fifth communication value is the similarity between the target user's recognition of other users' related operations and the associated user's recognition of other users' related operations to characterize the recognition between the target user and the associated user.

[0067] In some embodiments, determining the target interaction value corresponding to the target user and the associated user based on the first interaction value and the second interaction value includes: performing weighted summation of the first interaction value and the second interaction value to obtain a first fused interaction value corresponding to the current moment; obtaining a second fused interaction value corresponding to the previous moment, and calculating an interaction difference between the first fused interaction value and the second fused interaction value; determining a time adjustment factor between the target user and the associated user based on the interaction difference; and performing parameter fusion of the first fused interaction value and the second fused interaction value based on the time adjustment factor to obtain the target interaction value corresponding to the target user and the associated user.

[0068] Exemplarily, the weights corresponding to the first interaction value and the second interaction value are obtained according to expert experience or historical experience, so as to obtain the first fused interaction value corresponding to the current moment by weighted summing the first interaction value and the second interaction value according to the corresponding weights. The second fused interaction value corresponding to the associated user and the target user at the previous moment is obtained from the database.

[0069] Exemplarily, the difference between the first fused interaction value and the second fused interaction value is calculated to obtain the interaction difference. When the interaction difference is greater than 0, it indicates that the influence of the associated user on the target user is positive, which means that the relationship between the target user and the associated user is closer. When the interaction difference is less than 0, it indicates that the influence of the associated user on the target user is negative, which means that the relationship between the target user and the associated user is alienated, thereby determining the time adjustment factor between the target user and the associated user according to the interaction difference. When the interaction difference is greater than 0, the time adjustment factor should be rewarded to reduce the forgetting rate. When the interaction difference is less than 0, it should be punished to increase the forgetting rate.

[0070] Exemplarily, the time adjustment factor is multiplied with the second fused interaction value and then summed with the first fused interaction value to obtain the target interaction value corresponding to the target user and the associated user at the current moment, which is fused with the interaction value at the previous moment. The target interaction value not only considers the current moment but also the previous moment, and is controlled by the time adjustment factor to obtain the corresponding recognition of the target user and the associated user at the current moment.

[0071] Step S103: obtaining similar users corresponding to the target user from the associated users according to the target interaction value, and obtaining similar user behaviors corresponding to the similar users from the associated user behaviors.

[0072] Exemplarily, a preset value is set, the target interaction value is compared with the preset value, and when the target interaction value is greater than or equal to the preset value, the associated user is determined as a similar user corresponding to the target user, and the similar user behavior corresponding to the similar user is obtained from the associated user behavior.

[0073] Step S104: using a multi-target classification model to classify different features of the live content that the target user and the similar users pay attention to according to the target user behavior and the similar user behavior, to obtain neighboring users corresponding to the target user.

[0074] Exemplarily, key features are extracted from the target user behavior, such as the type preference, viewing frequency, interaction frequency, peak viewing time of live broadcasts, etc., and these features are quantified and converted into numerical form. And the features of live broadcast content are extracted from the target user behavior, such as the anchor style, content type, live broadcast duration, live broadcast theme, etc., to help understand the target user's attention to different content. And key features are extracted from similar user behaviors, such as the type preference, viewing frequency, interaction frequency, peak viewing time of live broadcasts, etc., and these features are quantified and converted into numerical form. And the features of live broadcast content are extracted from similar user behaviors, such as the anchor style, content type, live broadcast duration, live broadcast theme, etc., to help understand the similar users' attention to different content.

[0075] For example, a suitable multi-objective classification model is selected, such as random forest, support vector machine (SVM), neural network, etc., so as to learn multiple different features of each target user and similar users to distinguish the differences between the target user and similar users, and then find neighboring users with similar content types as the target user through the model classification results.

[0076] In some embodiments, the method of using a multi-objective classification model to classify different features of the target user and the similar users' attention to live content according to the target user behavior and the similar user behavior to obtain neighboring users corresponding to the target user includes: using a feature extraction network of the multi-objective classification model to perform feature processing on the target user behavior and the similar user behavior respectively to obtain a first feature vector corresponding to the target user behavior and a second feature vector corresponding to the similar user behavior; using a behavior feature network of the multi-objective classification model to perform feature processing on the target user behavior and the similar user behavior respectively to obtain a first behavior vector corresponding to the target user behavior and a second behavior vector corresponding to the similar user behavior; using a feature fusion network of the multi-objective classification model to perform feature fusion on the first feature vector and the first behavior vector to obtain a first target vector; using the multi-objective classification model to classify different features of the live content according to the target user behavior and the similar user behavior to obtain neighboring users corresponding to the target user, including: using a feature extraction network of the multi-objective classification model to perform feature processing on the target user behavior and the similar user behavior respectively to obtain a first feature vector corresponding to the target user behavior and a second behavior vector corresponding to the similar user behavior The feature fusion network of the class model performs feature fusion on the second feature vector and the second behavior vector to obtain a second target vector; uses the first classification network of the multi-target classification model to classify the first target vector into a category to obtain a first classification result; uses the second classification network of the multi-target classification model to classify the first target vector into a category to obtain a second classification result; uses the first classification network of the multi-target classification model to classify the second target vector into a category to obtain a third classification result; uses the second classification network of the multi-target classification model to classify the second target vector into a category to obtain a fourth classification result; calculates the attention difference value between the target user and the similar users based on the first classification result, the second classification result, the third classification result and the fourth classification result; screens the similar users based on the attention difference value to obtain the neighboring users corresponding to the target user.

[0077] Exemplarily, the multi-objective classification model includes a feature extraction network, a behavior feature network, a feature fusion network, a first classification network, and a second classification network.

[0078] Exemplarily, the feature extraction network of the multi-target classification model is used to perform feature processing on the target user behavior to obtain a first feature vector corresponding to the target user behavior, and the feature extraction network of the multi-target classification model is used to perform feature processing on similar user behaviors to obtain a second feature vector corresponding to similar user behaviors, the first feature vector is used to characterize the feature representation corresponding to the target user's evaluation of live content or purchase behavior in the target user behavior. The second feature vector is used to characterize the feature representation corresponding to the similar user's evaluation of live content or purchase behavior in similar user behaviors.

[0079] Exemplarily, the behavior of the target user is processed by the behavior feature network of the multi-target classification model to obtain a first behavior vector corresponding to the target user behavior, and the behavior feature network of the multi-target classification model is processed by the behavior feature of the similar user to obtain a second behavior vector corresponding to the similar user behavior. The first behavior vector is used to characterize the characteristic representation corresponding to the target user's viewing, forwarding or collection of the live content in the target user's behavior, and the second behavior vector is used to characterize the characteristic representation corresponding to the similar user's viewing, forwarding or collection of the live content in the similar user behavior.

[0080] It should be noted that the feature extraction network and the behavior feature network obtain the features corresponding to the target user or similar users from different angles, thereby providing support for the subsequent screening of neighboring users.

[0081] Exemplarily, a first target vector is obtained by weighted fusion of the first feature vector and the first behavior vector using a feature fusion network of a multi-target classification model; and a second target vector is obtained by weighted fusion of the second feature vector and the second behavior vector according to the feature fusion network.

[0082] Exemplarily, a first classification network of a multi-target classification model is used to classify a first target vector into a category to obtain a first classification result; and a second classification network of the multi-target classification model is used to classify the first target vector into a category to obtain a second classification result.

[0083] Exemplarily, the first classification network of the multi-target classification model is used to classify the second target vector to obtain a third classification result, and the second classification network of the multi-target classification model is used to classify the second target vector to obtain a fourth classification result.

[0084] Exemplarily, the first classification network is used to determine the classification results corresponding to the characteristics of the target user or similar users in the evaluation or purchase behavior of the live content. The second classification network is used to determine the classification results corresponding to the characteristics of watching, forwarding or collecting the live content. That is, the first classification network and the second classification network are the category information to which the target user or similar users belong when classifying the target user or similar users from different angles.

[0085] Exemplarily, the first classification difference between the target user and similar users in the first classification network is calculated according to the KL divergence using the first classification result and the third classification result, and the second classification difference between the target user and similar users in the second classification network is calculated according to the KL divergence using the second classification result and the fourth classification result. The smaller the distribution difference between the first classification result and the third classification result, the smaller the first classification difference, and the smaller the distribution difference between the second classification result and the fourth classification result, the smaller the second classification difference.

[0086] Exemplarily, the maximum value of the first classification difference and the second classification difference is determined as the attention difference value between the target user and the similar user, and the attention difference value is used to indicate the maximum difference value between the target user and the similar user at different angles.

[0087] Exemplarily, a preset difference value is determined, and when the attention difference value is less than or equal to the preset difference value, the similar user is determined as a neighboring user corresponding to the target user.

[0088] Step S105: recommending live content to the target user according to the neighboring user behaviors corresponding to the neighboring users to obtain an initial recommendation result.

[0089] Exemplarily, live broadcast content with high viewing times, long viewing time and high interaction volume by neighboring users is screened out from the live broadcast content in the neighboring user behaviors, and then the live broadcast content is recommended to the target user, thereby obtaining the initial recommendation result corresponding to the target user.

[0090] Step S106: using an emotion recognition model to identify the emotion of the target user according to the target user behavior to obtain a corresponding target emotion.

[0091] Exemplarily, emotion recognition is performed on the live video watched by the target user in the target user behavior according to the emotion recognition model, so as to determine the target emotion currently corresponding to the target user according to the emotion information corresponding to the live video watched by the target user.

[0092] In some embodiments, the use of an emotion recognition model to identify the emotion of the target user according to the target user behavior to obtain a corresponding target emotion includes: obtaining a target viewing video corresponding to the target user from the target user behavior; using the video processing layer of the emotion recognition model to extract video key frames from the target viewing video to obtain a corresponding target video key frame and a first time corresponding to the target video key frame; using the speech processing layer of the emotion recognition model to extract speech information from the target viewing video to obtain a corresponding target speech key frame and a second time corresponding to the target speech key frame; using the data alignment layer of the emotion recognition model to align information between the target video key frame and the target speech key frame according to the first time and the second time to obtain Target alignment result; using the weight determination layer of the emotion recognition model to determine the third weight corresponding to the target video key frame and the fourth weight corresponding to the target voice key frame according to the target alignment result; using the video feature representation layer of the emotion recognition model to extract features of the target video key frame to obtain target video features; using the voice feature representation layer of the emotion recognition model to extract features of the target voice key frame to obtain target voice features; using the information fusion layer of the emotion recognition model to fuse the target video features and the target voice features according to the third weight and the fourth weight to obtain a target fusion result; using the emotion classification layer of the emotion recognition model to identify the emotions of the target user according to the target fusion result to obtain the corresponding target emotion.

[0093] Exemplarily, the emotion recognition model includes: a video processing layer, a speech processing layer, a data alignment layer, a weight determination layer, a video feature representation layer, a speech feature representation layer, an information fusion layer, and an emotion classification layer.

[0094] Exemplarily, a target viewing video corresponding to the target user is obtained from the target user behavior, and then the video key frames of the target viewing video are extracted using the video processing layer of the emotion recognition model to obtain the corresponding target video key frames and the first time corresponding to the target video key frames in the target viewing video.

[0095] Exemplarily, the speech processing layer of the emotion recognition model is used to extract speech information from the target viewing video to obtain a corresponding target speech key frame and a second time corresponding to the target speech key frame in the target viewing video.

[0096] Exemplarily, the data alignment layer of the emotion recognition model is used to align information of the target video key frame and the target voice key frame according to the first time and the second time. When the time difference between the first time and the second time is within a preset range, the target video key frame and the target voice key frame are aligned. If the time difference between the first time and the second time is not within the preset range, the target video key frame and the target voice key frame cannot be aligned, thereby obtaining a target alignment result.

[0097] For example, the first time includes time 11, time 12, and time 13, the second time includes time 21, time 22, and time 23, the target video key frames include video key frame 1 corresponding to time 11, video key frame 2 corresponding to time 12, and video key frame 3 corresponding to time 13, and the target voice key frames include voice key frame 1 corresponding to time 21, voice key frame 2 corresponding to time 22, and voice key frame 3 corresponding to time 23. When the time difference between time 11 and time 13 is within a preset range, video key frame 2 and voice key frame 3 are aligned, and the remaining video key frames and voice key frames cannot be aligned, so that the target alignment result is that only video key frame 2 and voice key frame 3 are aligned.

[0098] Exemplarily, the weight determination layer of the emotion recognition model is used to set the weights of the video key frames included in the target video key frames to be larger according to the target alignment results, and the weights of the video key frames not included in the target video key frames to be smaller according to the target alignment results, so as to obtain the third weight corresponding to each video key frame in the target video key frames, and to set the weights of the speech key frames included in the target speech key frames to be larger according to the target alignment results, and the weights of the speech key frames not included in the target speech key frames to be smaller according to the target alignment results, so as to obtain the fourth weight corresponding to each speech key frame in the target speech key frames.

[0099] Exemplarily, the video feature representation layer in the emotion recognition model is used to extract features from the extracted target video key frames. Deep learning structures such as convolutional neural networks (CNNs) are usually used to extract emotion-related features from the key frames, such as expressions, postures, scenes, etc., and the extracted features are vectorized to obtain target video features. And the speech feature representation layer in the emotion recognition model is used to extract features from the extracted target speech key frames. Acoustic models, Mel-frequency cepstral coefficients (MFCCs) or long short-term memory networks (LSTMs) can be used to extract emotional features in speech, such as intonation, speed, volume, etc., and the extracted speech features are vectorized to obtain target speech features.

[0100] Exemplarily, the information fusion layer in the emotion recognition model is used to fuse the target video features and the target speech features according to the third weight and the fourth weight, so as to obtain the target fusion result by weighted average. Finally, the emotion classification layer of the emotion recognition model is used to process the target fusion result. The emotion classification layer may use a fully connected layer, an activation function (such as ReLU, Softmax, etc.) and other structures to map the fused feature vector to a specific emotion category. Thus, according to the output of the model, the target emotion of the target user, such as "happy", "sad", etc., is evaluated.

[0101] Step S107: adjusting the initial recommendation result according to the target emotion to obtain a target recommendation result.

[0102] Exemplarily, the corresponding emotional information in different live broadcast contents in the initial recommendation results is obtained, and then the similarity between the emotional information and the target emotion is calculated, so that the initial recommendation results are sorted according to the similarity between the emotional information and the target emotion, and then the target recommendation results with a similar sorting to the target emotion of the target user are obtained.

[0103] In some embodiments, the initial recommendation result is adjusted according to the target emotion to obtain the target recommendation result, including: when the target emotion is a preset emotion, obtaining the live broadcast style corresponding to the initial recommendation result and the associated emotion corresponding to the live broadcast style; performing data conversion on the associated emotion according to the score mapping table to obtain a first score corresponding to the associated emotion; performing data conversion on the preset emotion according to the score mapping table to obtain a second score corresponding to the preset emotion; filtering the initial recommendation result according to the second score to obtain an intermediate recommendation result; obtaining a third score corresponding to the intermediate recommendation result from the first score, and sorting the intermediate recommendation results according to the difference between the second score and the third score to obtain the target recommendation result.

[0104] Exemplarily, when the preset emotion is sadness or depression, the live broadcast style corresponding to the initial recommendation result and the associated emotions corresponding to the live broadcast style are obtained, and then the associated emotions are converted into data according to the score mapping table to obtain the first score corresponding to the associated emotions, and the preset emotions are converted into data according to the score mapping table to obtain the second score corresponding to the preset emotions.

[0105] Exemplarily, the initial recommendation results when the first score is lower than the second score are eliminated, so as to obtain the initial recommendation results corresponding to when the first score is greater than or equal to the second score and determine them as the intermediate recommendation results. That is, the intermediate recommendation results are live content whose associated emotions corresponding to the live style are higher than the target emotion of the target user.

[0106] Exemplarily, a third score corresponding to the intermediate recommendation result is obtained from the first score, and then the difference between the second score and the third score is calculated, and the intermediate recommendation results are sorted from small to large according to the difference, so as to obtain the target recommendation result.

[0107] For example, when the target emotion is a preset emotion, that is, the target user is sad or depressed at this time, in order to prevent the target user from falling into a sad emotional atmosphere, positive-style live broadcast content is gradually recommended to the target user, which neither affects the emotional jumps of the target user when watching the live content, nor can it gradually enable the target user to get rid of the sad emotional atmosphere.

[0108] See also Figure 2 , Figure 2 An AI-driven live content intelligent recommendation device 200 provided in an embodiment of the present application includes a data acquisition module 201, an interaction calculation module 202, a user screening module 203, a user determination module 204, an initial recommendation module 205, an emotion recognition module 206, and a recommendation adjustment module 207, wherein the data acquisition module 201 is used to obtain the target user behavior corresponding to the target user when watching the live broadcast, and obtain the associated users corresponding to the target user and the associated user behavior corresponding to the associated users when watching the live broadcast; the interaction calculation module 202 is used to calculate the target interaction value corresponding to the target user and the associated user according to the target user behavior and the associated user behavior; the user screening module 203 is used to obtain the target interaction value from the associated users according to the target interaction value similar users corresponding to the target user are obtained, and similar user behaviors corresponding to the similar users are obtained from the associated user behaviors; a user determination module 204 is used to classify the different features of the target user and the similar users' attention to live content according to the target user behavior and the similar user behavior by using a multi-target classification model, and obtain neighboring users corresponding to the target user; an initial recommendation module 205 is used to recommend live content to the target user according to the neighboring user behaviors corresponding to the neighboring users, and obtain an initial recommendation result; an emotion recognition module 206 is used to identify the emotion of the target user according to the target user behavior by using an emotion recognition model, and obtain a corresponding target emotion; a recommendation adjustment module 207 is used to adjust the initial recommendation result according to the target emotion, and obtain a target recommendation result.

[0109] It should be noted that technicians in the relevant field can clearly understand that for the convenience and brevity of description, the specific working process of the AI-driven live content intelligent recommendation device 200 described above can refer to the corresponding process in the aforementioned AI-driven live content intelligent recommendation method embodiment, and will not be repeated here.

[0110] See also Figure 3 , Figure 3 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention.

[0111] like Figure 3 As shown, the computer device 300 includes a processor 301 and a memory 302 , and the processor 301 and the memory 302 are connected via a bus 303 , such as an I2C (Inter-integrated Circuit) bus.

[0112] Specifically, the processor 301 is used to provide computing and control capabilities to support the operation of the entire computer device. The processor 301 can be a central processing unit (CPU), and the processor 301 can also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0113] Specifically, the memory 302 may be a Flash chip, a read-only memory (ROM) disk, an optical disk, a USB flash drive, or a mobile hard disk.

[0114] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the embodiment of the present invention, and does not constitute a limitation on the computer device to which the embodiment of the present invention is applied. The specific server may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0115] The processor is used to run a computer program stored in the memory, and implement any one of the AI-driven live content intelligent recommendation methods provided by the embodiments of the present invention when executing the computer program.

[0116] In one embodiment, the processor is used to run a computer program stored in the memory, and implements the following steps when executing the computer program:

[0117] Obtaining a target user behavior corresponding to the target user when watching the live broadcast, and obtaining associated users corresponding to the target user and associated user behaviors corresponding to the associated users when watching the live broadcast;

[0118] Calculate the target interaction values ​​corresponding to the target user and the associated user according to the target user behavior and the associated user behavior;

[0119] Obtain similar users corresponding to the target user from the associated users according to the target interaction value, and obtain similar user behaviors corresponding to the similar users from the associated user behaviors;

[0120] Using a multi-objective classification model to classify different features of the live content that the target user and the similar users pay attention to according to the target user behavior and the similar user behavior, to obtain neighboring users corresponding to the target user;

[0121] Recommending live content to the target user according to the neighboring user behaviors corresponding to the neighboring users to obtain an initial recommendation result;

[0122] Using an emotion recognition model to identify the emotion of the target user according to the target user behavior to obtain a corresponding target emotion;

[0123] The initial recommendation result is adjusted according to the target emotion to obtain a target recommendation result.

[0124] It should be noted that technicians in the relevant field can clearly understand that for the convenience and conciseness of description, the specific working process of the computer device described above can refer to the corresponding process in the aforementioned AI-driven live content intelligent recommendation method embodiment, and will not be repeated here.

[0125] An embodiment of the present invention also provides a storage medium for computer-readable storage, wherein the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of any AI-driven live content intelligent recommendation method provided in the description of the embodiment of the present invention.

[0126] The storage medium may be an internal storage unit of the computer device described in the above embodiment, such as a hard disk or memory of the computer device. The storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the computer device.

[0127] It will be appreciated by those skilled in the art that all or some of the steps, systems, and functional modules / units in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. In a hardware embodiment, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or transient medium). As is known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0128] It should be understood that the term "and / or" used in the present specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, including these combinations. It should be noted that, in this article, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "including a..." does not exclude the presence of other identical elements in the process, method, article or system including the element.

[0129] The serial numbers of the embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. The above description is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. An AI-driven live content intelligent recommendation method, characterized in that: The method comprises: Obtaining a target user behavior corresponding to the target user when watching the live broadcast, and obtaining associated users corresponding to the target user and associated user behaviors corresponding to the associated users when watching the live broadcast; Calculate the target interaction values ​​corresponding to the target user and the associated user according to the target user behavior and the associated user behavior; Obtain similar users corresponding to the target user from the associated users according to the target interaction value, and obtain similar user behaviors corresponding to the similar users from the associated user behaviors; Using a multi-objective classification model to classify different features of the live content that the target user and the similar users pay attention to according to the target user behavior and the similar user behavior, to obtain neighboring users corresponding to the target user; Recommending live content to the target user according to the neighboring user behaviors corresponding to the neighboring users to obtain an initial recommendation result; Using an emotion recognition model to identify the emotion of the target user according to the target user behavior to obtain a corresponding target emotion; Adjusting the initial recommendation result according to the target emotion to obtain a target recommendation result; The step of calculating the target interaction values ​​corresponding to the target user and the associated user according to the target user behavior and the associated user behavior includes: determining whether there is communication between the target user and the associated user according to the target user behavior and the associated user behavior, and obtaining a first communication value; Determine the same user behavior between the target user and the associated user according to the target user behavior and the associated user behavior, and determine a second communication value between the target user and the associated user according to the same user behavior; fusing the first communication value and the second communication value to obtain a first interaction value between the target user and the associated user; Obtaining a third communication value of the target user operating the relevant data of the associated user from the target user behavior; Obtaining, from the associated user behavior, a fourth communication value of the associated user operating the relevant data of the target user; Obtaining other users and other user behaviors corresponding to the other users, and determining a fifth communication value between the target user and the associated user according to the other user behaviors; fusing the third communication value, the fourth communication value, and the fifth communication value to obtain a second interaction value between the target user and the associated user; The target interaction values ​​corresponding to the target user and the associated user are determined according to the first interaction value and the second interaction value.

2. The method according to claim 1, characterized in that The determining the second communication value between the target user and the associated user according to the same user behavior includes: Splitting the same user behavior to obtain a first user behavior of the target user operating according to the corresponding behavior of the associated user and a second user behavior of the associated user operating according to the corresponding behavior of the target user; Obtaining a first behavior corresponding to the target user and a first weight corresponding to the first behavior from the first user behavior; Obtain a first quantity corresponding to the first behavior from the first user behavior and obtain a first total quantity corresponding to the target user under the first behavior from the target user behavior; Determine a first influence value of the associated user on the target user according to the first weight, the first number, and the first total number; Obtaining, from the second user behavior, a second behavior corresponding to the associated user and a second weight corresponding to the second behavior; Obtain a second quantity corresponding to the second behavior from the second user behavior and obtain a second total quantity corresponding to the associated user under the second behavior from the associated user behavior; determining a second influence value of the target user on the associated users according to the second weight, the second number, and the second total number; The first influence value and the second influence value are integrated to determine the second communication value between the target user and the associated user.

3. The method according to claim 1, characterized in that The determining the fifth communication value between the target user and the associated user according to the other user behaviors includes: Obtaining, from the other user behaviors, a third user behavior corresponding to the target user operating on the other user's information and a fourth user behavior corresponding to the associated user operating on the other user's information; Performing an intersection calculation on the third user behavior and the fourth user behavior to obtain a first behavior quantity, and performing a union calculation on the third user behavior and the fourth user behavior to obtain a second behavior quantity; Performing data statistics on the third user behavior to obtain the number of the third behavior and performing data statistics on the fourth user behavior to obtain the number of the fourth behavior; Calculate the degree of behavior overlap according to the first behavior number, the second behavior number, the third behavior number, and the fourth behavior number to obtain the fifth AC value; The fifth AC value is obtained according to the following formula: ; in, represents the fifth AC value, represents the number of the first behavior, Indicates the number of the second behavior, represents the number of the third line, Indicates that the fourth line is a quantity.

4. The method according to claim 1, characterized in that: The determining the target interaction value corresponding to the target user and the associated user according to the first interaction value and the second interaction value includes: Performing a weighted summation on the first interaction value and the second interaction value to obtain a first fused interaction value corresponding to the current moment; Obtaining a second fusion interaction value corresponding to a previous moment, and calculating an interaction difference between the first fusion interaction value and the second fusion interaction value; Determining a time adjustment factor between the target user and the associated user according to the interaction difference; The first fused interaction value and the second fused interaction value are subjected to parameter fusion according to the time adjustment factor to obtain the target interaction value corresponding to the target user and the associated user.

5. The method according to claim 1, characterized in that: The method of using a multi-target classification model to classify different features of the target user and the similar users' attention to live content according to the target user behavior and the similar user behavior, and obtaining neighboring users corresponding to the target user, includes: Using the feature extraction network of the multi-target classification model, feature processing is performed on the target user behavior and the similar user behavior respectively to obtain a first feature vector corresponding to the target user behavior and a second feature vector corresponding to the similar user behavior; Using the behavior feature network of the multi-target classification model, feature processing is performed on the target user behavior and the similar user behavior respectively to obtain a first behavior vector corresponding to the target user behavior and a second behavior vector corresponding to the similar user behavior; Using the feature fusion network of the multi-target classification model, the first feature vector and the first behavior vector are subjected to feature fusion to obtain a first target vector; Using the feature fusion network of the multi-target classification model, the second feature vector and the second behavior vector are subjected to feature fusion to obtain a second target vector; Using a first classification network of the multi-target classification model to classify the first target vector into a category to obtain a first classification result; Using a second classification network of the multi-target classification model to classify the first target vector into a category to obtain a second classification result; Using the first classification network of the multi-target classification model to classify the second target vector into a category to obtain a third classification result; Using the second classification network of the multi-target classification model to classify the second target vector into a category to obtain a fourth classification result; Calculating the attention difference value between the target user and the similar user according to the first classification result, the second classification result, the third classification result and the fourth classification result; The similar users are screened according to the attention difference value to obtain the neighboring users corresponding to the target user.

6. The method according to claim 1, characterized in that The using the emotion recognition model to identify the emotion of the target user according to the target user behavior to obtain the corresponding target emotion includes: Obtaining a target viewing video corresponding to the target user from the target user behavior; Using the video processing layer of the emotion recognition model to extract video key frames from the target viewing video, obtaining corresponding target video key frames and the first time corresponding to the target video key frames; Using the speech processing layer of the emotion recognition model to extract speech information from the target viewing video, obtaining a corresponding target speech key frame and a second time corresponding to the target speech key frame; Using the data alignment layer of the emotion recognition model to perform information alignment on the target video key frame and the target voice key frame according to the first time and the second time, to obtain a target alignment result; Determine, by using the weight determination layer of the emotion recognition model, a third weight corresponding to the target video key frame and a fourth weight corresponding to the target speech key frame according to the target alignment result; Using the video feature representation layer of the emotion recognition model to extract features of the target video key frames to obtain target video features; Using the speech feature representation layer of the emotion recognition model to extract features of the target speech key frame to obtain target speech features; Using the information fusion layer of the emotion recognition model to perform information fusion on the target video feature and the target speech feature according to the third weight and the fourth weight to obtain a target fusion result; The emotion classification layer of the emotion recognition model is used to identify the emotion of the target user according to the target fusion result to obtain the corresponding target emotion.

7. The method according to claim 1, characterized in that The adjusting the initial recommendation result according to the target emotion to obtain the target recommendation result includes: When the target emotion is a preset emotion, the live broadcast style corresponding to the initial recommendation result and the associated emotion corresponding to the live broadcast style are obtained; Performing data conversion on the associated emotion according to the score mapping table to obtain a first score corresponding to the associated emotion; Performing data conversion on the preset emotion according to the score mapping table to obtain a second score corresponding to the preset emotion; Filtering the initial recommendation results according to the second score to obtain intermediate recommendation results; A third score corresponding to the intermediate recommendation result is obtained from the first score, and the intermediate recommendation results are sorted according to a difference between the second score and the third score to obtain the target recommendation result.

8. An AI-driven live content intelligent recommendation device, characterized in that: include: A data acquisition module, used to obtain the target user behavior corresponding to the target user when watching the live broadcast, and obtain the associated users corresponding to the target user and the associated user behavior corresponding to the associated users when watching the live broadcast; The interaction calculation module is used to calculate the target interaction value corresponding to the target user and the associated user according to the target user behavior and the associated user behavior, wherein the calculation of the target interaction value corresponding to the target user and the associated user according to the target user behavior and the associated user behavior includes: determining whether there is communication between the target user and the associated user according to the target user behavior and the associated user behavior, and obtaining a first communication value; determining the same user behavior between the target user and the associated user according to the target user behavior and the associated user behavior, and determining a second communication value between the target user and the associated user according to the same user behavior; fusing the first communication value and the second communication value to obtain the a first interaction value between the target user and the associated user; a third interaction value of the target user operating the relevant data of the associated user from the target user behavior; a fourth interaction value of the associated user operating the relevant data of the target user from the associated user behavior; other users and other user behaviors corresponding to the other users are obtained, and a fifth interaction value between the target user and the associated user is determined according to the other user behaviors; a second interaction value between the target user and the associated user is obtained by fusing the third interaction value, the fourth interaction value and the fifth interaction value; and the target interaction value corresponding to the target user and the associated user is determined according to the first interaction value and the second interaction value; A user screening module, configured to obtain similar users corresponding to the target user from the associated users according to the target interaction value, and obtain similar user behaviors corresponding to the similar users from the associated user behaviors; A user determination module is used to classify different features of the target user and the similar users' attention to live content by using a multi-target classification model according to the target user behavior and the similar user behavior, and obtain neighboring users corresponding to the target user; An initial recommendation module, used to recommend live content to the target user according to the neighboring user behaviors corresponding to the neighboring users, to obtain an initial recommendation result; An emotion recognition module, used to use an emotion recognition model to recognize the emotion of the target user according to the target user's behavior to obtain a corresponding target emotion; The recommendation adjustment module is used to adjust the initial recommendation result according to the target emotion to obtain the target recommendation result.

9. A computer device, characterized in that: The computer device includes a processor and a memory; The memory is used to store computer programs; The processor is used to execute the computer program and implement the AI-driven live content intelligent recommendation method as described in any one of claims 1 to 7 when executing the computer program.

Images