Content recommendation method, device, equipment and computer storage medium

Abstract

Embodiments of the present invention relate to the technical field of computer data processing, and disclose a content recommendation method. The method includes: obtaining operation data of a user for initial recommended content; updating a sample pool of a recommendation model according to the user characteristics of the user and the operation data, where the recommendation model is used for content recommendation. When the operation data includes a preset conversion behavior, the positive sample pool of the recommendation model is updated according to the operation data. When the conversion behavior is not included, the negative sample pool of the recommendation model is updated according to the operation data. The conversion behavior includes an upgrade behavior and / or a payment behavior. The sample pool includes the positive sample pool and the negative sample pool. Training the recommendation model according to the updated sample pool to obtain an updated recommendation model. Recommending to the user according to the updated recommendation model. By the above method, the content recommendation effect is improved in the embodiments of the present invention.

Description

Technical Field

[0001] Embodiments of the present invention relate to the technical field of computer data processing, and in particular to a content recommendation method, apparatus, device, and computer storage medium. Background Art

[0002] Currently, information recommendation is generally used to improve the conversion rate of user actions such as upgrading and subscribing to memberships. However, existing information recommendations generally divide user groups through manual setting or data mining strategies, and recommend information to different user groups to promote conversion behaviors such as user upgrading and subscribing to memberships.

[0003] However, in the process of implementing the present invention, the inventors found that the accuracy of existing content recommendations is relatively low, resulting in a relatively low conversion rate of promoting user upgrading, subscribing to paid memberships, etc. Summary of the Invention

[0004] In view of the above problems, embodiments of the present invention provide a content recommendation to solve the problem of relatively low accuracy of content recommendation in the prior art, which leads to a relatively low user conversion rate.

[0005] According to one aspect of the embodiments of the present invention, a content recommendation method is provided, and the method includes:

[0006] Obtain operation data of a user for initial recommended content;

[0007] Update the sample pool of a recommendation model according to the user characteristics of the user and the operation data, where the recommendation model is used for content recommendation. When the operation data includes a preset conversion behavior, update the positive sample pool of the recommendation model according to the operation data. When the conversion behavior is not included, update the negative sample pool of the recommendation model according to the operation data; the conversion behavior includes an upgrade behavior and / or a payment behavior; the sample pool includes the positive sample pool and the negative sample pool;

[0008] Train the recommendation model according to the updated sample pool to obtain an updated recommendation model;

[0009] Recommend to the user according to the updated recommendation model.

[0010] In an optional manner, the sample pool of the recommendation model includes a recommended content sample pool and a user characteristic sample pool, and the method further includes:

[0011] When the operation data includes the conversion behavior, generate a recommended content positive sample and a user characteristic positive sample according to the operation behavior, the initial recommended content, and the user characteristics respectively;

[0012] Add the positive recommended content sample and the positive user feature sample to the positive sample pool of the recommended content sample pool and the positive sample pool of the user feature sample pool respectively;

[0013] When the operation data does not include the conversion behavior, generate a negative sample of the content questionnaire according to the operation behavior, the initial recommended content, and the user features;

[0014] Add the negative sample of the content questionnaire to the negative sample pool of the recommended content sample pool.

[0015] In an alternative manner, the recommendation model includes a user classification model and a content generation model, and the method further includes:

[0016] Train the user classification model according to the updated user feature sample pool;

[0017] Train the content generation model according to the updated recommended content sample pool.

[0018] In an alternative manner, the method further includes:

[0019] Input the user features into the updated user classification model to obtain a user feature embedding vector;

[0020] Input the user feature embedding vector into the updated content generation model to obtain a target recommended content;

[0021] Recommend the target recommended content to the user.

[0022] In an alternative manner, the method further includes:

[0023] When it is determined that the number of recommendations for the user is greater than a preset number threshold and the operation data does not include the conversion behavior, perform content recommendation for the user according to another recommendation model, where the another recommendation model is trained according to a new user feature sample pool and a new recommended content sample pool; the new user feature sample pool and the new recommended content sample pool are determined according to the user features, historical recommended content, and historical operation data corresponding to historical users whose number of recommendations is greater than the number threshold.

[0024] In an alternative manner, the operation data includes at least one of the residence duration, operation frequency, and trial duration of the user on the recommended content;

[0025] The user features include at least one of the current user level, user activity, and user content preference.

[0026] According to another aspect of the embodiments of the present invention, there is provided a content recommendation device, including:

[0027] An acquisition module, configured to acquire operation data of a user for initial recommended content;

[0028] An update module, configured to update a sample pool of a recommendation model according to user characteristics of the user and the operation data, where the recommendation model is used for content recommendation. When the operation data includes a preset conversion behavior, the positive sample pool of the recommendation model is updated according to the operation data; when the conversion behavior is not included, the negative sample pool of the recommendation model is updated according to the operation data; the conversion behavior includes an upgrade behavior and / or a payment behavior; the sample pool includes the positive sample pool and the negative sample pool;

[0029] A training module, configured to train the recommendation model according to the updated sample pool to obtain an updated recommendation model;

[0030] A recommendation module, configured to recommend to the user according to the updated recommendation model.

[0031] According to another aspect of the embodiments of the present invention, a content recommendation device is provided, including: a processor, a memory, a communication interface, and a communication bus, where the processor, the memory, and the communication interface complete communication with each other through the communication bus;

[0032] The memory is used to store at least one executable instruction, and the executable instruction causes the processor to execute the operations of the content recommendation method embodiment as described in any one of the above.

[0033] According to still another aspect of the embodiments of the present invention, a computer-readable storage medium is provided, where at least one executable instruction is stored in the storage medium, and the executable instruction causes a content recommendation device to execute the operations of the content recommendation method embodiment as described in any one of the above.

[0034] In an embodiment of the present invention, operation data of a user for initial recommended content is obtained; according to the user characteristics of the user and the operation data, a sample pool of a recommendation model is updated, where the recommendation model is used for content recommendation. When the operation data includes a preset conversion behavior, the positive sample pool of the recommendation model is updated according to the operation data. When the conversion behavior is not included, the negative sample pool of the recommendation model is updated according to the operation data; the conversion behavior includes an upgrade behavior and / or a payment behavior; the sample pool includes the positive sample pool and the negative sample pool; the recommendation model is trained according to the updated sample pool to obtain an updated recommendation model; and the user is recommended according to the updated recommendation model. Different from the prior art solution of directly performing content recommendation after clustering users based on big data, which has the problem of low accuracy resulting in low user conversion rate, in the embodiment of the present invention, the positive and negative samples in the sample pool of the recommendation model are correspondingly updated according to whether the user has a conversion behavior for the recommended content, that is, the process of simulating a questionnaire survey of user interests through the user's behavior on the recommended content. According to the operation behavior of the user for the recommended content, the demand and interest degree of the user for implementing the conversion behavior are determined, so as to convert the user into a member user more in line with the user's consumption needs and interests during the process of the user consuming the recommended content. While promoting the user to implement the conversion behavior, it is ensured that operation indicators such as the number of users and user duration do not decrease and the user experience is not affected. The embodiment of the present invention can improve the accuracy of content recommendation, thereby promoting the user to implement conversion behaviors such as payment and upgrade for the content recommended to the user.

[0035] The above description is only an overview of the technical solution of the embodiment of the present invention. In order to be able to understand the technical means of the embodiment of the present invention more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of the embodiment of the present invention more obvious and understandable, the specific embodiments of the present invention are specifically given below. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The drawings are only used to illustrate the embodiments and are not considered as a limitation of the present invention. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0037] Figure 1 A flowchart of the content recommendation method provided by the embodiment of the present invention is shown;

[0038] Figure 2 A flowchart of the content recommendation method provided by another embodiment of the present invention is shown;

[0039] Figure 3 A structural diagram of the content recommendation device provided by the embodiment of the present invention is shown;

[0040] Figure 4The structural schematic diagram of the content recommendation device provided by the embodiment of the present invention is shown. Detailed implementation manners

[0041] Hereinafter, the exemplary embodiments of the present invention will be described in more detail with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein.

[0042] Figure 1 The flowchart of the content recommendation method provided by the embodiment of the present invention is shown. This method is executed by a computer processing device. The computer processing device may include a mobile phone, a laptop computer, etc. As Figure 1 shown, the method includes the following steps:

[0043] Step 10: Obtain the operation data of the user for the initial recommended content.

[0044] In an embodiment of the present invention, the initial recommended content may be the content recommended to the user according to the prior art when the user logs in for the first time or when the user information is less.

[0045] In still another embodiment of the present invention, the initial recommended content may also be the historical recommended content recommended according to any embodiment of the method of the present invention, such as the most recent recommended content. Specifically, the initial recommended content may be multimedia content such as e-books, music, videos, and articles.

[0046] Step 20: Update the sample pool of the recommendation model according to the user characteristics of the user and the operation data. The recommendation model is used for content recommendation. Wherein, when the operation data includes a preset conversion behavior, update the positive sample pool of the recommendation model according to the operation data. When the conversion behavior is not included, update the negative sample pool of the recommendation model according to the operation data; the conversion behavior includes an upgrade behavior and / or a payment behavior; the sample pool includes the positive sample pool and the negative sample pool.

[0047] In an embodiment of the present invention, the conversion behavior may include payment behaviors such as purchasing content, paying for recharge, and upgrading membership, and may also include behaviors in which the user shows interest in the content such as following, forwarding, or liking. For some recommended content, only when the user has a conversion behavior can the user perform further specific types of operations on the content, such as downloading, using, forwarding, editing, etc.

[0048] It can be understood that when the content recommendation is more accurate and more in line with the user's content preferences, the user is more likely to perform a conversion behavior. Therefore, it can be reflected whether the current recommendation is accurate based on whether the operation data of the user for the initial recommended content includes a conversion behavior, so as to determine whether the current sharp reduction is effective for the user operation goal of promoting user conversion.

[0049] In the case where the recommendation is effective, according to the principle of collaborative filtering, similar content recommendations can be made for other users with similar user characteristics based on the content data corresponding to the current effective recommendation process, thereby improving the recommendation accuracy of other users and at the same time improving the user conversion rate.

[0050] Therefore, the sample pool can be divided into a positive sample pool that has a positive effect on promoting user conversion and a negative sample pool that has a negative effect. The positive sample pool and the negative sample pool are updated according to whether the operation data includes a conversion behavior and the corresponding user characteristics, so as to train the recommendation model based on the updated sample pool, and the trained recommendation model has a better effect of promoting user conversion and more accurate recommendation.

[0051] In another embodiment of the present invention, considering that usually users will browse and try the content to a certain extent before conversion. Among them, the trial refers to a certain restriction method for the user's rights to the content that is different from the complete rights, such as in terms of usage time, the completeness of the content used, and the types of executable operations for the content.

[0052] And the user's trial experience of the content will largely affect whether they perform formal conversion behaviors such as purchase and upgrade. For example, after the user listens to the free part of a certain song (such as the first 30s or the climax part) and feels very satisfied, but to listen to the whole song, they need to purchase the single copyright or upgrade the membership. Therefore, the user performs an upgrade or purchase, that is, completes the conversion behavior.

[0053] At the same time, even in the case of a satisfactory trial experience, whether the user performs a conversion is also related to their own factors, such as whether the user is active, whether they have a payment habit, and whether they are a user who has been using the current trial content for a long time and has high stickiness.

[0054] Therefore, in an embodiment of the present invention, the operation data includes at least one of the user's residence time, operation frequency, and trial time for the recommended content, and the user characteristics include at least one of the current user level, user activity, and user content preference.

[0055] In yet another embodiment of the present invention, the operation data may further include data such as the time, frequency, and specific operation parameters corresponding to various preset operation types of the user for the initial recommended content. Among them, the operation types may include clicking, browsing, and using, etc., and the specific operation parameters may include commenting and / or marking content, the number of bookmarks, the bookmark position, and favorite folder information, etc.

[0056] In yet another embodiment of the present invention, the user characteristics may further include personal attribute information of the user such as the user's gender, age, region, etc.

[0057] In one embodiment of the present invention, the sample pool of the recommendation model includes a recommended content sample pool and a user characteristic sample pool. Step 20 further includes: Step 201: When the operation data includes the conversion behavior, generate a recommended content positive sample and a user characteristic positive sample according to the operation behavior, the initial recommended content, and the user characteristics respectively.

[0058] In one embodiment of the present invention, feature extraction can be performed on the operation behavior and the initial recommended content respectively to obtain corresponding behavior feature vector samples and content feature vector samples, and the behavior feature vector samples, the content feature vector samples, and the user characteristics are combined to obtain a recommended content positive sample and a user characteristic positive sample. Among them, the combination methods can be methods such as vector splicing, embedding, and hybrid coding.

[0059] It should be noted that when generating the recommended content positive sample, the content feature vector sample includes at least one of the content identifier, type, author, release time, scene information, classification information such as style and genre, copyright information, popular group information, and included optional conversion behaviors, etc.

[0060] For example, use users who have successfully upgraded their membership levels and the songs they listened to, and develop similar users into higher-level members. Combine these songs into a "questionnaire", use the behavior data of the users on this questionnaire as the answers, and then correspondingly recommend songs that can develop the users into higher-level members. Concatenate the feature vectors of these songs with the user feature vectors and the behavior information of the users when using these songs, such as the stay duration, listening frequency, duration, etc. for this song to generate the recommended content sample.

[0061] When generating the recommended content positive sample, the content feature vector sample includes the content identifier.

[0062] For example, when recommending songs to users, the user feature vector corresponding to the user who has successfully upgraded the membership level can be concatenated with the song identifier recommended for this user and the operation information such as the stay duration, listening, actual usage frequency, and duration of this song by the user who has successfully upgraded to generate the recommended content positive sample.

[0063] Step 202: Add the positive recommended content sample and the positive user feature sample to the positive sample pool of the recommended content sample pool and the positive sample pool of the user feature sample pool respectively.

[0064] Step 203: When the operation data does not include the conversion behavior, generate negative samples of the content questionnaire according to the operation behavior, the initial recommended content, and the user features.

[0065] In an embodiment of the present invention, similar to the generation process of the positive samples of the content questionnaire, when the operation data does not include the conversion behavior, feature extraction can be performed on the operation behavior and the initial recommended content respectively to obtain corresponding behavior feature vector samples and content feature vector samples, and the behavior feature vector samples, the content feature vector samples, and the user features are combined to obtain negative samples of the recommended content and negative samples of the user features. Among them, the combination method can be ways such as vector splicing, embedding, and hybrid coding.

[0066] Among them, when generating negative samples of the recommended content, the content feature vector samples include at least one of the classification information such as the identifier, type, author, release time, scene information, style and genre, copyright information, popular group information, and optional conversion behaviors included in the content.

[0067] Step 204: Add the negative samples of the content questionnaire to the negative sample pool of the recommended content sample pool.

[0068] Step 30: Train the recommendation model according to the updated sample pool to obtain an updated recommendation model.

[0069] In an embodiment of the present invention, the recommendation model can be a neural network model, which can be trained according to the user features corresponding to the users with successful historical conversions and the content features of the historical recommended content recommended to these users, so as to obtain recommended content that can promote the conversion of other users similar to these users.

[0070] Optionally, the recommendation model can also be a clustering model, a logistic regression model, a genetic algorithm model, a decision tree model, or an Xgboost model, etc.

[0071] In still another embodiment of the present invention, the recommendation model can include a user sub-model for performing clustering analysis on users and a content sub-model for determining corresponding recommended content according to user features. Specifically, the user sub-model and the content sub-model can be neural network models. The user sub-model can be trained according to user features and behavior features, and output low-dimensional user feature vectors. The content sub-model is used to train according to the aforementioned output user feature vectors and content features to obtain recommended content.

[0072] In yet another embodiment of the present invention, the recommendation model can also be iteratively updated according to the conversion results of all historically recommended users for the recommended content. For example, positive samples are generated based on the user characteristics, operation behavior data, and historical recommended content corresponding to the target historical users, and the user sub-model and the content sub-model are trained based on the positive samples. At the same time, when the historical user fails to convert, negative samples are generated based on the user characteristics, recommended content, and historical operation behavior data corresponding to the historical user, the content sub-model is updated based on the negative samples, and content recommendations are continued for the users who have not successfully converted based on the updated user sub-model and content sub-model. Such a cycle continues until it is determined that the user has successfully converted or there are no optional conversion behaviors, at which point the recommendation stops.

[0073] The recommendation model includes a user classification model and a content generation model.

[0074] Therefore, in one embodiment of the present invention, step 30 further includes:

[0075] Step 301: Train the user classification model according to the updated user feature sample pool.

[0076] In one embodiment of the present invention, a certain number of user feature samples are taken from the updated user feature sample pool at a certain positive-negative sample ratio (such as 7:3) to form a training sample set and a test sample set.

[0077] Input the training sample set into the user classification model to train the user classification model.

[0078] Then input the test sample set into the trained user classification model to obtain the output user classification result.

[0079] Modify the model parameters of the user classification model according to the output user classification result and the labels (indicating whether the corresponding user has successfully converted) of each sample in the test sample set.

[0080] In yet another embodiment of the present invention, the user classification model can be one of a fully connected neural network, a convolutional neural network, a residual network, a generative adversarial network, an autoencoder, a recurrent neural network, and a long short-term memory network.

[0081] Optionally, the user classification model can also be a clustering model, a logistic regression model, a genetic algorithm model, a decision tree model, or an Xgboost model, etc.

[0082] Step 302: Train the content generation model according to the updated recommended content sample pool.

[0083] In one embodiment of the present invention, similar to the training process of the user classification model, a certain number of recommended content samples are taken from the updated recommended content sample pool at a certain positive-negative sample ratio (such as 7:3) to form a training sample set and a test sample set.

[0084] Input the training sample set into the content generation model to train the content generation model.

[0085] Then input the test sample set into the trained user classification model to obtain the output content recommendation result.

[0086] According to the output content recommendation result and the labels corresponding to each sample in the test sample set (indicating whether the user is successfully converted after the content is recommended), the model parameters of the content generation model are corrected.

[0087] In another embodiment of the present invention, the content generation model can be one of a fully connected neural network, a convolutional neural network, a residual network, a generative adversarial network, an autoencoder, a recurrent neural network, and a long short-term memory network.

[0088] Optionally, the content generation model can also be a clustering model, a logistic regression model, a genetic algorithm model, a decision tree model, or an Xgboost model, etc.

[0089] Step 40: Recommend to the user according to the updated recommendation model.

[0090] In one embodiment of the present invention, specifically, the neural network model can be one of a fully connected neural network, a convolutional neural network, a residual network, a generative adversarial network, an autoencoder, a recurrent neural network, and a long short-term memory network.

[0091] Specifically, the user classification model includes an input layer, a hidden layer, and an output layer connected in sequence, where each layer is composed of multiple neuron nodes connected.

[0092] Step 40 further includes: Step 401: Input the user features into the updated user classification model to obtain a user feature embedding vector.

[0093] In one embodiment of the present invention, embedding processing is performed on the outputs of all the processing layers. While reducing the data dimension, data features are extracted, which can accelerate the training efficiency of the recommendation model.

[0094] Step 403: Input the user feature embedding vector into the updated content generation model to obtain the target recommended content.

[0095] Step 404: Recommend the target recommended content to the user.

[0096] In one embodiment of the present invention, target recommended content is recommended to users so that they use the target recommended content. Since the target recommended content is obtained by updating the recommendation model based on user feature information and content information corresponding to users with successful historical conversions, users are more interested in converting to the target recommended content, thereby increasing the probability of completing user conversion during use. At the same time, it also avoids the existing problem of simply performing group recommendations based on big data for users, which has a low degree of personalization and targeting. By promoting user conversion during use, user experience is also guaranteed.

[0097] In another embodiment of the present invention, considering that the content preferences or consumption behavior habits of some users may be relatively niche, or the number of samples in the sample pool corresponding to some users is insufficient, resulting in that the recommendation effect of the recommendation model for such users may not reach the level of prompting the user to convert. In order to improve the content recommendation accuracy and conversion rate of the entire user group, after step 40, the following is further included:

[0098] When it is determined that the number of recommendations for the user is greater than a preset threshold and the operation data does not include the conversion behavior, content is recommended to the user according to another recommendation model, wherein the other recommendation model is trained based on a new user feature sample pool and a new recommended content sample pool; the new user feature sample pool and the new recommended content sample pool are determined based on user features, historical recommended content and historical operation data corresponding to historical users whose number of recommendations is greater than the threshold.

[0099] In one embodiment of the present invention, when it is determined that the number of recommendations for the user is greater than a preset number threshold and the operation data does not include the conversion behavior, it can be determined that the current recommendation model has a poor recommendation effect on the user. If the current recommendation model is still used to continue to recommend, it will not only fail to achieve the effect of promoting conversion, but may also cause user plans and reduce user experience. Therefore, it is necessary to replace a new recommendation model to update the user. Among them, the model type, structure and training process of another recommendation model can be the same or similar to the aforementioned recommendation model, and will not be repeated. The difference is that the construction of the training sample pool for another recommendation model is different from that of the aforementioned recommendation model.

[0100] When training another recommendation model, users who have failed to convert through repeated recommendations can actually be classified into a user group with similar characteristics, and a new user feature sample pool and a new recommended content sample pool can be determined based on the user features, historical recommended content, and historical operation data corresponding to historical users whose number of recommendations is greater than the threshold, thereby training the user classification model and content generation model in the other recommendation model based on the new user feature sample pool and the new recommended content sample pool.

[0101] In another embodiment of the present invention, another recommendation model can also be obtained by transfer learning of the recommendation model. Based on the training results of the recommendation model, transfer learning is performed on it by inputting an updated sample pool for niche populations, resulting in another recommendation model, so that the other recommendation model will perform better in scenarios where recommendations are made for users with difficult conversions.

[0102] In another embodiment of the present invention, the process of content recommendation for users can refer to Figure 2 .

[0103] Such as Figure 2 shown, during the training process of the recommendation model, content features and user features corresponding to users who have successfully upgraded to members are obtained.

[0104] According to the content features corresponding to users who have successfully upgraded to members, a questionnaire is constructed, that is, a recommended song list. Feature vectors of the involved songs are extracted, and the extracted feature vectors are fused with the user features corresponding to users who have successfully upgraded to members to form a content questionnaire matrix, and a questionnaire pool is formed according to multiple content questionnaire matrices.

[0105] The user features corresponding to users who have successfully upgraded to members and the user behavior data of these users on the recommended songs are fused, and the fused vector is input into the user neural network model to obtain the user Embedding vector corresponding to users who have successfully upgraded to members.

[0106] The questionnaire pool and the user Embedding vector corresponding to users who have successfully upgraded to members are used as inputs to the questionnaire neural network model for training to obtain a model that can select a questionnaire that can successfully influence users to upgrade to members from the questionnaire pool according to the user Embedding vector.

[0107] During the use of the model, first, the user features corresponding to the user and the user behavior data during the song listening process are obtained, and they are input into the user neural network model to obtain the user Embedding vector corresponding to the user. The user Embedding vector corresponding to the user is input into the questionnaire neural network model to obtain the questionnaire to be recommended to the user, that is, the recommended song list.

[0108] Record the operations of the user on the recommended song list, and record whether the user has performed a membership upgrade operation based on the recommended song list.

[0109] When the user is using the song content corresponding to the questionnaire matrix recommended to him and selects to upgrade the membership level to obtain better services, add the questionnaire data to the positive sample of the content questionnaire pool, and use the feature vector of the user to upgrade the user neural network model. When the recommended questionnaire content fails to upgrade the user's membership level, add the questionnaire data to the negative sample of the content questionnaire pool and regenerate a questionnaire content data to return to the user. In such a cycle, when it is determined that the user has been successfully upgraded to a higher membership level or the user's consumption needs no longer require a higher membership level, stop the update of the model and content recommendation.

[0110] The content recommendation method provided by the embodiments of the present invention obtains the operation data of the user for the initial recommended content; updates the sample pool of the recommendation model according to the user characteristics of the user and the operation data, where the recommendation model is used for content recommendation. When the operation data includes a preset conversion behavior, update the positive sample pool of the recommendation model according to the operation data. When the conversion behavior is not included, update the negative sample pool of the recommendation model according to the operation data; the conversion behavior includes an upgrade behavior and / or a payment behavior; the sample pool includes the positive sample pool and the negative sample pool; train the recommendation model according to the updated sample pool to obtain an updated recommendation model; recommend to the user according to the updated recommendation model. Different from the prior art solution of directly clustering users according to big data and then performing content recommendation, which has the problem of low accuracy resulting in low user conversion rate, the embodiments of the present invention update the positive and negative samples in the sample pool of the recommendation model according to whether the user has a conversion behavior for the recommended content, that is, simulate the questionnaire survey process of the user's interest through the user's behavior on the recommended content, and determine the user's need and interest degree for implementing the conversion behavior according to the user's operation behavior on the recommended content, so as to convert the user into a member user more in line with the user's consumption needs and interests during the process of the user consuming the recommended content, while promoting the user to implement the conversion behavior, ensuring that operation indicators such as the number of users and user duration do not decrease and do not affect the user experience. The embodiments of the present invention can improve the accuracy of content recommendation, thereby promoting the user to implement conversion behaviors such as payment and upgrade for the content recommended to him.

[0111] Figure 3 The structural schematic diagram of the content recommendation device provided by the embodiments of the present invention is shown. As Figure 3 shown, the device 400 includes: an acquisition module 401, an update module 402, a training module 403, and a recommendation module 404.

[0112] Among them, the acquisition module 401 is used to acquire the operation data of the user for the initial recommended content;

[0113] An update module 402, configured to update a sample pool of a recommendation model according to the user characteristics of the user and the operation data, where the recommendation model is used for content recommendation. When the operation data includes a preset conversion behavior, the positive sample pool of the recommendation model is updated according to the operation data; when the conversion behavior is not included, the negative sample pool of the recommendation model is updated according to the operation data; the conversion behavior includes an upgrade behavior and / or a payment behavior; the sample pool includes the positive sample pool and the negative sample pool;

[0114] A training module 403, configured to train the recommendation model according to the updated sample pool to obtain an updated recommendation model;

[0115] A recommendation module 404, configured to recommend to the user according to the updated recommendation model.

[0116] The operations performed by the content recommendation apparatus provided in the embodiment of the present invention are substantially the same as the operation processes of the foregoing method embodiments, and will not be described in detail.

[0117] The content recommendation apparatus provided in the embodiment of the present invention obtains operation data of a user for initial recommended content; updates a sample pool of a recommendation model according to the user characteristics of the user and the operation data, where the recommendation model is used for content recommendation. When the operation data includes a preset conversion behavior, the positive sample pool of the recommendation model is updated according to the operation data; when the conversion behavior is not included, the negative sample pool of the recommendation model is updated according to the operation data; the conversion behavior includes an upgrade behavior and / or a payment behavior; the sample pool includes the positive sample pool and the negative sample pool; trains the recommendation model according to the updated sample pool to obtain an updated recommendation model; recommends to the user according to the updated recommendation model. Different from the prior art solution of directly performing content recommendation after clustering users according to big data, which has the problem of low accuracy resulting in low user conversion rate, the embodiment of the present invention correspondingly updates the positive and negative samples in the sample pool of the recommendation model according to whether the user has a conversion behavior for the recommended content, that is, simulates the questionnaire survey process of user interests through the user's behavior on the recommended content, and determines the user's demand and interest degree for implementing the conversion behavior according to the user's operation behavior on the recommended content, so as to convert the user into a member user more in line with the user's consumption needs and interests during the process of the user consuming the recommended content, while promoting the user to implement the conversion behavior, ensuring that operation indicators such as the number of users and user duration do not decrease, and not affecting the user experience. The embodiment of the present invention can improve the accuracy of content recommendation, thereby promoting the user to implement conversion behaviors such as payment and upgrade for the content recommended to the user.

[0118] Figure 4The structural schematic diagram of the content recommendation device provided by the embodiments of the present invention is shown. The specific embodiments of the present invention do not limit the specific implementation of the content recommendation device.

[0119] As Figure 4 shown, the content recommendation device may include: a processor 502, a communications interface 504, a memory 506, and a communication bus 508.

[0120] Among them: The processor 502, the communications interface 504, and the memory 506 communicate with each other through the communication bus 508. The communications interface 504 is used to communicate with network elements of other devices such as clients or other servers. The processor 502 is used to execute the program 510, and specifically can execute the relevant steps in the above embodiments of the content recommendation method.

[0121] Specifically, the program 510 may include program code, and the program code includes computer-executable instructions.

[0122] The processor 502 may be a central processing unit CPU, or a specific integrated circuit ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present invention. One or more processors included in the content recommendation device may be of the same type of processor, such as one or more CPUs; or may be of different types of processors, such as one or more CPUs and one or more ASICs.

[0123] The memory 506 is used to store the program 510. The memory 506 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk memory.

[0124] The program 510 can specifically be called by the processor 502 to enable the content recommendation device to perform the following operations:

[0125] Obtain operation data of the user for the initial recommended content;

[0126] Update the sample pool of the recommendation model according to the user characteristics of the user and the operation data, where the recommendation model is used for content recommendation. When the operation data includes a preset conversion behavior, update the positive sample pool of the recommendation model according to the operation data. When the conversion behavior is not included, update the negative sample pool of the recommendation model according to the operation data; the conversion behavior includes an upgrade behavior and / or a payment behavior; the sample pool includes the positive sample pool and the negative sample pool;

[0127] Train the recommendation model according to the updated sample pool to obtain an updated recommendation model;

[0128] Recommend to the user according to the updated recommendation model.

[0129] The operations performed by the content recommendation device provided by the embodiments of the present invention are substantially the same as those in the foregoing method embodiments, and will not be elaborated herein.

[0130] The content recommendation device provided by the embodiments of the present invention obtains operation data of a user for initial recommended content; updates a sample pool of a recommendation model according to the user characteristics of the user and the operation data, where the recommendation model is used for content recommendation. When the operation data includes a preset conversion behavior, update the positive sample pool of the recommendation model according to the operation data. When the conversion behavior is not included, update the negative sample pool of the recommendation model according to the operation data; the conversion behavior includes an upgrade behavior and / or a payment behavior; the sample pool includes the positive sample pool and the negative sample pool; train the recommendation model according to the updated sample pool to obtain an updated recommendation model; recommend to the user according to the updated recommendation model. Different from the prior art solution of directly performing content recommendation after clustering users based on big data, which has the problem of low accuracy resulting in low user conversion rate, the embodiments of the present invention correspondingly update the positive and negative samples in the sample pool of the recommendation model according to whether the user has a conversion behavior for the recommended content, that is, simulate the process of a user interest questionnaire through the user's behavior on the recommended content, and determine the user's demand and interest degree for implementing the conversion behavior according to the user's operation behavior for the recommended content, so as to convert the user into a member user more in line with the user's consumption needs and interests during the process of the user consuming the recommended content, while promoting the user to implement the conversion behavior, ensuring that operation indicators such as the number of users and user duration do not decrease, and not affecting the user experience. The embodiments of the present invention can improve the accuracy of content recommendation, thereby promoting the user to implement conversion behaviors such as payment and upgrade for the content recommended to the user.

[0131] The embodiments of the present invention provide a computer-readable storage medium, and the storage medium stores at least one executable instruction. When the executable instruction runs on a content recommendation device, the content recommendation device is enabled to execute the content recommendation method in any of the foregoing method embodiments.

[0132] The executable instruction can specifically be used to enable the content recommendation device to perform the following operations:

[0133] Obtain operation data of a user for initial recommended content;

[0134] Update the sample pool of the recommendation model according to the user characteristics of the user and the operation data, where the recommendation model is used for content recommendation. When the operation data includes a preset conversion behavior, update the positive sample pool of the recommendation model according to the operation data. When the conversion behavior is not included, update the negative sample pool of the recommendation model according to the operation data; the conversion behavior includes an upgrade behavior and / or a payment behavior; the sample pool includes the positive sample pool and the negative sample pool;

[0135] Train the recommendation model according to the updated sample pool to obtain an updated recommendation model;

[0136] Recommend to the user according to the updated recommendation model.

[0137] The operations performed by the executable instructions stored in the computer storage medium provided in the embodiments of the present invention are substantially the same as the operation processes of the foregoing method embodiments, and will not be described in detail.

[0138] The executable instructions stored in the computer storage medium provided in the embodiments of the present invention obtain operation data of the user for the initial recommended content; update the sample pool of the recommendation model according to the user characteristics of the user and the operation data, where the recommendation model is used for content recommendation. When the operation data includes a preset conversion behavior, update the positive sample pool of the recommendation model according to the operation data. When the conversion behavior is not included, update the negative sample pool of the recommendation model according to the operation data; the conversion behavior includes an upgrade behavior and / or a payment behavior; the sample pool includes the positive sample pool and the negative sample pool; train the recommendation model according to the updated sample pool to obtain an updated recommendation model; recommend to the user according to the updated recommendation model. Different from the solution of directly performing content recommendation after clustering users according to big data in the prior art, which has the problem of low accuracy resulting in low user conversion rate, the embodiments of the present invention update the positive and negative samples in the sample pool of the recommendation model according to whether the user has a conversion behavior for the recommended content, that is, simulate the questionnaire survey process of user interests through the user's behavior on the recommended content, and determine the user's demand and interest degree for implementing the conversion behavior according to the user's operation behavior for the recommended content, so as to convert the user into a member user more in line with the user's consumption needs and interests during the process of the user consuming the recommended content, while promoting the user to implement the conversion behavior, ensuring that operation indicators such as the number of users and user duration do not decrease, and not affecting the user experience. The embodiments of the present invention can improve the accuracy of content recommendation, thereby promoting the user to implement conversion behaviors such as payment and upgrade for the content recommended to him.

[0139] The embodiments of the present invention provide a content recommendation device for executing the above content recommendation method.

[0140] An embodiment of the present invention provides a computer program, which can be called by a processor to cause a content recommendation device to execute the content recommendation method in any of the above method embodiments.

[0141] An embodiment of the present invention provides a computer program product. The computer program product includes a computer program stored on a computer-readable storage medium. The computer program includes program instructions. When the program instructions run on a computer, the computer is caused to execute the content recommendation method in any of the above method embodiments.

[0142] The algorithms or displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The structure required to construct such systems will be apparent from the above description. In addition, the embodiments of the present invention are not directed to any particular programming language. It should be understood that the content of the present invention described herein can be implemented using various programming languages, and the description of the specific language above is to disclose the best mode of the present invention.

[0143] In the specification provided herein, a large number of specific details are set forth. However, it can be understood that the embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0144] Similarly, it should be understood that, in order to streamline the present invention and assist in understanding one or more of the various inventive aspects, in the above description of the exemplary embodiments of the present invention, the various features of the embodiments of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, the disclosed method should not be construed as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim.

[0145] Those skilled in the art can understand that the modules in the devices in the embodiments can be adaptively changed and disposed in one or more devices different from the embodiments. The modules or units or components in the embodiments can be combined into a module or unit or component, and can be divided into multiple sub-modules or sub-units or sub-components. Except that at least some of such features and / or processes or units are mutually exclusive, any combination can be used to combine all the features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all the processes or units of any method or device so disclosed. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) can be replaced by an alternative feature that provides the same, equivalent, or similar purpose.

[0146] It should be noted that the above embodiments illustrate the present invention rather than limit the present invention, and those skilled in the art can design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in the claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention can be implemented by means of hardware including several different elements and by means of a suitably programmed computer. In a unit claim listing several devices, several of these devices can be embodied by the same item of hardware. The use of the words first, second, and third, etc. does not denote any order. These words can be interpreted as names. The steps in the above embodiments, unless otherwise specifically stated, should not be construed as limiting the order of execution.

Claims

1. A content recommendation method, characterized in that, The method includes: Obtaining operation data of the user for the initial recommended content; Updating the sample pool of the recommendation model according to the user characteristics of the user and the operation data; the sample pool includes a positive sample pool and a negative sample pool; the sample pool of the recommendation model further includes a recommended content sample pool and a user characteristics sample pool; the updating the sample pool of the recommendation model according to the user characteristics of the user and the operation data includes: when the operation data includes a preset conversion behavior, generating a recommended content positive sample and a user characteristics positive sample respectively according to the operation behavior, the initial recommended content, and the user characteristics; adding the recommended content positive sample and the user characteristics positive sample to the positive sample pool of the recommended content sample pool and the positive sample pool of the user characteristics sample pool respectively to update the positive sample pool of the recommendation model; when the operation data does not include the conversion behavior, generating a content questionnaire negative sample according to the operation behavior, the initial recommended content, and the user characteristics; adding the content questionnaire negative sample to the negative sample pool of the recommended content sample pool to update the negative sample pool of the recommendation model; the recommendation model is used for content recommendation; the conversion behavior includes an upgrade behavior and / or a payment behavior; Training the recommendation model according to the updated sample pool to obtain an updated recommendation model, the recommendation model includes a user classification model and a content generation model, the training the recommendation model according to the updated sample pool to obtain an updated recommendation model includes: training the user classification model according to the updated user characteristics sample pool; training the content generation model according to the updated recommended content sample pool; Recommending to the user according to the updated recommendation model, including: inputting the user characteristics into the updated user classification model to obtain a user characteristics embedding vector; inputting the user characteristics embedding vector into the updated content generation model to obtain a target recommended content; recommending the target recommended content to the user; When it is determined that the number of recommendations for the user is greater than a preset number threshold and the operation data does not include the conversion behavior, performing content recommendation for the user according to another recommendation model, the another recommendation model is trained according to a new user characteristics sample pool and a new recommended content sample pool; the new user characteristics sample pool and the new recommended content sample pool are determined according to the user characteristics, the historical recommended content, and the historical operation data of historical users whose number of recommendations is greater than the number threshold; When it is determined that the user has been successfully converted or there is no optional conversion behavior, stop recommending.

2. The method according to claim 1, wherein The operation data includes at least one of the residence duration, operation frequency, and trial duration of the user on the recommended content; The user characteristics include at least one of the current user level, user activity, and user content preference.

3. A content recommendation device, characterized in that, The device includes: An obtaining module, configured to obtain operation data of the user for the initial recommended content; An update module, configured to update a sample pool of a recommendation model according to the user characteristics of the user and the operation data; the sample pool includes a positive sample pool and a negative sample pool; the sample pool of the recommendation model further includes a recommended content sample pool and a user characteristic sample pool; the updating the sample pool of the recommendation model according to the user characteristics of the user and the operation data includes: when the operation data includes a preset conversion behavior, generating a recommended content positive sample and a user characteristic positive sample according to the operation behavior, the initial recommended content, and the user characteristics respectively; adding the recommended content positive sample and the user characteristic positive sample to the positive sample pool of the recommended content sample pool and the positive sample pool of the user characteristic sample pool respectively to update the positive sample pool of the recommendation model; when the operation data does not include the conversion behavior, generating a content questionnaire negative sample according to the operation behavior, the initial recommended content, and the user characteristics; adding the content questionnaire negative sample to the negative sample pool of the recommended content sample pool to update the negative sample pool of the recommendation model; the recommendation model is used for content recommendation; the conversion behavior includes an upgrade behavior and / or a payment behavior; A training module, configured to train the recommendation model according to the updated sample pool to obtain an updated recommendation model, the recommendation model includes a user classification model and a content generation model, the training the recommendation model according to the updated sample pool to obtain an updated recommendation model includes: training the user classification model according to the updated user characteristic sample pool; training the content generation model according to the updated recommended content sample pool; A recommendation module, configured to recommend to the user according to the updated recommendation model, when it is determined that the number of recommendations for the user is greater than a preset number threshold and the operation data does not include the conversion behavior, performing content recommendation for the user according to another recommendation model, the another recommendation model is trained according to a new user characteristic sample pool and a new recommended content sample pool; the new user characteristic sample pool and the new recommended content sample pool are determined according to the user characteristics, historical recommended content, and historical operation data of historical users whose recommendation times are greater than the number threshold; when it is determined that the user is successfully converted or there is no optional conversion behavior, the recommendation is stopped; the recommendation module includes: a first input unit, configured to input the user characteristics into the updated user classification model to obtain a user characteristic embedding vector; a second input unit, configured to input the user characteristic embedding vector into the updated content generation model to obtain a target recommended content; a recommendation unit, configured to recommend the target recommended content to the user.

4. A content recommendation device, characterized in that, Comprising: A processor, a memory, a communication interface, and a communication bus, and the processor, the memory, and the communication interface complete communication with each other through the communication bus; The memory is used to store at least one executable instruction, and the executable instruction causes the processor to execute the content recommendation method according to any one of claims 1-2.

5. A computer-readable storage medium, characterized in that, At least one executable instruction is stored in the storage medium, and the executable instruction causes the processor to execute the content recommendation method according to any one of claims 1-2.

Images