Content recommendation method, device, system, equipment, medium and program product

Abstract

The present disclosure provides a content recommendation method, device, system, electronic device, computer-readable storage medium and computer program product in accordance with an embodiment of the present disclosure, and relates to the field of computer technology. The method includes: receiving a model selection operation of a first user, selecting at least one first recommendation model for display; responding to a model merging instruction of the first user, merging at least one first recommendation model corresponding to the model merging instruction into the current recommendation model to generate a first merged model; responding to a merge completion instruction of the first user, determining target recommended content for the first user based on the obtained first merged model. The present disclosure allows users to autonomously and purposefully conduct model training in combination with recommendation models of other users, so as to perform content recommendations based on the recommendation model obtained after merge training, which can effectively improve the matching degree and accuracy of content recommendations.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a content recommendation method, a content recommendation device, a content recommendation system, an electronic device, a computer-readable storage medium, and a computer program product. Background Art

[0002] With the rapid development of computer networks, computer networks have been widely used in various fields of people's lives. For example, online shopping, watching videos and browsing news information through the Internet have become people's daily lives. Due to the explosion of network information content, people are increasingly hoping to mine the information content they are interested in from massive network resources. Therefore, it is very important and urgent to accurately mine user preferences and make efficient and accurate content recommendations.

[0003] At present, algorithm recommendation systems can be divided into three categories: collaborative filtering systems, content-based recommendation systems, and hybrid recommendation models; among them, collaborative filtering systems only use the interaction information between users and recommended content to generate recommendations, content-based recommendation systems use user preferences and / or recommended content preferences to generate recommendations, and hybrid recommendation models use interaction information, user and recommended content metadata to generate recommendations. Models in the above categories have corresponding limitations, such as data sparsity, cold start of users and recommended content. Summary of the invention

[0004] The present disclosure provides a content recommendation method, a content recommendation device, a content recommendation system and an electronic device, a computer-readable storage medium and a computer program product, so as to at least solve the problem that the recommendation system in the related art is too dependent on user historical data, which increases the user's usage cost, and the user cannot intervene in the recommendation model, resulting in low accuracy of recommended content and easy increase in user loss. The technical solution of the present disclosure is as follows:

[0005] According to a first aspect of an embodiment of the present disclosure, a content recommendation method is provided, comprising: receiving a model selection operation of a first user, and selecting at least one first recommendation model for display; in response to a model merging instruction of the first user, merging at least one first recommendation model corresponding to the model merging instruction into a current recommendation model to generate a first merged model; in response to a merge completion instruction of the first user, determining target recommended content for the first user based on the obtained first merged model.

[0006] In an exemplary embodiment of the present disclosure, the above method also includes: in response to a model removal instruction from the first user, removing the merged recommendation model in the first merged model to generate a second merged model; in response to a removal completion instruction from the first user, determining target recommended content for the first user based on the obtained second merged model.

[0007] In an exemplary embodiment of the present disclosure, the method further includes: the first recommendation model is a recommendation model corresponding to a second user; and the second user is a user who turns on a model sharing option.

[0008] In an exemplary embodiment of the present disclosure, a model selection operation of a first user is received, and at least one first recommended model is selected for display, and the method also includes: receiving a model selection operation performed by the first user based on a model selection area, and determining a model keyword corresponding to the model selection operation; and selecting at least one first recommended model from a model library according to the model keyword to display on the interactive interface.

[0009] In an exemplary embodiment of the present disclosure, target recommended content is determined for the first user based on the obtained first merged model, including: determining the model weight of each of the recommendation models in the first merged model; determining the content acquisition proportion of the first user for each of the recommendation models according to each of the model weights; and allocating a corresponding content request quantity to each of the recommendation models according to the content acquisition proportion, so as to determine the target recommended content according to the content request quantity.

[0010] In an exemplary embodiment of the present disclosure, determining the model weight of each recommendation model in the first merged model includes: obtaining historical recommendation data of each recommendation model in the first merged model; the historical recommendation data includes multiple recommendation influencing factors; determining the factor weight of each recommendation influencing factor based on the historical recommendation data; and performing weighted calculation based on multiple factor weights to determine the model weight.

[0011] In an exemplary embodiment of the present disclosure, determining the model weight of each of the recommendation models in the first merged model includes: determining the content label of each of the recommendation models in the first merged model; determining the model priority of each of the recommendation models according to the content label; and determining the model weight corresponding to each of the recommendation models according to the model priority.

[0012] In an exemplary embodiment of the present disclosure, the above method also includes: determining associated users associated with the first user, and generating an associated user group based on the first user and the associated users; in response to a model training operation of any user in the associated user group, determining associated group recommended content corresponding to the associated user group; and displaying the associated group recommended content to any user in the associated user group.

[0013] According to a second aspect of an embodiment of the present disclosure, a content recommendation system is provided, comprising: a user end, configured to provide a model operation interface and display target recommended content, and receive a user's model operation based on the model operation interface; the model operation includes a model selection operation, a model removal operation, and a model sharing operation; a model training end, configured to merge a first recommendation model determined based on the model selection operation into a current recommendation model to generate a first merged model; and according to a model removal operation of the first merged model, remove at least one recommended model in the first merged model to generate a second merged model; and a content recommendation platform, configured to determine target recommended content for the user based on the merged model, and send the target recommended content to the user end.

[0014] According to a third aspect of an embodiment of the present disclosure, there is provided a content recommendation device, comprising: a model display module, configured to execute a model selection operation received from a first user, and select at least one first recommendation model for display; a model merging module, configured to execute a model merging instruction in response to the first user, and merge at least one first recommendation model corresponding to the model merging instruction into a current recommendation model to generate a first merged model; and a first content recommendation module, configured to execute a merge completion instruction in response to the first user, and determine target recommended content for the first user based on the obtained first merged model.

[0015] In an exemplary embodiment of the present disclosure, the content recommendation device also includes a second content recommendation module, which is configured to execute: in response to a model removal instruction from the first user, remove the merged recommendation model in the first merged model to generate a second merged model; in response to a removal completion instruction from the first user, determine target recommended content for the first user based on the obtained second merged model.

[0016] In an exemplary embodiment of the present disclosure, the model display module also includes a model display unit, which is configured to perform: receiving a model selection operation performed by the first user based on a model selection area, and determining a model keyword corresponding to the model selection operation; and selecting at least one of the first recommended models from a model library based on the model keyword and displaying it on the interactive interface.

[0017] In an exemplary embodiment of the present disclosure, the first content recommendation module includes a first content recommendation unit, which is configured to perform: determining the model weight of each recommendation model in the first merged model; determining the content acquisition proportion of the first user for each recommendation model according to each model weight; and allocating a corresponding content request quantity to each recommendation model according to the content acquisition proportion, so as to determine the target recommended content according to the content request quantity.

[0018] In an exemplary embodiment of the present disclosure, the first content recommendation unit includes a first weight determination subunit, and the first weight determination subunit is configured to perform: obtaining historical recommendation data of each recommendation model in the first merged model; the historical recommendation data includes multiple recommendation influencing factors; determining the factor weight of each recommendation influencing factor based on the historical recommendation data; and performing weighted calculation based on multiple factor weights to determine the model weight.

[0019] In an exemplary embodiment of the present disclosure, the first content recommendation unit includes a second weight determination subunit, and the second weight determination subunit is configured to perform: determining the content label of each of the recommendation models in the first merged model; determining the model priority of each of the recommendation models according to the content label; and determining the model weight corresponding to each of the recommendation models according to the model priority.

[0020] In an exemplary embodiment of the present disclosure, the content recommendation device also includes a third content recommendation module, which is configured to perform: determining associated users associated with the first user, and generating an associated user group based on the first user and the associated users; determining associated group recommended content corresponding to the associated user group in response to a model training operation of any user in the associated user group; and displaying the associated group recommended content to any user in the associated user group.

[0021] According to a fourth aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement any one of the content recommendation methods described above.

[0022] According to a fifth aspect of the present disclosure, a computer-readable storage medium is provided. When instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute any one of the content recommendation methods described above.

[0023] According to a sixth aspect of an embodiment of the present disclosure, a computer program product is provided, including a computer program / instruction, wherein the computer program / instruction implements any one of the above-mentioned content recommendation methods when executed by a processor.

[0024] The technical solution provided by the embodiments of the present disclosure brings at least the following beneficial effects:

[0025] The content recommendation method disclosed in the present invention, on the one hand, allows users to independently select a recommendation model for model training, and improves the accuracy of recommended content distribution by autonomously and quickly learning the recommendation model. On the other hand, users are no longer unaware of the recommendation model, and providing users with the ability to autonomously control the recommendation algorithm model learning can greatly improve the user's sense of control. On the other hand, by allowing users to autonomously and quickly learn the recommendation algorithm, the adaptation cost of cold-start users is greatly reduced, which can greatly reduce the churn rate of related users.

[0026] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description are used to explain the principles of the present disclosure, and do not constitute improper limitations on the present disclosure.

[0028] Figure 1 The present invention is a flowchart of a content recommendation method according to an exemplary embodiment.

[0029] Figure 2 The present invention is an interface diagram showing a user opening a model sharing option according to an exemplary embodiment.

[0030] Figure 3 The diagram is a schematic diagram showing a first user selecting a recommendation model based on a second user for model training according to an exemplary embodiment.

[0031] Figure 4 The diagram is a schematic diagram showing a first user selecting a recommendation model through a user interaction interface for model training according to an exemplary embodiment.

[0032] Figure 5 This is an interface diagram showing a first user and other users forming an associated user group according to an exemplary embodiment.

[0033] Figure 6 The diagram is a structural diagram of a content recommendation system according to an exemplary embodiment.

[0034] Figure 7 The present invention is a block diagram of a content recommendation device according to an exemplary embodiment.

[0035] Figure 8 A block diagram of an electronic device according to an exemplary embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0036] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.

[0037] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0038] Existing algorithm recommendation systems can include collaborative filtering systems, content-based recommendation systems, and hybrid recommendation model systems. The two major types of recommendation systems can include: content-based recommendation systems and collaborative filtering systems. Collaborative filtering systems can build models based on user historical behaviors, such as content that users have browsed, liked, and commented on, combined with similar decisions of other users. This model can be used to predict what content users may be interested in or the degree of interest in the content. Content-based recommendation systems can use some discrete features of the content to recommend similar content with similar properties.

[0039] However, the recommendation algorithms currently used are very dependent on the user's historical data as a basis for judgment. It is difficult to determine the real needs of "cold start" users and users whose "hobbies have changed". The recommendation algorithm can only continuously collect user usage habits to continuously train the algorithm model, which increases the user's usage cost and also increases the possibility of user loss. For users, algorithm training is imperceptible to users, but it has a great impact on the content they see. Many users do not know how their recommendation algorithm models are formed, and they cannot intervene in the models themselves. Users have a weak sense of control over the models.

[0040] The following uses two user groups as examples to illustrate: For initial users of a certain platform, the content they see is completely recommended according to the cold start content model. For such users, these contents may not be what they want to see, but the users cannot intervene and control it. For users who switch platforms, cold start content is often popular content in the past, but for users who switch platforms, many contents are content they have already seen. In the current situation where content homogeneity is serious, this recommendation model is very likely to cause user loss.

[0041] Based on this, according to the embodiments of the present disclosure, a content recommendation method, a content recommendation device, a content recommendation system, an electronic device, a computer-readable storage medium, and a computer program product are proposed.

[0042] Figure 1 is a flow chart of a content recommendation method according to an exemplary embodiment. Figure 1 As shown, the content recommendation method can be used in a computer device, wherein the computer device described in the present disclosure may include mobile terminal devices such as mobile phones, tablet computers, laptop computers, PDAs, and fixed terminal devices such as desktop computers. This exemplary embodiment uses the method applied to a computer device as an example, and it can be understood that the method can also be applied to a server, and can also be applied to a system including a computer device and a server, and is implemented through the interaction between the computer device and the server. Specifically, the following steps are included.

[0043] In step S110, a model selection operation of a first user is received, and at least one first recommended model is selected for display.

[0044] In some exemplary embodiments of the present disclosure, the first user may be a user who autonomously performs a recommendation model training operation, for example, the first user may be a new user who has just started using a certain recommendation platform. The model selection operation may be a user operation in which the first user selects a recommendation model. The first recommendation model may be a recommendation model that can be selected by the first user.

[0045] For a certain recommendation platform, if the first user is not interested in the current recommended content and wants to change the content recommended to him by the recommendation platform, he can independently perform the training operation of the recommendation model. At this time, the first user can select one or more first recommendation models to display in the user interaction interface. For example, the first user can be a new user who has recently completed the platform registration operation of the recommendation platform; the first user can also be a user who wants to update the relevant recommended content in the recommendation platform. Specifically, the first user can obtain the recommendation model of other users as the first recommendation model through the user homepage of other users. The first user can also select a specific user from the personal user homepage and use the recommendation model of the specific user as the first recommendation model. When receiving the model selection operation of the first user, the user interaction interface can display the selected first recommendation model.

[0046] In step S120, in response to the model merging instruction of the first user, at least one first recommendation model corresponding to the model merging instruction is merged into the current recommendation model to generate a first merged model.

[0047] In some exemplary embodiments of the present disclosure, the model merging instruction may be a processing instruction corresponding to a model merging operation of multiple recommendation algorithm models. The current recommendation model may be a recommendation algorithm model initially corresponding to the first user. The first merged model may be a recommendation model obtained after model merging training of the first recommendation model and the current recommendation model.

[0048] After the first user performs a model selection operation, one or more first recommended models will be displayed in the user interaction interface. At this time, the first user can select other recommended models that he wants to merge and train with the current recommended model. When the model merging operation of the first user is received, a corresponding model merging instruction will be generated according to the model merging operation. In response to the model merging instruction of the first user, the recommendation platform merges one or more first recommended models corresponding to the model merging instruction into the current recommended model, so that the first recommended model and the current recommended model are trained for model merging to generate a first merged model.

[0049] For example, the current recommendation model corresponding to the first user is "Recommendation Algorithm 1". The first user selects the recommendation algorithm model of other users, such as "Recommendation Algorithm 2" as the first recommendation model. Then, "Recommendation Algorithm 2" and "Recommendation Algorithm 1" can be merged and trained to obtain the model after merged training, that is, the first merged model "Recommendation Algorithm 1 + Recommendation Algorithm 2".

[0050] In step S130 , in response to the merge completion instruction of the first user, target recommended content is determined for the first user based on the obtained first merge model.

[0051] In some exemplary embodiments of the present disclosure, the merge completion instruction may be an instruction corresponding to multiple recommendation algorithm models after completing model merge training. The target recommended content may be the recommended content pushed to the first user by the recommendation platform according to the merge model. For example, the target recommended content may be the recommended content determined according to the first merge model. The target recommended content may include various types of information content such as video, graphic, and text.

[0052] After the first user selects the first recommendation model, the first recommendation models can be merged into the current recommendation model one by one. When the first user merges all the selected first recommendation models into the current recommendation model, a merge completion instruction is generated. In response to the first user's merge completion instruction, the recommendation platform performs model merge training based on the first recommendation model and the current recommendation model to obtain a first merged model, and determines target recommendation content for the user based on the obtained first merged model.

[0053] For example, when the recommended content corresponding to "recommendation algorithm 1" is sports content, and the recommended content corresponding to "recommendation algorithm 2" is music content, after the above-mentioned model merging process, the first merged model obtained can push "sports + music" recommended content to the first user, that is, the target recommended content.

[0054] According to the content recommendation method in this example embodiment, on the one hand, users can independently select recommendation models for model training, and improve the accuracy of recommended content distribution by autonomously and quickly learning recommendation models. On the other hand, users are no longer unaware of the recommendation model, and providing users with the ability to autonomously control the recommendation algorithm model learning can greatly improve the user's sense of control. On the other hand, by allowing users to autonomously and quickly learn the recommendation algorithm, the adaptation cost of cold-start users is greatly reduced, which can greatly reduce the churn rate of related users.

[0055] Next, the content recommendation method in this exemplary embodiment will be further described.

[0056] In an exemplary embodiment of the present disclosure, in response to a model removal instruction from a first user, the merged recommendation model in the first merged model is removed to generate a second merged model; in response to a removal completion instruction from the first user, target recommended content is determined for the first user based on the obtained second merged model.

[0057] The model removal instruction may be an operation instruction to remove a specified recommendation model from the first merged model. The second merged model may be a recommendation model obtained by removing one or more merged recommendation models in the first merged model through the model removal instruction. The second merged model may include only one recommendation algorithm model. The target recommended content may also be the recommended content pushed to the first user by the recommendation platform based on the second merged model.

[0058] When the recommendation platform uses the first merged model to recommend content to the first user, if the first user wants the recommendation platform to stop pushing certain categories of recommended content to him / her in the future, one or more merged recommendation models in the first merged model can be removed from the first merged model through a model removal operation. When the first user performs a model removal operation, a corresponding model removal instruction will be generated. In response to the model removal instruction, the recommendation platform can remove the merged recommendation models in the first merged model from the first merged model to obtain a second merged model. When the first user can remove multiple merged recommendation models from the first merged model through multiple model removal operations, when the first user completes the model removal operation, a corresponding removal completion instruction can be generated. In response to the first user's removal completion instruction, target recommended content can be determined for the first user based on the generated second merged model.

[0059] For example, after multiple model merging trainings by the first user, the current first merged model pushes recommended content of music + film + sports + cute pets to the first user. At this time, when the first user no longer wants to receive recommended content of film and television and cute pets, the recommendation algorithm models corresponding to the film and television and cute pets can be removed from the first merged model through the model removal instruction. Then, the second merged model currently corresponding to the first user only includes the recommendation algorithm models of music + sports. At this time, the target recommended content can be determined for the first user based on the recommendation algorithm models of music + sports. By responding to the relevant processing operations of the model removal instruction, the matching degree of the recommended content can be further improved.

[0060] In an exemplary embodiment of the present disclosure, the first recommendation model is a recommendation model corresponding to the second user; the second user is a user who turns on the model sharing option.

[0061] The second user may be a user who has a certain association with the first user. For example, the second user may be a user in the first user's friend list, the second user may also be a user followed by the first user, or the second user may be a user with a high following in the recommendation platform (such as a popular user). The recommendation model corresponding to the second user may be a recommendation algorithm model used by the recommendation platform to push recommended content to the second user. The model sharing option may be a configuration option for sharing the recommendation algorithm model used by the user with other users for model merging training.

[0062] In order to solve the problem that in the existing content recommendation solutions, during the process of model training and learning, it is necessary to continuously analyze user historical data for model training, resulting in a long model optimization cycle, the present disclosure provides an active training content recommendation solution. Specifically, the recommendation platform can provide users with the option to choose whether to open the configuration item of the content recommendation algorithm model they use, that is, the model sharing option. If a user chooses to open the model sharing option, other users will be able to choose to learn the user's recommendation model to train their own algorithm recommendation model. Figure 2 , Figure 2 This is an interface diagram of a user turning on a model sharing option according to an exemplary embodiment. For example, the first user in the recommendation platform can click the operation control 210 in his or her "Personal Center" page, and choose to turn on the model sharing option in the "Privacy Settings" 220 in the "Personal Center" section, such as turning off the switch 230 of the "Do not open my recommendation algorithm to others" configuration item. The switch of the "Do not open my recommendation algorithm to others" configuration item is usually turned on by default in the recommendation platform. If the first user turns off this option, the switch of the configuration item will be converted to the display style of the control 240. When a user turns on the model sharing option, other users in the recommendation platform will be able to obtain the user's recommendation model for model training.

[0063] When the first user performs a model selection operation through the second user, the first user can determine the second user from his or her friend list or follow list. The first user can also select some users with high attention through the recommendation platform as the second user. If the first user wants to obtain the recommended model of the second user, the second user must be a user who has enabled the model sharing option in the recommendation platform, otherwise the first user cannot obtain the recommended model of the user. At this time, the first user can perform a model selection operation through the second user. When the first user selects the recommended model of the second user, the recommended model of the second user obtained can be used as the first recommended model, and the first recommended model can be displayed.

[0064] refer to Figure 3 , Figure 3It is a schematic diagram of a first user selecting a recommendation model based on a second user for model training according to an exemplary embodiment. The first user can enter his friend list through the "Personal Center", select a user from the friend list as the second user, and obtain the recommendation model of the second user, for example, by clicking the "Get Recommendation Algorithm" control 320 to obtain the recommendation model of the second user for display. When the first user completes the model merging operation, in response to the user's merge completion instruction, the recommendation platform performs the model merging training process, and the user interaction interface will display the model training prompt information 330 of "Recommendation algorithm model learning, please wait". When the first recommendation model selected by the first user is merged with the current recommendation model, the user interface will display the update completion prompt information 340 such as "Congratulations, model update is completed".

[0065] In an exemplary embodiment of the present disclosure, a model selection operation performed by a first user based on a model selection area is received, and a model keyword corresponding to the model selection operation is determined; and at least one first recommended model is selected from a model library based on the model keyword and displayed on an interactive interface.

[0066] The model selection area may be an operation area provided by the recommendation platform to the first user for model selection. The model keyword may be a keyword used to determine a certain recommendation model. The model library may be a database for storing recommendation models.

[0067] The interactive interface of the recommendation platform may also provide a model selection area for the first user, and the first user may perform a model selection operation through the model selection area. Figure 4 , Figure 4 FIG. 1 is a schematic diagram showing a first user selecting a recommended model through a user interaction interface for model training according to an exemplary embodiment. Figure 4 As can be seen in the figure, the first user can trigger the display of the model selection area 420 through the "add model" control 410 in the user interaction interface. The first user can directly select one or more recommended models from the model selection area 420. For example, the model selection area 420 displays recommended models such as "sports", "cute pets", "animation", and "movies" for the first user to select.

[0068] In addition, the model selection area also provides an input area, and the first user can also perform a model selection operation by inputting text. When the model selection operation of the first user is completed, the recommendation platform can determine the model keywords in the model selection operation performed by the first user. For example, the model keywords may include "cute pets", "table tennis", "food", "rap singer", etc. The model selection area 430 displays the model keywords selected by the first user after the model selection operation is completed. After determining the model keywords, the recommendation platform will select the corresponding recommended model from the model library and display it in the interactive interface so that the first user can select one or more recommended models for model merging operations. When the first user completes the model merging operation, in response to the first user's merging completion instruction, the recommendation platform will perform a model merging training process. At this time, the user interface can also display a model training prompt message 440 of "recommended algorithm model learning, please wait". By providing a model selection area, the autonomy of the first user in selecting a recommended model and the sense of control over the recommended model can be further improved, while reducing the user's adaptation cost and reducing the user's churn rate.

[0069] In an exemplary embodiment of the present disclosure, the model weight of each recommendation model in the first merged model is determined; the content acquisition proportion of the first user for each recommendation model is determined according to each model weight; and the corresponding content request quantity is allocated to each recommendation model according to the content acquisition proportion to determine the target recommended content according to the content request quantity.

[0070] The model weight may be the importance of each recommendation model in the first merged model. The content acquisition ratio may be the ratio by which the recommendation platform obtains the recommended content according to the recommendation model. The content request quantity may be the number of requests corresponding to the recommended content pushed by the recommendation platform to the first user.

[0071] After determining the first merged model, the model weights corresponding to each recommendation model in the first merged model can be further determined, so as to determine the proportion of content acquisition of the first user from each recommendation model in the first merged model according to the determined model weights. For example, when the first merged model includes four recommendation models, the model weights corresponding to the four recommendation models are determined respectively. Since the recommendation platform may include a large number of recommendation models, the model weights of the recommendation models may be determined based on historical recommendation effect data. Therefore, the sum of the model weights of the four recommendation models may not be 1. At this time, the model weights of the four recommendation models can be converted and calculated to determine the proportion of content acquisition corresponding to each recommendation model by the first user. For example, the determined content acquisition proportions can be 0.5, 0.3, 0.1, and 0.1. At this time, if the recommendation platform needs to push one hundred recommended contents to the first user, the number of content requests corresponding to the four recommendation models is 50, 30, 10, and 10 respectively. For another example, when the model weights corresponding to the four recommendation models are the same, the number of content requests corresponding to each recommendation model is 25. Determining the final number of content requests by model weights can further improve the accuracy of the determined recommended content.

[0072] In an exemplary embodiment of the present disclosure, historical recommendation data of each recommendation model in the first merged model is obtained; the historical recommendation data includes multiple recommendation influencing factors; the factor weight of each recommendation influencing factor is determined based on the historical recommendation data; and weighted calculation is performed based on the multiple factor weights to determine the model weight.

[0073] The historical recommendation data may be the recommendation effect data generated when the recommendation model historically recommends content. The recommendation influencing factors may be related factors that affect the recommendation effect. For example, the recommendation influencing factors may include the user's viewing time, interaction frequency, viewing volume, number of comments, number of likes, etc. for the video. The factor weight may be the weight corresponding to each recommendation influencing factor.

[0074] When calculating the model weight of the recommendation model, the historical recommendation data of each recommendation model in the first merged model can be obtained first. The historical recommendation data can include relevant data of all recommendation influencing factors. For example, the corresponding recommended content can be determined according to each recommendation model, and users can view, comment on, and forward these recommended contents. Therefore, relevant data such as viewing time, interaction frequency, number of views, number of likes, and number of comments can be generated. When determining the model weight, the factor weight corresponding to each recommendation influencing factor can be determined first. For example, the number of views and viewing time have a greater impact on the recommendation effect. Therefore, the viewing time and number of views can be configured with a larger weight value. The number of comments, number of likes, and interaction frequency have a smaller impact on the recommendation effect, so a relatively small weight value can be configured accordingly. After determining the factor weight of each recommendation influencing factor, a weighted calculation can be performed based on all recommendation influencing factors and the corresponding factor weights, and finally the corresponding model weight is determined.

[0075] In an exemplary embodiment of the present disclosure, content labels of each recommendation model in the first merged model are determined; model priorities of each recommendation model are determined according to the content labels; and model weights corresponding to each recommendation model are determined according to the model priorities.

[0076] The content label may be a label of the recommended content corresponding to the recommendation model, for example, the content label may be games, fitness, movies, comedy, variety shows, etc. The model priority may be the priority used when the recommendation model is used to recommend content.

[0077] When determining the model weights of each recommendation model in the first merged model, the content tags of each recommendation model can be obtained. The recommendation platform can set default priorities for various content tags, and then determine the model priorities of each recommendation model based on the content tags. For example, the recommendation platform can set higher priorities for content tags such as "fitness", "movies", and "funny". After determining the model priorities of different recommendation models, the model priorities can be sorted, and the model weights of each recommendation model can be determined based on the sorting results. In addition, during the process of content recommendation by the recommendation platform, the first user can change the user preference settings. For example, the first user can reset the preferences to variety shows and games, etc., then the recommendation platform can change the model priority of the corresponding recommendation model, and recommend content based on the changed model priority. The merged model will continue to sort and filter the recommended content to improve the accuracy of the recommended content.

[0078] Furthermore, when a user initially uses a platform, the initial model weights of the recommendation models are the same. At this time, the final model weights of the recommendation models can be determined based on the order of label priorities of the default content labels in the system. The default label priority can be determined by the recommendation platform based on the user preferences of most users, or it can be determined based on the order of selection when the user selects the recommendation model. For example, there are three recommendation models in the first merged model, and the label priorities of the three recommendation models are high, medium, and low, respectively. The model weights of the three recommendation models can be configured to 50%, 30%, and 20%, respectively.

[0079] In an exemplary embodiment of the present disclosure, associated users associated with a first user are determined, and an associated user group is generated based on the first user and the associated users; in response to a model training operation of any user in the associated user group, associated group recommended content corresponding to the associated user group is determined; and the associated group recommended content is displayed to any user in the associated user group.

[0080] The associated user may be another user having a specific association relationship with the first user. The associated user group may be a user set consisting of multiple users having a specific association relationship. The model training operation may be an operation in which any user selects a recommendation model for combined training. The associated group recommended content may be the recommended content pushed to the associated group by the recommendation platform.

[0081] For the first user, the associated users with a specific association relationship with the first user can be determined from the recommendation platform, and the first user and the associated users are regarded as the same group of users to generate a corresponding associated user group. If any user in the associated user group performs a model training operation, such as selecting one or more recommendation models for a merge training operation, the associated group recommendation content pushed to the associated user group is determined according to the model training operation, and the associated recommendation content is displayed to all users in the associated group. Figure 5 , Figure 5This is an interface diagram of a first user and other users forming an associated user group according to an exemplary embodiment. The first user can enter the friend interface through the operation control 510 of the "Personal Center", and the friend list 520 corresponding to the first user is displayed in the friend interface. When the first user wants to use friend 1 as an associated user to jointly generate an associated user group, he can click the interactive button 521 corresponding to friend 1. At this time, the associated friend operation interface 530 will be displayed. The associated friend operation interface 530 displays prompt information on whether to add friend 1 as an associated user, such as "Add as an associated friend". The first user can establish an association relationship with friend 1 by clicking the "Confirm" control 531 to generate a corresponding associated user group. At this time, all users in the associated user group will share the recommendation model. At this time, the prompt message 540 "Recommendation algorithm model is learning, please wait" can be displayed to the first user. After the model merge training is completed, the recommendation platform will push the same target recommended content to any user in the associated user group.

[0082] For example, in a content push scenario, if two accounts are bound as couple accounts to obtain recommended content, the two accounts will be users in the same associated user group. When any user in the associated user group performs a model training operation, the corresponding recommended content will be generated according to the corresponding model training operation, and the generated recommended content will be pushed to any user in the couple account, and the recommended content of the two can always be consistent. By providing the associated user group function, user interaction in the recommendation platform can be improved, effectively improving user stickiness.

[0083] In summary, the content recommendation method of the present disclosure receives a model selection operation of a first user, selects at least one first recommendation model for display; in response to a model merging instruction of the first user, merges at least one first recommendation model corresponding to the model merging instruction into the current recommendation model to generate a first merged model; in response to a merge completion instruction of the first user, determines the target recommended content for the first user based on the obtained first merged model. On the one hand, the user can independently select a recommendation model for model training, and improve the accuracy of recommended content distribution by autonomously and quickly learning the recommendation model. On the other hand, the user is no longer unaware of the recommendation model, and the ability to autonomously control the recommendation algorithm model learning is provided to the user, which can greatly improve the user's sense of control. On the other hand, by allowing users to autonomously and quickly learn the recommendation algorithm, the adaptation cost of cold-start users is greatly reduced, and the churn rate of related users can be greatly reduced.

[0084] According to a second aspect of the present disclosure, a content recommendation system is provided. Figure 6 , Figure 66 is a structural diagram of a content recommendation system according to an exemplary embodiment. The content recommendation system 600 includes: a user terminal 610, a model training terminal 620 and a content recommendation platform 630. Specifically:

[0085] The user end 610 is used to provide a model operation interface and display target recommended content, and receive user model operations based on the model operation interface; model operations include model selection operations, model removal operations and model sharing operations; the model training end 620 is used to merge the first recommendation model determined based on the model selection operation into the current recommendation model to generate a first merged model; and according to the model removal operation of the first merged model, remove at least one recommended model in the first merged model to generate a second merged model; the content recommendation platform 630 is used to determine the target recommended content for the user based on the merged model, and send the target recommended content to the user end.

[0086] The model operation interface may be a user interaction interface used to perform relevant model operations. Model operations may be user operations performed on recommendation models. Model selection operations may be operations related to user selection of one or more recommendation models. Model removal operations may be operations related to user removal of one or more recommendation models from a merged model. Model sharing operations may be operations related to a user sharing a recommendation model used by him / her with other users so that other users can perform model merging training based on the shared recommendation model.

[0087] The user end can provide the user with a model operation interface, and the user performs model sharing operations through the model operation interface and turns on the model sharing option so that other users can obtain their own recommendation algorithm models. Users can also perform model selection operations through the model operation interface to obtain recommendation models of other users, so as to quickly and purposefully train their own recommendation models. When the user selects the first recommendation model through the model selection operation, the model training end can merge the first recommendation model into the current recommendation model to generate a first merged model. After generating the first merged model, the content recommendation platform will determine the target recommended content to the user based on the first merged model. In the process of receiving recommended content, if the user wants to reduce the push of certain recommended content, the corresponding recommendation model can be removed through the model removal operation to obtain the second merged model. At this time, the content recommendation platform will determine the target recommended content to the user based on the second merged model.

[0088] Figure 7 is a block diagram of a content recommendation device according to an exemplary embodiment. Figure 7 The content recommendation device 700 includes a model display module 710, a model merging module 720 and a first content recommendation module 730. Specifically:

[0089] The model display module 710 is configured to receive a model selection operation from a first user and select at least one first recommended model for display.

[0090] The model merging module 720 is configured to execute, in response to a model merging instruction from the first user, merging at least one first recommendation model corresponding to the model merging instruction into the current recommendation model to generate a first merged model.

[0091] The first content recommendation module 730 is configured to execute in response to the merge completion instruction of the first user and determine target recommended content for the first user based on the obtained first merge model.

[0092] In an exemplary embodiment of the present disclosure, the content recommendation device 700 also includes a second content recommendation module, which is configured to execute: in response to a model removal instruction from the first user, remove the merged recommendation model in the first merged model to generate a second merged model; in response to a removal completion instruction from the first user, determine target recommended content for the first user based on the obtained second merged model.

[0093] In an exemplary embodiment of the present disclosure, the model display module 710 also includes a model display unit, which is configured to execute: receiving a model selection operation performed by a first user based on a model selection area, and determining a model keyword corresponding to the model selection operation; and selecting at least one first recommended model from a model library based on the model keyword and displaying it on an interactive interface.

[0094] In an exemplary embodiment of the present disclosure, the first content recommendation module 730 includes a first content recommendation unit, which is configured to perform: determining the model weight of each recommendation model in the first merged model; determining the content acquisition proportion of the first user for each recommendation model according to each model weight; allocating a corresponding content request quantity to each recommendation model according to the content acquisition proportion, so as to determine the target recommended content according to the content request quantity.

[0095] In an exemplary embodiment of the present disclosure, the first content recommendation unit includes a first weight determination subunit, which is configured to perform: obtaining historical recommendation data of each recommendation model in the first merged model; the historical recommendation data includes multiple recommendation influencing factors; determining the factor weight of each recommendation influencing factor based on the historical recommendation data; and performing weighted calculation based on multiple factor weights to determine the model weight.

[0096] In an exemplary embodiment of the present disclosure, the first content recommendation unit includes a second weight determination subunit, which is configured to perform: determining the content label of each recommendation model in the first merged model; determining the model priority of each recommendation model according to the content label; and determining the model weight corresponding to each recommendation model according to the model priority.

[0097] In an exemplary embodiment of the present disclosure, the content recommendation device 700 also includes a third content recommendation module, which is configured to perform: determining associated users associated with the first user, and generating an associated user group based on the first user and the associated users; determining associated group recommended content corresponding to the associated user group in response to a model training operation of any user in the associated user group; and displaying the associated group recommended content to any user in the associated user group.

[0098] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0099] Reference below Figure 8 800 according to this embodiment of the present disclosure is described. Figure 8 The electronic device 800 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0100] like Figure 8 As shown, the electronic device 800 is in the form of a general computing device. The components of the electronic device 800 may include, but are not limited to: the at least one processing unit 810, the at least one storage unit 820, a bus 830 connecting different system components (including the storage unit 820 and the processing unit 810), and a display unit 840.

[0101] The storage unit stores program codes, which can be executed by the processing unit 810, so that the processing unit 810 executes the steps according to various exemplary embodiments of the present disclosure described in the above “Exemplary Method” section of this specification.

[0102] The storage unit 820 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 821 and / or a cache memory unit 822 , and may further include a read-only memory unit (ROM) 823 .

[0103] The storage unit 820 may include a program / utility 824 having a set (at least one) of program modules 825, such program modules 825 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0104] Bus 830 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0105] The electronic device 800 may also communicate with one or more external devices 870 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 800, and / or communicate with any device that enables the electronic device 800 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 850. Furthermore, the electronic device 800 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 860. As shown, the network adapter 860 communicates with other modules of the electronic device 800 via a bus 830. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 800, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.

[0106] In an exemplary embodiment, a computer-readable storage medium including instructions is also provided, such as a memory including instructions, and the above instructions can be executed by a processor of the device to complete the above method. Optionally, the computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0107] In an exemplary embodiment, a computer program product is also provided, including a computer program / instruction, wherein when the computer program / instruction is executed by a processor, any one of the above-mentioned content recommendation methods is implemented.

[0108] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0109] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A content recommendation method, characterized in that: include: Receive a model selection operation from a first user, select at least one first recommended model for display, the first recommended model is a recommended model corresponding to a second user, the second user is a user who turns on a model sharing option, and the model sharing option is a configuration option for sharing the recommended model; In response to a model merging instruction from the first user, merging at least one first recommendation model corresponding to the model merging instruction into a current recommendation model to generate a first merged model; In response to the merge completion instruction of the first user, target recommended content is determined for the first user based on the obtained first merge model.

2. The method according to claim 1, characterized in that: The method further comprises: In response to a model removal instruction from the first user, removing the merged recommendation model in the first merged model to generate a second merged model; In response to the removal completion instruction of the first user, target recommended content is determined for the first user based on the obtained second merge model.

3. The method according to claim 1, characterized in that: Receiving a model selection operation of a first user, selecting at least one first recommended model for display, further comprising: receiving a model selection operation performed by the first user based on the model selection area, and determining a model keyword corresponding to the model selection operation; At least one of the first recommended models is selected from a model library according to the model keyword and displayed on an interactive interface.

4. The method according to claim 1, characterized in that: Determining target recommended content for the first user based on the obtained first merged model includes: Determining a model weight of each of the recommendation models in the first combined model; Determining a content acquisition proportion of each recommendation model for the first user according to each model weight; A corresponding content request quantity is allocated to each of the recommendation models according to the content acquisition proportion, so as to determine the target recommended content according to the content request quantity.

5. The method according to claim 4, characterized in that Determining a model weight of each of the recommendation models in the first combined model includes: Acquire historical recommendation data of each of the recommendation models in the first combined model; the historical recommendation data includes a plurality of recommendation influencing factors; Determining the factor weight of each of the recommendation influencing factors according to the historical recommendation data; A weighted calculation is performed based on a plurality of the factor weights to determine the model weight.

6. The method according to claim 4, characterized in that Determining a model weight of each of the recommendation models in the first combined model includes: Determining a content label of each of the recommendation models in the first merged model; Determining a model priority of each of the recommendation models according to the content tag; The model weight corresponding to each of the recommended models is determined according to the model priority.

7. The method according to any one of claims 1 to 6, characterized in that: The method further comprises: Determine an associated user associated with the first user, and generate an associated user group based on the first user and the associated user; In response to a model training operation of any user in the associated user group, determining an associated group recommended content corresponding to the associated user group; The associated group recommended content is displayed to any user in the associated user group.

8. A content recommendation system, characterized in that: include: The user end is used to provide a model operation interface and display target recommended content, and receive a model operation performed by a first user based on the model operation interface; the model operation includes a model selection operation, a model removal operation, and a model sharing operation; A model training end, configured to merge the first recommendation model determined based on the model selection operation into the current recommendation model to generate a first merged model; and according to the model removal operation of the first merged model, remove at least one recommendation model in the first merged model to generate a second merged model, wherein the first recommendation model is a recommendation model corresponding to the second user; The second user is a user who turns on the model sharing option, where the model sharing option is a configuration option for sharing the recommendation model; The content recommendation platform is used to determine target recommended content for the first user according to the merged model, and send the target recommended content to the user terminal.

9. A content recommendation device, characterized in that: include: a model display module configured to receive a model selection operation of a first user, select at least one first recommended model for display, wherein the first recommended model is a recommended model corresponding to a second user, the second user is a user who turns on a model sharing option, and the model sharing option is a configuration option for sharing the recommended model; a model merging module, configured to execute, in response to a model merging instruction of the first user, merging at least one first recommendation model corresponding to the model merging instruction into a current recommendation model to generate a first merged model; The first content recommendation module is configured to execute a merging completion instruction in response to the first user and determine target recommended content for the first user based on the obtained first merging model.

10. The device according to claim 9, characterized in that The content recommendation device further includes a second content recommendation module, wherein the second content recommendation module is configured to execute: In response to a model removal instruction from the first user, removing the merged recommendation model in the first merged model to generate a second merged model; In response to the removal completion instruction of the first user, target recommended content is determined for the first user based on the obtained second merge model.

11. The device according to claim 9, characterized in that The model display module further includes a model display unit, wherein the model display unit is configured to execute: receiving a model selection operation performed by the first user based on the model selection area, and determining a model keyword corresponding to the model selection operation; At least one of the first recommended models is selected from a model library according to the model keyword and displayed on an interactive interface.

12. The device according to claim 9, characterized in that The first content recommendation module includes a first content recommendation unit, wherein the first content recommendation unit is configured to execute: Determining a model weight of each of the recommendation models in the first combined model; Determining a content acquisition proportion of each recommendation model for the first user according to each model weight; A corresponding content request quantity is allocated to each of the recommendation models according to the content acquisition proportion, so as to determine the target recommended content according to the content request quantity.

13. The device according to claim 12, characterized in that The first content recommendation unit includes a first weight determination subunit, which is configured to perform: Acquire historical recommendation data of each of the recommendation models in the first combined model; the historical recommendation data includes a plurality of recommendation influencing factors; Determining the factor weight of each of the recommendation influencing factors according to the historical recommendation data; A weighted calculation is performed based on a plurality of the factor weights to determine the model weight.

14. The device according to claim 12, characterized in that The first content recommendation unit includes a second weight determination subunit, wherein the second weight determination subunit is configured to perform: Determining a content label of each of the recommendation models in the first merged model; Determining a model priority of each of the recommendation models according to the content tag; The model weight corresponding to each of the recommended models is determined according to the model priority.

15. The device according to claim 9, characterized in that The content recommendation device further includes a third content recommendation module, wherein the third content recommendation module is configured to execute: Determine an associated user associated with the first user, and generate an associated user group based on the first user and the associated user; In response to a model training operation of any user in the associated user group, determining an associated group recommended content corresponding to the associated user group; The associated group recommended content is displayed to any user in the associated user group.

16. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the content recommendation method as described in any one of claims 1 to 7.

17. A computer-readable storage medium, when instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to execute the content recommendation method as described in any one of claims 1 to 7.

18. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the content recommendation method according to any one of claims 1 to 7 is implemented.

Images