Content recommendation method, device, electronic device and storage medium

By obtaining user base and behavior information, calculating similarity with content attribute information and generating a recall set, filtering content recommendations based on user tags, the problem that collaborative filtering algorithms are difficult to recommend content is solved, and user retention time is improved.

CN113392329BActive Publication Date: 2024-11-26西安力传智能技术有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202110818646.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-07-20
Publication Date
2024-11-26
Estimated Expiration
2041-07-20

AI Technical Summary

Technical Problem

The existing collaborative filtering algorithms are difficult to accurately recommend content to users who do not have clear requirements, resulting in the retention time of these users being too short or even lost.

Method used

By obtaining the user's basic information and behavioral information, combining the content attribute information, calculating the similarity between contents, generating a recall set, and filtering content recommendations based on user tags and tag weights, improving the accuracy of recommendations.

Benefits of technology

Improve the accuracy of recommending content to users who do not have clear requirements and extend the retention time of users.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113392329B_ABST
    Figure CN113392329B_ABST
Patent Text Reader

Abstract

The embodiments of the present application provide a content recommendation method, device, electronic device and storage medium, which relate to the field of data analysis. The method includes: obtaining the user's basic information, behavior information and attribute information of multiple contents; determining the similarity between each content according to the attribute information of each content, and generating multiple recall sets according to the similarity, each recall set including multiple contents with similarities greater than a preset threshold; generating user tags and tag weights according to the basic information and behavior information; filtering out at least one target recall set from multiple recall sets according to the user's tags; filtering the contents in the target recall set according to the user's tag weight, and generating and outputting content recommendation results according to the filtering results. The content recommendation method provided by the embodiments of the present application can improve the accuracy of content recommended to users who do not have clear needs, which is conducive to improving the retention time of these users.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of data analysis technology, and in particular to a content recommendation method, device, electronic device and storage medium. Background Art

[0002] With the development of the Internet, people have entered an era of information explosion. In order to meet people's demand for information, information websites have emerged. Information websites usually divide the collected information into multiple types, and users can freely choose the type to conveniently obtain the information they are interested in.

[0003] In the prior art, many information websites use collaborative filtering algorithms to calculate content that users may be interested in and recommend it to users. In simple terms, collaborative filtering algorithms use the preferences of similar user groups to recommend content that users may be interested in, thereby increasing user retention time.

[0004] However, collaborative filtering algorithms are more dependent on similar user groups, and it is difficult to accurately recommend content to users who do not have clear needs, that is, users who do not have similar user groups, which can easily cause the retention time of these users to be too short or even cause them to leave. Summary of the invention

[0005] In view of this, the purpose of the present application is to provide a content recommendation method, device, electronic device and storage medium to improve the accuracy of content recommendation to users who do not have clear needs and increase the retention time of these users.

[0006] In a first aspect, an embodiment of the present application provides a content recommendation method, comprising:

[0007] Obtaining basic information, behavior information, and attribute information of multiple contents of the user, wherein the basic information includes at least one of the user's age, gender, preference, and location, the behavior information includes at least one of the user's click behavior, browsing behavior, collection behavior, sharing behavior, and comment behavior, and the attribute information includes at least one of the title and type of the content and the behavior weight of the content;

[0008] Determine the similarity between the contents according to the attribute information of the contents, and generate multiple recall sets according to the similarity, each recall set including multiple contents whose similarity is greater than a preset threshold;

[0009] Generate user tags and tag weights based on basic information and behavior information;

[0010] Filtering at least one target recall set from multiple recall sets according to the user's label;

[0011] The content in the target recall set is filtered according to the user's tag weight, and the content recommendation results are generated and output based on the filtering results.

[0012] Optionally, in a specific implementation manner, generating a user tag and a tag weight according to the basic information and the behavior information includes:

[0013] extracting a first preset number of keywords from each content respectively;

[0014] Filter out keywords related to behavioral information and basic information from the keywords as user tags;

[0015] The product of the sum of the behavior weights of the content corresponding to the user's tag and the preset time decay function is calculated as the tag weight corresponding to the tag.

[0016] Optionally, in a specific implementation, the content in the target recall set is screened according to the user's tag weight, and a content recommendation result is generated and output according to the screening result, including:

[0017] The contents in the target recall set are sorted according to the user's tag weight, the contents with the first second preset number of sequence numbers in the sorted result are extracted as content recommendation results, and the content recommendation results are output.

[0018] Optionally, in a specific implementation, the method further includes:

[0019] Recommending content recommendation results to a first preset proportion of users, and obtaining evaluation data of the first preset proportion of users;

[0020] Adjusting the label and label weight of each user in the first preset proportion of users according to the evaluation data;

[0021] Generate new content recommendation results based on the adjusted user tags and tag weights.

[0022] Optionally, in a specific implementation, adjusting the user's tag and tag weight according to the evaluation data includes:

[0023] When the user is offline, build the user's training sample, which includes the user's label, label weight and corresponding evaluation data;

[0024] A machine learning model is obtained based on training samples;

[0025] Adjust user tags and tag weights based on machine learning models.

[0026] Optionally, in a specific implementation manner, determining the similarity between the contents according to the attribute information of the contents includes:

[0027] extracting a first preset number of keywords from each content respectively;

[0028] The similarity of each content is calculated according to a first preset number of keywords extracted from each content.

[0029] Optionally, in a specific implementation manner, extracting a first preset number of keywords from each content respectively includes:

[0030] Based on the first keyword extraction algorithm, extracting a third preset number of keywords and a weight corresponding to each keyword from each content;

[0031] Based on the second keyword extraction algorithm, extracting a fourth preset number of keywords and a weight corresponding to each keyword from each content;

[0032] Based on the weight corresponding to each keyword, the second preset number of keywords and the third preset number of keywords corresponding to each content are sorted, and the first preset number of keywords are screened out.

[0033] Optionally, in a specific implementation manner, calculating the similarity of each content according to a first preset number of keywords extracted from each content includes:

[0034] Generate a word vector corresponding to each content according to a first preset number of keywords corresponding to each content;

[0035] Based on the local sensitive hashing algorithm, the similarity of each content is calculated according to the word vector corresponding to each content.

[0036] Optionally, in a specific implementation, the similarity of each content includes at least one of similarity of keywords and similarity of topics.

[0037] In a second aspect, an embodiment of the present application provides a content recommendation device, including:

[0038] An information acquisition module, for acquiring basic information, behavior information of a user and attribute information of a plurality of contents, wherein the basic information includes at least one of the user's age, gender, preference and location, the behavior information includes at least one of the user's click behavior, browsing behavior, collection behavior, sharing behavior and comment behavior, and the attribute information includes at least one of the title and type of the content and the behavior weight of the content;

[0039] A recall set generation module, used to determine the similarity between the contents according to the attribute information of the contents, and generate multiple recall sets according to the similarity, each recall set including multiple contents with similarity greater than a preset threshold;

[0040] The tag generation module is used to generate user tags and tag weights based on basic information and behavior information;

[0041] A first screening module, used to screen out at least one target recall set from multiple recall sets according to user tags;

[0042] The second screening module is used to screen the content in the target recall set according to the user's tag weight, and generate and output content recommendation results according to the screening results.

[0043] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the storage medium communicate through the bus, and the processor executes the machine-readable instructions to perform the steps of any method of the first aspect.

[0044] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of any method of the first aspect are executed.

[0045] The content recommendation method, device, electronic device and storage medium provided in the embodiment of the present application first determine the similarity between each content according to the attribute information of each content, and generate multiple recall sets according to the similarity; then generate the user's label and label weight according to the basic information and behavior information; and filter out at least one target recall set from multiple recall sets according to the user's label; finally, filter the content in the target recall set according to the user's label weight, and generate and output the content recommendation result according to the filtering result. The content recommendation method provided in the embodiment of the present application can improve the accuracy of content recommended to users who do not have clear needs, which is conducive to improving the retention time of these users.

[0046] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0048] Figure 1 A flow chart showing a content recommendation method provided in an embodiment of the present application is shown;

[0049] Figure 2 A flowchart of generating content recommendation results provided by an embodiment of the present application is shown;

[0050] Figure 3 A schematic diagram showing the structure of a content recommendation device provided in an embodiment of the present application is shown;

[0051] Figure 4 A schematic structural diagram of an electronic device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0052] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0053] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0054] In addition, the terms “first”, “second”, etc., if used, are merely used to distinguish between the descriptions and should not be understood as indicating or implying relative importance.

[0055] It should be noted that, in the absence of conflict, the features in the embodiments of the present application may be combined with each other.

[0056] Before explaining the present application in detail, the application scenario of the present application is first introduced.

[0057] With the development of the Internet, people have entered an era of information explosion. In order to meet people's demand for information, information websites have emerged. Information websites usually divide the collected information into multiple types, and users can freely choose the type to conveniently obtain the information they are interested in. In the prior art, many information websites will calculate the content that users may be interested in through collaborative filtering algorithms and recommend it to users. In simple terms, collaborative filtering algorithms use the preferences of similar user groups to recommend content that users may be interested in, so as to increase user retention time. However, collaborative filtering algorithms are more dependent on similar user groups, and it is difficult to accurately recommend content to users who do not have clear needs, that is, users who do not have similar user groups, which can easily cause the retention time of these users to be too short or even loss.

[0058] In order to solve this problem, the present application provides a content recommendation method, which can improve the accuracy of content recommended to users who do not have clear needs, and is conducive to increasing the retention time of these users.

[0059] See also Figure 1 , the embodiment of the present application provides a content recommendation method, comprising:

[0060] S101: Obtaining basic information, behavior information, and attribute information of multiple contents of the user, wherein the basic information includes at least one of the user's age, gender, preference, and region; the behavior information includes at least one of the user's click behavior, browsing behavior, collection behavior, sharing behavior, and comment behavior; and the attribute information includes at least one of the title and type of the content and the behavior weight of the content.

[0061] Specifically, the content in the embodiments of the present application may include articles, news, and other content containing text information, and the present application does not make any specific limitations on this.

[0062] Business system data usually includes basic information of users and attribute information of multiple contents. Optionally, business system data can be imported into Hive database in advance through Sqoop tool, so as to avoid direct operation of business system data and improve data security. Sqoop tool is an open source tool, which is mainly used to transfer data between Hadoop-based database and traditional database. Hadoop-based database is Hive, and traditional database is MySQL, Oracle, Postgres, etc. Sqoop can import data from traditional databases such as MySQL, Oracle, Postgres into Hive-based database and other Hadoop-based database, and can also import data from Hive-based database into traditional databases such as MySQL, Oracle, Postgres, etc.

[0063] Optionally, you can use the log collection tool Flume to collect user behavior logs, obtain user behavior information, and store the user behavior information in the Hive database. Flume is a log collection tool that can collect log data from the data source and transmit it to the storage system of the big data ecosystem, such as the Hive database.

[0064] It should be noted that the behavioral weight of content refers to the weights corresponding to users clicking, browsing, collecting, sharing, and commenting on the content.

[0065] S102: Determine the similarity between the contents according to the attribute information of the contents, and generate multiple recall sets according to the similarity, each recall set including multiple contents whose similarity is greater than a preset threshold.

[0066] Specifically, the preset threshold may be a value set as needed, for example, the preset threshold may be 80%, which is not limited in the present application.

[0067] Optionally, in a specific implementation, the similarity of each content includes at least one of the similarity of keywords and the similarity of themes. When the similarity of each content includes the similarity of keywords and the similarity of themes, determining the similarity between each content includes: determining the similarity of keywords between each content and determining the similarity of themes between each content. Then, multiple recall sets can be generated based on the similarity, such as a recall set with keyword similarity greater than a preset threshold and a recall set with subject similarity greater than a preset threshold.

[0068] Optionally, the specific process of determining the similarity between the contents can be: constructing a portrait of each content, that is, extracting keywords from each content, and then processing each keyword into a multi-dimensional word vector through Word2vec in NLP (Natural Language Processing); determining the similarity of each dimension of the keywords in each content. Furthermore, multiple recall sets can be generated according to the similarity of each dimension of the keywords in each content, such as a recall set whose similarity in a certain dimension is greater than a preset threshold.

[0069] S103: Generate user tags and tag weights based on basic information and behavior information.

[0070] Specifically, the user's behavior information and keywords in the basic information can be extracted as user tags. For example, if the user's behavior information shows that the user has browsed a lot of mobile phone-related content, or the user's basic information shows that the user prefers to browse digital products, then mobile phone or digital product enthusiasts can be used as tags for the user. Of course, this is only an exemplary description and does not mean that the present application is limited to this.

[0071] The user's tag weight refers to the product of the sum of the behavior weights of the content corresponding to the user's tag and a preset time decay function.

[0072] S104: Filter out at least one target recall set from multiple recall sets according to the user's tag.

[0073] For example, if the user's tags are digital product enthusiast and travel enthusiast, then the recall sets related to digital products and the recall sets related to travel can be filtered out from multiple recall sets.

[0074] S105: Filter the contents in the target recall set according to the user's tag weight, and generate and output content recommendation results according to the filtering results.

[0075] Optionally, in a specific implementation, the content in the target recall set is screened according to the user's label weight, and a content recommendation result is generated and output based on the screening result, including: sorting the content in the target recall set according to the user's label weight, extracting the content with the first second preset number of sequence numbers in the sorting result as the content recommendation result, and outputting the content recommendation result.

[0076] Specifically, the second preset number can be a value set as needed, which is not limited in this application. In addition, different contents in the content recommendation results can be recommended to the user at different time points, or all contents in the content recommendation results can be recommended to the user at the same time point, which is not limited in this application.

[0077] The content recommendation method provided in the embodiment of the present application first determines the similarity between each content according to the attribute information of each content, and generates multiple recall sets according to the similarity; then generates the user's label and label weight according to the basic information and behavior information; and filters out at least one target recall set from the multiple recall sets according to the user's label; finally, the content in the target recall set is filtered according to the user's label weight, and a content recommendation result is generated and output according to the filtering result, thereby improving the accuracy of content recommended to users who do not have clear needs, which is conducive to improving the retention time of these users.

[0078] Optionally, in a specific implementation, a user label and label weight are generated based on basic information and behavior information, including: extracting a first preset number of keywords from each content; filtering out keywords associated with the behavior information and basic information from the keywords as the user's label; calculating the product of the sum of the behavior weights of the content corresponding to the user's label and a preset time decay function as the label weight corresponding to the label.

[0079] Specifically, the second preset number may be a value set as needed, such as 20, which is not limited in the present application.

[0080] In addition, the user's tag corresponds to certain keywords among the first preset number of keywords extracted from each content, and these keywords all have corresponding content. The content corresponding to the user's tag refers to the content corresponding to the keyword corresponding to the user's tag.

[0081] The time decay function may be a function pre-set as needed, as long as it is a function that decays with time, and this application does not make any specific limitation on this.

[0082] Optionally, filtering out keywords associated with the behavior information and the basic information from the keywords specifically refers to determining whether the keywords are included in the behavior information and the basic information, and if so, using the keywords as keywords associated with the behavior information and the basic information. Of course, this is only an exemplary description and does not mean that the present application is limited thereto.

[0083] Optionally, in a specific implementation, the method also includes: recommending content recommendation results to a first preset proportion of users, and obtaining evaluation data of the first preset proportion of users; adjusting the labels and label weights of each user in the first preset proportion of users according to the evaluation data; and generating new content recommendation results based on the adjusted user labels and label weights.

[0084] If the content recommendation results are recommended directly to all users, some users may be lost due to various reasons, which poses a high risk. Therefore, in an embodiment of the present application, the content recommendation results are first recommended to a first preset proportion of users, and the evaluation data of the first preset proportion of users is obtained; the label and label weight of each user in the first preset proportion of users are adjusted according to the evaluation data; and new content recommendation results are generated according to the adjusted user labels and label weights. In this way, the risk of user loss is reduced. Specifically, ABTest multi-layer diversion can be performed to recommend content recommendation results to the first preset proportion of users. Of course, this is only an exemplary description and does not mean that the present application is limited to this.

[0085] See also Figure 2 It should be noted that generating a new content recommendation result here may be a cyclic process, that is, recommending the content recommendation result to a first preset proportion of users, and obtaining the evaluation data of the first preset proportion of users; judging whether the evaluation data meets the preset conditions; if so, recommending the corresponding content recommendation result to the user; if not, adjusting the label and label weight of each user in the first preset proportion of users according to the evaluation data; generating a new content recommendation result according to the adjusted user label and label weight; after a cycle, recommending the new content recommendation result to the first preset proportion of users to start a new cycle.

[0086] Optionally, in a specific implementation, the user's label and label weight are adjusted according to the evaluation data, including: when the user is offline, constructing a training sample for the user, the training sample including the user's label, label weight and corresponding evaluation data; obtaining a machine learning model based on the training sample training; and adjusting the user's label and label weight based on the machine learning model.

[0087] The process of adjusting user labels and label weights based on evaluation data is time-consuming, and it is helpful to improve efficiency when the user is offline.

[0088] Optionally, in a specific implementation, determining the similarity between each content according to the attribute information of each content includes: extracting a first preset number of keywords from each content; and calculating the similarity of each content according to the first preset number of keywords extracted from each content.

[0089] The similarity of each content may refer to the overlap rate of keywords in each content. For example, content A includes two keywords, mobile phone and computer, and content B includes two keywords, mobile phone and clothes. Then the similarity between content A and content B is 50%.

[0090] Optionally, in a specific implementation, a first preset number of keywords is extracted from each content, including: based on a first keyword extraction algorithm, extracting a third preset number of keywords and the weight corresponding to each keyword from each content; based on a second keyword extraction algorithm, extracting a fourth preset number of keywords and the weight corresponding to each keyword from each content; based on the weight corresponding to each keyword, sorting the third preset number of keywords and the fourth preset number of keywords corresponding to each content, and screening out the first preset number of keywords.

[0091] The third preset number and the fourth preset number can be values ​​set as needed, and this application does not limit this.

[0092] In an embodiment of the present application, when the first keyword extraction algorithm is a TF-IDF (term frequency-inverse document frequency) algorithm, the second keyword extraction algorithm may be a TextRank algorithm; when the first keyword extraction algorithm is a TextRank algorithm, the second keyword extraction algorithm may be a TF-IDF algorithm, and the present application does not limit this. Among them, TF-IDF is a statistical algorithm used to evaluate the importance of a word to a file set or one of the files in a corpus. The importance of a word increases in direct proportion to the number of times it appears in the file, but at the same time it decreases in inverse proportion to the frequency of its appearance in the corpus. The TextRank algorithm is an algorithm that can be separated from the background of the corpus and can extract the keywords of the document by analyzing only a single document.

[0093] Optionally, in a specific implementation, the similarity of each content is calculated based on a first preset number of keywords extracted from each content, including: generating a word vector corresponding to each content based on the first preset number of keywords corresponding to each content; and calculating the similarity of each content based on the word vector corresponding to each content based on a local sensitive hashing algorithm.

[0094] Specifically, each keyword can be processed into a multi-dimensional word vector through Word2vec in NLP. The basic idea of ​​the locality sensitive hashing algorithm, namely the LSH (Locality Sensitive Hashing) algorithm, is that after two adjacent data points in the original data space are transformed by the same mapping or projection, the probability that the two data points are still adjacent in the new data space is very high, while the probability that non-adjacent data points are mapped to the same bucket is very small. The locality sensitive hashing algorithm can help quickly obtain the word vector corresponding to each content and calculate the similarity of each content.

[0095] In addition, it should be noted that, since the user's behavior information is constantly updated, in the embodiment of the present application, the user's behavior information can be updated regularly, so that a new content recommendation list can be generated regularly. The specific generation method will not be repeated here.

[0096] Optionally, in an embodiment of the present application, popular articles within a preset time period, newly published articles, etc. may be added to a content recommendation list in a certain proportion and recommended to the user, and the present application does not limit this.

[0097] See also Figure 3 , the embodiment of the present application provides a content recommendation device 30, including:

[0098] The information acquisition module 301 is used to obtain the user's basic information, behavior information and attribute information of multiple contents. The basic information includes at least one of the user's age, gender, preferences and region. The behavior information includes at least one of the user's click behavior, browsing behavior, collection behavior, sharing behavior and comment behavior. The attribute information includes at least one of the title and type of the content and the behavior weight of the content.

[0099] The recall set generation module 302 is used to determine the similarity between the contents according to the attribute information of each content, and generate multiple recall sets according to the similarity, each recall set including multiple contents with similarity greater than a preset threshold.

[0100] The tag generation module 303 is used to generate a user tag and a tag weight according to the basic information and behavior information.

[0101] The first screening module 304 is used to screen out at least one target recall set from multiple recall sets according to the user's tag.

[0102] The second screening module 305 is used to screen the contents in the target recall set according to the user's tag weight, and generate and output content recommendation results according to the screening results.

[0103] Optionally, in a specific implementation, the label generation module 303 is specifically used to extract a first preset number of keywords from each content; filter out keywords associated with behavioral information and basic information from the keywords as user labels; calculate the product of the sum of behavioral weights of the content corresponding to the user's label and a preset time decay function as the label weight corresponding to the label.

[0104] Optionally, in a specific implementation, the second screening module 305 is specifically used to sort the contents in the target recall set according to the user's tag weight, extract the contents with the first second preset number of sequence numbers in the sorting result as the content recommendation results, and output the content recommendation results.

[0105] Optionally, in a specific implementation, the device also includes: an adjustment module, used to recommend content recommendation results to a first preset proportion of users, and obtain evaluation data of the first preset proportion of users; adjust the labels and label weights of each user in the first preset proportion of users according to the evaluation data; and generate new content recommendation results based on the adjusted user labels and label weights.

[0106] Optionally, in a specific implementation, the adjustment module is specifically used to construct a training sample for the user when the user is offline, the training sample including the user's label, label weight and corresponding evaluation data; obtain a machine learning model based on the training sample; and adjust the user's label and label weight based on the machine learning model.

[0107] Optionally, in a specific implementation, the recall set generation module 302 is specifically configured to extract a first preset number of keywords from each content respectively; and calculate the similarity of each content according to the first preset number of keywords extracted from each content.

[0108] Optionally, in a specific implementation, the recall set generation module 302 is specifically used to extract a third preset number of keywords and the weight corresponding to each keyword from each content based on the first keyword extraction algorithm; extract a fourth preset number of keywords and the weight corresponding to each keyword from each content based on the second keyword extraction algorithm; sort the third preset number of keywords and the fourth preset number of keywords corresponding to each content based on the weight corresponding to each keyword, and filter out the first preset number of keywords.

[0109] Optionally, in a specific implementation, the recall set generation module 302 is specifically used to generate a word vector corresponding to each content according to a first preset number of keywords corresponding to each content; based on a local sensitive hashing algorithm, the similarity of each content is calculated according to the word vector corresponding to each content.

[0110] Optionally, in a specific implementation, the similarity of each content includes at least one of similarity of keywords and similarity of topics.

[0111] The content recommendation device provided in the embodiment of the present application can be specific hardware on the device or software or firmware installed on the device. The implementation principle and technical effects of the device provided in the embodiment of the present application are the same as those of the aforementioned method embodiment. For the sake of brief description, the parts not mentioned in the device embodiment can refer to the corresponding contents in the aforementioned method embodiment. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, devices and units described above can all refer to the corresponding processes in the aforementioned method embodiment, and will not be repeated here.

[0112] See also Figure 4 An embodiment of the present application provides an electronic device 40, including: a processor 401, a storage medium 402 and a bus 403, the storage medium 402 stores machine-readable instructions executable by the processor 401, when the electronic device 40 is running, the processor 401 and the storage medium 402 communicate through the bus 403, and the processor 401 executes the machine-readable instructions to perform the steps of any of the aforementioned embodiments.

[0113] An embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the aforementioned implementations are executed.

[0114] For the specific implementation, please refer to the method embodiment, which will not be described in detail here.

[0115] The content recommendation method, device, electronic device and storage medium provided by the embodiment of the present application first determine the similarity between the contents according to the attribute information of each content, and generate multiple recall sets according to the similarity; then generate the user's label and label weight according to the basic information and behavior information; and filter out at least one target recall set from the multiple recall sets according to the user's label; finally, filter the content in the target recall set according to the user's label weight, and generate and output the content recommendation result according to the filtering result. Thereby improving the accuracy of content recommended to users who do not have clear needs, which is conducive to improving the retention time of these users.

[0116] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0117] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0118] In addition, each functional unit in the embodiments provided in the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0119] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0120] It should be noted that similar numbers and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description and are not to be understood as indicating or implying relative importance.

[0121] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application. They should all be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A content recommendation method, characterized in that: include: Obtaining basic information, behavior information, and attribute information of multiple contents of a user, wherein the basic information includes at least one of the user's age, gender, preference, and region; the behavior information includes at least one of the user's click behavior, browsing behavior, collection behavior, sharing behavior, and comment behavior; and the attribute information includes at least one of the title and type of the content and the behavior weight of the content; Determining the similarity between the contents according to the attribute information of the contents, and generating a plurality of recall sets according to the similarity, each recall set including a plurality of contents whose similarity is greater than a preset threshold; Generate a tag and a tag weight of the user according to the basic information and behavior information; Filtering at least one target recall set from the multiple recall sets according to the user's tag; The content in the target recall set is screened according to the tag weight of the user, and a content recommendation result is generated and output according to the screening result; wherein, The method of generating the user's label and label weight based on the basic information and the behavior information includes: extracting a first preset number of keywords from each content; filtering out keywords associated with the behavior information and the basic information from the keywords as the user's label; and calculating the product of the sum of the behavior weights of the content corresponding to the user's label and a preset time decay function as the label weight corresponding to the label.

2. The method according to claim 1, characterized in that The screening of the content in the target recall set according to the tag weight of the user, and generating and outputting content recommendation results according to the screening results, includes: The contents in the target recall set are sorted according to the tag weights of the user, the contents with sequence numbers before a second preset number in the sorting result are extracted as content recommendation results, and the content recommendation results are output.

3. The method according to claim 1, characterized in that The method further comprises: Recommending the content recommendation result to a first preset proportion of users, and obtaining evaluation data of the first preset proportion of users; Adjusting the label and label weight of each user in the first preset proportion of users according to the evaluation data; Generate new content recommendation results based on the adjusted user tags and tag weights.

4. The method according to claim 3, characterized in that The step of adjusting the user's label and label weight according to the evaluation data includes: When the user is offline, constructing a training sample for the user, the training sample including a label, a label weight, and corresponding evaluation data of the user; Obtaining a machine learning model based on the training samples; The user's tags and tag weights are adjusted based on the machine learning model.

5. The method according to claim 1, characterized in that The determining the similarity between the contents according to the attribute information of the contents includes: extracting a first preset number of keywords from each content respectively; The similarity of each content is calculated according to a first preset number of keywords extracted from each content.

6. The method according to claim 5, characterized in that The extracting a first preset number of keywords from each content respectively includes: Based on the first keyword extraction algorithm, extracting a third preset number of keywords and a weight corresponding to each keyword from each content; Based on the second keyword extraction algorithm, extracting a fourth preset number of keywords and a weight corresponding to each keyword from each content; Based on the weight corresponding to each keyword, the third preset number of keywords and the fourth preset number of keywords corresponding to each content are sorted, and the first preset number of keywords are screened out.

7. A content recommendation device, characterized in that: include: An information acquisition module, used to acquire basic information, behavior information of a user and attribute information of a plurality of contents, wherein the basic information includes at least one of the user's age, gender, preference and location, the behavior information includes at least one of the user's click behavior, browsing behavior, collection behavior, sharing behavior and comment behavior, and the attribute information includes at least one of the title and type of the content and the behavior weight of the content; A recall set generation module, used to determine the similarity between the contents according to the attribute information of the contents, and generate multiple recall sets according to the similarity, each recall set including multiple contents with similarity greater than a preset threshold; A tag generation module, used to generate a tag and a tag weight of the user according to the basic information and behavior information; A first screening module, configured to screen out at least one target recall set from the multiple recall sets according to the user's tag; A second screening module is used to screen the content in the target recall set according to the tag weight of the user, and generate and output content recommendation results according to the screening results; The label generation module is used to extract a first preset number of keywords from each content respectively; filter out keywords associated with the behavior information and basic information from the keywords as labels of the user; calculate the product of the sum of the behavior weights of the content corresponding to the user's label and a preset time decay function as the label weight corresponding to the label.

8. An electronic device, characterized in that: The electronic device comprises: a processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of any method as claimed in claims 1-6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of any method according to claim 1 are executed.