Content recommendation method and system based on user attention type

Abstract

The invention discloses a content recommendation method and system based on a user attention type, and the method comprises the steps: obtaining the overall tolerance of a user to the content, and the overall tolerance comprises a plurality of levels; performing user group recommendation distribution for each type of trial push content according to the content label to form a trial push content recommendation queue; inserting the trial push content queue into a normal recommendation content queue preset by the user according to the overall tolerance of the user; and the user performs complete attention or type attention on the creator. According to the method, the user can have two attention options, namely attention to the field of the creator or complete attention to the creator. The homepage of the creator has obvious field classification to distinguish the content that the user wants to watch; a creator can try innovation in a new field and does not need to worry about powder falling, and meanwhile, the possibility of ring breaking is increased; and the platform obtains more high-quality contents, so that cross-domain innovation of high-quality creators is easier to realize.

Description

Technical Field

[0001] The present invention belongs to the technical field of content recommendation, and in particular, relates to a content recommendation method and system based on user attention types. Background Art

[0002] Existing content recommendation methods, such as short video recommendations, are dominated by user interests, similar users, local, hot content, and large accounts, making it difficult for new creators to obtain traffic. The solidification of traffic levels has led to a lack of innovation in platform content, and talented people dare not innovate rashly. New creators may need to try different directions, but the platform's attention + recommendation model means that users can only follow vertical accounts, so once they are not vertical, it is difficult for new creators' accounts to attract traffic.

[0003] For example, the prior art Chinese patent application CN202111450249.9 discloses a video recommendation method, device, storage medium and computer equipment. In this method, since the basic data of each video in the recommendation pool includes basic features such as anchor type, content label, video rating, and listing time, after sorting each video in the recommendation pool by these basic features, the high-quality anchor's high-quality videos can be recommended to users; and, in the secondary sorting, the application takes into account the popularity data of each video in the first video set, such as the traffic level, attention rate, and dwell time. Therefore, after secondary sorting of each video in the first video set using the popularity data of each video, videos with more traffic levels, higher attention rates, and longer dwell times can be preferentially recommended to new users. Since such videos are more popular, they are easier to catch the attention of new users in a short time, thereby stimulating new users to consume video content and improving the retention rate of new users.

[0004] In addition, the content recommended to users will be mixed with advertisements and test content. The acceptance ratio of each user to advertisements and content that they are not interested in varies greatly. When the sample size is small, the test content will be greatly affected, and a relatively fair test mechanism is needed. The platform needs new creators and new content. How to encourage content innovation, let the platform flourish, and fairly recommend test information to interested people is an area that needs improvement.

[0005] Therefore, the existing technology has defects. The recommendation mechanism solidifies the ecosystem. The recommendation model after following does not allow accounts to have the opportunity to trial and error. It lacks innovation and needs to be improved and optimized. Summary of the invention

[0006] In view of this, the present invention provides a content recommendation method and system based on user attention types, wherein users can pay attention to segmented fields of video creators, so that the system can recommend content for the segmented fields used to pay attention to video creators.

[0007] In order to solve the above technical problems, the technical solution of the present invention is to adopt a content recommendation method based on user attention types, comprising: Obtaining the user's overall tolerance to the content, where the overall tolerance includes several levels; According to the content tags, user groups are recommended and allocated for each type of trial content to form a trial content recommendation queue; Inserting the trial content queue into the normal recommended content queue preset by the user according to the user's overall tolerance; If a user follows a certain content, the user will be marked as a type follower of the content creator in the type of the content; if the user follows the creator, the user will be marked as a full follower of the creator; If the user is a follower of a certain creator, when the user enters the creator's homepage, the creator's content of this type will be displayed separately from other types of content, and the unread and watched parts of the content of this type will be displayed separately; When a user is a full follower of a creator, when the user enters the creator's homepage, the viewed content and the unviewed content are displayed separately.

[0008] As an improvement, methods for obtaining the user's overall tolerance to content include: Crawl logs to obtain the user's behavior patterns on the content in the logs; Label the user's overall tolerance based on their behavior patterns towards the content. As a further improvement, the behavior mode includes: one or more of watching, liking, commenting, collecting, and swiping. As another further improvement, the method of forming the trial content recommendation queue includes: Assign an initial number of recommended users to each type of trial content; Obtain the proportion of people with each overall tolerance level among the online population, and allocate the number of recommended users according to the proportion of people with each overall tolerance level to form a trial content recommendation queue; An adjustment coefficient is set based on the number of recommendations and actual number of views of a certain category of trial content, and the number of recommended users is adjusted using the adjustment coefficient.

[0009] As an improvement, the method for allocating the number of initialized recommended users according to the proportion of people with different overall tolerance levels includes: Using the formula S i =N*(k i *m i ) /

[0010] Calculate the number of users of a certain overall tolerance level; where Si is the number of users of overall tolerance level i, ki is the allocation coefficient of users of overall tolerance level i, mi is the number of online users of users of overall tolerance level i, and n is the number of overall tolerance levels; the overall tolerance user allocation coefficient ki is proportional to the overall tolerance value, and =1.

[0011] As an improvement, the method for adjusting the number of recommended users is to use the formula N'=(N*L+N) / 2 Adjust the number of recommended users; where N' is the number of recommended users after adjustment, N is the number of recommended users before adjustment, and L is the adjustment coefficient; Using the formula L=X / Y Calculate the adjustment coefficient; where L is the adjustment coefficient, Y is the number of recommendations, and X is the actual number of views.

[0012] As an improvement, the method of inserting the trial content queue into the normal recommended content queue preset by the user according to the overall tolerance of the user includes: Obtaining the preset allocation ratio according to the user's overall tolerance level; A content recommendation queue is formed by setting placeholders in the normal recommendation content queue according to the allocation ratio; When the user views the station symbol according to the content recommendation queue, the trial content in the trial content recommendation queue will be pushed to the user.

[0013] As an improvement, when the user's interest in the trial content is less than a threshold, when the placeholder is swiped, a random number is used to determine whether to recommend it to reduce the recommendation ratio; the random number is between 0 and 1, and it is recommended if it is greater than 0.5, so the recommendation ratio is reduced to approximately 50%. After a certain number of plays, the user behavior eases and normal recommendations can be restored.

[0014] As an improvement, when a user is a follower of a certain creator, the content of this creator and this type of content is recommended in the normal recommended content queue; When a user is a complete follower of a creator, all types of content of the creator are recommended in the normal recommended content queue.

[0015] The present invention also provides a content recommendation system based on user-focused types, comprising: A user overall tolerance acquisition module, used to acquire the user's overall tolerance to the content, wherein the overall tolerance includes several levels; A trial push content queue construction module is used to recommend and allocate user groups for each type of trial push content according to content tags to form a trial push content recommendation queue; A content recommendation queue construction module is used to insert the trial content queue into the normal recommended content queue preset by the user according to the user's overall tolerance; Following module: if a user follows a certain content, the user will be marked as a type follower of the content creator in the type of the content; if the user follows the creator, the user will be marked as a full follower of the creator; If the user is a follower of a certain creator, when the user enters the creator's homepage, the creator's content of this type will be displayed separately from other types of content, and the unread and watched parts of the content of this type will be displayed separately; When a user is a full follower of a creator, when the user enters the creator's homepage, the viewed content and the unviewed content are displayed separately.

[0016] The present invention is beneficial in that: The present invention obtains user portraits through overall tolerance, and constructs a trial content queue based on the user portrait. The normal content recommendation queue and the trial content queue are combined into a content recommendation queue to recommend content to users. And according to the user's real-time operation response, the preset trial content can be dynamically reduced to improve the user experience.

[0017] In the present invention, users have two options for following: following the creator's field, or following completely, or partially following and only accepting content in the favorite field; the creator's homepage has obvious field classifications to distinguish the content that users want to see; creators can try innovations in new fields without worrying about losing fans, while increasing the possibility of breaking the circle; the platform obtains more high-quality content, and cross-field innovation by high-quality creators is easier to achieve. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of the present invention.

[0019] Figure 2 It is a structural diagram of the present invention. DETAILED DESCRIPTION

[0020] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below in conjunction with specific implementation methods.

[0021] Example 1 This embodiment takes short video recommendation as an example. Figure 1 As shown, the present invention provides a content recommendation method based on user attention types, comprising: S1 obtains the user's overall tolerance to the content, where the overall tolerance includes several levels.

[0022] The purpose of this step is to obtain user portraits. Users have different tolerance levels for recommended content. For example, some users can carefully read certain recommended content and comment and like it. However, some users do not have enough patience for certain recommended content and directly swipe past it (referred to as "second swipe" in this invention).

[0023] For new content, it is more likely to be viewed, liked, and even followed if it is recommended to users with high overall tolerance.

[0024] The method for obtaining the user's overall tolerance to the content in this step includes: S11 crawls the logs to obtain the user's behavior patterns on the content in the logs.

[0025] Specifically, logs in the App are captured through Logstash and stored using ES.

[0026] Logstash is an open source data processing pipeline tool that is mainly used to collect, process and transform data from various sources (such as log files, databases, APIs, etc.), and then send this data to specified targets (such as Elasticsearch, Kafka, files, etc.). Its core functions include data input, filtering, conversion and output, and it can process structured, unstructured and semi-structured data. In the log processing scenario, Logstash can crawl data from log files, format, filter and extract fields, and then send the processed data to Elasticsearch for storage.

[0027] Elasticsearch (ES) is a distributed search engine and data analysis engine built on Apache Lucene, with powerful full-text retrieval, data storage and analysis capabilities. It supports real-time indexing, complex queries, aggregate analysis and other functions, and is widely used in log analysis, search services, monitoring and data visualization scenarios. In the log storage scenario, Elasticsearch can store log data processed by Logstash, and provide efficient query and analysis functions, supporting the retrieval and analysis of user behavior data.

[0028] Through the combination of Logstash and Elasticsearch, user behavior data can be efficiently collected, processed and stored, providing strong support for subsequent analysis and application.

[0029] S12 labels the overall tolerance of the user based on the user's behavior pattern towards the content.

[0030] In this embodiment, user behavior patterns include but are not limited to: watching, liking, commenting, collecting, swiping, etc. Through statistics and analysis of various user behavior patterns, user profiles can be created based on overall tolerance, for example, users can be divided into three levels: high tolerance, medium tolerance, and low tolerance.

[0031] S2 recommends and allocates user groups for each type of trial content according to content tags to form a trial content recommendation queue.

[0032] The so-called trial content includes content that the platform wants to recommend to users, such as new videos by new authors, advertisements, etc. Such content is generally not in the user's normal recommendation queue (the normal recommendation queue is usually a content list generated based on the user's historical behavior, preferences and the platform's recommendation algorithm. These contents are mainly derived from the user's interest map and behavior data, and are designed to provide users with personalized and highly relevant content).

[0033] Specifically, the method of forming the trial content recommendation queue in this step includes: S21 assigns an initialized number of recommended users to each type of trial content.

[0034] The test push content can be divided into multiple categories according to the type label, such as games, food, emotions, etc. During initialization, each type of test push content is assigned a preset number of initial recommended users, such as 200. In other words, the game-type test push content will be recommended to 200 users during initialization.

[0035] S22 obtains the proportion of people with each overall tolerance level among the online population, and allocates the number of recommended users according to the proportion of people with each overall tolerance level to form a trial content recommendation queue.

[0036] According to the analysis in step S1, for new content, it is more likely to be viewed, liked, or even followed if it is recommended to users with high overall tolerance. Therefore, when allocating users, each type of user needs to be allocated a certain number of proportions.

[0037] More specifically, the method for allocating the number of initialized recommended users according to the proportion of people with different overall tolerance levels in this step includes: Using the formula S i =N*(k i *m i ) /

[0038] Calculate the number of users of a certain overall tolerance level; where Si is the number of users of overall tolerance level i, ki is the allocation coefficient of users of overall tolerance level i, mi is the number of online users of users of overall tolerance level i, and n is the number of overall tolerance levels; the overall tolerance user allocation coefficient ki is proportional to the overall tolerance value, and =1.

[0039] For example, the number of online users of various tags is updated once in a certain period of time, and the number of people with high tolerance M is calculated according to the overall tolerance level. 1 , medium tolerance number M 2 , low tolerance number M 3 The distribution coefficients in the high, medium and low tolerance user groups are k 1 , k 2 , k 3 Finally, the number of recommended users in each group is obtained according to the above formula. For example, among 200 recommended users, 100 are high tolerance, 60 are medium tolerance, and 40 are low tolerance.

[0040] For a certain user, in addition to the normal content recommendation queue, there is also a trial content queue. The normal content recommendation queue and the trial content queue are pre-built when the user logs in. In this step, the system will randomly add the trial recommended content to the trial content queues of 200 users.

[0041] S23 sets an adjustment coefficient according to the number of recommendations and the actual number of views of the trial content of a certain category, and uses the adjustment coefficient to adjust the number of recommended users.

[0042] Specifically, the method for adjusting the number of recommended users in this step is to use the formula N'=(N*L+N) / 2 Adjust the number of recommended users; where N' is the number of recommended users after adjustment, N is the number of recommended users before adjustment, and L is the adjustment coefficient; Using the formula L=X / Y Calculate the adjustment coefficient; where L is the adjustment coefficient, Y is the number of recommendations, and X is the actual number of views.

[0043] If there are 5 trial videos in the game category within a 1-hour period, and a total of 5*200 (initial base) = 1000 recommendations are given to users, and 1 hour is set as a statistical span and statistics are collected every few minutes, and 800 videos are viewed in the past hour (since the statistical span is greater than the recommendation period, X may also be greater than Y). Then the adjustment coefficient L = 800 / 1000 = 0.8.

[0044] The adjusted number of users who recommend each video = (200*0.8+200) / 2 = 180 people. The purpose of this formula is to avoid numerical fluctuations.

[0045] When the modulation coefficient L is greater than or equal to 1, it means that the acceptance of the trial content is high or the number of online users is large, and the delivery strength can be increased. When the modulation coefficient L is less than or equal to 1, it means that the acceptance of the trial content is low or the number of online users is small, and the delivery strength can be appropriately reduced to improve user experience.

[0046] S3 inserts the trial content queue into the normal recommended content queue preset by the user according to the user's overall tolerance. The specific steps include: S31 obtains a distribution ratio preset according to the user's overall tolerance level.

[0047] Each user's overall tolerance level is preset with a corresponding allocation ratio. For example, the allocation ratio for high tolerance level is 1 / 5, which means that one trial content is added after every 4 normal recommended content. It is foreseeable that the allocation ratio of medium tolerance level and low tolerance level users will be lower than that of high tolerance level, such as 1 / 10 and 1 / 15.

[0048] S32 sets placeholders in the normal recommended content queue according to the allocation ratio to form a content recommendation queue.

[0049] The normal recommended content queue and the trial content queue both contain content IDs, such as the normal content recommendation queue is [3016, 4527, 6537, 2413, 3856, ...], and the trial content queue is [5732, 5821 ...]. If 0 is used as a placeholder and the user lists them for high tolerance, the content recommendation queue will be [3016, 4527, 6537, 2413, 0, 3856, ...].

[0050] S33 When the user views the placeholder according to the content recommendation queue, the trial content in the trial content recommendation queue is pushed to the user.

[0051] The system recommends content to users according to the content recommendation queue. Users watch content IDs 3016, 4527, 6537, and 2413 in the above queue. When the user swipes to placeholder 0, the content with ID 5732 in the trial content queue is recommended to the user. When the user swipes to the next placeholder, the content with ID 5821 is recommended to the user, and so on.

[0052] In addition, when the user's interest in the trial content is less than a threshold, the trial content is recommended according to a random recommendation ratio; the random recommendation ratio is less than 100%.

[0053] That is to say, when a user is not interested in the test content, in order to improve the user experience, it is necessary to reduce the delivery of the test content to the user. Every time a placeholder is swiped, the user's recent second swipe ratio is counted. If the second swipe ratio is higher than the threshold, a random number between 0 and 1 is generated and judged whether it is greater than 0.5. If it is greater than 0.5, the test content is recommended, and if it is less than or equal to 0.5, the test content is skipped, thereby reducing the number of test pushes and improving the user experience.

[0054] The user's interest in the trial content can be obtained through analysis of their behavior patterns, such as the percentage of seconds swiped for the trial content, etc., which is not limited in this embodiment.

[0055] S4: If the user follows a certain content, the user is marked as a type follower of the content creator in the type of the content; if the user follows the creator, the user is marked as a full follower of the creator; If the user is a follower of a certain creator, when the user enters the creator's homepage, the creator's content of this type will be displayed separately from other types of content, and the unread and watched parts of this type of content will be displayed separately; When a user is a full follower of a creator, when the user enters the creator's homepage, the viewed content and the unviewed content are displayed separately.

[0056] In this embodiment, the user's attention includes two types: the first is attention to the creator himself, and the second is attention to a certain type of content of the creator.

[0057] The first type of following is a common way of following in existing apps, that is, full following. The user is a full follower of the creator. In the database code, the field for full following is allfollow. When this field is 1, it means that the user is a full follower of the creator.

[0058] The second type of attention is that the user is not interested in all the content created by the creator, but only in a few types of content. The user is a type follower of the creator. In the database code, the attention field is follow, which stores the ID of the type of attention, such as the game type ID is 1001. Each additional type is separated by a comma, such as 1001, 1003, 1005.

[0059] When a user enters a creator's homepage, if the user is a follower of the creator's type, for example, a creator of games or movies, the unwatched parts of the games and movies on the creator's homepage will be framed with a silver border, the watched parts will be framed with a gold border, and other content will not be framed.

[0060] If the user is a full follower of the creator, the unread parts of the content on the creator's homepage will be framed with a gold border, and the viewed parts will be framed with a silver border.

[0061] Of course, the various color frame selections are merely examples of different display methods and are not intended to be a limitation of the present invention.

[0062] In addition, if the user is a follower of a certain creator, the content of this creator and this type will be recommended in the normal recommended content queue; When a user is a complete follower of a creator, all types of content of the creator are recommended in the normal recommended content queue.

[0063] Example 2 like Figure 2 As shown, the present invention also provides a content recommendation system based on user-focused types, including: A user overall tolerance acquisition module, used to acquire the user's overall tolerance to the content, wherein the overall tolerance includes several levels; A trial push content queue construction module is used to recommend and allocate user groups for each type of trial push content according to content tags to form a trial push content recommendation queue; A content recommendation queue construction module is used to insert the trial content queue into the normal recommended content queue preset by the user according to the user's overall tolerance; Following module: if a user follows a certain content, the user will be marked as a type follower of the content creator in the type of the content; if the user follows the creator, the user will be marked as a full follower of the creator; If the user is a follower of a certain creator, when the user enters the creator's homepage, the creator's content of this type will be displayed separately from other types of content, and the unread and watched parts of this type of content will be displayed separately; When a user is a full follower of a creator, when the user enters the creator's homepage, the viewed content and the unviewed content are displayed separately.

[0064] The above are only preferred embodiments of the present invention. It should be noted that the above preferred embodiments should not be regarded as limiting the present invention, and the protection scope of the present invention should be based on the scope defined by the claims. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the spirit and scope of the present invention, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A content recommendation method based on user attention types, characterized in that: include: Obtaining the user's overall tolerance to the content, where the overall tolerance includes several levels; According to the content tags, user groups are recommended and allocated for each type of trial content to form a trial content recommendation queue; Inserting the trial content queue into the normal recommended content queue preset by the user according to the user's overall tolerance; If a user follows a certain content, the user will be marked as a type follower of the content creator in the type of the content; if the user follows the creator, the user will be marked as a full follower of the creator; If the user is a follower of a certain creator, when the user enters the creator's homepage, the creator's content of this type will be displayed separately from other types of content, and the unread and watched parts of this type of content will be displayed separately; When a user is a full follower of a creator, when the user enters the creator's homepage, the viewed content and the unviewed content are displayed separately.

2. A method for content recommendation based on user attention type according to claim 1, characterized in that: Methods for obtaining users' overall tolerance for content include: Crawl logs to obtain the user's behavior patterns on the content in the logs; Label the user's overall tolerance based on their behavior patterns towards the content.

3. A method for content recommendation based on user attention type according to claim 2, characterized in that: The behavior mode includes: one or more of watching, liking, commenting, collecting, and swiping.

4. The method for content recommendation based on user attention type according to claim 1, characterized in that: Methods for forming a test content recommendation queue include: Assign an initial number of recommended users to each type of trial content; Obtain the proportion of people with each overall tolerance level among the online population, and allocate the number of recommended users according to the proportion of people with each overall tolerance level to form a trial content recommendation queue; An adjustment coefficient is set based on the number of recommendations and actual number of views of a certain category of trial content, and the number of recommended users is adjusted using the adjustment coefficient.

5. A method for content recommendation based on user attention type according to claim 4, characterized in that: The method of allocating the initialization dynamic flow value according to the proportion of people with different overall tolerance levels includes: Using the formula S i =N*(k i *m i ) / Calculate the number of users of a certain overall tolerance level; where Si is the number of users of overall tolerance level i, ki is the overall tolerance user allocation coefficient of level ki, mi is the number of online users of overall tolerance level i, and n is the number of overall tolerance levels; the overall tolerance user allocation coefficient ki is proportional to the overall tolerance value, and =1.

6. A method for content recommendation based on user attention type according to claim 4, characterized in that: The method to adjust the number of recommended users is to use the formula N'=(N*L+N) / 2 Adjust the number of recommended users; where N' is the number of recommended users after adjustment, N is the number of recommended users before adjustment, and L is the adjustment coefficient; Using the formula L=X / Y Calculate the adjustment coefficient; where L is the adjustment coefficient, Y is the number of recommendations, and X is the actual number of views.

7. The method for content recommendation based on user attention type according to claim 1, characterized in that: The method of inserting the trial content queue into the normal recommended content queue preset by the user according to the overall tolerance of the user includes: Obtaining the preset allocation ratio according to the user's overall tolerance level; A content recommendation queue is formed by setting placeholders in the normal recommendation content queue according to the allocation ratio; When the user views the station symbol according to the content recommendation queue, the trial content in the trial content recommendation queue will be pushed to the user.

8. The method for content recommendation based on user attention type according to claim 7, characterized in that: When the user's interest in the trial content is less than a threshold, when the user views the placeholder, a random number is used to determine whether to recommend it to reduce the recommendation ratio; the random number is between 0 and 1, and it is recommended if it is greater than 0.

5.

9. The method for content recommendation based on user attention type according to claim 1, characterized in that: If the user is a follower of a certain creator, the content of this creator and this type will be recommended in the normal recommended content queue; When a user is a complete follower of a creator, all types of content of the creator are recommended in the normal recommended content queue.

10. A content recommendation system based on user attention types, characterized in that: include: A user overall tolerance acquisition module, used to acquire the user's overall tolerance to the content, wherein the overall tolerance includes several levels; A trial push content queue construction module is used to recommend and allocate user groups for each type of trial push content according to content tags to form a trial push content recommendation queue; A content recommendation queue construction module is used to insert the trial content queue into the normal recommended content queue preset by the user according to the user's overall tolerance; Following module: if a user follows a certain content, the user will be marked as a type follower of the content creator in the type of the content; if the user follows the creator, the user will be marked as a full follower of the creator; If the user is a follower of a certain creator, when the user enters the creator's homepage, the creator's content of this type will be displayed separately from other types of content, and the unread and watched parts of this type of content will be displayed separately; When a user is a full follower of a creator, when the user enters the creator's homepage, the viewed content and the unviewed content are displayed separately.

Images