Learning interest assessment method and system based on large model

Abstract

The invention discloses a learning interest assessment method and system based on a large model, and the method comprises the steps: firstly obtaining the multi-level identity information of an initial user from a multi-social platform, obtaining an identity tag through an identity assessment large model trained based on a mode clustering and emotional tendency analysis dual verification mechanism, and marking a teenager user; and then teenager users are classified into test and reference groups, contents are sent in different learning content publishing modes to obtain learning and interaction process data, and finally learning interest evaluation results corresponding to different publishing modes are determined based on the data.

Description

Technical Field

[0001] The present invention relates to the field of smart education, and in particular to a learning interest assessment method and system based on a large model. Background Art

[0002] In the field of modern education and learning technology, accurate assessment of learning interests is crucial to optimizing the dissemination of learning content and improving learning outcomes. With the widespread use of social platforms, users have left a wealth of information on these platforms. However, how to effectively mine content related to learning interests from this massive amount of information is a challenge. Traditional learning interest assessment methods often lack accurate identification and classification of user identities, and cannot fully consider the learning and interaction of different user groups under different learning content release methods. Moreover, a single assessment method is difficult to fully and accurately reflect learning interests. Summary of the invention

[0003] The purpose of the present invention is to provide a learning interest assessment method and system based on a large model.

[0004] In a first aspect, an embodiment of the present invention provides a learning interest assessment method based on a large model, comprising:

[0005] Obtain multi-level identity information of multiple initial users from multiple social platforms;

[0006] Inputting the multi-level identity information of the multiple initial users into the identity assessment model trained by the dual verification mechanism based on pattern clustering and sentiment tendency analysis to obtain the identity label corresponding to each of the initial users;

[0007] Marking the initial user whose identity tag is characterized as a teenager as a teenager user;

[0008] Divide multiple teenage users into a test user group and a reference user group, send learning content to the multiple teenage users according to different learning content publishing methods, obtain corresponding learning process data and interaction process data, and determine the learning interest evaluation results corresponding to the different learning content publishing methods based on the learning process data and the interaction process data.

[0009] In a second aspect, an embodiment of the present invention provides a server system, including a server, wherein the server is used to execute the method described in the first aspect.

[0010] Compared with the prior art, the beneficial effects provided by the present invention include: adopting a learning interest assessment method and system based on a large model disclosed by the present invention, by obtaining multi-level identity information of initial users from multiple social platforms, and obtaining identity labels through an identity assessment large model trained based on a dual verification mechanism of pattern clustering and sentiment tendency analysis, and marking young users. Then, the young users are divided into test and reference groups, and the learning and interaction process data are obtained by sending content in different learning content publishing methods, and finally the learning interest assessment results corresponding to different publishing methods are determined based on these data. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can also be obtained based on these drawings without creative work.

[0012] Figure 1 A schematic diagram of the steps of a large model-based learning interest assessment method provided in an embodiment of the present invention;

[0013] Figure 2 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0014] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0015] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings.

[0016] In order to solve the technical problems in the aforementioned background technology, Figure 1 A flowchart of a large-model-based learning interest assessment method provided in an embodiment of the present disclosure is provided below. The large-model-based learning interest assessment method is introduced in detail.

[0017] Step S201, obtaining multi-level identity information of multiple initial users from multiple social platforms;

[0018] Step S202, inputting the multi-level identity information of the multiple initial users into the identity evaluation model trained by the dual verification mechanism based on pattern clustering and sentiment tendency analysis, to obtain the identity label corresponding to each of the initial users;

[0019] Step S203, marking the initial user whose identity tag is characterized as a teenager as a teenager user;

[0020] Step S204, divide multiple youth users into a test user group and a reference user group, send learning content to the multiple youth users according to different learning content publishing methods, obtain corresponding learning process data and interaction process data, and determine the learning interest evaluation results corresponding to the different learning content publishing methods based on the learning process data and the interaction process data.

[0021] In an embodiment of the present invention, illustratively, the server first determines multiple social platforms from which data is to be obtained, such as selecting three social platforms: Weibo, Douyin and Zhihu. For the Weibo platform, the server obtains user information through the open API (application programming interface) provided by Weibo. It can obtain basic information of the user, such as user name, user profile, follow list, fan list, etc. On the Douyin platform, the server reaches a data cooperation agreement with Douyin (under the premise of complying with the privacy policy and relevant laws and regulations) to obtain information such as the user's video release information, the content type of the likes and comments, and the type of account followed. On the Zhihu platform, the server can obtain information such as the user's questions, answers, topics of interest, and likes. This information constitutes part of the multi-level identity information of the initial user. Taking Weibo users as an example, the server obtains a user whose username is "Sunny Boy". His profile mentions that he is a student who likes technology and sports. He follows many technology bloggers and sports stars, and his fans are mainly peers. The video release of another user is obtained from the Douyin platform. The user often posts videos of his own learning painting process. The videos he likes are mostly artistic creation and learning skills, and he follows many painting teaching accounts. The user information obtained from the Zhihu platform shows that a certain user frequently answers questions about mathematics learning, and the topics he is interested in include mathematics competitions and learning methods. Multi-level identity information includes the user's basic personal information (such as age range speculation, gender, etc.), interest tendencies (reflected in terms of attention content, likes content, etc.), social relationships (following list and fan list), and behavior patterns on different platforms (such as the frequency of posts on Weibo, the type of video releases on Douyin, and the activeness of answers and questions on Zhihu). For example, for a certain user, by analyzing his information on three platforms, it is found that he often reposts technology news on Weibo, likes many technology product review videos on Douyin, and answers questions about computer programming learning on Zhihu. Combining this information can construct a more comprehensive multi-level identity information of the user. When constructing a large identity evaluation model, the server first collects a large number of labeled user data samples, which contain multi-level identity information of users with known identity labels (such as teenagers, adults, and the elderly). For the pattern clustering part, the server uses a clustering algorithm (such as the KMeans clustering algorithm) to cluster these samples and cluster users with similar behavior patterns and information characteristics into one category. For example, clustering users who follow more entertainment stars on social platforms, have a light-hearted and lively posting style, and mostly like content related to pop culture can be a characteristic of the teenager or young people group.

[0022] At the same time, for the sentiment analysis part, the server uses natural language processing technology to analyze the sentiment of the content and comments posted by users. For example, if a user often posts positive and energetic content on Weibo and expresses a positive attitude towards learning-related topics, this is also a characteristic of teenagers. By combining pattern clustering and sentiment analysis, a double verification mechanism is constructed to train the identity evaluation model. The server inputs the multi-level identity information of the initial user obtained from the social platform into the trained identity evaluation model. Take an initial user as an example. The user often posts photos of himself participating in school club activities on Weibo, likes many videos of campus life on Douyin, and asks questions about how to improve academic performance on Zhihu. The identity evaluation model finally labels the user with the identity label of "teenager" by analyzing the user's behavior pattern (in line with the pattern of teenagers being active in school-related affairs) and sentiment tendency (positive tendency to be passionate about learning). For example, another user mainly focuses on stock investment and workplace dynamics on Weibo, mostly likes business news interpretations on Douyin, and answers questions about career advancement on Zhihu. The identity assessment model marks him with an identity tag of "adult" based on this information and the training model. After obtaining the identity tags of all initial users, the server marks the initial users with the identity tag of "teenager". For example, after processing the identity tags of 1,000 initial users, the server finds that 300 of them are marked as "teenagers". The server will create a separate list of teen users for these 300 users in the database to distinguish them from other types of users. And add a special identifier to each teen user so that subsequent operations can quickly identify them as teen users. The server divides the marked teen users. Assuming there are 300 teen users in total, the server can divide them into test user groups and reference user groups according to a certain ratio (such as 1:1). For example, 150 teen users are divided into the test user group and the other 150 are divided into the reference user group. The server can use a random division method, or divide according to some other factors (such as previous activity on social platforms, etc.). For example, 150 teenage users with high activity are divided into a test user group, and 150 with relatively low activity are divided into a reference user group. Assuming that the learning content is about mathematical knowledge, the test learning content can be released in the form of fun animation videos, in which mathematical concepts are explained through the interaction of animated characters. The reference learning content is released in the form of traditional text handouts. For teenage users in the test user group, the server sends mathematical learning content in the form of fun animation videos to these 150 teenage users through a special learning platform or private messages on social platforms.For adolescent users in the reference user group, mathematics learning content in the form of traditional text handouts is sent through the same channel. After sending the learning content, the server starts to collect relevant data. Taking a adolescent user in the test user group as an example, the server can obtain his learning process data for watching animated videos, such as the frequency of his video watching, and found that this user watched the video 3 times in a week; and the duration of the viewing, the average duration of each viewing of this user is 15 minutes. This is the learning process data of this user for the fun animation video learning content. At the same time, the server will also obtain the interactive process data. If this user shares the video with his 5 friends (these 5 friends are also adolescent users) after watching the video, these 5 friends become secondary adolescent users. The server will obtain the learning behavior data of these 5 secondary adolescent users for the shared content. For example, one of the secondary adolescent users watched the shared video, and his viewing time was 10 minutes, which is part of the interactive process data. The server first infers the learning interest coefficient formed by the learning process status of all adolescent users for the learning content. For example, through the learning interest coefficient learning interest inference model, for a young user who watches animated video learning content, based on his viewing frequency, viewing time, and his previous interest in learning content on the social platform and other user characteristics, it is inferred that his learning interest coefficient is 0.6. Then the learning interest coefficients of all young users are averaged. Assuming that the learning interest coefficients of 300 young users are [0.6, 0.5, 0.7, ...], the average learning interest mean coefficient is calculated to be 0.55. For a target young user, his learning interest coefficient is 0.6, and the coefficient difference between it and the average learning interest mean coefficient of 0.55 is 0.05. This coefficient difference value is used as the user feature of the target young user to reflect the difference between him and other users in learning interest. Then, the server performs difference optimization processing on the real learning process data and the real interaction process data according to the user behavior differences corresponding to multiple young users. For example, for a target young user, his viewing frequency and duration are quite different from those of other young users, and the server determines that his corresponding difference optimization coefficient is 0.8. Assume that the average viewing time of the remaining adolescent users (adult users in the same group except the target adolescent users) is 12 minutes (mean data of the learning process), and the average interaction time is 8 minutes (mean data of the interaction process). The target adolescent users watch for 10 minutes, and interact for 6 minutes. The difference between the target adolescent users' viewing time and the mean data of the learning process is 2 minutes, and the difference between the interaction time and the mean data of the interaction process is 2 minutes. The difference optimization coefficient of 0.8 is used to optimize these two difference values, and the optimized difference value of the learning data is 1.6 minutes, and the optimized difference value of the interaction data is 1.6 minutes.Then, the optimization difference value of the learning data and the mean data of the learning process are cumulatively calculated, and the optimized learning process data is 10.4 minutes. The optimization difference value of the interactive data and the mean data of the interactive process are cumulatively calculated, and the optimized interactive process data is 6.4 minutes. For the test user group and the reference user group, the server calculates the multiple optimized learning process state values ​​and optimized interactive state values ​​corresponding to the young users in the group. For example, in the test user group, the viewing time (an optimized learning process state value) of all young users is cumulatively calculated to obtain the learning process state cumulative value. Similarly, the interaction time (an optimized interactive state value) is cumulatively calculated to obtain the interactive process state cumulative value. The same operation is performed on the reference user group. Then the learning process state difference value (such as the difference value of the viewing time) and the interactive process state difference value (such as the difference value of the interactive time) corresponding to the same type of optimized learning process state values ​​between the two groups are calculated, and these difference values ​​are used as learning deviation data and interactive deviation data. Assume that the cumulative viewing time of the test user group is 300 minutes, the cumulative viewing time of the reference user group is 200 minutes, and the learning process state difference value is 100 minutes. According to these difference values, the correlated learning process state difference values ​​and interactive process state difference values ​​are cumulatively calculated to obtain multiple first interest coefficients for the test learning content publishing method and the reference learning content publishing method for the learning content. For example, suppose that the two first interest coefficients are 0.8 and 0.7 respectively. Finally, the learning interest evaluation result is determined based on these first interest coefficients. If the first preset threshold is set to 0.6 and the second preset threshold is 1. Among the two first interest coefficients obtained, the number of coefficients greater than 0.6 is 2, which exceeds 1, then it can be determined that the learning interest evaluation result corresponding to the test learning content publishing method (fun animation video form) is better than the reference learning content publishing method (traditional text handout form).

[0023] In an embodiment of the present invention, multiple teenage users are divided into a test user group and a reference user group, learning content is sent to the multiple teenage users according to different learning content publishing methods, corresponding learning process data and interaction process data are obtained, and learning interest evaluation results corresponding to different learning content publishing methods are determined based on the learning process data and the interaction process data. This can be implemented through the following examples.

[0024] Dividing a plurality of adolescent users into a test user group and a reference user group, sending a first learning content to the adolescent users in the test user group according to a test learning content publishing method, and sending a second learning content to the adolescent users in the reference user group according to a reference learning content publishing method;

[0025] Acquire real learning process data of each of the multiple young users for target learning content, and acquire real interactive process data of shared learning content transmitted by each of the young users; the target learning content is the learning content transmitted to each of the young users after being sent, and the target learning content and the shared learning content are derived from a learning content set consisting of the first learning content and the second learning content;

[0026] Determine the user characteristics of each of the youth users, and based on the user behavior differences formed by the learning process states of each of the youth users with respect to the learning content, perform difference optimization processing on the real learning process data and the real interaction process data corresponding to the multiple youth users, respectively, according to the user behavior differences corresponding to the multiple youth users, to obtain the optimized learning process data and the optimized interaction process data corresponding to the multiple youth users, respectively;

[0027] Based on the optimized learning process data of the adolescent users in the test user group and the learning deviation data between the optimized learning process data of the adolescent users in the reference user group, as well as the optimized interaction process data of the adolescent users in the test user group and the interaction deviation data between the optimized interaction process data of the adolescent users in the reference user group, the learning interest assessment results corresponding to the test learning content publishing method and the reference learning content publishing method are determined respectively.

[0028] In an embodiment of the present invention, illustratively, it is assumed that the server has marked 1,000 young users. In order to conduct an effective comparative evaluation, the server divides these 1,000 young users into a test user group and a reference user group according to certain rules. For example, the server can be divided according to geographical location, and 500 young users from certain specific areas (such as the eastern coastal areas) are divided into a test user group, and 500 young users from other areas (such as the central and western regions) are divided into a reference user group. The purpose of this is to test the effect of different learning content publishing methods in different sample groups and reduce the deviations that may be caused by factors such as regions. Assuming that the first learning content is about historical and cultural learning content, the test learning content publishing method adopts the form of virtual reality (VR) experience. In this form, young users can experience historical scenes immersively through VR devices, such as scenes of ancient wars, internal structures of ancient buildings, etc., and there are voice explanations and interactive elements in VR scenes, such as gestures that can be used to select different details of cultural relics or listen to stories of different characters. The second learning content is also about history and culture, but the reference learning content is published in the traditional book reading form, presented in the form of electronic documents or paper books, and the content includes text descriptions, pictures, and simple charts. The server sends the first learning content presented in the form of VR experience to the young users in the test user group through a dedicated learning platform. For these 500 young users, the server will send them the download link or online experience entrance of the VR experience content based on their account information registered on the platform. At the same time, the server sends the second learning content in the form of traditional book reading to the 500 young users in the reference user group, which can be to send electronic documents to the document library under their learning platform account, or to provide a purchase link for paper books (if this is the case, the corresponding electronic version will also be provided for data collection). For the young users in the test user group, the server begins to collect their real learning process data for the first learning content (i.e., the target learning content) in the form of VR experience. Taking Xiao Ming, one of the teenage users, as an example, the server can obtain the following data through the built-in sensors of the VR device and the monitoring function of the learning platform: Learning frequency: Xiao Ming entered the VR history and culture learning scene a total of 4 times in a week, which shows that he is quite interested in this learning content and is willing to experience it many times. Learning duration: After entering the VR scene each time, Xiao Ming stayed for an average of 30 minutes. This includes the time he spent watching explanations in the scene, interacting with interactive elements, and exploring different scenes. Learning progress: The server can also obtain Xiao Ming's progress in the VR learning content. For example, he has completed the study of the origin of ancient civilizations and is entering the medieval history part.For the adolescent users in the reference user group, taking Xiaohong as an example, for the second learning content in the form of traditional book reading (also the target learning content), the server obtains her learning process data: Learning frequency: Xiaohong opened the electronic document to read historical and cultural books 3 times in a week. Learning duration: The average time for each reading is 20 minutes. Learning progress: Xiaohong has read the first few chapters of the book and completed about 30% of the content. Taking Xiaoming as an example, after experiencing the VR historical and cultural learning content, he found it very interesting, so he shared a wonderful clip of this learning content (sharing learning content) to his social group, which has 10 other adolescent users. The server will obtain the interaction of these secondary adolescent users with Xiaoming's shared content as real interaction process data; for one of the secondary adolescent users, Xiaogang, he clicked to watch the VR clip shared by Xiaoming, and the viewing time was 15 minutes. And he also had a brief discussion with Xiaoming after watching it, and left a comment on the learning platform, expressing his interest in this historical scene and hope to learn more about it. Another secondary adolescent user, Xiaoli, although she did not click to watch the shared content, she liked the shared content in the social group, indicating that she had a certain interest in the content. For Xiaohong in the reference user group, after reading a historical and cultural book, she shared an interesting story in the book (sharing learning content) with her friend Xiaomei. Xiaomei read the story and spent 10 minutes reading the related annotations, which is also part of the real interaction process data. The server analyzes multiple user characteristics of each adolescent user. Taking learning historical and cultural content as an example, for Xiaoming, his user characteristics may include: Historical knowledge foundation: Through the previous learning records on the learning platform, the server found that Xiaoming had completed some basic history courses and had a certain understanding of history, which would affect his interest in and learning methods of new historical and cultural learning content. Technology acceptance: Since the first learning content is in the form of VR experience, Xiaoming is more interested in and easy to accept new technologies. He often pays attention to new technological developments on social platforms, which makes him more willing to try this novel learning method. Social influence: Xiaoming has a high degree of activity in social groups, and his sharing is easy to attract the attention of other adolescent users, which is also an important user feature. For Xiaohong, her user characteristics include: Reading habits: From her previous learning records, Xiaohong prefers to read paper books, which will affect her reading experience of electronic documents. Historical interest direction: Xiaohong is more interested in the history of ancient literature, and this historical and cultural learning content covers a wider range of fields, which may affect her learning motivation. Social activity: Xiaohong is relatively low-key in social aspects, and her sharing range is relatively small, which will affect the dissemination range of shared learning content.Taking the test user group as an example, suppose that there are 10 adolescent users whose learning duration data are [30, 25, 35, 20, 40, 15, 28, 32, 22, 38] minutes respectively. The server will calculate the average value of these data as 28.5 minutes. For the adolescent user Xiao Li whose learning duration is 15 minutes, his difference from the average value is large. The server determines the difference optimization coefficient based on Xiao Li's user characteristics. Considering that Xiao Li's acceptance of new technologies is average and his historical knowledge base is relatively weak, he may spend less time in VR experience learning. The server determines his difference optimization coefficient as 0.7 based on these factors. For the learning duration in Xiao Li's real learning process data, the difference between him and the average value is calculated to be 1528.5 = 13.5 minutes. The difference optimization coefficient of 0.7 is used to optimize this difference value, and the optimized difference value is 9.45 minutes. Then the optimized difference value is added to the average value, and the learning duration in Xiao Li's optimized learning process data is 19.05 minutes. Similar processing is performed on the interactive process data. Assume that 5 secondary adolescent users interacted with the VR learning content shared by Xiao Li, and the data of interaction (such as viewing time, etc.) are [10, 8, 12, 6, 9] minutes, with an average of 9 minutes. Among them, the interaction time of one secondary adolescent user is 6 minutes, which is quite different from the average. According to the user characteristics of this secondary adolescent user (such as he has a low interest in history and culture, but is curious about new technologies), the difference optimization coefficient is determined to be 0.8. The difference value is calculated to be 69 = 3 minutes, the difference value after optimization is 2.4 minutes, and the interaction time after optimization is 6.6 minutes, which is part of the optimized interaction process data. The adolescent users in the reference user group are also subjected to difference optimization processing in the same way. For example, for the historical story shared by Xiao Hong, the reading time of the secondary adolescent user Xiao Mei is subjected to difference optimization processing to obtain the optimized interaction process data. For the test user group, the server accumulates the learning duration in the optimized learning process data of all adolescent users. Assume that after optimization processing, the cumulative learning duration of 500 adolescent users is 10,000 minutes. For the reference user group, after the same calculation, the cumulative learning duration of 500 adolescent users is 8000 minutes. Then the learning deviation data is 100008000=2000 minutes. For the interaction deviation data, assume that the cumulative interaction duration in the optimized interaction process data of all adolescent users in the test user group is 5000 minutes, and the cumulative interaction duration in the reference user group is 3000 minutes. Then the interaction deviation data is 50003000=2000 minutes. The server will determine the learning interest evaluation results based on the learning deviation data and the interaction deviation data. A simple method is to calculate a comprehensive interest coefficient.For example, the learning bias data and the interactive bias data are given certain weights respectively (assuming that the learning bias data weight is 0.6 and the interactive bias data weight is 0.4). For the test learning content release method (VR experience form), the comprehensive interest coefficient is calculated as (2000*0.6+2000*0.4)=2000. For the reference learning content release method (traditional book reading form), the comprehensive interest coefficient is 0 (assuming it is used as a benchmark). Since 2000>0, it can be concluded that the learning interest evaluation result corresponding to the test learning content release method (VR experience form) is better than that of the reference learning content release method (traditional book reading form). This means that in the process of disseminating and learning this historical and cultural learning content, the use of VR experience form can better stimulate the learning interest of young users, both in terms of the degree of involvement in the learning process (reflected by the learning bias data) and in terms of the interaction after sharing (reflected by the interactive bias data).

[0029] In an embodiment of the present invention, based on the differences in user behaviors corresponding to the multiple teenage users, difference optimization processing is performed on the real learning process data and the real interaction process data corresponding to the multiple teenage users, so as to obtain the optimized learning process data and the optimized interaction process data corresponding to the multiple teenage users, which can be implemented through the following examples.

[0030] Determining a difference optimization coefficient corresponding to a target teenage user among the plurality of teenage users according to a user behavior difference corresponding to the target teenage user;

[0031] Performing an average calculation on the real learning process data of the remaining young users to obtain learning process mean data, and performing an average calculation on the real interaction process data of the remaining young users to obtain interaction process mean data; the remaining young users are young users in the group to which the target young user belongs except the target young user;

[0032] According to the difference optimization coefficient corresponding to the target youth user, the learning process mean data and the interaction process mean data, difference optimization processing is performed on the real learning process data and the real interaction process data corresponding to the target youth user to obtain the optimized learning process data and the optimized interaction process data corresponding to the target youth user.

[0033] In an embodiment of the present invention, illustratively, it is assumed that the server is processing an evaluation project on scientific knowledge learning content, in which a test user group includes 500 young users. In this group, there is a target young user Xiao A. The server first analyzes the differences in user behavior of Xiao A. In terms of Xiao A's learning frequency, he only learned scientific knowledge content once in a week, while the average learning frequency of other young users in the same group was 3 times a week. By checking Xiao A's historical records on the learning platform, the server found that Xiao A also participated in many artistic creation activities, which may have distracted his energy from learning scientific knowledge. In terms of learning duration, the average duration of Xiao A's learning of scientific knowledge content each time was 15 minutes, while the average duration of other young users in the group was 30 minutes. Further analysis found that Xiao A's slow understanding of the theoretical part of scientific knowledge may be due to his lack of solid foundation in the previous basic science courses. In terms of interaction, after Xiao A shared the learning content, only 2 secondary young users interacted with his shared content (watched or commented), while the average number of interactions after other young users shared content was 5 people. Xiao A has relatively few friends on the social platform, and his friends are not very interested in scientific knowledge, which affects the interaction of his shared content. The server determines the difference optimization coefficient based on these differences in user behavior of Xiao A. In terms of learning frequency, Xiao A has a large difference from the average learning frequency of the group. Considering that he is distracted by participating in other activities and has a relatively low level of investment in scientific knowledge, this factor reduces his difference optimization coefficient by 0.2 in this regard. In terms of learning duration, slow understanding due to weak foundation is an important factor, which reduces his difference optimization coefficient by 0.3 in this regard. In terms of interaction, the small number of people interacting due to social factors reduces his difference optimization coefficient by 0.1 in this regard. Taking these factors into consideration, the server determines that Xiao A's difference optimization coefficient is 0.4 (assuming the initial coefficient is 1, 10.20.30.1=0.4). For the test user group (a total of 500 people) where Xiao A is located, the learning process data of the remaining 499 adolescent users except Xiao A are used by the server to calculate the mean data of the learning process. In terms of learning frequency, the server counts the total learning frequency of these 499 adolescent users within a week. Assuming that the total learning frequency of these 499 young users is 1497 times (that is, an average of about 3 times per person, which is consistent with the average learning frequency mentioned above), then the average learning process data of learning frequency is 3 times / week. In terms of learning duration, the server counts the total duration of each time these 499 young users learn scientific knowledge content. Assuming that this total is 14970 minutes (that is, an average of 30 minutes per person), then the average learning process data of learning duration is 30 minutes / time. In terms of interaction, the server counts the total number of people who interact after these 499 young users share learning content.Assuming that the total number is 2495 people (i.e. the average number of interactions per person is 5 people), then the mean data of the interaction process of the number of interactions is 5 people.

[0034] At the same time, the server can also count the total interaction time of secondary youth users after these 499 youth users share learning content (assuming that the total interaction time is 12475 minutes), and the average interaction time is 12475÷2495=5 minutes, which is also part of the mean data of the interaction process. Xiao A's actual learning frequency is once a week, and the mean data of the learning process is 3 times a week, with a difference of 13=2 times. Xiao A's difference optimization coefficient is 0.4, and the difference value is optimized to obtain an optimized difference value of 2×0.4=0.8 times. Perform a cumulative operation on the optimized difference value and the mean data of the learning process, and obtain Xiao A's optimized learning frequency of 3+(0.8)=2.2 times / week, which is the learning frequency in Xiao A's optimized learning process data. Xiao A's actual learning duration is 15 minutes / time, and the mean data of the learning process is 30 minutes / time, with a difference of 1530=15 minutes. The difference optimization coefficient of 0.4 is used to optimize the difference value, and the optimized difference value is 15×0.4=6 minutes. The optimized difference value and the mean data of the learning process are cumulatively added, and the optimized learning duration of Xiao A is 30+(6)=24 minutes / time, which is the learning duration in the optimized learning process data of Xiao A. The actual number of people who interacted with Xiao A is 2, the mean data of the interaction process is 5, and the difference is 25=3. The difference optimization coefficient of 0.4 is used to optimize the difference value, and the optimized difference value is 3×0.4=1.2. The optimized difference value and the mean data of the interaction process are cumulatively added, and the optimized number of people who interacted with Xiao A is 5+(1.2)=3.8, which is the number of people who interacted with Xiao A in the optimized interaction process data. Assume that after Xiao A shares the content, the average interaction time of secondary youth users is 3 minutes (actual situation), the mean data of the interaction process is 5 minutes, and the difference is 35=2 minutes. The difference value is optimized with a difference optimization coefficient of 0.4, and the optimized difference value is 2×0.4=0.8 minutes. The optimized difference value and the mean data of the interaction process are added together to obtain the optimized interaction time of Xiao A, which is 5+(0.8)=4.2 minutes. This is the interaction time in Xiao A's optimized interaction process data. Through such difference optimization processing, the server can more accurately evaluate the performance of each adolescent user in the learning process and the interaction process, and takes into account the differences between individuals and groups, thereby providing a more reliable data basis for subsequent learning interest evaluation. When processing a large amount of adolescent user data, the server will repeat the above steps for each target adolescent user to obtain complete optimized learning process data and optimized interaction process data for the entire group.

[0035] In an embodiment of the present invention, based on the difference optimization coefficient corresponding to the target youth user, the learning process mean data and the interaction process mean data, difference optimization processing is performed on the real learning process data and the real interaction process data corresponding to the target youth user to obtain the optimized learning process data and the optimized interaction process data corresponding to the target youth user, which can be implemented through the following examples.

[0036] Obtaining a learning data difference value between the real learning process data corresponding to the target adolescent user and the learning process mean data, and an interaction data difference value between the real interaction process data corresponding to the target adolescent user and the interaction process mean data;

[0037] Using the difference optimization coefficient corresponding to the target youth user, performing a difference optimization operation on the learning data difference value to obtain a learning data optimized difference value, and using the difference optimization coefficient corresponding to the target youth user, performing a difference optimization operation on the interaction data difference value to obtain an interaction data optimized difference value;

[0038] An accumulation operation is performed on the optimization difference value of the learning data and the mean data of the learning process to obtain the optimized learning process data corresponding to the target youth user; an accumulation operation is performed on the optimization difference value of the interaction data and the mean data of the interaction process to obtain the optimized interaction process data corresponding to the target youth user.

[0039] In the embodiment of the present invention, for example, it is assumed that the server is processing an evaluation task on foreign language learning content, in which a test user group includes several young users. There is a target young user Xiao B. In terms of learning frequency, the server counts that Xiao B has learned foreign languages ​​8 times in a month. The server counts the learning frequencies of other young users in the group where Xiao B is located (except Xiao B) and calculates the mean data of the learning process. Assume that after calculation, the average number of times other young users in this group learn foreign languages ​​in a month is 12 times. Then the learning data difference between Xiao B's learning frequency and the learning process mean data is: 812=4 times. In terms of learning duration, the average duration of Xiao B's foreign language learning each time is 35 minutes. The server calculates that the average duration of foreign language learning for other young users in the group is 45 minutes each time (learning process mean data). The learning data difference between Xiao B's learning duration and the learning process mean data is: 3545=10 minutes. In the process of learning a foreign language, Xiao B shared some learning materials in the social group, and 5 secondary adolescent users interacted with his shared content (watched, commented or liked, etc.). The server counted that after other adolescent users in Xiao B's group (except Xiao B) shared foreign language learning materials, an average of 8 secondary adolescent users interacted (interaction process mean data). The difference in interaction data between the number of people who interacted after Xiao B's sharing and the mean data of the interaction process was: 58 = 3 people. The average interaction time of secondary adolescent users after Xiao B shared the content was 12 minutes. The server calculated that after other adolescent users in the group shared the content, the average interaction time of secondary adolescent users was 15 minutes (interaction process mean data). The difference in interaction data between the interaction time of Xiao B's secondary adolescent users and the mean data of the interaction process was: 1215 = 3 minutes. Based on Xiao B's multiple user behavior characteristics (such as previous foreign language learning foundation, preference for learning methods, and activity on social platforms, etc.), the server determined that Xiao B's difference optimization coefficient was 0.6. For the learning data difference value of Xiao B's learning frequency of 4 times, the difference optimization coefficient is used to perform the difference optimization operation, and the learning data optimization difference value is: 4×0.6=2.4 times. For the learning data difference value of Xiao B's learning duration of 10 minutes, the difference optimization coefficient of 0.6 is used to perform the difference optimization operation, and the learning data optimization difference value is: 10×0.6=6 minutes. For the interactive data difference value of the number of people who interacted after Xiao B shared, 3 people, the difference optimization coefficient of 0.6 is used to perform the difference optimization operation, and the interactive data optimization difference value is: 3×0.6=1.8 people. For the interactive data difference value of the interactive time of Xiao B's secondary youth users of 3 minutes, the difference optimization coefficient of 0.6 is used to perform the difference optimization operation, and the interactive data optimization difference value is: 3×0.6=1.8 minutes.For learning frequency, the mean data of Xiao B's learning process is 12 times, and the optimized difference value of learning data is 2.4 times. The optimized difference value of learning data and the mean data of learning process are cumulatively added, and the learning frequency in Xiao B's optimized learning process data is: 12 + (2.4) = 9.6 times. For learning duration, the mean data of Xiao B's learning process is 45 minutes, and the optimized difference value of learning data is 6 minutes. The optimized difference value of learning data and the mean data of learning process are cumulatively added, and the learning duration in Xiao B's optimized learning process data is: 45 + (6) = 39 minutes. For the number of people who interact after sharing, the mean data of Xiao B's interaction process is 8 people, and the optimized difference value of interaction data is 1.8 people. The optimized difference value of interaction data and the mean data of interaction process are cumulatively added, and the number of people who interact after sharing in Xiao B's optimized interaction process data is: 8 + (1.8) = 6.2 people. For the interaction time of secondary adolescent users, the average data of Xiao B's interaction process is 15 minutes, and the optimized difference value of the interaction data is 1.8 minutes. The optimized difference value of the interaction data and the average data of the interaction process are added together to obtain the interaction time of secondary adolescent users in Xiao B's optimized interaction process data: 15+(1.8)=13.2 minutes. Through the above detailed steps, the server can accurately optimize the data for each target adolescent user, so that the evaluation results can better reflect the actual situation of each adolescent user in the learning process and the interaction process, while also taking into account the differences between individuals and groups. When processing large-scale adolescent user data, the server will operate on each target adolescent user one by one according to this process, thereby providing a reliable data basis for a more comprehensive and accurate evaluation of the impact of different learning content release methods on adolescent learning interests.

[0040] In the embodiment of the present invention, the determination of the user characteristics of each teenage user and the differences in user behaviors formed by the learning process status of each teenage user with respect to the learning content may be implemented through the following examples.

[0041] Using a learning interest coefficient learning interest inference model, inferring learning interest coefficients formed by the learning process states of the plurality of young users for the learning content according to the user characteristics corresponding to the plurality of young users respectively;

[0042] Performing an average calculation on the learning interest coefficients respectively corresponding to the plurality of adolescent users to obtain an average learning interest mean coefficient;

[0043] Obtain the learning interest coefficient corresponding to the target youth user, and the coefficient difference value between it and the average learning interest mean coefficient, and use the coefficient difference value as the user feature of the target youth user, and the user behavior differences formed by the learning process status of each youth user for the learning content.

[0044] In an embodiment of the present invention, illustratively, it is assumed that the server is evaluating a project on art appreciation learning content, involving a group of 1,000 young users. For one of the young users, Xiao C, the server first analyzes his multi-dimensional user characteristics. In terms of art foundation, Xiao C has previously participated in school painting courses and has a certain understanding of color and composition, which is a positive factor. From his behavior on the social platform, he follows many accounts of art museums and well-known painters, and often likes posts related to artworks, which shows that he has a strong external interest in art. In terms of learning habits, Xiao C is usually more active at night, he likes to study in a quiet environment, and when he learns other knowledge, he tends to obtain information through video tutorials. In addition, his family environment attaches great importance to art, and there are many art books and painting tools at home, which will subtly affect his attitude towards art appreciation learning. The server inputs these user characteristics of Xiao C into the learning interest coefficient learning interest inference model. This model is trained by a large amount of labeled learning data of young users (including known user characteristics and corresponding learning interest levels). Based on Xiao C's artistic foundation, social platform behavior, learning habits, and family environment, the model comprehensively infers that Xiao C's learning interest coefficient for the learning process state of art appreciation learning content is 0.7. This means that Xiao C has a high interest in art appreciation learning content and will show high enthusiasm in the learning process. Let's look at another young user, Xiao D. Xiao D does not have much artistic foundation and rarely pays attention to art-related content on social platforms, but he has a wide range of interests in cultural knowledge. His learning habit is to like reading text materials, and his school often organizes cultural activities, which makes him curious about all kinds of knowledge learning. After inputting these user characteristics of Xiao D into the learning interest coefficient learning interest inference model, the model infers that Xiao D's learning interest coefficient for art appreciation learning content is 0.3, indicating that his initial interest in art appreciation learning content is relatively low. The server performs similar user feature analysis and learning interest coefficient inference on these 1,000 young users one by one. For example, for the young user Xiao E, he is a photography enthusiast, has a unique insight into visual arts, and often participates in photography exhibitions. When learning other knowledge, he is good at communicating and discussing with others, and his family also supports him to participate in various art activities. After the model inference, Xiao E's learning interest coefficient is 0.8. As for the young user Xiao F, he is mainly interested in science and technology and sports, has almost no involvement in art, and does not follow art-related accounts on social platforms. His learning habit is to focus on practical operations, and he may not be very interested in theoretical art appreciation. The model infers that Xiao F's learning interest coefficient is 0.1. The server sums the corresponding learning interest coefficients of these 1,000 young users.Assume that the 1000 learning interest coefficients are [0.7, 0.3, 0.8, 0.1, ...], and add them up to get a total of 500 (this is just an assumed sum, the actual calculation is based on all 1000 coefficients). Then, the server divides this sum by the total number of young users 1000, and gets the average learning interest mean coefficient of 0.5. This average learning interest mean coefficient represents the average interest level of these 1000 young users in art appreciation learning content as a whole. For Xiao C mentioned above, his learning interest coefficient is 0.7, and the average learning interest mean coefficient is 0.5. The coefficient difference between the two is calculated as: 0.70.5 = 0.2. This coefficient difference of 0.2 will be used as a user feature of Xiao C to reflect the difference in learning interest between Xiao C and the overall young users. This difference will affect Xiao C's behavior in the process of art appreciation learning. For example, he may participate in learning activities more actively than other teenagers, such as actively looking for more art appreciation materials or participating in related online discussions. Since Xiao C's coefficient difference is positive, it means that his learning interest is higher than the average level. In the actual learning process, Xiao C will show different behaviors from other teenagers. For example, in terms of learning resource acquisition, Xiao C will actively search for some high-end art appreciation courses or lecture videos of well-known foreign art critics, while other teenagers may only be satisfied with basic learning materials. In terms of learning time arrangement, Xiao C will take more time to learn art appreciation every week, which may reach 5 hours, while the average teenager may only spend 3 hours. In terms of interaction, Xiao C is more likely to take the initiative to initiate discussions on art appreciation topics and actively respond to other users' comments. His interaction frequency may be about 30% higher than other teenagers. Xiao D's learning interest coefficient is 0.3, and the average learning interest mean coefficient is 0.5. The coefficient difference between the two is calculated as: 0.30.5=0.2. This coefficient difference of 0.2 will be used as a user feature of Xiao D, indicating that Xiao D's learning interest is lower than the average level. In terms of learning resource acquisition, Xiao D may only obtain the most basic art appreciation learning materials and is unlikely to take the initiative to explore more in-depth content. In terms of learning time arrangement, he may only spend 1 hour a week learning art appreciation, which is far lower than the average of 3 hours. In terms of interaction, Xiao D may rarely actively participate in discussions related to art appreciation, and rarely respond to other users' comments. His interaction frequency may be about 40% lower than the average level. Through the above steps, the server can accurately determine the user characteristics of each adolescent user, especially through the coefficient difference between the learning interest coefficient and the average learning interest mean coefficient to characterize the user behavior difference, which helps to more carefully analyze the reactions of different adolescent users to learning content and the impact of different learning content release methods on their learning interest in the subsequent learning interest evaluation process.

[0045] In the embodiment of the present invention, the acquisition of the real learning process data of each of the multiple teenage users for the target learning content may be implemented through the following examples.

[0046] Acquire first learning behavior data of a target youth user among the plurality of youth users for the corresponding target learning content;

[0047] Acquire, from the first learning behavior data corresponding to the target adolescent user, a plurality of learning process state values ​​of the target adolescent user for the corresponding target learning content;

[0048] The multiple learning process status values ​​of the target teenage user are used as the real learning process data of the target teenage user for the target learning content.

[0049] In an embodiment of the present invention, illustratively, it is assumed that the server is processing data for an evaluation project on programming learning content, which involves a group of 2,000 young users. The server selects a target young user, Xiao G, from this group as the analysis object. For the case of Xiao G learning programming, the server obtains his first learning behavior data in a variety of ways. First, Xiao G uses a special programming learning platform, which records Xiao G's login information, operation records, etc. The server has a data interaction protocol with this learning platform and can obtain Xiao G's relevant data on the platform. For example, the server can obtain Xiao G's login time and find that Xiao G usually logs in to the platform to learn programming in the afternoon after school (approximately 17:0019:00), which is part of his first learning behavior data. At the same time, the learning platform also records Xiao G's operation track in the platform. Xiao G completed a series of programming tasks on the platform, and the server can obtain information such as the time he started each task, the time he completed the task, the difficulty level of the task, and the number of times he viewed the help document during the task. In addition, Xiao G's school also has an online learning management system, which records some activities related to Xiao G's participation in programming courses in the school. The server obtains Xiao G's data in the school's online learning management system through legal means (such as data sharing authorized by the school), such as Xiao G's participation in online discussions of programming courses, including the number of times he spoke, the content of his questions, and the situation of answering other students' questions. From Xiao G's login records on the programming learning platform, the server can determine Xiao G's learning process status value related to his learning frequency. Xiao G logged into the programming learning platform 15 times in a month, which shows that he has a certain regularity and enthusiasm for programming learning. After further analysis, the server found that Xiao G logged into the platform more frequently on Tuesdays, Thursdays and Saturdays every week, which is also a state value reflection of learning frequency, indicating that Xiao G may have more free time for programming learning on these days. For the duration of Xiao G's learning programming on the platform each time, the server can calculate the time interval from the start of the task to the completion of the task. For example, Xiao G spent an average of 30 minutes to complete a simple programming task, and an average of 60 minutes to complete a medium-difficulty task. By analyzing the duration of tasks of different difficulty levels, the server obtained Xiao G's learning process status value in terms of learning duration. In addition, the server also found that during a continuous learning process, Xiao G's longest learning duration was 90 minutes and the shortest was 15 minutes. These data were also recorded as status values ​​related to the learning duration. The programming learning platform sets different learning stages and tasks for each programming course. The server can determine the learning process status value related to Xiao G's learning progress based on the number and difficulty of tasks completed by him.Xiao G has completed 80% of the tasks in the introductory programming course and has begun to dabble in some basic content of the intermediate programming course, which shows that he has made some progress in programming learning. The server can also further refine the learning progress status value according to Xiao G's stay time at different learning stages. For example, Xiao G spends more time on some key knowledge points in the introductory course (such as loop structure and function definition), which may mean that he needs more learning and understanding on these knowledge points. The number of times Xiao G checks the help document during the process of learning programming is an important learning process status value related to learning resource utilization. The server found that when Xiao G completed medium-difficulty tasks, he checked the help document an average of 3 times per task, and when he completed simple tasks, he checked the help document an average of 1 time per task. In addition, whether Xiao G uses the additional learning resources provided by the platform (such as sample code library, programming forum, etc.) is also part of the learning resource utilization status value. According to the survey, Xiao G occasionally browses the sample code library, but rarely participates in discussions in the programming forum, which reflects his characteristics in learning resource utilization. The server integrates Xiao G's learning frequency (15 logins in a month, with high login frequencies on Tuesdays, Thursdays, and Saturdays), learning duration (average time for tasks of different difficulty levels, longest and shortest continuous learning time), learning progress (80% completion of introductory courses, involvement in the basic content of intermediate courses, and time spent on key knowledge points), and learning resource utilization (number of times help documents were viewed, and utilization of additional learning resources). These integrated learning process status values ​​constitute Xiao G's real learning process data for programming learning content (target learning content). For example, the server can store these data in a structured form in the database. A record can be: {"teenage user": "Little G", "learning content": "programming", "learning frequency": "15 times / month, high frequency on Tuesdays, Thursdays, and Saturdays", "learning duration": "simple tasks average 30 minutes, medium tasks average 60 minutes, longest 90 minutes, shortest 15 minutes", "learning progress": "80% for introductory courses, involved in the basics of intermediate courses, and stay on key knowledge points", "learning resource utilization": "view help documents 3 times for each medium task, 1 time for simple tasks, occasionally browse sample code libraries, rarely participate in forum discussions"}. Through the above detailed steps, the server can accurately obtain the real learning process data of each teenage user for the target learning content, which is of great significance for the subsequent analysis of the learning interests of teenage users and the evaluation of the effects of different learning content publishing methods. When processing large-scale teenage user data, the server will operate each target teenage user according to this process to fully grasp the learning situation of the entire teenage user group.

[0050] In the embodiment of the present invention, the acquisition of multiple learning process status values ​​of the target adolescent user for the corresponding target learning content from the first learning behavior data corresponding to the target adolescent user can be implemented through the following examples.

[0051] Obtaining the learning frequency of the target young user for the corresponding target learning content from the first learning behavior data corresponding to the target young user;

[0052] Obtaining, from the first learning behavior data corresponding to the target adolescent user, a learning duration of the target adolescent user for the corresponding target learning content;

[0053] The learning frequency and learning duration corresponding to the target adolescent user are used as multiple learning process state values ​​of the target adolescent user for the corresponding target learning content.

[0054] In an embodiment of the present invention, illustratively, it is assumed that the server is processing a project about online math course learning, and the target youth user is Xiao H. The server obtains Xiao H's first learning behavior data on the online math course learning platform. This learning platform records Xiao H's login information in detail, including the timestamp of each login. The server first analyzes these login records to determine Xiao H's learning frequency. In one day, the server found that Xiao H's login situation was as follows: Xiao H logged into the platform at 8:00-8:30 in the morning to check the course schedule for the day, which can be regarded as a learning-related login behavior. Then at 14:00-15:30 in the afternoon, Xiao H logged into the platform again, took an online math test, and watched the explanation video after the test. This is the second learning-related login behavior. At 20:00-21:00 in the evening, Xiao H logged into the platform again, reviewed the math knowledge points learned that day, and did some after-class exercises. This is the third learning-related login behavior. Therefore, Xiao H's daily learning frequency for the online math course (target learning content) on this day is 3 times. The server continues to analyze Xiao H's login situation within a week. After statistics, the number of learning-related logins of Xiao H from Monday to Sunday is 3, 2, 1, 3, 2, 1, and 2 respectively. Adding these times, Xiao H's weekly learning frequency for the online mathematics course in this week is 14 times. For the monthly learning frequency, the server uses the same method to count the number of learning-related logins of Xiao H in a month (assuming 30 days). Assuming that after detailed statistics, the total number of logins of Xiao H in this month is 60 times, then Xiao H's monthly learning frequency for the online mathematics course is 60 times. The server will also consider the impact of special circumstances on the learning frequency. For example, in a certain week, Xiao H's school organized an online simulation exam for a mathematics competition, which led to a significant increase in the number of learning-related logins of Xiao H that week. The server will mark and analyze such special events to more accurately understand the reasons for the changes in Xiao H's learning frequency. In this case, the server found that Xiao H logged in 5 times on the day of the simulation exam, 23 more times than usual. This shows that special learning activities will prompt Xiao H to increase the frequency of learning, but this increase is temporary. Assume that the target youth user is Xiao I, and the learning content is an online English learning course. The server obtains Xiao I's first learning behavior data on the English learning platform, including the start time and end time records of each learning process. For example, in a vocabulary learning course, Xiao I started learning at 10:00 in the morning and finished this part of the learning at 10:30, so Xiao I's learning duration in this vocabulary learning is 30 minutes. Online English courses contain multiple learning modules, such as listening practice, oral practice, reading and writing. The server counts Xiao I's learning duration in each module separately. In the listening practice module, Xiao I performed listening practice at 14:00-14:20 in the afternoon, and the learning duration was 20 minutes.In the oral practice module, Xiao I started at 14:30 and ended at 15:00, with a learning duration of 30 minutes. In the reading module, Xiao I studied at 15:10-15:40, with a duration of 30 minutes, and in the writing module, Xiao I started at 15:50 and ended at 16:30, with a learning duration of 40 minutes. For a complete online English learning process of Xiao I, the server calculates the total duration of a single learning. Add the learning duration of each of the above modules: 30+20+30+40=120 minutes, which is the total duration of a single learning of Xiao I in this complete learning process. In order to have a more comprehensive understanding of Xiao I's learning habits, the server also calculates Xiao I's average learning duration. Assume that Xiao I conducted 10 complete online English learning processes in a month, and the total duration of each time was 120, 100, 130, 90, 110, 140, 100, 120, 110, and 130 minutes respectively. Adding these times together, we get 1150 minutes, and then dividing by 10, Xiao I's average study duration is 115 minutes. The server further analyzes Xiao I's study duration trend over a longer period of time (such as several months or a semester). For example, in the first month of the semester, Xiao I's average study duration is 100 minutes. As the semester progresses, in the second month, Xiao I's average study duration increases to 115 minutes, and in the third month, it increases to 125 minutes. By analyzing this trend, the server can find that Xiao I's investment time in online English learning is gradually increasing, which may reflect Xiao I's increasing interest in English learning or an increase in learning needs. For Xiao H's online math course learning, the server uses Xiao H's learning frequency (daily learning frequency, weekly learning frequency, monthly learning frequency, and changes in learning frequency under special circumstances) and learning duration (if there are statistics on learning duration for different math knowledge points or question types, they can also be included) as Xiao H's multiple learning process state values ​​for online math courses. For example, the set of learning process status values ​​of Xiao H can be expressed as {"learning frequency":{"daily frequency":3 times, "weekly frequency":14 times, "monthly frequency":60 times, "special circumstances impact":"23 times added on the day of the mathematics competition simulation test"},"learning duration":"based on the time statistics of different learning tasks"}. For Xiao I's online English learning, the set of learning process status values ​​of Xiao I is {"learning frequency":"based on the statistics of the number of learning logins in different time periods","learning duration":{"total duration of a single time":120 minutes,"average duration":115 minutes,"long-term trend":"gradually increases as the semester progresses"}}. Through the above detailed steps, the server can accurately obtain the learning frequency and learning duration from the first learning behavior data corresponding to the target youth user, and use them as multiple learning process status values.These status values ​​are important for gaining a deeper understanding of the learning process of young users, assessing their learning interests, and optimizing the way learning content is released. When processing a large amount of young user data, the server will repeat these steps for each target young user to fully understand the learning situation of the entire group of young people.

[0055] In the embodiment of the present invention, the acquisition of the real interactive process data of the shared learning content transmitted by each of the youth users may be implemented through the following examples.

[0056] Determine a secondary youth user corresponding to a target youth user among the plurality of youth users; the secondary youth user is a youth user among the plurality of youth users who receives the shared learning content transmitted by the target youth user;

[0057] Acquire second learning behavior data of the shared learning content transmitted by the secondary youth user to the target youth user;

[0058] Analyzing and obtaining, from the second learning behavior data, a plurality of learning process status values ​​of the shared learning content transmitted by the secondary youth user to the target youth user as a plurality of interaction status values ​​corresponding to the target youth user;

[0059] The multiple interaction status values ​​corresponding to the target youth user are used as the real interaction process data of the shared learning content transmitted by the target youth user.

[0060] In an embodiment of the present invention, illustratively, it is assumed that the server is processing interactive data about online music course learning content. There is a target young user Xiao J. After completing an online course on classical music appreciation, he felt that the course content was very exciting, so he decided to share a clip of the course (shared learning content) to his social learning group. This social learning group is composed of multiple young users, all of whom are interested in various types of knowledge learning. The server determines which young users are the secondary young users corresponding to Xiao J through the interaction records of the social learning group. The course clip shared by Xiao J is seen by other members in the group, and the server records the young users who click to view the shared content. For example, there are five young users, Xiao K, Xiao L, Xiao M, Xiao N and Xiao O, who click to view the classical music appreciation course clip shared by Xiao J, then these five young users are the secondary young users corresponding to Xiao J. If Xiao J does not share in the social learning group, but shares this course clip on a special music learning forum. The server will determine the secondary young users through the browsing records and click records of the forum. For example, in this music learning forum, three young users, Xiao P, Xiao Q, and Xiao R, clicked to view Xiao J's shared content, and they also became Xiao J's corresponding secondary young users. In this case, the server needs to integrate the secondary young user information from different platforms (social learning groups and music learning forums) to fully obtain the interactive situation of Xiao J's shared content. For the case where Xiao J shared a clip of a classical music appreciation course in a social learning group, the server obtained the second learning behavior data of secondary young users (Xiao K, Xiao L, Xiao M, Xiao N, and Xiao O) from the background data of the social learning group. These data include the timestamps when the secondary young users viewed the shared content. For example, Xiao K clicked to view it within 10 minutes after Xiao J shared it, and Xiao L viewed it 30 minutes later. At the same time, the server can also obtain the device information when they viewed the shared content, such as Xiao M used a mobile phone to view it, and Xiao N used a tablet to view it. The server can also obtain the operation records of secondary young users when viewing shared content. For example, when Xiao O was viewing a clip of a classical music appreciation course shared by Xiao J, he clicked on some annotation links in the clip to get more information about the musicians. Xiao K paused several times during the viewing process, perhaps because he was thinking or taking notes (although it cannot be directly determined, these operation records are an important basis for analyzing learning behavior). In addition to viewing the shared content, the server also records the interactive operations of secondary youth users. After watching the shared content, Xiao L made a comment in the group, expressing his great interest in a certain movement in classical music and hoping to learn more about related works. Xiao M liked Xiao J's sharing, which is also a record of interactive operations.For viewing timeliness, the server uses the time when Xiao J shared the content as the benchmark and calculates the time difference of the secondary adolescent users' viewing as the learning process status value related to viewing timeliness. Xiao K's viewing timeliness status value is 10 minutes, and Xiao L's is 30 minutes. Generally speaking, viewing timeliness can reflect the initial interest of secondary adolescent users in shared content. The shorter the viewing time, the higher the initial interest may be. From the device information of secondary adolescent users viewing shared content, the server can analyze the learning process status value related to device use. For example, Xiao M uses a mobile phone to view, and the mobile phone screen is relatively small, which will affect the viewing experience. Xiao N uses a tablet computer to view, and the tablet computer has a larger screen, which may be more conducive to viewing some details in the music appreciation course, such as the display of music scores. This can be used as a reference status value for the possible impact of different devices on learning effects. Xiao L commented in the group, and Xiao M liked the operation. These interactive operations can be converted into learning process status values ​​related to content interaction. The server can quantify based on the content depth of the comment, the number of likes, etc. For example, Xiao L's comments are relatively in-depth, involving the understanding of the music work and the need for further exploration. The server can give a higher content interaction status value, such as 3 (assuming a quantitative range of 0.5, with 5 being the highest). Xiao M's like operation is relatively simple, and is given a content interaction status value of 1. From Xiao O's click on the annotation link and Xiao K's pause operation, the server can try to infer the learning process status value related to the learning concentration of the secondary adolescent user. Xiao O's click on the annotation link shows that he has a deep desire to understand some knowledge points in the shared content, which may indicate a high learning concentration, and is given a status value of 3 (assuming a quantitative range of 0.5). Xiao K paused several times, and although the specific reasons are uncertain, a learning concentration status value of 2 can be given. The server integrates the various interaction status values ​​of the secondary adolescent users corresponding to Xiao J (Xiao K, Xiao L, Xiao M, Xiao N, and Xiao O) as the real interaction process data of the shared learning content (classical music appreciation course clips) delivered by Xiao J.For example, the real interactive process data of Xiao J's sharing content can be expressed as {"target youth user":"Xiao J","shared content":"classical music appreciation course clips","secondary youth user interaction status value":[{"secondary youth user":"Xiao K","view timeliness":10 minutes,"device usage":"mobile phone","content interaction":1,"learning concentration":2},{"secondary youth user":"Xiao L","view timeliness":30 minutes,"device usage":"mobile phone","content interaction":3,"learning concentration":unspecified (can be based on Other factors are speculated)},{"Secondary adolescent user":"Little M","Viewing timeliness":Unclear,"Device use":"Mobile phone","Content interaction":1,"Learning concentration":Unclear},{"Secondary adolescent user":"Little N","Viewing timeliness":Unclear,"Device use":"Tablet computer","Content interaction":Unclear,"Learning concentration":Unclear},{"Secondary adolescent user":"Little O","Viewing timeliness":Unclear,"Device use":"Mobile phone","Content interaction":Unclear,"Learning concentration":3}]}. Through the above detailed steps, the server can accurately obtain the real interactive process data of sharing learning content transmitted by each adolescent user. When processing a large amount of sharing and interaction data of adolescent users, the server will operate according to this process for each target adolescent user and its corresponding secondary adolescent user to fully grasp the situation of the entire adolescent group in sharing and interacting with learning content, which is of great significance for evaluating learning interests and optimizing the way of publishing learning content.

[0061] In an embodiment of the present invention, the real learning process data corresponding to each youth user includes a variety of learning process status values ​​of each youth user for the target learning content, the optimized learning process data corresponding to each youth user includes a variety of optimized learning process status values ​​obtained by performing differential optimization processing on the multiple learning process status values, the real interaction process data corresponding to each youth user includes a variety of interaction status values ​​of the shared learning content of each youth user, and the optimized interaction process data corresponding to each youth user includes a variety of optimized interaction status values ​​obtained by performing differential optimization processing on the multiple interaction status values; the embodiment of the present invention also provides the following implementation method.

[0062] Performing a cumulative operation on similar optimized learning process state values ​​among the multiple optimized learning process state values ​​respectively corresponding to the adolescent users in the test user group, to obtain the learning process state cumulative values ​​respectively corresponding to the multiple optimized learning process state values ​​of the test user group; performing a cumulative operation on similar optimized interactive state values ​​among the multiple optimized interactive state values ​​respectively corresponding to the adolescent users in the test user group, to obtain the interactive process state cumulative values ​​respectively corresponding to the multiple optimized interactive state values ​​of the test user group;

[0063] Obtaining the learning process state accumulated values ​​corresponding to the multiple optimized learning process state values ​​of the reference user group, and the interaction process state accumulated values ​​corresponding to the multiple optimized interaction state values ​​of the reference user group;

[0064] Performing difference value calculation on the learning process state accumulation values ​​corresponding to the same type of optimized learning process state values ​​in the test user group and the reference user group to obtain learning process state difference values ​​corresponding to the multiple optimized learning process state values, and using the multiple learning process state difference values ​​as the learning deviation data;

[0065] A difference value operation is performed on the interaction process state accumulated values ​​corresponding to the same type of optimized interaction state values ​​within the test user group and the reference user group to obtain interaction process state difference values ​​corresponding to the multiple optimized interaction state values ​​respectively, and the multiple interaction process state difference values ​​are used as the interaction deviation data.

[0066] In an embodiment of the present invention, illustratively, it is assumed that the server is processing an evaluation project on the content of online history course learning. The test user group includes 100 young users. For each young user, there is a corresponding optimized learning process state value and an optimized interactive process state value. Taking the optimized learning process state value as an example, these state values ​​include learning frequency, learning duration, and learning progress. For the optimized learning process state value of learning frequency, each young user has a value after differential optimization processing. For learning frequency, the optimized learning process state value of young user 1 is 4 times per week, young user 2 is 3 times per week, young user 3 is 5 times per week, and so on. The server performs cumulative calculations on the learning frequencies (similar optimized learning process state values) of these 100 young users. Assume that after accumulation, the cumulative value of the learning process state corresponding to the learning frequency is 350 times per week. For learning duration, it is in minutes. The optimized learning process state value of young user 1 is an average of 40 minutes per study, and the average learning duration of young user 2 is 35 minutes per study, and so on. The server accumulates the average duration of each study of these 100 young users. Assume that after calculation, the cumulative value of the learning process state corresponding to the learning duration is 3000 minutes. In terms of learning progress, assume that the optimized learning process state value is expressed as the percentage of completed courses. The optimized learning process state value of youth user 1 is 30% of the completed courses, youth user 2 is 25% of the completed courses, and so on. The server accumulates the percentage of completed courses for these 100 youth users. Assume that after accumulation, the cumulative value of the learning process state corresponding to the learning progress is 2800% (here it is just for the convenience of expressing the cumulative situation, the actual meaning is the sum of the progress of all youth users to complete the course). In terms of interaction, the optimized interactive process state value includes the number of interactions after sharing, the interaction time of secondary youth users, etc. For the number of interactions after sharing, the optimized interactive process state value of youth user 1 is that there are 3 secondary youth users interacting after each sharing, and youth user 2 is that there are 2 secondary youth users interacting after each sharing, and so on. The server accumulates the number of interactions after sharing (the same type of optimized interactive process state value) of these 100 youth users. Assume that after accumulation, the cumulative value of the interactive process state corresponding to the number of interactions after sharing is 200 people. The interaction time of secondary adolescent users is measured in minutes. The optimized interaction process status value of adolescent user 1 is 15 minutes for the total interaction time of secondary adolescent users after each sharing, and that of adolescent user 2 is 10 minutes for each sharing, and so on. The server accumulates the total interaction time of secondary adolescent users after each sharing of these 100 adolescent users (the same type of optimized interaction process status value). Assume that after calculation, the interaction process status accumulation value corresponding to the interaction time of secondary adolescent users is 1200 minutes. The reference user group also includes 100 adolescent users.The server also needs to obtain the cumulative learning process state values ​​corresponding to the various optimized learning process state values ​​of this group, as well as the cumulative interactive process state values ​​corresponding to the various optimized interactive state values. For the learning frequency, it is assumed that the learning frequency of the adolescent users in the reference user group after differential optimization is 2 times a week, 3 times a week, etc. The server performs cumulative calculations on the learning frequency of these 100 adolescent users. It is assumed that the cumulative learning process state value obtained is 250 times a week. For the learning duration, it is assumed that the average duration of each learning of the adolescent users in the reference user group is 30 minutes, 32 minutes, etc. The server accumulates the average duration of each learning of these 100 adolescent users, and assumes that the cumulative learning process state value obtained is 2500 minutes. In terms of learning progress, it is assumed that the percentage of completed courses of adolescent users in the reference user group is 20%, 22%, etc. The server accumulates the percentage of completed courses of these 100 adolescent users, and assumes that the cumulative learning process state value obtained is 2300%. For the number of interactions after sharing, it is assumed that the adolescent users in the reference user group have 1 or 2 secondary adolescent users interacting after each sharing. The server accumulates the number of people who interacted after sharing by these 100 young users, assuming that the accumulated value of the interactive process status is 150 people. For the interaction time of secondary young users, it is assumed that the total interaction time of secondary young users after each sharing by young users in the reference user group is 8 minutes, 10 minutes, etc. The server accumulates the total interaction time of secondary young users after each sharing by these 100 young users, assuming that the accumulated value of the interactive process status is 1000 minutes. For the learning frequency, the accumulated value of the learning process status of the test user group is 350 times per week, and the accumulated value of the learning process status of the reference user group is 250 times per week. Perform the difference value calculation: 350-250=100 times. This 100 times is the learning process status difference value corresponding to the learning frequency. It represents the difference between the test user group and the reference user group in terms of learning frequency, reflecting the possible impact of different learning content publishing methods (assuming that the test user group and the reference user group adopt different historical course publishing methods) on the learning frequency. The cumulative value of the learning process status corresponding to the learning duration of the test user group is 3000 minutes, and that of the reference user group is 2500 minutes. Calculate the difference value: 3000-2500=500 minutes. These 500 minutes are the difference value of the learning process status corresponding to the learning duration, indicating that there is a certain difference between the two groups in terms of learning duration, which may be caused by the different ways of publishing learning content. The cumulative value of the learning process status corresponding to the learning progress of the test user group is 2800%, and that of the reference user group is 2300%. Calculate the difference value: 2800-2300=500%.This 500% is the difference value of the learning process state corresponding to the learning progress, which reflects the gap between the test user group and the reference user group in terms of learning progress. The cumulative value of the interactive process state corresponding to the number of people who interacted after sharing in the test user group is 200 people, and the reference user group is 150 people. Perform the difference value calculation: 200-150=50 people. The 50 people are the difference value of the interactive process state corresponding to the number of people who interacted after sharing, which reflects the difference between the test user group and the reference user group in terms of the number of people who interacted after sharing. It may be because different learning content publishing methods affect the attractiveness and dissemination effect of shared content. The cumulative value of the interactive process state corresponding to the interaction time of secondary adolescent users in the test user group is 1200 minutes, and the reference user group is 1000 minutes. Calculate the difference value: 1200-1000=200 minutes. These 200 minutes are the difference value of the interactive process state corresponding to the interaction time of secondary adolescent users, indicating that there are differences between the two groups in the interaction time of secondary adolescent users, which is also related to the way of publishing learning content. Through the above detailed steps, the server can accurately calculate the learning bias data and interaction bias data, which are of great significance for evaluating the impact of different learning content publishing methods on teenagers' learning interests. When processing large-scale youth user data, the server will operate on different learning content and user groups according to this process to comprehensively and accurately evaluate learning effects and learning interests.

[0067] In the embodiment of the present invention, there is a correlation between one of the multiple learning process state difference values ​​and one of the multiple interactive process state difference values;

[0068] The method of determining the learning interest assessment results corresponding to the test learning content publishing method and the reference learning content publishing method respectively based on the optimized learning process data of the adolescent users in the test user group and the learning deviation data between the optimized learning process data of the adolescent users in the reference user group, as well as the optimized interaction process data of the adolescent users in the test user group and the interaction deviation data between the optimized interaction process data of the adolescent users in the reference user group, can be implemented through the following examples.

[0069] According to the multiple learning process state difference values ​​and the multiple interactive process state difference values, cumulatively calculating the correlated learning process state difference values ​​and interactive process state difference values, to obtain multiple first interest coefficients of the test learning content publishing method and the reference learning content publishing method for the learning content;

[0070] According to the multiple first interest coefficients, the learning interest evaluation results corresponding to the test learning content publishing mode and the reference learning content publishing mode are determined.

[0071] In an embodiment of the present invention, illustratively, it is assumed that the server is processing an evaluation of the learning content of an online programming course. The test user group and the reference user group respectively adopt different learning content publishing methods (for example, the test user group adopts a project-driven publishing method, and the reference user group adopts a traditional chapter publishing method). The server has calculated the learning deviation data (learning process state difference value) and the interaction deviation data (interaction process state difference value). In terms of the learning process state difference value, there are learning frequency difference value, learning duration difference value and learning progress difference value, etc.; in terms of the interaction process state difference value, there are differences in the number of people who interact after sharing and differences in the interaction time of secondary youth users, etc. Analysis shows that there is a correlation between the learning frequency and the number of people who interact after sharing, because teenagers with higher learning frequencies may have more opportunities to share learning content, resulting in more people who interact after sharing. Similarly, there is a correlation between learning progress and the interaction time of secondary youth users. When teenagers make more learning progress in programming learning, they will attract secondary youth users to interact for a longer time, such as asking more questions about programming projects or seeking suggestions for code improvements. For the related difference values ​​of learning frequency and the number of people who interact after sharing, it is assumed that the learning process state difference value of learning frequency is 20 times per week (i.e., the test user group learns 20 more times per week than the reference user group), and the interaction process state difference value of the number of people who interact after sharing is 10 people (i.e., the test user group interacts with 10 more people after each sharing than the reference user group). The server performs cumulative calculations on these two related difference values: 20+10=30. This 30 can be used as a component of a preliminary first interest coefficient. For the related difference values ​​of learning progress and the interaction time of secondary adolescent users, it is assumed that the learning process state difference value of learning progress is 20% (i.e., the test user group has completed 20% more course progress than the reference user group), and the interaction process state difference value of the interaction time of secondary adolescent users is 30 minutes (i.e., the secondary adolescent users after the test user group shares interact for 30 minutes more than the secondary adolescent users after the reference user group shares). Perform cumulative calculations: 20+30=50. This 50 is also a component of a preliminary first interest coefficient. Then, the server will combine these preliminary coefficients according to certain weight distribution. Assume that the weight of the related value of learning frequency and the number of people who interact after sharing is 0.4, and the weight of the related value of learning progress and the interaction time of secondary youth users is 0.6. Calculate the first first interest coefficient: (30×0.4+50×0.6)=42. Assuming that there are other related difference value pairs, calculate other preliminary coefficients in the same way, and then perform comprehensive calculations based on their respective weights to obtain more first interest coefficients. For example, suppose there is another pair of related difference values ​​(such as learning duration and another interaction status value) and the preliminary coefficients calculated are comprehensively calculated to obtain another first interest coefficient of 35.The server sets some thresholds to determine the learning interest evaluation results. For example, the first preset threshold is set to 30, and the second preset threshold is set to 2. This means that if a sufficient number of coefficients (exceeding the second preset threshold) among the multiple first interest coefficients are greater than the first preset threshold, then it can be determined that the test learning content publishing method is more conducive to stimulating the learning interest of young people. In the case where the first interest coefficients calculated above are 42 and 35, because 42>30 and 35>30, and the number of coefficients (2) exceeds the second preset threshold (2). Therefore, the server determines that the learning interest evaluation result corresponding to the test learning content publishing method (project-driven publishing method) is better than the reference learning content publishing method (traditional chapter-based publishing method). This shows that in online programming course learning, the project-driven publishing method has more advantages in stimulating the learning interest of young people, whether in the performance in the learning process (learning frequency, learning progress, etc.) or in the performance in the interactive process (number of interactions after sharing, interaction time of secondary youth users, etc.). If in another evaluation of online art course learning, the first interest coefficients are assumed to be 25 and 28, although 28>30 does not hold and 25>30 does not hold, if another judgment rule is set, for example, as long as one coefficient is close to the first preset threshold (assuming the proximity range is ±5), and the other coefficients are not 0 or negative, it is determined that the test learning content release method has a certain advantage. In this case, 28 is within the range close to 30, so the server can determine that the test learning content release method (assuming an interactive art work display release method) has a certain advantage in stimulating the learning interest of teenagers compared to the reference learning content release method (assuming a traditional picture + text explanation release method), although this advantage may not be very obvious. Suppose again that in an evaluation of online physics course learning, the first interest coefficients are 15 and 20. Since 15<30 and 20<30, and other special judgment rules are not met, the server determines that the learning interest evaluation result corresponding to the reference learning content release method (assuming a traditional theoretical derivation release method) is better than the test learning content release method (assuming a simulation experiment + simplified theory release method). This means that in online physics course learning, the traditional theoretical derivation publishing method performs better in stimulating young people's interest in learning. This may be because in this course, young people need more systematic theoretical knowledge construction, and the traditional method can better meet this need. Through the above detailed steps, the server can calculate the first interest coefficient based on the correlated learning process state difference value and the interactive process state difference value, and accurately determine the learning interest evaluation results corresponding to the test learning content publishing method and the reference learning content publishing method based on these coefficients. When dealing with different types of learning content and large-scale youth user data, the server will operate according to this process to comprehensively and objectively evaluate the impact of different learning content publishing methods on young people's learning interests.

[0072] In the embodiment of the present invention, determining the learning interest evaluation results corresponding to the test learning content publishing mode and the reference learning content publishing mode respectively according to the multiple first interest coefficients can be implemented through the following examples.

[0073] Obtaining the number of coefficients corresponding to the first interest coefficients greater than the first preset threshold among the plurality of first interest coefficients;

[0074] In the case where the number of coefficients exceeds the second preset threshold, it is determined that the learning interest evaluation result corresponding to the test learning content publishing method is better than the learning interest evaluation result corresponding to the reference learning content publishing method;

[0075] When the number of coefficients does not exceed the second preset threshold, it is determined that the learning interest evaluation result corresponding to the reference learning content publishing method is better than the learning interest evaluation result corresponding to the test learning content publishing method.

[0076] In the embodiment of the present invention, illustratively, it is assumed that the server is evaluating two publishing methods of online foreign language learning content: one is a multimedia interactive publishing method (as a test learning content publishing method), and the other is a traditional text reading publishing method (as a reference learning content publishing method). After the previous calculation, the server obtains multiple first interest coefficients, for example, 40, 35, 25, 18, and 45. At the same time, the server sets the first preset threshold to 30. The server starts to check the coefficients of these first interest coefficients that are greater than the first preset threshold (30). In this example, the three first interest coefficients of 40, 35 and 45 are greater than 30. The server counts that the number of coefficients corresponding to the first interest coefficients greater than the first preset threshold is 3. For this online foreign language learning content evaluation, the server sets the second preset threshold to 2. Since the number of coefficients corresponding to the first interest coefficients greater than the first preset threshold (30) is 3, and 3>2 (the second preset threshold). This means that in online foreign language learning, the multimedia interactive publishing method (test learning content publishing method) shows the characteristics of being more able to stimulate the learning interest of young people in multiple aspects (aspects related to the learning process and the interactive process, which are combined to form the first interest coefficient). For example, the multimedia interactive publishing method may include elements such as video explanations, voice dialogue exercises, and interactive games. These elements make teenagers more actively participate in learning foreign languages, learn more frequently, and interact more frequently after sharing, so that when compared with the traditional text reading publishing method, it performs better in more relevant aspects. Therefore, the server determines that the learning interest evaluation result corresponding to the multimedia interactive publishing method is better than the learning interest evaluation result corresponding to the traditional text reading publishing method. Assume that the server is evaluating two publishing methods for online mathematics course learning content: one is a fun animation demonstration publishing method (as a test learning content publishing method), and the other is a classic example explanation publishing method (as a reference learning content publishing method). After calculation, the first interest coefficients obtained by the server are 25, 28, 15, 22, and 18 respectively. The first preset threshold is set to 30. The server checks the coefficients of these first interest coefficients that are greater than the first preset threshold (30) and finds that no coefficient is greater than 30. Therefore, the number of coefficients corresponding to the first interest coefficients greater than the first preset threshold is 0. The server sets the second preset threshold to 1. Since the number of coefficients 0 does not exceed the second preset threshold 1. This shows that in online mathematics course learning, the classic example explanation-style publishing method (reference learning content publishing method) shows the characteristics of being more able to stimulate teenagers' interest in learning when comprehensively considering factors related to the learning process and the interactive process (these factors are reflected in the first interest coefficient).The classic example explanation publishing method may be more suitable for subjects such as mathematics that require rigorous logic and a lot of practice. Teenagers may be more accustomed to learning mathematics knowledge in depth through this method. Although the fun animation demonstration publishing method may have certain advantages in attracting initial attention, it is not as good as the classic example explanation publishing method in terms of comprehensive learning interest stimulation. Therefore, the server determines that the learning interest evaluation result corresponding to the classic example explanation publishing method is worse than the learning interest evaluation result corresponding to the fun animation demonstration publishing method. Assume that two publishing methods for online history course learning content are evaluated: one is the story series publishing method (as a test learning content publishing method), and the other is the chronological explanation publishing method (as a reference learning content publishing method). The first interest coefficients obtained by the server are 20, 22, 28, 18, and 25. The first preset threshold is set to 30. It is found that the number of coefficients corresponding to the first interest coefficient greater than the first preset threshold (30) is 0. The second preset threshold is set to 1. Because the number of coefficients 0 does not exceed the second preset threshold 1. In the online history course learning, the chronological explanation-style publishing method (reference learning content publishing method) can better stimulate the learning interest of young people when various factors (reflected in the first interest coefficient) are comprehensively considered. The chronological explanation-style publishing method may be more helpful for young people to construct the time context of history, and it is helpful for in-depth understanding of the sequence and causal relationship of historical events. Although the story series publishing method makes historical knowledge more vivid and interesting, it is slightly inferior in terms of overall learning interest stimulation. Therefore, the server determines that the learning interest evaluation result corresponding to the chronological explanation-style publishing method is better than the learning interest evaluation result corresponding to the story series publishing method. Through the above detailed steps, the server can accurately determine the learning interest evaluation results corresponding to the test learning content publishing method and the reference learning content publishing method according to multiple first interest coefficients and preset thresholds. When processing a large number of different types of learning content and different user group data, this method can provide an objective and quantitative way to evaluate which learning content publishing method can better stimulate the learning interest of young people.

[0077] In the embodiments of the present invention, the following implementation modes are also provided.

[0078] Dividing a plurality of adolescent users into a verification user group and a benchmark user group, sending a third learning content to adolescent users in the verification user group according to the test learning content publishing method, and sending a fourth learning content to adolescent users in the benchmark user group according to the reference learning content publishing method;

[0079] Determine a plurality of second interest coefficients of the test learning content publishing method and the reference learning content publishing method for the learning content according to the optimized learning process data of the adolescent users in the verification user group, the optimized learning process data of the adolescent users in the benchmark user group, the optimized interaction process data of the adolescent users in the verification user group, and the optimized interaction process data of the adolescent users in the benchmark user group;

[0080] The determining, according to the plurality of first interest coefficients, learning interest evaluation results corresponding to the test learning content publishing mode and the reference learning content publishing mode, respectively, includes:

[0081] According to the multiple first interest coefficients and the multiple second interest coefficients, the learning interest evaluation results corresponding to the test learning content publishing mode and the reference learning content publishing mode are determined.

[0082] In an embodiment of the present invention, illustratively, it is assumed that the server is processing an evaluation of online science course learning content. The server has a total sample of 1,000 adolescent users. In order to further verify the previous evaluation results on the learning content publishing method, the server divides the 1,000 adolescent users into a verification user group and a benchmark user group. The server divides according to certain rules, such as stratifying according to the average scores of adolescent users in previous science course learning, dividing 500 adolescent users with scores in the top 50% into a verification user group, and dividing 500 adolescent users with scores in the bottom 50% into a benchmark user group. The purpose of this division is to test the effectiveness of the learning content publishing method among adolescent users at different learning levels. Assume that the test learning content publishing method is an inquiry-based learning content publishing method (for example, guiding adolescents to learn scientific knowledge by setting a series of scientific inquiry questions), and the reference learning content publishing method is a traditional teaching-based learning content publishing method (teachers directly explain scientific knowledge concepts and principles). For the 500 adolescent users in the verification user group, the server sends them the third learning content according to the inquiry-based learning content publishing method. This third learning content can be an online science course on plant photosynthesis, which is presented in the form of exploratory questions, such as "Why do plants grow better in sunlight?" "How to design experiments to verify the conditions for photosynthesis?", etc., and provides relevant experimental videos, data collection tools and other resources, allowing young users to explore the answers themselves. For the 500 young users in the benchmark user group, the server sends them the fourth learning content according to the traditional teaching-style learning content publishing method. The fourth learning content is also an online science course on plant photosynthesis, but it is presented in the form of a video explained by a teacher. The video details the definition, reaction formula, influencing factors and other knowledge of photosynthesis, and is accompanied by simple example pictures. For young users in the verification user group, the server obtains their optimized learning process data. Taking learning frequency as an example, the server determines the learning frequency by recording the number of times young users log in to the course platform to learn photosynthesis courses. After a period of observation (for example, one month), it is found that there is a young user Xiao Ming in the verification user group, and his learning frequency is 4 times a week. This is the value after the previous difference optimization processing, which may have taken into account the difference between his individual differences (such as his own interest in science, study time arrangement, etc.) and the group average level. In terms of learning duration, Xiao Ming's average duration of each photosynthesis course is 40 minutes. In terms of learning progress, Xiao Ming has completed 80% of the photosynthesis experiment design part of the course, which are part of his optimized learning process data. The server collects and organizes such data for all 500 young users in the verification user group to obtain the optimized learning process data for the entire verification user group.For adolescent users in the benchmark user group, the optimized learning process data is also obtained. For example, there is a adolescent user named Xiaohong in the benchmark user group. Her learning frequency is 3 times a week (this is also the value after differential optimization). The average duration of her photosynthesis course is 30 minutes each time. In terms of learning progress, she has completed 60% of the photosynthesis definition and reaction part of the course. The server collects similar data for 500 adolescent users in the benchmark user group to obtain the optimized learning process data of the benchmark user group. In terms of interaction, for adolescent users in the verification user group, the server obtains their optimized interaction process data. Taking the number of people who interact after sharing as an example, Xiaoming shared some of his ideas on the design of photosynthesis experiments with the learning group during the learning process. There are 8 secondary adolescent users who interacted with his shared content (such as asking questions, discussing, etc.), which is part of his optimized interaction process data. In terms of the interaction time of secondary adolescent users, the total interaction time of these 8 secondary adolescent users with Xiaoming on his shared content is 60 minutes. These data have been processed by differential optimization, taking into account the differences between individuals and groups in interaction. The server collects and organizes the interaction data of 500 adolescent users in the verification user group to obtain the optimized interaction process data of the entire verification user group. For adolescent users in the benchmark user group, for example, Xiaohong shared her memory skills for the photosynthesis reaction formula to the learning group, and 3 secondary adolescent users interacted with her shared content. The total interaction time of these 3 secondary adolescent users and Xiaohong about her shared content was 20 minutes. The server collects similar interaction data for 500 adolescent users in the benchmark user group to obtain the optimized interaction process data of the benchmark user group. For learning frequency, the cumulative value of learning frequency of the verification user group (obtained by summing the learning frequency of 500 adolescent users) is assumed to be 1500 times / week, and the cumulative value of learning frequency of the benchmark user group is assumed to be 1200 times / week. The difference in learning frequency is 1500-1200=300 times / week. In terms of learning duration, the cumulative value of learning duration of the verification user group (summing the average duration of each learning session of 500 young users) is assumed to be 18,000 minutes, the cumulative value of learning duration of the benchmark user group is assumed to be 15,000 minutes, and the difference in learning duration is 18,000-15,000=3,000 minutes. For learning progress, the cumulative value of learning progress of the verification user group (summing the learning progress of 500 young users, expressed as the percentage of completed courses) is assumed to be 3,500%, the cumulative value of learning progress of the benchmark user group is assumed to be 3,000%, and the difference in learning progress is 3,500-3,000=500%.In terms of the number of people who interacted after sharing, the cumulative value of the number of people who interacted after sharing for the verification user group (summing the number of people who interacted after each sharing for 500 young users) is assumed to be 2,500 people, and the cumulative value of the number of people who interacted after sharing for the benchmark user group is assumed to be 1,500 people. The difference in the number of people who interacted after sharing is 2,500-1,500=1,000 people. For the interaction time of secondary young users, the cumulative value of the interaction time of secondary young users for the verification user group (summing the total interaction time of secondary young users after each sharing for 500 young users) is assumed to be 15,000 minutes, and the cumulative value of the interaction time of secondary young users for the benchmark user group is assumed to be 10,000 minutes. The difference in the interaction time of secondary young users is 15,000-10,000=5,000 minutes. The server sets weights based on the importance of these difference values. Assume that the weight of learning frequency is 0.1, the weight of learning duration is 0.2, the weight of learning progress is 0.2, the weight of the number of people who interact after sharing is 0.3, and the weight of the interaction time of secondary youth users is 0.2. Calculate the second interest coefficient: (300×0.1+3000×0.2+500×0.2+1000×0.3+5000×0.2)=2030. This is just an example of a comprehensive second interest coefficient. In fact, there may be multiple calculations of second interest coefficients based on different correlation analyses and weight adjustments. Assume that in the previous evaluation step, for this online science course learning content, the first interest coefficient calculated based on other relevant data is [400,350,280] (here is just an example value). The server comprehensively considers the first interest coefficient and the second interest coefficient to determine the final learning interest evaluation result. One way is to sum the first interest coefficient and the second interest coefficient respectively, and then compare the size of the two sums. The sum of the first interest coefficient is 400+350+280=1030. Assume that the second interest coefficient has only one value of 2030 calculated previously (if there are multiple values, they are summed according to certain rules). Because 2030>1030, the server determines that the learning interest evaluation result corresponding to the test learning content publishing method (inquiry-based learning content publishing method) is better than the reference learning content publishing method (traditional teaching-based learning content publishing method). This means that in online science course learning, the inquiry-based learning content publishing method performs better in stimulating the learning interest of young people, whether from the previous preliminary evaluation (reflected by the first interest coefficient) or from this verification evaluation of young users with different learning levels (reflected by the second interest coefficient). If the sum of the first interest coefficients is greater than the sum of the second interest coefficients, then it will be concluded that the reference learning content publishing method is better; if the two are quite close, further analysis or adjustment of the evaluation parameters (such as weights, etc.) is needed to more accurately determine which learning content publishing method can better stimulate the learning interest of young people.Through the above detailed steps, the server can calculate the second interest coefficient by introducing the verification user group and the benchmark user group, and combine it with the first interest coefficient to more comprehensively and accurately determine the learning interest evaluation results corresponding to the test learning content publishing method and the reference learning content publishing method. This multi-round evaluation and multi-group verification method helps to improve the reliability of the evaluation results, especially when dealing with a large number of young users and different types of learning content.

[0083] In the embodiments of the present invention, the following implementation modes are also provided.

[0084] According to a user knowledge graph between a plurality of adolescent users, clustering processing is performed on the plurality of adolescent users to obtain a plurality of user cluster groups; the user knowledge graph is used to characterize the learning interactive association relationship between the plurality of adolescent users, and an interactive association relationship exists between the adolescent users in each user cluster group in the plurality of user cluster groups;

[0085] A plurality of user clustering groups that meet preset test criteria are extracted from the plurality of user clustering groups, and the adolescent users in the plurality of user clustering groups are taken as the plurality of adolescent users.

[0086] In an embodiment of the present invention, for example, it is assumed that the server is processing adolescent user data on an online learning platform, which provides learning content in multiple subjects, such as mathematics, science, language arts, etc. The server first collects various interaction data of multiple adolescent users (assuming 1,000) on the platform to build a user knowledge graph. For each adolescent user, the server collects their learning records, including the courses studied, the time of study, the homework completed, the test scores, etc. For example, adolescent user A has studied 3 math courses in the past month, each study lasting about 45 minutes, completed 2 math homework, and the math test score was 80 points. The server also collects interactive data between users, such as asking questions, answering, sharing learning resources, and discussions on forums. For example, adolescent user A asked adolescent user B a question about mathematical functions, and adolescent user B gave a detailed answer; adolescent user C shared an article about scientific experiments, and adolescent users D, E, and F liked and commented on this sharing. The server builds a user knowledge graph based on these data. In this graph, each adolescent user is a node. If there is a learning interaction between two young users (such as asking questions and answering, sharing and liking, etc.), an edge is established between their corresponding nodes. For the weight of the edge, the server can determine it according to the frequency and depth of the interaction. For example, if young user A often asks young user B questions and their discussions are in-depth, the weight of the edge between them is higher; if young user C only occasionally receives a like from young user D, the weight of the edge between them is lower. At the same time, the nodes themselves also have attributes, which include the learning performance of young users, learning preferences (such as whether they prefer theoretical learning or practical learning), etc. For example, young user A has good grades in mathematics learning and prefers to learn by doing exercises. These attributes are associated with node A. The server selects a suitable clustering algorithm to cluster these 1,000 young users. For example, the server selects a graph-based clustering algorithm, such as a spectral clustering algorithm. This algorithm is suitable for processing graph-structured data such as user knowledge graphs. The server takes the user knowledge graph as input and runs the spectral clustering algorithm. The algorithm clusters young users with similar learning interaction associations into one category based on the weights of the edges between nodes and the attributes of the nodes. For example, after clustering, five user clusters are obtained. The first cluster may include young users who have good grades in mathematics learning, often ask and answer math questions to each other, and share a lot of math learning resources; the second cluster may include young users who are interested in scientific experiments and often discuss experimental results and experimental designs together. It is assumed that the preset test criteria are based on the relevance of learning content and the activeness of interaction.For the relevance of learning content, it is required that the learning content that the adolescent users in the cluster group mainly focus on is the subject related to the learning content to be evaluated (for example, a newly launched set of online programming courses), such as basic computer science knowledge, logical thinking training, etc. For the activity of interaction, it is required that the interaction frequency (the total number of questions, answers, sharing, etc.) of the adolescent users in the cluster group in the past month exceeds a certain threshold (assuming it is 50 times). The server starts to analyze the five user cluster groups obtained above. For the first cluster group, the server checks the learning content they mainly focus on. It is found that the adolescent users in this group mainly focus on mathematics learning, which has a low relevance to online programming courses and does not meet the learning content relevance criteria. At the same time, checking their interaction frequency in the past month, it is found that the total number is 30 times, which is also lower than the threshold of interaction activity, so this cluster group does not meet the preset test criteria. For the second cluster group, the adolescent users in this group mainly focus on scientific experiments, which also have a low relevance to online programming courses and do not meet the learning content relevance criteria. Although their interaction frequency is 60 times, which meets the interaction activity criteria, this cluster group does not meet the requirements because it does not meet the learning content relevance criteria. For the third cluster group, the server found that the young users in this group mainly focused on computer basics and some simple programming logic, and had a high correlation with online programming courses. And their interaction frequency in the past month was 80 times, exceeding the threshold of 50 times, so this cluster group met the preset test criteria. For the fourth cluster group, although the young users in this group have a certain interest in logical thinking training and have a certain correlation with programming courses, their interaction frequency in the past month was 40 times, which is lower than the threshold of 50 times and does not meet the preset test criteria. For the fifth cluster group, the young users in this group mainly focus on electronic circuit knowledge, which has a low correlation with online programming courses and does not meet the learning content relevance criteria. The interaction frequency is 55 times. Although it meets the interactive activity criteria, it does not meet the preset test criteria as a whole. After analysis, only the third cluster group meets the preset test criteria. The server uses the young users in this cluster group as multiple young users to be used for subsequent learning interest evaluation. For example, there are 200 young users in this cluster group, and these 200 young users will participate in the subsequent learning content publishing, data collection, and learning interest evaluation steps. Through the above detailed steps, the server can perform clustering based on the user knowledge graph and extract appropriate youth user groups according to the preset test criteria, so as to conduct more targeted learning interest assessment and improve the accuracy and effectiveness of the assessment results. When dealing with different types of learning content and large-scale youth user data, this method helps to focus on users with similar characteristics and active interactive relationships, and better understand the impact of different learning content publishing methods on the learning interests of specific youth user groups.

[0087] The embodiment of the present invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the aforementioned large model-based learning interest assessment method. Figure 2 As shown, Figure 2 The block diagram of the computer device 100 provided in the embodiment of the present invention. The computer device 100 includes a memory 111, a processor 112 and a communication unit 113. To achieve data transmission or interaction, the memory 111, the processor 112 and the communication unit 113 are directly or indirectly electrically connected to each other. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0088] For illustrative purposes, the foregoing description is made with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the present disclosure to the precise form disclosed. Numerous modifications and variations are possible in accordance with the above teachings. These embodiments are selected and described in order to best illustrate the principles of the present disclosure and its practical application, so that those skilled in the art can best utilize the present disclosure and utilize various embodiments with different modifications to suit the intended specific application.

Claims

1. A learning interest assessment method based on a large model, characterized in that: include: Obtain multi-level identity information of multiple initial users from multiple social platforms; Inputting the multi-level identity information of the multiple initial users into the identity assessment model trained by the dual verification mechanism based on pattern clustering and sentiment tendency analysis to obtain the identity label corresponding to each of the initial users; Marking the initial user whose identity tag is characterized as a teenager as a teenager user; Divide multiple teenage users into a test user group and a reference user group, send learning content to the multiple teenage users according to different learning content publishing methods, obtain corresponding learning process data and interaction process data, and determine the learning interest evaluation results corresponding to the different learning content publishing methods based on the learning process data and the interaction process data.

2. The method according to claim 1, characterized in that The method of dividing a plurality of adolescent users into a test user group and a reference user group, sending learning content to the plurality of adolescent users according to different learning content publishing methods, obtaining corresponding learning process data and interaction process data, and determining learning interest evaluation results corresponding to different learning content publishing methods based on the learning process data and the interaction process data, includes: Dividing a plurality of adolescent users into a test user group and a reference user group, sending a first learning content to the adolescent users in the test user group according to a test learning content publishing method, and sending a second learning content to the adolescent users in the reference user group according to a reference learning content publishing method; Acquire first learning behavior data of a target youth user among the plurality of youth users for the corresponding target learning content; Obtaining the learning frequency of the target young user for the corresponding target learning content from the first learning behavior data corresponding to the target young user; Obtaining, from the first learning behavior data corresponding to the target adolescent user, a learning duration of the target adolescent user for the corresponding target learning content; The learning frequency and learning duration corresponding to the target youth user are used as multiple learning process state values ​​of the target youth user for the corresponding target learning content; The multiple learning process status values ​​of the target adolescent user are used as the real learning process data of the target adolescent user for the target learning content, and the secondary adolescent users corresponding to the target adolescent user among the multiple adolescent users are determined; the secondary adolescent users are adolescent users among the multiple adolescent users who receive the shared learning content transmitted by the target adolescent user; Acquire second learning behavior data of the shared learning content transmitted by the secondary youth user to the target youth user; Analyzing and obtaining, from the second learning behavior data, a plurality of learning process status values ​​of the shared learning content transmitted by the secondary youth user to the target youth user as a plurality of interaction status values ​​corresponding to the target youth user; The multiple interaction state values ​​corresponding to the target youth user are used as real interaction process data of the shared learning content transmitted by the target youth user; the target learning content is the learning content transmitted to each youth user after being sent, and the target learning content and the shared learning content are derived from a learning content set consisting of the first learning content and the second learning content; Using a learning interest coefficient learning interest inference model, inferring learning interest coefficients formed by the learning process states of the plurality of young users for the learning content according to the user characteristics corresponding to the plurality of young users respectively; Performing an average calculation on the learning interest coefficients respectively corresponding to the plurality of adolescent users to obtain an average learning interest mean coefficient; Obtaining a learning interest coefficient corresponding to a target youth user and a coefficient difference value between the coefficient of the average learning interest mean, taking the coefficient difference value as a user feature of the target youth user, and performing difference optimization processing on the real learning process data and the real interaction process data corresponding to the multiple youth users according to the user behavior differences respectively corresponding to the multiple youth users based on the user behavior differences formed by the learning process states of the multiple youth users for the learning content, so as to obtain optimized learning process data and optimized interaction process data respectively corresponding to the multiple youth users; Based on the optimized learning process data of the adolescent users in the test user group and the learning deviation data between the optimized learning process data of the adolescent users in the reference user group, as well as the optimized interaction process data of the adolescent users in the test user group and the interaction deviation data between the optimized interaction process data of the adolescent users in the reference user group, the learning interest assessment results corresponding to the test learning content publishing method and the reference learning content publishing method are determined respectively.

3. The method according to claim 2, characterized in that The method of performing difference optimization processing on the real learning process data and the real interaction process data respectively corresponding to the multiple young users according to the differences in user behaviors respectively corresponding to the multiple young users, and obtaining the optimized learning process data and the optimized interaction process data respectively corresponding to the multiple young users, comprises: Determining a difference optimization coefficient corresponding to a target teenage user among the plurality of teenage users according to a user behavior difference corresponding to the target teenage user; Performing an average calculation on the real learning process data of the remaining young users to obtain learning process mean data, and performing an average calculation on the real interaction process data of the remaining young users to obtain interaction process mean data; the remaining young users are young users in the group to which the target young user belongs except the target young user; According to the difference optimization coefficient corresponding to the target youth user, the learning process mean data and the interaction process mean data, difference optimization processing is performed on the real learning process data and the real interaction process data corresponding to the target youth user to obtain the optimized learning process data and the optimized interaction process data corresponding to the target youth user.

4. The method according to claim 3, characterized in that The method of performing difference optimization processing on the real learning process data and the real interaction process data corresponding to the target youth user according to the difference optimization coefficient corresponding to the target youth user, the learning process mean data and the interaction process mean data, to obtain the optimized learning process data and the optimized interaction process data corresponding to the target youth user, includes: Obtaining a learning data difference value between the real learning process data corresponding to the target adolescent user and the learning process mean data, and an interaction data difference value between the real interaction process data corresponding to the target adolescent user and the interaction process mean data; Using the difference optimization coefficient corresponding to the target youth user, performing a difference optimization operation on the learning data difference value to obtain a learning data optimized difference value, and using the difference optimization coefficient corresponding to the target youth user, performing a difference optimization operation on the interaction data difference value to obtain an interaction data optimized difference value; An accumulation operation is performed on the optimization difference value of the learning data and the mean data of the learning process to obtain the optimized learning process data corresponding to the target youth user; an accumulation operation is performed on the optimization difference value of the interaction data and the mean data of the interaction process to obtain the optimized interaction process data corresponding to the target youth user.

5. The method according to claim 2, characterized in that: The real learning process data corresponding to each youth user includes a variety of learning process state values ​​of each youth user for target learning content, the optimized learning process data corresponding to each youth user includes a variety of optimized learning process state values ​​obtained by performing differential optimization processing on the multiple learning process state values, the real interaction process data corresponding to each youth user includes a variety of interaction state values ​​of the shared learning content of each youth user, and the optimized interaction process data corresponding to each youth user includes a variety of optimized interaction state values ​​obtained by performing differential optimization processing on the multiple interaction state values; The method further comprises: Performing a cumulative operation on similar optimized learning process state values ​​among the multiple optimized learning process state values ​​respectively corresponding to the adolescent users in the test user group, to obtain the learning process state cumulative values ​​respectively corresponding to the multiple optimized learning process state values ​​of the test user group; performing a cumulative operation on similar optimized interactive state values ​​among the multiple optimized interactive state values ​​respectively corresponding to the adolescent users in the test user group, to obtain the interactive process state cumulative values ​​respectively corresponding to the multiple optimized interactive state values ​​of the test user group; Obtaining the learning process state accumulated values ​​corresponding to the multiple optimized learning process state values ​​of the reference user group, and the interaction process state accumulated values ​​corresponding to the multiple optimized interaction state values ​​of the reference user group; Performing difference value calculation on the learning process state accumulation values ​​corresponding to the same type of optimized learning process state values ​​in the test user group and the reference user group to obtain learning process state difference values ​​corresponding to the multiple optimized learning process state values, and using the multiple learning process state difference values ​​as the learning deviation data; A difference value operation is performed on the interaction process state accumulated values ​​corresponding to the same type of optimized interaction state values ​​within the test user group and the reference user group to obtain interaction process state difference values ​​corresponding to the multiple optimized interaction state values ​​respectively, and the multiple interaction process state difference values ​​are used as the interaction deviation data.

6. The method according to claim 5, characterized in that There is a correlation between one of the plurality of learning process state difference values ​​and one of the plurality of interactive process state difference values; The method of determining the learning interest evaluation results corresponding to the test learning content publishing mode and the reference learning content publishing mode respectively based on the optimized learning process data of the adolescent users in the test user group and the learning deviation data between the optimized learning process data of the adolescent users in the reference user group, and the interaction deviation data between the optimized interaction process data of the adolescent users in the test user group and the optimized interaction process data of the adolescent users in the reference user group, comprises: According to the multiple learning process state difference values ​​and the multiple interactive process state difference values, cumulatively calculating the correlated learning process state difference values ​​and interactive process state difference values, to obtain multiple first interest coefficients of the test learning content publishing method and the reference learning content publishing method for the learning content; According to the multiple first interest coefficients, the learning interest evaluation results corresponding to the test learning content publishing mode and the reference learning content publishing mode are determined.

7. The method according to claim 6, characterized in that The determining, according to the plurality of first interest coefficients, learning interest evaluation results corresponding to the test learning content publishing mode and the reference learning content publishing mode, respectively, includes: Obtaining the number of coefficients corresponding to the first interest coefficients greater than the first preset threshold among the plurality of first interest coefficients; In the case where the number of coefficients exceeds the second preset threshold, it is determined that the learning interest evaluation result corresponding to the test learning content publishing method is better than the learning interest evaluation result corresponding to the reference learning content publishing method; When the number of coefficients does not exceed the second preset threshold, it is determined that the learning interest evaluation result corresponding to the reference learning content publishing method is better than the learning interest evaluation result corresponding to the test learning content publishing method.

8. The method according to claim 6, characterized in that The method further comprises: Dividing a plurality of adolescent users into a verification user group and a benchmark user group, sending a third learning content to adolescent users in the verification user group according to the test learning content publishing method, and sending a fourth learning content to adolescent users in the benchmark user group according to the reference learning content publishing method; Determine a plurality of second interest coefficients of the test learning content publishing method and the reference learning content publishing method for the learning content according to the optimized learning process data of the adolescent users in the verification user group, the optimized learning process data of the adolescent users in the benchmark user group, the optimized interaction process data of the adolescent users in the verification user group, and the optimized interaction process data of the adolescent users in the benchmark user group; The determining, according to the plurality of first interest coefficients, learning interest evaluation results corresponding to the test learning content publishing mode and the reference learning content publishing mode, respectively, includes: According to the multiple first interest coefficients and the multiple second interest coefficients, the learning interest evaluation results corresponding to the test learning content publishing mode and the reference learning content publishing mode are determined.

9. The method according to claim 2, characterized in that: The method further comprises: According to a user knowledge graph between a plurality of adolescent users, clustering processing is performed on the plurality of adolescent users to obtain a plurality of user cluster groups; the user knowledge graph is used to characterize the learning interactive association relationship between the plurality of adolescent users, and an interactive association relationship exists between the adolescent users in each user cluster group in the plurality of user cluster groups; A plurality of user clustering groups that meet preset test criteria are extracted from the plurality of user clustering groups, and the adolescent users in the plurality of user clustering groups are taken as the plurality of adolescent users.

10. A server system, characterized in that: The method comprises a server, wherein the server is used to execute the method described in any one of claims 1 to 9.

Images