Recommendation method for AIDS prevention and control knowledge propaganda and education

Abstract

The invention relates to the field of intelligent recommendation, and provides an AIDS prevention and control knowledge propaganda and education recommendation method, which comprises the following steps of: carrying out semantic coding on personal basic information and each piece of browsing information of a user by adopting a natural language processing technology based on artificial intelligence; on the basis of each browsing information semantic feature, performing portrait updating on the personal basic information semantic feature of the user, and then performing semantic coding on to-be-pushed AIDS prevention and control knowledge information, and intelligently judging whether to recommend the to-be-pushed AIDS prevention and control knowledge information to the user according to the user information portrait level updating semantic features and the AIDS prevention and control knowledge information semantic features. Therefore, by deeply knowing the real interest of the user in real time, the accuracy and timeliness of AIDS prevention and control knowledge recommendation can be improved.

Description

Technical Field

[0001] This application relates to the field of intelligent recommendation, and more specifically, to a recommendation method for AIDS prevention and control knowledge education and publicity. Background Art

[0002] AIDS (Acquired Immunodeficiency Syndrome) is a serious infectious disease caused by the human immunodeficiency virus (HIV). Despite significant progress in medical research and treatment methods, the spread of AIDS cannot be underestimated. In the prevention and control work, improving the public's awareness and understanding of AIDS is a key link.

[0003] In this regard, the invention with the publication number CN111538913A proposes a personalized recommendation method for AIDS prevention and control knowledge education and publicity. It collects user basic information and browsing records, uses KL divergence to calculate the similarity between information to solve the problem of data sparsity. An improved K-medoids algorithm is used for information clustering to break the symmetric mode of traditional distance methods. Based on the Top-n recommendation mechanism, unrated information is predicted and recommended to users. According to the interest level of enterprising people, more AIDS prevention and treatment content is pushed personalized to improve the effectiveness and accuracy of knowledge education and publicity.

[0004] This invention conducts personalized recommendation by calculating KL divergence, clustering analysis on user browsing information, and using the Top-n recommendation algorithm. The patent method mainly relies on the probability distribution of user ratings (such as KL divergence calculation) for similarity evaluation, and fails to deeply mine the semantic information of the content itself. Although this method can handle the problem of data sparsity, it ignores the deep meaning of the content, which may lead to inaccurate recommendations. In addition, users' interests and preferences change over time, and the recommendation system based on historical browsing behavior and rating records is difficult to capture these dynamic changes in real time, resulting in the recommended content lagging behind the user's current interests.

[0005] Therefore, an optimized recommendation method for AIDS prevention and control knowledge education and publicity is needed. Summary of the Invention

[0006] This application aims at the deficiencies in the prior art and provides a recommendation method for AIDS prevention and control knowledge education and publicity.

[0007] According to one aspect of this application, a recommendation method for AIDS prevention and control knowledge education and publicity is provided, which includes: Collecting user personal basic information; collecting user historical browsing information; based on the user personal basic information and the user historical browsing information, recommending AIDS prevention and control knowledge information to the user; Among them, based on the user personal basic information and the user historical browsing information, recommending AIDS prevention and control knowledge information to the user includes: Semantically encode the user's personal basic information to obtain the semantic features of the user's personal basic information; Semantically encode each browsing information in the user's historical browsing information to obtain a set of semantic features of the browsing information; Based on the set of semantic features of the browsing information, perform portrait update with modal perception on the semantic features of the user's personal basic information to obtain the updated semantic features at the user information portrait level, including: extracting the semantic kernel features of the browsing information from the set of semantic features of the browsing information to obtain the semantic kernel features of the browsing information; based on the semantic kernel features of the browsing information, perform clustering fine-grained update on the set of semantic features of the browsing information and the semantic features of the user's personal basic information to obtain the updated semantic features at the user information portrait level; Obtain the AIDS prevention and control knowledge information to be pushed; Semantically encode the AIDS prevention and control knowledge information to be pushed to obtain the semantic features of the AIDS prevention and control knowledge information; Based on the updated semantic features at the user information portrait level and the semantic features of the AIDS prevention and control knowledge information, determine whether to recommend the AIDS prevention and control knowledge information to be pushed to the user.

[0008] Further, semantically encoding the user's personal basic information to obtain the semantic features of the user's personal basic information includes: using a semantic encoder based on BiRNN to semantically encode the user's personal basic information to obtain a semantic encoding vector of the user's personal basic information as the semantic features of the user's personal basic information.

[0009] Further, semantically encoding each browsing information in the user's historical browsing information to obtain a set of semantic features of the browsing information includes: using the semantic encoder based on BiRNN to semantically encode each browsing information in the user's historical browsing information to obtain a set of semantic encoding vectors of the browsing information as the set of semantic features of the browsing information.

[0010] Further, extracting the semantic kernel features of the browsing information from the set of semantic features of the browsing information to obtain the semantic kernel features of the browsing information includes: Calculating the semantic difference values between each semantic encoding vector of the browsing information in the set of semantic encoding vectors of the browsing information and all other semantic encoding vectors of the browsing information to obtain a set of semantic difference values of the browsing information; Performing normalization processing on the set of semantic difference values of the browsing information to obtain a set of normalized semantic difference values of the browsing information; Using the set of the normalized semantic difference values of the browsing information as a set of weights, calculate the weighted sum of the set of the semantic encoding vectors of the browsing information to obtain a semantic kernel feature vector of the browsing information as the semantic kernel feature of the browsing information.

[0011] Further, based on the semantic kernel feature of the browsing information, perform clustering fine-grained update on the set of the semantic features of the browsing information and the semantic features of the user's personal basic information to obtain the updated semantic features at the user information portrait level, including: Based on the semantic encoding vector of the user's personal basic information and the semantic kernel feature vector of the browsing information, determine a prior clustering center encoding vector of the basic information - browsing information; Perform cross-domain fine-grained update of the clustering center on the prior clustering center encoding vector of the basic information - browsing information and the set of the semantic encoding vectors of the browsing information to obtain an updated semantic encoding vector at the user information portrait level as the updated semantic feature at the user information portrait level.

[0012] Further, perform cross-domain fine-grained update of the clustering center on the prior clustering center encoding vector of the basic information - browsing information and the set of the semantic encoding vectors of the browsing information to obtain an updated semantic encoding vector at the user information portrait level, including: Use a semantic metric module of the inverse hyperbolic cosine function to respectively perform semantic metric on the prior clustering center encoding vector of the basic information - browsing information and each semantic encoding vector in the set of the semantic encoding vectors of the browsing information to obtain a set of basic information - browsing information semantic metric values; Use a binary function to perform clustering judgment on the set of the basic information - browsing information semantic metric values to obtain a set of basic information - browsing information semantic feature clustering coefficients; Based on the set of the basic information - browsing information semantic feature clustering coefficients, perform key semantic enhancement on the prior clustering center encoding vector of the basic information - browsing information and the set of the semantic encoding vectors of the browsing information to obtain the updated semantic encoding vector at the user information portrait level.

[0013] Further, perform semantic encoding on the AIDS prevention and control knowledge information to be pushed to obtain semantic features of the AIDS prevention and control knowledge information, including: using a semantic understanding encoder based on a BERT - bidirectional LSTM hybrid model to perform semantic encoding on the AIDS prevention and control knowledge information to be pushed to obtain a semantic encoding vector of the AIDS prevention and control knowledge information as the semantic feature of the AIDS prevention and control knowledge information.

[0014] Further, based on the updated semantic features at the user information portrait level and the semantic features of the AIDS prevention and control knowledge information, determine whether to recommend the AIDS prevention and control knowledge information to be pushed to the user, including: Calculate the KL divergence between the user information portrait-level updated semantic coding vector and the AIDS prevention and control knowledge information semantic coding vector; Based on the comparison between the KL divergence and a preset threshold, determine whether to recommend the AIDS prevention and control knowledge information to be pushed to the user.

[0015] Due to the adoption of the above technical solutions, this application has remarkable technical effects: The recommended method for AIDS prevention and control knowledge education provided by this application uses natural language processing technology based on artificial intelligence to perform semantic coding on the user's personal basic information and each browsing information. Then, based on the semantic features of each browsing information, the semantic features of the user's personal basic information are updated in the portrait. Then, semantic coding is performed on the AIDS prevention and control knowledge information to be pushed. In this way, according to the user information portrait-level updated semantic features and the AIDS prevention and control knowledge information semantic features, it is possible to intelligently determine whether to recommend the AIDS prevention and control knowledge information to be pushed to the user. In this way, by deeply understanding the user's true interests in real time, the accuracy and timeliness of AIDS prevention and control knowledge recommendation can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] By describing the embodiments of the present application in more detail in conjunction with the drawings, the above and other objects, features, and advantages of the present application will become more obvious. The drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0017] Figure 1 It is a flowchart of the recommended method for AIDS prevention and control knowledge education according to an embodiment of the present application.

[0018] Figure 2 It is a flowchart of step S3 in the recommended method for AIDS prevention and control knowledge education according to an embodiment of the present application.

[0019] Figure 3 It is a schematic diagram of data flow in step S3 of the recommended method for AIDS prevention and control knowledge education according to an embodiment of the present application.

[0020] Figure 4 It is a flowchart of step S33 in the recommended method for AIDS prevention and control knowledge education according to an embodiment of the present application.

[0021] Figure 5 It is a flowchart of step S332 in the recommended method for AIDS prevention and control knowledge education according to an embodiment of the present application.

[0022] Figure 6It is a flowchart of step S36 in the recommended method for AIDS prevention and control knowledge education according to an embodiment of the present application. Detailed implementation manners

[0023] Hereinafter, exemplary embodiments of the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0024] AIDS, namely Acquired Immune Deficiency Syndrome (AIDS), is a serious infectious disease caused by the infection of the Human Immunodeficiency Virus (HIV). Despite significant progress in medical research and treatment methods, the spread of AIDS still poses a severe public health challenge. To effectively contain its spread, it is particularly important to enhance the public's awareness and understanding of AIDS.

[0025] In this context, the invention with the publication number CN111538913A proposes a personalized recommendation method for AIDS prevention and control knowledge education. This method first collects the personal information and browsing records of users, and then applies the KL divergence to measure the similarity between different pieces of information to overcome the problem of data sparsity. Next, an improved K-medoids algorithm is used for information clustering to break the symmetric pattern of traditional distance methods. Finally, based on the Top-n recommendation mechanism, unrated information is predicted and recommended to users. According to the interest level of aggressive people, more AIDS prevention and treatment content is pushed personalized to improve the effectiveness and accuracy of knowledge education.

[0026] This invention combines KL divergence calculation, clustering analysis, and the Top-n recommendation algorithm to achieve personalized information recommendation. However, this method mainly relies on the similarity evaluation of the user rating probability distribution and fails to fully explore the semantic level of the content itself. This means that although it can effectively handle the problem of data sparsity, it has deficiencies in understanding and capturing the deep meaning of the content, which may result in less than ideal recommendation accuracy. In addition, considering that users' interests and preferences evolve over time, the recommendation mechanism relying on past browsing behaviors and rating records is difficult to immediately reflect the latest dynamics of users, so the recommended content may lag behind the actual interests of users.

[0027] Based on this, the present application proposes a recommended method for AIDS prevention and control knowledge education. Figure 1 It is a flowchart of the recommended method for AIDS prevention and control knowledge education according to an embodiment of the present application. As Figure 1As shown, the recommended method for AIDS prevention and control knowledge education according to the embodiments of the present application includes: S1, collecting the user's personal basic information; S2, collecting the user's historical browsing information; S3, based on the user's personal basic information and the user's historical browsing information, recommending AIDS prevention and control knowledge information to the user.

[0028] In step S1, the user's personal basic information is collected. It should be understood that the user's personal basic information includes basic identity information such as the user's name, gender, age, geographical location information, occupation information, health status information, etc. By collecting this user's personal basic information, a basic portrait of the user can be established, and thus a basis for subsequent information recommendation can be provided. For example, by combining the user's age, occupation, health status, etc., the possible needs and preferences of the user for AIDS prevention and control knowledge can be roughly outlined, which is conducive to achieving more accurate AIDS prevention and control knowledge recommendation.

[0029] In step S2, the user's historical browsing information is collected. It should be understood that the user's historical browsing information specifically includes the theme information of the AIDS prevention and control knowledge articles or videos browsed by the user, the browsing time and frequency information, and some interactive behavior information during browsing, such as liking, commenting, sharing, etc. Specifically, the content browsed by the user in different aspects of AIDS prevention and control can reflect the key points of the user's attention to knowledge in different fields of AIDS. For example, if the user frequently browses articles on the progress of AIDS treatment, it indicates that the user may be interested in the latest treatment technologies and drug research and development. The length of browsing time can reflect the user's attention and in-depth reading degree to a specific content. Long-term browsing may indicate that the user is seriously studying and researching this content. The browsing frequency reflects the user's repeated attention to different themes, and the themes with high browsing frequencies are often the ones that the user cares more about. The user's interactive behaviors such as liking, commenting, and sharing can further reveal the user's preferences and attitudes towards the content. If the user has made positive comments and sharing on an AIDS prevention and control knowledge article, it indicates that the content of this article is very likely to meet the user's needs. Generally speaking, by analyzing the themes, time, frequency, and interactive behaviors in the user's historical browsing information, the user's interest points and needs in AIDS prevention and control knowledge can be accurately determined, which is conducive to realizing personalized recommendation of AIDS prevention and control knowledge.

[0030] In today's digital age, collecting the user's historical browsing information is of crucial significance for the recommendation of AIDS prevention and control knowledge education. Its specific implementation process covers multiple levels and complex technical means, and operates closely through multiple channels in real life. The following is a specific elaboration of its implementation process: All kinds of information websites play a crucial role in this process. When users visit these websites and browse content related to AIDS prevention and control, the log system in the website background will meticulously capture every browsing action of the users. It will accurately record the URLs of the pages visited by the users, and these URLs contain rich information. Through complex algorithms and data analysis, keywords and page classification information can be extracted from them, so as to precisely determine whether the themes of the AIDS prevention and control knowledge articles or videos browsed by the users focus on the transmission mechanism of AIDS, the latest research on prevention methods, or the sharing of clinical cases during the treatment process, etc.

[0031] At the same time, the recording of timestamps is also extremely crucial. The moment when a user enters a page each time, the time will be accurately recorded, and the time data accurate to the second or even millisecond level can provide a solid foundation for subsequent analysis. With the help of these time data, not only can the browsing duration of each page by the user be counted to understand the energy and attention invested by the user in different contents, but also through the analysis of browsing behaviors within different time periods, the changing trend of the user's attention frequency to AIDS prevention and control knowledge can be clearly presented. For example, if it is found that the user frequently browses AIDS-related content within a certain specific time period, this may mean that the user has a more urgent need for knowledge in this area during that period.

[0032] In the case of user registration and login, the value of this data collection is further enhanced. The website can closely associate the browsing information with the user account, realizing long-term, systematic, and comprehensive tracking and integration of the historical browsing information of individual users. This is like establishing a dedicated knowledge browsing archive for each user. As time goes by, this archive keeps enriching and can clearly show the evolution path of the user's interests in the field of AIDS prevention and control knowledge.

[0033] Mobile applications are also important positions for collecting users' historical browsing information. In mobile applications related to AIDS prevention and control, developers have embedded special data collection codes. From the moment when a user opens the application, these codes will comprehensively capture the user's operation behaviors. When the user browses articles or videos within the application, the application will automatically record key information such as its title, content summary, and playing duration. These data not only help to determine the browsing theme and time information of the user, but also through the text analysis of the content summary, specific knowledge points and potential interest points that the user may be concerned about can be mined. For example, if the prevention measures for mother-to-child transmission of AIDS are repeatedly mentioned in the content summary and the user stays on this page for a long time, this may imply that the user has a high degree of attention to this specific area.

[0034] For users' interaction behaviors, mobile applications have set up dedicated mechanisms for recording. Operations such as liking, commenting, and sharing have all become important bases for analyzing users' interest preferences. When a user clicks the like button, the application backend will quickly respond and record this behavior and the corresponding content. Users' comments directly reflect their thoughts and concerns. By analyzing these comments through natural language processing technology, we can deeply understand users' understanding of AIDS prevention and control knowledge, their doubts, and the further information they expect to obtain. The sharing behavior indicates that users think the content has certain value and are willing to spread it to others, which also reflects the degree of fit between the content and users' interests from the side. These data are usually securely stored in the application's cloud server for subsequent in-depth data analysis and processing.

[0035] Cross-platform data integration elevates this information collection work to a more macroscopic and comprehensive level. With the powerful processing capabilities of big data technology and advanced user identification algorithms, it can break down the information barriers between different platforms and devices. By conducting correlation analysis on the account information logged in by users on different devices and combining the identification of unique identifiers such as IP address tracking, the system can accurately determine which browsing behaviors belong to the same user. For example, when a user browses some basic introductions to AIDS prevention and control knowledge on a computer in the office and then continues to delve into relevant treatment progress content on their mobile phone on the way home, the system can seamlessly associate these scattered browsing information on different devices with its powerful integration ability to form a complete and coherent user historical browsing record. This cross-platform data integration greatly enriches the dimension of user information, can more comprehensively and accurately reflect the changes in users' interests and the trajectory of their knowledge acquisition, and provides a more solid and rich data support for subsequent precise recommendation of AIDS prevention and control knowledge.

[0036] In short, through the coordinated operation of various methods such as network platforms, mobile applications, and cross-platform data integration, the collection of users' historical browsing information can be effectively implemented, laying a solid foundation for the precise operation of AIDS prevention and control knowledge education and recommendation, and thus powerfully promoting the popularization and dissemination of AIDS prevention and control knowledge.

[0037] In step S3, based on the user's personal basic information and the user's historical browsing information, AIDS prevention and control knowledge information is recommended to the user. It should be understood that by analyzing the user's personal basic information and historical browsing information, the recommended AIDS prevention and control knowledge can be highly consistent with the user's life background and interest points. This personalized recommendation method can make it easier for users to understand and accept the recommended knowledge, which is conducive to enhancing the acceptability of AIDS prevention and control knowledge among the user group, thereby strengthening users' mastery of AIDS prevention and control knowledge.

[0038] Accordingly, in the process of recommending AIDS prevention and control knowledge information to users based on the user's personal basic information and the user's historical browsing information, the technical concept of this application is to use natural language processing technology based on artificial intelligence to perform semantic encoding on the user's personal basic information and each browsing information. Then, based on the semantic features of each browsing information, the semantic features of the user's personal basic information are updated for profiling. Then, the AIDS prevention and control knowledge information to be pushed is obtained and its semantic encoding is performed. Based on this, it is intelligently determined whether to recommend the AIDS prevention and control knowledge information to be pushed to the user according to the updated semantic features at the user information profiling level and the semantic features of the AIDS prevention and control knowledge information. This application can capture the deep meaning of the content, rather than simply relying on the scoring probability distribution. This enables the system to more accurately understand the user's true interests. Moreover, it can promptly capture the changes in the user's interests. Even if the user's interests suddenly shift to a new field or the interest in a specific topic deepens, the system can quickly adjust the recommendation strategy and provide the latest relevant information.

[0039] Specifically, Figure 2 FIG. is a flowchart of step S3 in the AIDS prevention and control knowledge education and promotion recommendation method according to an embodiment of the present application. Figure 3 FIG. is a schematic diagram of data flow in step S3 in the AIDS prevention and control knowledge education and promotion recommendation method according to an embodiment of the present application. As Figure 2 and Figure 3 shown, the step S3 includes: S31, performing semantic encoding on the user's personal basic information to obtain semantic features of the user's personal basic information; S32, performing semantic encoding on each browsing information in the user's historical browsing information to obtain a set of semantic features of the browsing information; S33, based on the set of semantic features of the browsing information, performing modality-aware profiling update on the semantic features of the user's personal basic information to obtain updated semantic features at the user information profiling level; S34, obtaining the AIDS prevention and control knowledge information to be pushed; S35, performing semantic encoding on the AIDS prevention and control knowledge information to be pushed to obtain semantic features of the AIDS prevention and control knowledge information; S36, based on the updated semantic features at the user information profiling level and the semantic features of the AIDS prevention and control knowledge information, determining whether to recommend the AIDS prevention and control knowledge information to be pushed to the user.

[0040] In step S31, semantic encoding is performed on the user's personal basic information to obtain the semantic features of the user's personal basic information. Specifically, in the embodiment of the present application, step S31 includes: using a semantic encoder based on BiRNN to perform semantic encoding on the user's personal basic information to obtain a semantic encoding vector of the user's personal basic information as the semantic features of the user's personal basic information. It should be understood that considering that the user's personal basic information includes various types, such as age, gender, occupation, region, etc., key semantic meaning information is included among these information. Based on this, semantic encoding is performed on the user's personal basic information to capture the semantic features and potential semantic correlations in the user's personal basic information, and a semantic encoding vector of the user's personal basic information is obtained. In particular, in a specific embodiment of the present application, a semantic encoder based on BiRNN is used to perform semantic encoding on the user's personal basic information to obtain a semantic encoding vector of the user's personal basic information. Those of ordinary skill in the art should know that BiRNN (Bidirectional Recurrent Neural Network) is a special RNN structure that processes the forward and reverse information of the input data simultaneously. For text data such as the user's personal basic information, BiRNN can capture the relationship before and after words, better understand the meaning of the entire sentence or paragraph, and thus help build a more refined user profile.

[0041] The following is a detailed elaboration of a specific implementation process of "using a semantic encoder based on BiRNN to perform semantic encoding on the user's personal basic information to obtain a semantic encoding vector of the user's personal basic information": First is the data arrangement link. The user's personal basic information is obtained from various channels, including but not limited to the user registration information form, questionnaire feedback, etc. For example, on a platform for promoting knowledge about AIDS prevention and control, information such as the user's name, gender, age, geographical location, occupation, and health status filled in during registration will be collected by the system. The collected data needs to be cleaned and standardized to remove incomplete, incorrect, or inconsistent data. For example, unifying data in different formats into text format, standardizing geographical location information to be accurate to the city or region, etc., to ensure the quality and usability of the data.

[0042] Next is the model building environment. The construction of a semantic encoder based on BiRNN requires selecting a suitable development framework, such as TensorFlow or PyTorch. After determining the framework, design the network structure of BiRNN. According to the characteristics and data volume of the user's personal basic information, determine parameters such as the number of layers of BiRNN, the number of neurons in each layer, and the activation function of the hidden layer. If the user's personal basic information is relatively rich and complex, a BiRNN structure with 2 - 3 layers may be selected, with 128 - 256 neurons in each layer, and activation functions such as ReLU are used to enhance the nonlinear expression ability of the model. At the same time, the dimensions of the input layer and the output layer also need to be determined. The dimension of the input layer should match the dimension of the vector representation of the user's personal basic information after preprocessing, while the dimension of the output layer is set according to the dimension of the required semantic encoding vector.

[0043] Before model training, the user's personal basic information needs to be preprocessed to adapt to the model input. As mentioned before, first perform text cleaning to remove interference information such as special characters and stop words in the text. For occupational information like "I am a teacher, working at [school name], and in good health.", common meaningless words such as "I am a" and "working at..." will be removed. Then, use word embedding technology to convert the text into a vector representation. Common Word2Vec or GloVe models can map each word to a vector with a fixed dimension, such as a 100 - dimensional or 300 - dimensional vector. In this way, the text sequence in the user's personal basic information is converted into a vector sequence as the input of the BiRNN model.

[0044] Model training is the core step. Prepare a large amount of labeled data or samples of the user's personal basic information with known characteristics. Divide these samples into a training set, a validation set, and a test set, usually in a ratio of 7:2:1 or 8:1:1. During training, input the preprocessed vector sequence of the user's personal basic information into the BiRNN model. The model performs forward propagation calculations according to its internal structure and parameters to obtain the output result. By comparing with the real semantic labels or the expected output, calculate the value of the loss function, such as using the cross - entropy loss function. Then, use the backpropagation algorithm to adjust the parameters of the model according to the value of the loss function, and continuously iterate this process until the performance of the model on the validation set reaches the optimal or meets the preset stop conditions. For example, when the accuracy on the validation set no longer improves or the value of the loss function no longer decreases, stop training.

[0045] After the model training is completed, the actual semantic encoding application can be carried out. The new user's personal basic information is input into the trained BiRNN model after passing through the same preprocessing steps. The model will process the input information according to the semantic patterns and rules learned during the training process. Under the bidirectional propagation mechanism of BiRNN, the forward and backward information flows can fully capture the semantic relationships before and after words in the user's personal basic information. For example, for information like "a 40-year-old woman engaged in the medical industry, living in a coastal city, with a habit of regular physical examinations", the model can accurately understand the associations among age, occupation, gender, geographical location, and health behaviors, and encode them into a representative semantic encoding vector. This vector can comprehensively reflect the user's basic characteristics and potential semantic information, providing strong support for subsequent user portrait construction and AIDS prevention and control knowledge recommendation.

[0046] To ensure the long-term effectiveness of the model, model monitoring and updating are also required. The model is evaluated regularly using new data. If it is found that the performance of the model on the new data has declined, such as a decrease in the accuracy of semantic encoding or a deterioration in the adaptability to new patterns of user personal basic information, it is necessary to re-collect data to retrain or fine-tune the model to ensure that the model can always accurately perform semantic encoding on user personal basic information and adapt to the changing requirements of real-world applications.

[0047] In step S32, semantic encoding is performed on each browsing information in the user's historical browsing information to obtain a set of browsing information semantic features. Specifically, in the embodiment of the present application, step S32 includes: using the BiRNN-based semantic encoder to perform semantic encoding on each browsing information in the user's historical browsing information to obtain a set of browsing information semantic encoding vectors as the set of browsing information semantic features. It should be understood that the user's historical browsing information usually includes various types, such as news reports, popular science articles, forum posts, and other different types of texts, and each type of text contains complex semantic patterns, such as different topics and keywords. Therefore, in order to extract meaningful information from the browsing records to infer the user's interest preferences for different topics and fields, the present application performs semantic encoding on each browsing information in the user's historical browsing information to fully understand the relationships between these information and extract key and meaningful semantic information, obtaining a set of browsing information semantic encoding vectors. In particular, in a specific embodiment of the present application, the BiRNN-based semantic encoder is used to perform semantic encoding on each browsing information in the user's historical browsing information to process the elements in the sequence both from front to back and from back to front, so as to capture the sequential relationships in each browsing content and their mutual influences, obtaining a set of browsing information semantic encoding vectors.

[0048] In step S33, based on the set of browsing information semantic features, the semantic features of the user's personal basic information are updated with a modality-aware portrait to obtain user information portrait-level updated semantic features. Specifically, Figure 4 FIG. 1 is a flowchart of step S33 in the method for recommending AIDS prevention and control knowledge education according to an embodiment of the present application. Figure 4 As shown, the step S33 includes: S331, extracting browsing information semantic kernel features from the set of browsing information semantic features to obtain browsing information semantic kernel features; S332, based on the browsing information semantic kernel features, clustering and fine-grained updating the set of browsing information semantic features and the semantic features of the user's personal basic information to obtain the user information portrait-level updated semantic features.

[0049] It should be understood that the semantic features of the user's personal basic information reflect the inherent characteristics of the user, while the semantic features of each browsing information reflect the interests and concerns of the user based on the browsing behavior. That is, the user's interests and needs will change over time, and the browsing information can capture these short-term changes. Therefore, in order to continuously update the user portrait so that the portrait is more in line with the user's current state, the present application performs a modality-aware portrait update on the semantic features of the user's personal basic information based on the set of browsing information semantic features to obtain user information portrait-level update semantic features. In other words, the modality-aware portrait update can tap into the subtle correlation between the two, and discover user features that are difficult to detect from only a single information source to more accurately integrate the user's basic attributes and behavioral interests, so that the portrait is more in line with the user's actual situation.

[0050] Specifically, in the embodiment of the present application, the step S331 includes: calculating the semantic difference value between each browsing information semantic coding vector and all other browsing information semantic coding vectors in the set of browsing information semantic coding vectors to obtain a set of browsing information semantic difference values; normalizing the set of browsing information semantic difference values ​​to obtain a set of normalized browsing information semantic difference values; using the set of normalized browsing information semantic difference values ​​as a set of weights, calculating the weighted sum of the set of browsing information semantic coding vectors to obtain a browsing information semantic kernel feature vector as the browsing information semantic kernel feature. The above process can be expressed as: ; in, is a set of the browsing information semantic encoding vectors, , , , and are the first, second, and third in the set of browsing information semantic encoding vectors. The first, the The first and the first browsing information semantic coding vectors, For performing kernel feature extraction, represents the L1 norm of the feature vector, represents the number of vectors in the set of the browsing information semantic coding vectors minus one, is the corresponding browsing information semantic difference value, represents the exponential function with the natural constant as the base, is the browsing information semantic kernel feature vector.

[0051] It should be understood that the user's personal basic information and browsing information belong to different modal data respectively. The user's personal basic information is relatively stable, while the browsing information is dynamically changing. In order to enable these two modal data to work effectively together to better represent the user portrait, it is necessary to find the connection link between them, that is, the cross-modal data interaction anchoring bridge. By performing browsing information semantic kernel feature extraction on the set of the browsing information semantic features of the present application, the key elements closely related to the user's inherent attributes in the browsing information can be mined, and then the two modal data can be organically combined to enable the model to better understand the user's overall needs and behavior patterns.

[0052] Specifically, Figure 5 is the flowchart of step S332 in the recommended method for AIDS prevention and control knowledge education according to the embodiment of the present application. As Figure 5 shown, the step S332 includes: S3321, determining the basic information-browsing information prior clustering center coding vector based on the user's personal basic information semantic coding vector and the browsing information semantic kernel feature vector; S3322, performing cross-domain fine-grained update of the clustering center on the basic information-browsing information prior clustering center coding vector and the set of the browsing information semantic coding vectors to obtain the user information portrait-level updated semantic coding vector as the user information portrait-level updated semantic feature.

[0053] Next, based on the user's personal basic information semantic coding vector and the browsing information semantic kernel feature vector, the basic information-browsing information prior clustering center coding vector is determined. The above process can be expressed as: ; Wherein, is the browsing information semantic kernel feature vector, is the user's personal basic information semantic coding vector, is the concatenation operation, is the weight transformation matrix, is the bias vector, is the basic information - browsing information prior clustering center coding vector.

[0054] It should be understood that the user's personal basic information and browsing information are two key modal data sources. The semantic coding vector of the user's personal basic information reflects the relatively stable inherent attributes of the user, while the semantic kernel feature vector of the browsing information reflects the user's dynamic interests and concerns. In this application, by determining the basic information - browsing information prior clustering center coding vector, these two modal information can be effectively fused, enabling the model to better understand the user. That is, the basic information - browsing information prior clustering center coding vector serves as an anchor bridge for cross - modal data interaction, establishing a close connection between the user's personal basic information and browsing information, and guiding the interaction of these two cross - modal data. By combining the advantages of the two modalities, namely the directly extracted basic information features and the processed browsing information features, it can provide strong guidance for subsequent analysis, making the results of subsequent analysis more capable of reflecting the user's needs and behavior patterns in different dimensions.

[0055] More specifically, in the embodiment of this application, the step S3322 includes: using the semantic metric module of the inverse hyperbolic cosine function to perform semantic metric on each browsing information semantic coding vector in the set of the basic information - browsing information prior clustering center coding vector and the browsing information semantic coding vector respectively to obtain a set of basic information - browsing information semantic metric values. This process can be expressed as: ; wherein, is the th browsing information semantic coding vector in the set of the browsing information semantic coding vectors, is the basic information - browsing information prior clustering center coding vector, is the square of the Euclidean norm of the vector, is the inverse hyperbolic cosine function, is and the basic information - browsing information semantic metric value between; using a binary function to perform clustering judgment on the set of the basic information - browsing information semantic metric values to obtain a set of basic information - browsing information semantic feature clustering coefficients. This process can be expressed as: ; wherein, is and the basic information - browsing information semantic metric value between, is the preset threshold, is The corresponding basic information - browsing information semantic feature clustering coefficient; Based on the set of the basic information - browsing information semantic feature clustering coefficients, perform key semantic enhancement on the set of the basic information - browsing information prior clustering center encoding vectors and the browsing information semantic encoding vectors to obtain the user information portrait - level updated semantic encoding vectors. This process can be expressed as: ; wherein, is the th browsing information semantic encoding vector in the set of the browsing information semantic encoding vectors, is the basic information - browsing information prior clustering center encoding vector, is the corresponding basic information - browsing information semantic feature clustering coefficient, is the number of vectors in the set of the browsing information semantic encoding vectors, is the user information portrait - level updated semantic encoding vector.

[0056] It should be understood that there may be noise and redundancy in the set of browsing information semantic features and the user's personal basic information semantic features. Direct use may lead to inaccurate construction of the user portrait. By using the basic information - browsing information prior clustering center encoding vector as a cross - modal data interaction anchoring bridge for cross - domain fine - grained update of the clustering center, this application can focus on the most valuable information part of the browsing information semantic features, remove interference factors, and thus generate a new feature representation that is more accurate and can better reflect the user's essential features, namely the user information portrait - level updated semantic encoding vector. The finally obtained user information portrait - level updated semantic encoding vector can highly accurately reflect the user's comprehensive situation in terms of AIDS prevention and control knowledge, including their stable basic attributes and dynamic interest changes. This enables subsequent recommendations to closely fit the user's actual needs, thereby improving the accuracy of recommendations.

[0057] In step S34, obtain the AIDS prevention and control knowledge information to be pushed. It should be understood that the AIDS prevention and control knowledge information to be pushed should cover comprehensive and diverse topics to meet the interests and needs of different users. The AIDS prevention and control knowledge information generally includes basic education knowledge about AIDS, prevention measure knowledge, treatment progress knowledge, scientific research dynamic knowledge, etc. By performing semantic analysis on the AIDS prevention and control knowledge information to be pushed, the key content in the AIDS prevention and control knowledge information to be pushed can be understood, and it can be matched with the user interests reflected in the user portrait constructed previously, so as to realize personalized recommendation of the AIDS prevention and control knowledge information. For example, if the user portrait shows that the user is more concerned about AIDS treatment, and if the AIDS prevention and control knowledge information to be pushed expounds on the content of AIDS treatment, then this AIDS prevention and control knowledge information is very likely to be recommended to the user.

[0058] In step S35, semantic encoding is performed on the AIDS prevention and control knowledge information to be pushed to obtain the semantic features of the AIDS prevention and control knowledge information. Specifically, in the embodiment of the present application, step S35 includes: using a semantic understanding encoder based on a BERT-biLSTM hybrid model to perform semantic encoding on the AIDS prevention and control knowledge information to be pushed to obtain a semantic encoding vector of the AIDS prevention and control knowledge information as the semantic features of the AIDS prevention and control knowledge information. It should be understood that the AIDS prevention and control knowledge information to be pushed contains the deep meaning and context relationship in the text. Therefore, in order to more clearly understand and analyze the semantic information contained therein, the present application performs semantic encoding on the AIDS prevention and control knowledge information to be pushed to mine the semantic features and relationships therein, and obtain a semantic encoding vector of the AIDS prevention and control knowledge information. In particular, in a specific embodiment of the present application, a semantic understanding encoder based on a BERT-biLSTM hybrid model can be used to perform semantic encoding on the AIDS prevention and control knowledge information to be pushed to obtain a semantic encoding vector of the AIDS prevention and control knowledge information. Specifically, BERT is pre-trained on large-scale text data and can learn rich language knowledge and semantic information. The AIDS prevention and control knowledge information contains a large number of professional terms, complex sentence patterns, and subtle semantic relationships, and BERT can effectively capture these information and understand the semantics of each vocabulary and the overall sentence. BiLSTM is particularly good at processing sequence data. It can capture the long-term dependence relationship between elements in the sequence. For the knowledge content expressed in sequence, such as the description of the process from AIDS infection to onset, BiLSTM can effectively process the information flow in this time series or logical sequence. Therefore, a BERT-biLSTM hybrid model is adopted. First, BERT processes the input prevention and control knowledge text to extract global features and semantics; then, the output of BERT is used as the input of BiLSTM, which deeply mines the word-level context semantic dependence and generates a more representative encoding vector, providing a reliable basis for accurate recommendation of prevention and control knowledge.

[0059] In step S36, based on the user information portrait-level updated semantic features and the semantic features of the AIDS prevention and control knowledge information, it is determined whether to recommend the AIDS prevention and control knowledge information to be pushed to the user. Specifically, Figure 6 is a flowchart of step S36 in the recommendation method for AIDS prevention and control knowledge education according to the embodiment of the present application. As Figure 6 shown, step S36 includes: S361, calculating the KL divergence between the user information portrait-level updated semantic encoding vector and the semantic encoding vector of the AIDS prevention and control knowledge information; S362, based on the comparison between the KL divergence and a preset threshold, determining whether to recommend the AIDS prevention and control knowledge information to be pushed to the user.

[0060] In step S361, the KL divergence between the user information portrait-level updated semantic encoding vector and the AIDS prevention and control knowledge information semantic encoding vector is calculated. It should be understood that the KL divergence is a statistic used to measure the difference between two probability distributions, which can provide a numerical index to describe the difference degree between the two distributions. In this scenario, the user information portrait-level updated semantic encoding vector and the AIDS prevention and control knowledge information semantic encoding vector can be regarded as the "distribution" representations of different semantic information. By calculating the KL divergence, the similarity degree of these two vectors in the semantic space can be quantified, and the difference size of the semantic content they represent can be judged. In particular, since the semantic encoding vector contains rich semantic information, the KL divergence provides an analysis method applicable to this vector space. It can effectively process the complex semantic relationships in the vector. Different from simple distance measurement methods that only consider the geometric distance of the vector, the KL divergence pays more attention to the semantic distribution characteristics represented by the vector, which highly fits the requirements of semantic analysis.

[0061] In a preferred example, since the user information portrait-level updated semantic encoding vector represents the aggregated semantic encoding features after the prior cross-domain update of the semantic encoding features of the user's personal basic information based on the semantic encoding features of the user's historical browsing information, and the AIDS prevention and control knowledge information semantic encoding vector represents the semantic encoding features of the AIDS prevention and control knowledge information to be pushed, when calculating the KL divergence between them for mapping towards a common KL divergence space, there will be insufficient correlation correspondence of different-order semantic encoding features under different-source semantic representations, thus causing the sparse correlation mapping of the user information portrait-level updated semantic encoding vector and the AIDS prevention and control knowledge information semantic encoding vector towards the common KL divergence space, and thus reducing the accuracy of the calculated KL divergence due to the lack of common mapping reasoning degree.

[0062] Based on this, before calculating the KL divergence between the user information portrait-level updated semantic encoding vector and the AIDS prevention and control knowledge information semantic encoding vector, the present application first performs common correlation optimization on the user information portrait-level updated semantic encoding vector and the AIDS prevention and control knowledge information semantic encoding vector, specifically including the steps: Calculate the user information portrait-level updated semantic encoding vector of the eigenvalue and the AIDS prevention and control knowledge information semantic encoding vector of the eigenvalue between the distance sum to respectively obtain the user information - AIDS prevention and control knowledge information common coding first similarity matrix The Second Similarity Matrix Co - encoded with User Information - AIDS Prevention and Control Knowledge Information : ; ; Among them, and respectively represent distance and distance, represents the value at the position of the First Similarity Matrix Co - encoded with User Information - AIDS Prevention and Control Knowledge Information, represents the value at the position of the Second Similarity Matrix Co - encoded with User Information - AIDS Prevention and Control Knowledge Information; Calculate the weighted sum of the First Similarity Matrix Co - encoded with User Information - AIDS Prevention and Control Knowledge Information and the Second Similarity Matrix Co - encoded with User Information - AIDS Prevention and Control Knowledge Information to obtain the Multi - level Similarity Clustering Matrix Co - encoded with User Information - AIDS Prevention and Control Knowledge Information , among them, represents element - wise multiplication by position, represents element - wise addition by position, and are weight hyperparameters; Multiply the Multi - level Similarity Clustering Matrix Co - encoded with User Information - AIDS Prevention and Control Knowledge Information respectively with the Semantic Encoding Vector of User Information Portrait - level Update and the Semantic Encoding Vector of AIDS Prevention and Control Knowledge Information to obtain the Multi - level Similarity Clustering Vector of User Information Portrait - level Update Semantic Encoding and the Multi - level Similarity Clustering Vector of AIDS Prevention and Control Knowledge Information Semantic Encoding , among them, represents matrix multiplication; Multiply the Multi - level Similarity Clustering Matrix Co - encoded with User Information - AIDS Prevention and Control Knowledge Information with the Self - correlation Matrix of the Semantic Encoding Vector of User Information Portrait - level Update and the Semantic Encoding Vector of AIDS Prevention and Control Knowledge Information to obtain the Clustering Association Matrix Co - encoded with User Information - AIDS Prevention and Control Knowledge Information , among them, represents the transpose of a vector; After multiplying the Multi - level Similarity Clustering Vector of User Information Portrait - level Update Semantic Encoding and the Multi - level Similarity Clustering Vector of AIDS Prevention and Control Knowledge Information Semantic Encoding respectively with the Clustering Association Matrix Co - encoded with User Information - AIDS Prevention and Control Knowledge Information, calculate their weighted sum to obtain the Clustering Association Vector Co - encoded with User Information - AIDS Prevention and Control Knowledge Information , among them, and are different weight hyperparameters. Here, those of ordinary skill in the art should know that the weight hyperparameters can be determined using existing automated machine learning tools; Multiply the jointly encoded clustering association vectors of the user information - AIDS prevention and control knowledge information with the user information portrait - level updated semantic encoding vector and the AIDS prevention and control knowledge information semantic encoding vector respectively to obtain an optimized user information portrait - level updated semantic encoding vector and an optimized AIDS prevention and control knowledge information semantic encoding vector.

[0063] That is, for the jointly encoded first similarity matrix of user information - AIDS prevention and control knowledge information and the jointly encoded second similarity matrix of user information - AIDS prevention and control knowledge information based on different spatial similarity orders for the user information portrait - level updated semantic encoding vector and the AIDS prevention and control knowledge information semantic encoding vector, generate a multi - level similarity clustering representation. Then, perform mapping processing on the relationships between different associated clusters for the user information portrait - level updated semantic encoding vector, the AIDS prevention and control knowledge information semantic encoding vector, and their self - associated representations respectively, to simulate the unit activation pattern of the associated system based on sparse similarity response. And use the query feature expression to regulate the sparse characteristics of the jointly encoded clustering association, and achieve a compensation mechanism for the sparsity of the association response at each order through the association mapping feature expression of the multi - level similarity clustering representation, so as to avoid the decline in the inference efficiency of the joint mapping caused by the weak association due to the sparse characteristics, and improve the calculation accuracy of the KL divergence between the user information portrait - level updated semantic encoding vector and the AIDS prevention and control knowledge information semantic encoding vector.

[0064] In step S362, based on the comparison between the KL divergence and a preset threshold, determine whether to recommend the to - be - pushed AIDS prevention and control knowledge information to the user. That is, the preset threshold serves as a criterion, which can help the system screen out the knowledge information that best matches the user's interests, avoid pushing irrelevant or low - relevant content, and thus improve the quality and accuracy of the recommendation. Specifically, if the KL divergence is less than the preset threshold, it indicates that the AIDS prevention and control knowledge information is highly relevant to the user's interests and needs. At this time, recommending this information to the user can accurately meet the user's specific needs for AIDS prevention and control knowledge, improve the user's attention and acceptance of the recommended content, and enhance the practicality of the recommendation system. When the KL divergence is greater than the preset threshold, it means that the knowledge information has a low matching degree with the user's current interests and needs. Not making a recommendation can avoid pushing a large amount of irrelevant information to the user, reduce the possibility of user interference, improve the user experience, and at the same time optimize the resource utilization efficiency of the recommendation system. In particular, the preset threshold can be adjusted according to actual needs. By setting an appropriate threshold, a balance can be found between the two to adapt to different application scenarios and user groups.

[0065] In summary, the recommended method for AIDS prevention and control knowledge education based on the embodiments of the present application is elucidated. It uses natural language processing technology based on artificial intelligence to perform semantic encoding on the user's personal basic information and various browsing information. Then, based on the semantic features of each browsing information, the semantic features of the user's personal basic information are updated for profiling. Next, the AIDS prevention and control knowledge information to be pushed is semantically encoded, and thus, according to the updated semantic features of the user information profile and the semantic features of the AIDS prevention and control knowledge information, it is intelligently determined whether to recommend the AIDS prevention and control knowledge information to be pushed to the user. In this way, by deeply understanding the user's true interests in real time, the accuracy and timeliness of AIDS prevention and control knowledge recommendation can be improved.

Claims

1. A recommended method for AIDS prevention and control knowledge education, characterized in that: include: Collect basic personal information of users; Collect user browsing history information; Recommending AIDS prevention and control knowledge information to the user based on the user's personal basic information and the user's historical browsing information; Wherein, based on the user's personal basic information and the user's historical browsing information, recommending AIDS prevention and control knowledge information to the user includes: Performing semantic coding on the user's basic personal information to obtain semantic features of the user's basic personal information; Semantically encoding each browsing information in the user's historical browsing information to obtain a set of browsing information semantic features; Based on the set of browsing information semantic features, the semantic features of the user's personal basic information are updated with modality perception to obtain the user information portrait-level updated semantic features, including: extracting browsing information semantic kernel features from the set of browsing information semantic features to obtain browsing information semantic kernel features; based on the browsing information semantic kernel features, clustering fine-grained updates on the set of browsing information semantic features and the semantic features of the user's personal basic information to obtain the user information portrait-level updated semantic features; Obtain AIDS prevention and control knowledge information to be pushed; Performing semantic coding on the AIDS prevention and control knowledge information to be pushed to obtain semantic features of the AIDS prevention and control knowledge information; Based on the user information portrait-level updated semantic features and the AIDS prevention and control knowledge information semantic features, it is determined whether to recommend the AIDS prevention and control knowledge information to be pushed to the user.

2. The method for recommending AIDS prevention and control knowledge education according to claim 1, characterized in that: The user's personal basic information is semantically encoded to obtain a semantic feature of the user's personal basic information, including: using a BiRNN-based semantic encoder to semantically encode the user's personal basic information to obtain a user's personal basic information semantic encoding vector as the user's personal basic information semantic feature.

3. The method for recommending AIDS prevention and control knowledge education according to claim 2, characterized in that: Semantically encoding each browsing information in the user's historical browsing information to obtain a set of browsing information semantic features, including: using the BiRNN-based semantic encoder to semantically encode each browsing information in the user's historical browsing information to obtain a set of browsing information semantic encoding vectors as the set of browsing information semantic features.

4. The method for recommending AIDS prevention and control knowledge education according to claim 3, characterized in that: Extracting browsing information semantic kernel features from the set of browsing information semantic features to obtain browsing information semantic kernel features includes: Calculating the semantic difference value between each browsing information semantic coding vector and all other browsing information semantic coding vectors in the set of browsing information semantic coding vectors to obtain a set of browsing information semantic difference values; Normalizing the set of browsing information semantic difference values ​​to obtain a set of normalized browsing information semantic difference values; The set of normalized browsing information semantic difference values ​​is used as a set of weights, and the weighted sum of the set of browsing information semantic encoding vectors is calculated to obtain a browsing information semantic kernel feature vector as the browsing information semantic kernel feature.

5. The method for recommending AIDS prevention and control knowledge education according to claim 4, characterized in that: Based on the browsing information semantic core feature, clustering and fine-grained updating are performed on the set of browsing information semantic features and the semantic features of the user's personal basic information to obtain the user information portrait-level updated semantic features, including: Determine a basic information-browsing information priori cluster center encoding vector based on the user personal basic information semantic encoding vector and the browsing information semantic kernel feature vector; The cluster center is updated cross-domain fine-grainedly on the set of the basic information-browsing information prior cluster center encoding vector and the browsing information semantic encoding vector to obtain a user information portrait-level updated semantic encoding vector as the user information portrait-level updated semantic feature.

6. The method for recommending AIDS prevention and control knowledge education according to claim 5, characterized in that: The cluster center is updated cross-domain fine-grainedly on the set of the basic information-browsing information prior cluster center encoding vector and the browsing information semantic encoding vector to obtain a user information portrait-level updated semantic encoding vector, including: Using a semantic measurement module of an inverse hyperbolic cosine function, semantic measurement is performed on each browsing information semantic coding vector in the set of the basic information-browsing information prior clustering center coding vector and the browsing information semantic coding vector to obtain a set of basic information-browsing information semantic measurement values; Using a binary function to perform clustering judgment on the set of basic information-browsing information semantic measurement values ​​to obtain a set of basic information-browsing information semantic feature clustering coefficients; Based on the set of basic information-browsing information semantic feature clustering coefficients, key semantic enhancement is performed on the set of basic information-browsing information prior clustering center encoding vectors and the browsing information semantic encoding vectors to obtain the user information portrait-level updated semantic encoding vector.

7. The method for recommending AIDS prevention and control knowledge education according to claim 6, characterized in that: The AIDS prevention and control knowledge information to be pushed is semantically encoded to obtain semantic features of the AIDS prevention and control knowledge information, including: using a semantic understanding encoder based on a BERT-bidirectional LSTM hybrid model to semantically encode the AIDS prevention and control knowledge information to be pushed to obtain a semantic encoding vector of the AIDS prevention and control knowledge information as the semantic feature of the AIDS prevention and control knowledge information.

8. The method for recommending AIDS prevention and control knowledge education according to claim 7, characterized in that: Determining whether to recommend the AIDS prevention and control knowledge information to be pushed to the user based on the user information portrait-level updated semantic features and the AIDS prevention and control knowledge information semantic features includes: Calculating the KL divergence between the user information portrait-level updated semantic coding vector and the AIDS prevention and control knowledge information semantic coding vector; Based on the comparison between the KL divergence and a preset threshold, it is determined whether to recommend the AIDS prevention and control knowledge information to be pushed to the user.

Images