User interest intelligent recommendation method and system based on Internet

Abstract

The invention provides a user interest intelligent recommendation method and system based on the Internet, and relates to the technical field of the Internet, the system collects and integrates behavior data of a user on different platforms in real time to form a multi-source data set; by utilizing natural language processing and time sequence analysis technologies, the system deeply digs user interests and interest change trends, and identifies potential interest sudden increase trends; through a clustering algorithm, the system divides users with similar interest characteristics into dynamic user interest clusters, and dynamically adjusts parameters of the user interest clusters according to instantaneous data; the system predicts future demands of the user through a deep learning model, generates personalized recommendation content in combination with a time sequence analysis model and user interest cluster features, and pushes the personalized recommendation content to the user through multiple channels. According to the method and the device, the related content is accurately pushed under the condition that the user suddenly changed interest characteristics are improved.

Description

Technical Field

[0001] The present invention relates to the field of Internet technology, and in particular to an Internet-based user interest intelligent recommendation method and system. Background Art

[0002] With the rapid development of Internet technology and the widespread application of big data, personalized recommendation systems have become one of the key technologies to improve user experience and satisfaction; traditional recommendation systems mainly rely on collaborative filtering, content-based recommendations and rule-based recommendations; however, these methods have significant shortcomings in dealing with dynamic changes in user interests and integrating multi-source data.

[0003] Collaborative filtering methods rely on the similarity between users or items to recommend items that similar users like; however, this method is sensitive to data sparsity and cold start problems, and it is difficult to respond quickly to the addition of new users or new items; content-based recommendation methods rely on the characteristics of items and the historical preferences of users for recommendations, but often find it difficult to capture the complexity and diversity of user interests; in addition, rule-based recommendation methods are usually manually set and find it difficult to automatically adapt to changes in user interests. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the deficiencies in the prior art, the present invention provides an Internet-based method and system for intelligent recommendation of user interests, which solves the problem of how to accurately recommend information to users by capturing information on user interests and interest mutations.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method and system for intelligent recommendation of user interests based on the Internet, comprising:

[0008] Real-time multi-source data collection: Obtain user-generated behavior data and news data, collect and integrate instantaneous data from multiple data sources in real time, and combine user historical behavior data to form a multi-source data set;

[0009] Deep interest analysis: Through natural language processing technology combined with time series analysis models, we conduct in-depth analysis of instantaneous data in multi-source data sets, and identify potential interest surge trends by detecting changes in interest levels among user groups;

[0010] User interest cluster construction: Based on the sudden increase trend of user interests and multi-source data sets, dynamic user interest clusters are constructed, including interest weights, preference indexes and multiple dimensions, and the parameters of user interest clusters are dynamically adjusted according to instantaneous data;

[0011] Predictive recommendation generation: Use deep learning models to predict users' potential needs based on dynamic user interest clusters and interest surge trends, and generate and push recommended content based on user interests.

[0012] Preferably, real-time multi-source data collection collects instantaneous data from multiple data sources in real time, and the data sources include social media platforms, news websites, video platforms, forums, and e-commerce websites; obtains instantaneous data in various forms including user-generated content, comments, likes, shares, search queries, and click records through data crawlers and API interfaces; includes text information, as well as images, videos, and audio multimedia data; while collecting data, performs data cleaning and preprocessing to remove noise data and redundant information, and combines the collected instantaneous data with the user's historical behavior data to form a multi-source data set, and the user's historical behavior data includes the user's browsing history, search history, purchase history, and favorites; stores the multi-source data set in a database to describe the user's interest preferences and behavior patterns; the multi-source data set reflects the user's long-term interests and captures the user's interest changes in a short period of time; the multi-source data set uses natural language processing technology combined with a time series analysis model for in-depth interest analysis.

[0013] Preferably, deep interest analysis uses natural language processing technology to process instantaneous data in multi-source data sets and extract features, including text segmentation and annotation, sentiment analysis, and extraction of text from videos and audios; the extracted features are analyzed by a timing analysis model, which includes a time series model to analyze changes in user interest within a specific time period, where interest changes include the frequency and intensity of user attention to certain topics or content; when it is detected that user interest in a certain topic or content has increased significantly in a short period of time, it is identified as a potential interest surge trend; combined with recent news events and hot topics on social media, it is determined whether the user's interest surge trend is affected by external factors; through analysis, the dynamic changes in user interests are captured in a timely manner, so as to recommend content in real time.

[0014] Preferably, user interest cluster construction divides users with similar interests and behavioral characteristics into the same cluster through a clustering algorithm; the user's historical behavior data is used as the main basis for clustering to construct the initial user interest cluster, and the weight of the suddenly increased interest point is incorporated into the construction process of the user interest cluster through the interest sudden increase trend identified in step S2; after the user interest cluster is constructed, instantaneous data is collected in real time, and the parameters of each user interest cluster are dynamically adjusted. By adjusting the user's latest behavior on different platforms in the instantaneous data, such as clicks, likes, comments, etc., as well as current hot information and social events, the instantaneous data is monitored to identify the latest changes in user interests, and the latest changes are reflected in the user interest cluster. In terms of adjustment, for example, when a user in a user interest cluster shows a strong interest in an emerging topic, the system dynamically increases the weight of the emerging topic in the user interest cluster, reflecting the current interest tendency of the user interest cluster; the system adjusts the composition and boundaries of the user interest cluster in a timely manner according to the changes in instantaneous data; for example, when the interests of some users change significantly and are inconsistent with the characteristics of the user interest cluster to which they currently belong, the system re-evaluates the affiliation of these users and re-classifies them into user interest clusters that are more in line with their latest interests; through dynamic adjustment, the system can reflect changes in user interests in real time, thereby improving the accuracy and relevance of recommendations, and ensuring that user interest clusters always remain highly consistent with the user's real interests.

[0015] Preferably, the clustering algorithm is implemented by the DBSCAN clustering algorithm. The system collects historical behavior data of users from various platforms, including browsing records, purchase records, and click behaviors, and extracts feature vectors representing the interests and behavior patterns of each user. In the process of constructing the initial user interest cluster, the weights of the sudden increase points of interest are taken into consideration by analyzing the feature vectors and combining the sudden increase trend of interest identified in step S2, and the users are clustered using the DBSCAN algorithm. The DBSCAN algorithm defines the formation of user interest clusters by setting a radius parameter ε and a minimum number of samples MinPts. For each user, the DBSCAN clustering algorithm calculates the number of other users in its ε neighborhood. If the number of users in the neighborhood reaches M i nPts, the user is regarded as the core point, and the users directly connected by the core point form a user interest cluster. For users who do not belong to any user interest cluster and are not in the neighborhood of the core point, they are marked as noise points; after the initial user interest cluster is built, the user's instantaneous data is collected in real time, including the latest browsing history, search terms, likes and comments; by monitoring the instantaneous data, the latest changes in user interests are identified, and the latest changes are reflected in the adjustment of the user interest cluster; specifically, when the user's interest characteristics change, the system re-evaluates its position in the current user interest cluster and reallocates the user to a more appropriate user interest cluster; the collection and analysis of instantaneous data enables the parameters of the user interest cluster to be dynamically adjusted. The system adjusts the composition and boundaries of the cluster in a timely manner according to the changes in the user's instantaneous data to ensure that the user grouping accurately reflects its latest interest changes. Through the dynamic adjustment mechanism, the DBSCAN algorithm effectively adapts to changes in user interests during implementation and provides more accurate user clustering.

[0016] Preferably, predictive recommendation generation utilizes the constructed dynamic user interest clusters and the identified interest surge trends to extract user features and interest change trends in each user interest cluster; the user features and interest change trends are used as model inputs to train machine learning and deep learning models; the deep learning model learns the historical behavior patterns, interest change rules, and interest surge characteristics of users in the user interest clusters, thereby predicting the user's future potential needs; during the training process, the user behavior and feedback information in the user interest clusters are used to optimize the model parameters so that the deep learning model can capture the user's potential interests and needs; the system combines the latest trends and hot spot information in the external data source to further enhance the model's perception of sudden interests; after the deep learning model is trained, The system makes online predictions on new data and generates personalized recommendations in real time. The recommendations include news, videos, products, and social posts, covering all areas of user interest. When pushing personalized recommendations, the system considers multiple factors, such as recommendation diversity, user preferences, content freshness, and news contextual information, to ensure that the recommended content is in line with user interests and is attractive and timely. Recommended content is presented to users through multiple channels, such as homepage recommendations, notification push, and personalized advertising. The system continuously collects user feedback information and adjusts the model and recommendation strategy based on the feedback data. The prediction and recommendation method based on dynamic user interest clusters and interest surge trends can more accurately meet users' personalized needs and improve user satisfaction and loyalty.

[0017] Preferably, the deep learning model predicts user potential needs through the RNN model, uses the time series analysis model as the input layer of the RNN model, detects user interest changes through the time series analysis model, identifies the user's interest surge trend, and uses the dynamic user interest cluster as the input layer; the hidden layer is composed of LSTM units, which can remember long-term and short-term information and choose to update, save or output information through the input gate, forget gate and output gate mechanism; at each time step, the LSTM unit receives input data and integrates current and past instantaneous data in the hidden layer, and the output of the hidden layer is further processed by the fully connected layer to convert the high-dimensional LSTM The output is converted into a predicted value; using the softmax function, the output is converted into a probability distribution to represent the possibility of each category; output predictions of the user's future behavior or interests, such as the interest level in specific content or click-through rate prediction; by combining the time features extracted by the time series analysis model with the instantaneous data and dynamic user interest cluster features and inputting them into the LSTM network, LSTM generates prediction results by learning the relationship and time dependency between the features; in actual applications, when the instantaneous data collected in real time arrives, the deep learning model can process the input data in real time and predict the user's future behavior and interest changes, and generate personalized recommendation content based on the prediction results.

[0018] (III) Beneficial effects

[0019] The present invention provides an Internet-based user interest intelligent recommendation method and system, which has the following beneficial effects:

[0020] 1. The present invention can analyze and identify changes in user interests in real time, especially potential sudden increases in interest, so that the recommendation system can quickly respond to changes in user interests and provide the freshest and most relevant recommended content, greatly improving the timeliness of recommendations and user satisfaction.

[0021] 2. The present invention adopts a dynamic user interest cluster construction and adjustment mechanism. Through clustering algorithms and real-time monitoring of instantaneous data, the system updates the composition and parameters of user interest clusters in real time, so that user interest clusters can accurately reflect the user's latest interest tendencies. Even if the user's interests change significantly, the system can quickly make adjustments to ensure the high relevance of the recommended content.

[0022] 3. The system of the present invention collects and integrates multi-source data in real time, including social media, news websites, video platforms, etc., which not only enriches the data sources, but also enables the recommended content to cover multiple fields and forms; the system can provide recommended content based on user interests and diversity, meet the multi-level needs of users, and enhance user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] During the system construction process, the system collects and integrates instantaneous data from multiple data sources such as news websites and social media platforms in real time. Specifically, the system uses data crawler technology and API interfaces to obtain this data. Data crawler technology accesses different websites through automated scripts to extract content from the pages, such as news articles, comments, and user interactions. For social media platforms and other websites with stricter restrictions, the system obtains data through API interfaces. API interfaces usually provide authorized access rights and can obtain information such as user-posted content, interaction records, and metadata. For example, when a user browses multiple articles about sports events on a news website, the system's crawler program will capture the titles, texts, publication times, and related tags of these articles in real time. At the same time, the API interfaces on social media platforms will provide data such as user-shared content, comments, and likes. Through data crawler technology and API interfaces, the system can capture users' real-time behavior and interest dynamics.

[0026] The system collects users' historical behavior data, which includes the categories of articles they have read in the past, the keywords they have searched for, and the content they have liked or commented on. The system constructs a multi-source data set by integrating historical data and instantaneous data.

[0027] After acquiring the data, the system performs data preprocessing, which includes data cleaning, deduplication and standardization. The system uses natural language processing technology to remove irrelevant information, such as advertisements and noise data. Specific cleaning steps include removing stop words, performing stem extraction and normalizing characters. For structured data, the system uses standardized methods to uniformly process data from different sources to form a consistent data format. The system stores the processed data in a database to form a multi-source data set containing user behavior characteristics. This multi-source data set includes structured data tables and unstructured text data. The system uses feature extraction technology to convert text data into numerical feature vectors, such as TF-IDF and word vector representation. Feature vectors describe user interests, preferences and behavior patterns, providing a basis for subsequent recommendation algorithms.

[0028] The system uses natural language processing technology to conduct in-depth analysis of multi-source data sets to identify users' interests and sentiment tendencies; it breaks down the text into analyzable units, such as words or phrases, by performing word segmentation on the titles and content of articles that users have read; then, the system applies sentiment analysis to evaluate the user's emotional response to a topic or entity. For example, by analyzing users' comments on sports events, the system determines that the user has positive emotions towards a certain team; if the user frequently reads articles about football and shows a high degree of engagement in news about a certain football team, such as liking and making positive comments, the system marks these behaviors as evidence of a strong interest in football; sentiment analysis will analyze keywords and phrases in these comments, such as "wonderful" and "supportive", to determine that the user's emotional tendency is positive.

[0029] The system uses a time series analysis model to track changes in user interest over time. By analyzing users' attention to sports news in a specific time period, the system discovers periodic patterns in user interest or a surge in interest triggered by an event. For example, as an important football match day approaches, users' search and reading frequency for related news increases significantly. The system captures this change through time series analysis and identifies the increase in user interest. The system combines contextual information of the news, such as upcoming sporting events or hot topics on social media, to evaluate the reasons behind the surge in user interest, which helps the system understand users' interest changes more accurately and predict their future behavior patterns.

[0030] Through these analyses, the system builds dynamic user interest clusters, groups users with similar interests and emotional tendencies, and dynamically adjusts the parameters of user interest clusters based on instantaneous data and interest surge trends; for example, if the system finds that most users in a user interest cluster show great interest in an upcoming football match, the system will increase the weight of this topic in the user interest cluster to ensure that the recommended content matches the user's current hot interests.

[0031] Based on an in-depth analysis of user data, the system uses the DBSCAN clustering algorithm to classify users and form different user interest clusters based on interest characteristics; the DBSCAN algorithm processes outliers in user behavior data and data sets of different density areas; the system collects user historical behavior data from multiple data sources, including but not limited to user reading records, search queries, likes and comments, and these data are used to extract user interest characteristics; the system generates a feature vector for each user, which contains the user's interest measure in a certain topic, such as reading frequency, positive scores of sentiment analysis results, etc.; the DBSCAN algorithm defines the formation of user interest clusters by setting two key parameters, namely the radius parameter ε and the minimum number of samples MinPts; the system sets ε to a reasonable range that can cover the user's interest characteristics in a certain topic, and MinPts is set to the minimum number of users sufficient to identify an interest group.

[0032] When the DBSCAN clustering algorithm is running, for each user, the system DBSCAN clustering algorithm calculates the number of other users in its ε neighborhood. If the number of users in a user's neighborhood reaches or exceeds Mi nPts, the user is regarded as a core point, and together with all users in its neighborhood, a user interest cluster is formed; through the connection between core points, user interest clusters are gradually formed; in this process, those users who do not belong to any user interest cluster are marked as noise points, and users who do not belong to any user interest cluster have low interest in the topic or show inconsistent behavior patterns; for example, as a football match progresses, the user's focus may change, such as shifting from pre-match prediction to post-match commentary, and the system captures these changes by monitoring the user's latest behavior data in real time; when the user's interest point shifts, the system re-evaluates the user's feature vector and adjusts the composition of the user interest cluster according to the new behavior data; for example, if the user shows a higher interest in the commentary content after the game, the system will increase the weight of these behaviors in the feature vector, which may reclassify the user into a new user interest cluster that pays more attention to game analysis and commentary; in this way, the system can dynamically reflect changes in user interests and regularly update clustering results to ensure that the composition of user interest clusters is always consistent with the user's latest interests.

[0033] After completing the construction of user interest clusters and the identification of interest spikes, the system further uses this information to generate predictive recommendations. By extracting user features and interest change trends in each user interest cluster, these user features and trends are then used as input to the deep learning model. The deep learning model, especially the recurrent neural network (RNN) model, predicts the user's future interests and needs by learning the user's historical behavior patterns, interest change patterns, and interest spike characteristics. During the model training phase, the system not only relies on the user's historical data, but also combines the latest trends and hot information in external data sources to enhance the model's perception of the user's sudden interests. In this way, the RNN model can capture the user's potential interests and needs while maintaining sensitivity to news hotspots. When a user interest cluster shows a strong interest in a certain content, the system combines this interest spike trend with the current news hotspots and optimizes the model parameters to more accurately predict the user's interest in the relevant news hotspot content. After the model training is completed, the system performs online predictions on new data and generates personalized recommendations in real time. These recommendations are not limited to news articles, but also include videos, products, social posts and other forms, covering all areas of user interest.

[0034] The RNN model is used to predict user potential needs, and the time series analysis model is used as the input layer of the RNN model. The user interest changes detected by the time series analysis model are used to identify the user's interest surge trend, and the dynamic user interest cluster is also used as the input layer; the hidden layer is composed of LSTM units, which can remember long-term and short-term information and choose to update, save or output information through the input gate, forget gate and output gate mechanism; at each time step, the LSTM unit receives input data and integrates current and past instantaneous data in the hidden layer. The output of the hidden layer is further processed by the fully connected layer to convert the high-dimensional LSTM output into The model converts the output into a predicted value; uses the softmax function to convert the output into a probability distribution to represent the possibility of each category; outputs predictions of the user's future behavior or interests, such as the interest level in a specific content or click-through rate prediction; combines the time features extracted by the time series analysis model with the instantaneous data and the dynamic user interest cluster features and inputs them into the LSTM network. The LSTM generates prediction results by learning the relationship and time dependency between the features; in actual applications, when the instantaneous data collected in real time arrives, the model can process the input data in real time and predict the user's future behavior and interest changes, and generate personalized recommendation content based on the prediction results.

[0035] In actual applications, the system can process the collected instantaneous data in real time, and predict the user's future behavior and interest changes based on the instantaneous data; this real-time prediction capability enables the system to quickly respond to changes in user interests and generate personalized recommendations based on user interests based on the prediction results; the system presents recommended content to users through multiple channels, such as homepage recommendations, notification push, and personalized advertising; at the same time, the system continuously collects user feedback information, and adjusts the model and recommendation strategy based on this feedback to ensure that the recommended content always accurately meets the user's personalized needs and improves user satisfaction; through this prediction and recommendation method based on dynamic user interest clusters and interest surge trends, the system provides users with a highly personalized and dynamically updated recommendation experience.

[0036] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A user interest intelligent recommendation method based on the Internet, characterized in that: The following steps are involved: S1. Collect and integrate instantaneous data from multiple data sources in real time, and combine it with user historical behavior data to form a multi-source data set; S2. Through natural language processing technology combined with time series analysis models, we conduct in-depth analysis of instantaneous data in multi-source data sets, and identify potential interest surge trends by analyzing changes in user interest; S3. Based on the sudden increase trend of user interests and multi-source data sets, dynamic user interest clusters are constructed, and the parameters of the user interest clusters are dynamically adjusted according to instantaneous data; S4. Use deep learning models to predict users’ potential needs based on dynamic user interest clusters and interest surge trends, and generate and push recommended content based on user interests.

2. The method for intelligently recommending user interests based on the Internet according to claim 1, characterized in that: The S1 step collects instantaneous data from multiple data sources in real time, and obtains instantaneous data generated by users through data crawlers and API interfaces; while collecting instantaneous data, data cleaning and preprocessing are performed to remove noise data and redundant information, and the collected instantaneous data are combined with the user's historical behavior data to form a multi-source data set, and the multi-source data set is stored in a database to describe the user's interest preferences and behavior patterns; Multi-source datasets reflect users’ long-term interests and capture the changing trends of users’ interests in a short period of time; The multi-source data set is deeply analyzed through natural language processing technology combined with a time series analysis model to identify changes in user interests; the instantaneous data includes user-generated content, comments, likes, shares, search queries, and click records.

3. The method for intelligently recommending user interests based on the Internet according to claim 2, characterized in that: The step S2 extracts features from instantaneous data in the multi-source data set by natural language processing technology, wherein the features are part-of-speech features, sentiment features, image and audio features, and user behavior features; The extracted features are analyzed through a time series analysis model to analyze the changes in users' interest within a certain period of time, where the change in interest refers to the frequency and intensity of users' attention to certain topics or content. When it is detected that the user's interest in a certain topic or content has increased significantly in a short period of time, it is identified as a potential interest surge trend. Combined with recent news events and hot topics on social media, it is determined whether the user's interest surge trend is affected by external factors, so as to capture the dynamic changes of user interests in a timely manner.

4. The method for intelligently recommending user interests based on the Internet according to claim 3, characterized in that: In step S3, users with similar interests and behavioral characteristics in multi-source data sets are divided into the same cluster through a clustering algorithm; the user's historical behavior data is used as the main basis for clustering to construct the initial user interest cluster, and the weight of the sudden increase in interest points is incorporated into the construction process of the user interest cluster through the interest surge trend identified in step S2; after the user interest cluster is constructed, instantaneous data is collected in real time, and the parameters of each user interest cluster are dynamically adjusted; the system identifies the latest changes in user interests by monitoring instantaneous data, and reflects the latest changes in the adjustment of user interest clusters; according to the changes in instantaneous data, the parameters of the user interest cluster are adjusted in a timely manner, and the parameters include composition and boundaries; the changes in user interests are reflected by dynamically adjusting the parameters of the user interest cluster.

5. The method for intelligently recommending user interests based on the Internet according to claim 4, characterized in that: The clustering algorithm adopts the DBSCAN clustering algorithm; historical behavior data of users, including browsing records, purchase records and click behaviors, are collected from various platforms, and feature vectors representing the interests and behavior patterns of each user are extracted; in the process of constructing the initial user interest cluster, the weights of the sudden increase interest points are taken into consideration by analyzing the feature vectors and combining the sudden increase trend of interest identified in step S2, and the users are clustered using the DBSCAN algorithm; the DBSCAN algorithm defines the formation of user interest clusters by setting the radius parameter ε and the minimum number of samples MinPts; for each user, the DBSCAN clustering algorithm calculates the number of other users in its ε neighborhood, and if the number of users in the neighborhood reaches MinPts, the user is regarded as a core point, and users directly connected through the core point form a user interest cluster, and users who do not belong to any user interest cluster and are not in the neighborhood of the core point are marked as noise points; after the initial user interest cluster is constructed, the instantaneous data of the user is collected in real time; by viewing the instantaneous data, the latest changes in the user's interests are identified, and the latest changes are reflected in the adjustment of the user interest cluster.

6. The method for intelligently recommending user interests based on the Internet according to claim 4, characterized in that: The S4 step utilizes the construction of dynamic user interest clusters and the identification of interest surge trends to extract user characteristics and interest change trends in each user interest cluster; uses the user characteristics and interest change trends as model input to train the deep learning model; uses the deep learning model to learn the user's historical behavior patterns, interest change rules, and interest surge characteristics from the user interest cluster, thereby predicting the user's future potential needs; during the training process, optimizes the model parameters using user behavior and feedback information in historical data so that the model can capture the user's potential interests and needs; combines the latest trends and hot spot information in external data sources to further enhance the model's perception of sudden interests; after the model training is completed, new data is predicted online to generate personalized recommendation content in real time.

7. The method for intelligently recommending user interests based on the Internet according to claim 6, characterized in that: The deep learning model predicts user potential needs through the RNN model, uses the time series analysis model as the input layer of the RNN model, identifies the sudden increase trend of user interest through the changes in user interests detected by the time series analysis model, and also uses the dynamic user interest cluster as the input layer; the hidden layer is composed of LSTM units, which can remember long-term and short-term information and choose to update, save or output information through the input gate, forget gate and output gate mechanism; at each time step, the LSTM unit receives input data and integrates current and past instantaneous data in the hidden layer, and the output of the hidden layer is further processed by the fully connected layer to convert the high-dimensional LSTM output into a predicted value; using the softmax function, the output is converted into a probability distribution to represent the possibility of each category, and the prediction of the user's future behavior or interest is output; by combining the time features extracted by the time series analysis model with the instantaneous data and the dynamic user interest cluster features and inputting them into the LSTM network, the LSTM generates prediction results by learning the relationship and time dependency between the features.

8. An Internet-based user interest intelligent recommendation system, characterized by: include, Real-time multi-source data collection: Obtain user-generated behavior data and news data, collect and integrate instantaneous data from multiple data sources in real time, and combine user historical behavior data to form a multi-source data set; Deep interest analysis: Through natural language processing technology combined with time series analysis models, we conduct in-depth analysis of instantaneous data in multi-source data sets, and identify potential interest surge trends by detecting changes in interest levels among user groups; User interest cluster construction: Based on the sudden increase trend of user interests and multi-source data sets, dynamic user interest clusters are constructed, including interest weights, preference indexes and multiple dimensions, and the parameters of user interest clusters are dynamically adjusted according to instantaneous data; Predictive recommendation generation: Use deep learning models to predict users' potential needs based on dynamic user interest clusters and interest surge trends, and generate and push recommended content based on user interests.

Images