Intelligent information pushing system and method based on user behavior analysis

Abstract

The invention discloses an intelligent information pushing system and method based on user behavior analysis, and the system can generate highly personalized content recommendation on the basis of obtaining the current behavior of a user and predicting the future behavior through the dynamic analysis capability of a real-time behavior prediction module, so that the recommended content meets the instant demand of the user, and the user experience is improved. The potential requirements of the user can be predicted, so that the recommendation accuracy and the user experience are improved; through the mode of combining real-time and prediction, the content recommendation and generation module can comprehensively analyze interest points and behavior modes of the user and effectively generate personalized content highly related to the current situation and future expectation of the user, and the information pushing module can push the personalized content to the user through optimizing the pushing path and time. The personalized content is ensured to be accurately sent to the user at the most appropriate time in the most appropriate mode, finally, the behavior model is further optimized through the feedback learning module, a closed-loop feedback mechanism is formed, and the intelligence of the system and the satisfaction degree of the user are continuously improved.

Description

Technical Field

[0001] The present invention belongs to the field of information push, and in particular, relates to an intelligent information push system and method based on user behavior analysis. Background Art

[0002] With the rapid development of mobile Internet and big data technology, personalized content recommendation systems have been widely used in e-commerce, social media, content distribution and other fields. Traditional recommendation systems usually rely on collaborative filtering, content analysis and other technologies. By analyzing the user's historical behavior data, the user's interest preference model is generated, and based on this, the content that the user may be interested in is recommended to the user. However, traditional recommendation systems have the following main limitations:

[0003] First, most traditional recommendation systems rely on static historical behavior data, which makes it difficult to capture users’ immediate needs and context changes in real time. The limitations of this static analysis often make recommendation systems inadequate in responding to users’ dynamic needs and unable to provide content that is highly relevant to the user’s current context.

[0004] Secondly, although some recommendation systems in the prior art use rule-based or simple time series analysis methods to make predictions to a certain extent, these methods often perform poorly when dealing with complex user behavior patterns, and the prediction accuracy is not high, making it difficult to accurately predict users' potential needs. In addition, the existing systems usually use fixed push strategies when pushing content, and fail to fully consider factors such as the user's device status, location information, and network conditions, resulting in poor effectiveness and timeliness of the pushed content when it reaches the user, affecting the user experience.

[0005] In view of this, the present invention is proposed. Summary of the invention

[0006] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide an intelligent information push system and method based on user behavior analysis, which solves the problems raised in the above-mentioned background technology.

[0007] In order to solve the above technical problems, the basic concept of the technical solution adopted by the present invention is:

[0008] An intelligent information push system based on user behavior analysis includes: a user behavior data collection module for collecting behavior data from multiple devices and channels of a user in real time;

[0009] The historical behavior analysis module is used to receive and process the data from the user behavior data collection module, analyze the user's historical behavior data based on the DBSCAN density clustering algorithm and association rule mining technology, and identify the user's key points of interest POI and behavior patterns;

[0010] The real-time behavior prediction module is connected to the historical behavior analysis module. It uses the results of historical behavior analysis, long short-term memory network (LSTM) and real-time data stream processing technology to predict the user's current and future behaviors and generate a dynamic behavior model of the user;

[0011] Content recommendation and generation module: The content recommendation and generation module generates personalized content related to the user's current needs and future expectations by analyzing the user's interests and behavior patterns based on the dynamic behavior model generated by the real-time behavior prediction module and the user's key interests;

[0012] The information push module is connected to the content recommendation and generation module. It optimizes the push path and time by combining the user's location information, device status and network conditions. When the user enters the predefined point of interest, the generated personalized content is pushed to the user through multiple channels;

[0013] Feedback learning module: The feedback learning module is connected to the information push module and is used to collect user feedback on the pushed content and send the feedback information back to the historical behavior analysis module and the real-time behavior prediction module. The user's behavior model is continuously updated through machine learning algorithms to optimize future content recommendation and push strategies.

[0014] The data security and privacy protection module runs through all modules of the entire system. The data security and privacy protection module protects the privacy of user data through differential privacy and encryption technology.

[0015] Optionally, in the process of DBSCAN cluster analysis, first, the neighborhood density of each data point is calculated and the core points in the data are identified based on the density estimation results. Then, the core points and the points in their neighborhood are aggregated together to form density clusters, thereby identifying the key points of interest POIs of users at different time periods and locations. The expression is: Finally, the points that do not belong to any cluster are marked as noise points, where POI represents the i-th key interest point, which is a point set, where X represents the entire data set, and N ∈ (x) represents the set of neighboring points of point x in the ∈ neighborhood, δ represents the density threshold, which is usually set to a specific value to distinguish core points from non-core points, |N ∈ (x)| represents the number of neighboring points of point x in the ∈ neighborhood, and |X| represents the total number of points in the entire data set;

[0016] Then, the user's behavior frequency at each point of interest POI is analyzed based on the obtained key points of interest POI Identify high-frequency behavior patterns and use the Apriori algorithm to extract the behavioral correlation between users at different points of interest; then, screen the generated association rules according to the confidence and support of the behavior pattern, retain the rules with high confidence and high support to form the final user behavior pattern, and finally output the key points of interest POIs and behavior patterns identified by DBSCAN clustering analysis and association rule mining, where: represents the user's behavior frequency at the i-th key interest point, T represents the total length of observation time, and N i represents the number of user behavior events at key interest point i, w j represents the weight of the jth behavior event, σ j represents the normalized frequency of the jth behavioral event, Represents the total weight of all behaviors at the key interest point i.

[0017] Optionally, the steps of generating a user's dynamic behavior model are:

[0018] Obtain the user's historical behavior data and real-time data stream from the user behavior data collection module, then organize the user's historical behavior data and real-time behavior data into sequence input in chronological order, and convert the original data into a feature vector suitable for model processing as the input of the LSTM model;

[0019] When building an LSTM model, you need to first set the structure of the input layer, hidden layer, and output layer and use historical behavior data to train the LSTM model;

[0020] The user behavior data received in real time is input into the trained LSTM model, and the model updates its internal state based on the input data. Finally, the LSTM model combines historical behavior data with current real-time data to predict the user's possible future behavior and generate a dynamic behavior model of the user.

[0021] Optionally, when designing the structure of the LSTM model, first, set the size of the feature vector in the input layer corresponding to each time step and set the number of time steps processed by the LSTM model in a time series; set the hidden layer and set a 3-layer LSTM structure in the hidden layer, and use different functions in the output layer according to the situation, where if the output is a classification problem, use the softmax activation function; if it is a regression problem, use the linear activation function.

[0022] When generating personalized content related to the user's current needs and future expectations, first, obtain the user's dynamic behavior model from the real-time behavior prediction module and obtain the user's key points of interest POIs and related behavior patterns from the historical behavior analysis module. By analyzing the user's dynamic behavior model, identify the user's current needs, and then analyze the user's possible future needs in combination with the user's future behavior predicted by the dynamic behavior model. The expression is: Among them, D future : represents the future needs of users. m: represents the number of features involved in predicting future user behavior. j It represents the weight coefficient related to the jth future behavior prediction feature, reflecting the influence of the feature on future demand. j represents the predicted feature vector in the dynamic behavior model associated with the jth feature, t future represents the predicted future time point, ψ(h j， t future ) represents the feature h j At future time t future The response function under the condition of outputting the predicted impact value related to future demand, θ represents the weight coefficient of historical behavior pattern on future demand prediction, ω(H, P j , Δt) represents the historical behavior pattern H and key interest points P j The impact function on future demand under time change Δt is used to predict the impact of historical behavior on future demand.

[0023] Optionally, when future needs are obtained, it is also necessary to match the user's current needs and future expectations with the content library in the system and personalize the matching content according to the user's needs, interests and behavior patterns. Then, the generated personalized content is prioritized according to the relevance of the content to the user's needs, the urgency and the user's preferences. Finally, different types of personalized content are integrated into one or more information packages.

[0024] Optionally, when the generated personalized content is pushed to the user, the user's current geographic location is first obtained in real time through GPS, Wi-Fi positioning, and Bluetooth beacons and compared with a pre-defined range of key points of interest. When the user's current location information matches a range of points of interest, it is recognized that the user has entered the point of interest area. Then, based on the user's current location information and network conditions, the optimal data transmission path is selected and the most appropriate push time is selected in combination with the user's device status and behavior patterns. Finally, when the user enters the predefined point of interest range and the device status and network conditions are suitable, the system immediately pushes the personalized content to the user through the selected channel.

[0025] Optionally, when continuously updating the user's behavior model through machine learning algorithms and optimizing future content recommendation and push strategies, first, monitor and record the user's interactive behavior with the pushed content in real time and classify the collected feedback information by type, then transmit the processed user feedback data back to the historical behavior analysis module, update the user's historical behavior record to reflect the user's latest behavior changes and interest point adjustments, and simultaneously transmit the feedback data to the real-time behavior prediction module to adjust the prediction results of the dynamic behavior model. Subsequently, based on the updated user behavior data set, use the machine learning algorithm to retrain and optimize the user behavior model. During the model training process, automatically adjust the model parameters to improve the prediction accuracy and relevance of the content recommendation. Finally, based on the updated behavior model, re-optimize the content recommendation strategy and optimize the content push path based on the latest model results.

[0026] An intelligent information push method based on user behavior analysis comprises the following steps:

[0027] Collect behavioral data from users across multiple devices and channels in real time;

[0028] Receive and process user behavior data, analyze the user's historical behavior data based on the DBSCAN density clustering algorithm and association rule mining technology, and identify the user's key points of interest and behavior patterns;

[0029] Using the results of historical behavior analysis, long short-term memory network (LSTM) and real-time data stream processing technology are used to predict the user's current and future behavior and generate a dynamic behavior model of the user;

[0030] Based on the dynamic behavior model generated by the real-time behavior prediction module and the user's key interests, the module generates personalized content related to the user's current needs and future expectations by analyzing the user's interests and behavior patterns;

[0031] By combining the user's location information, device status and network conditions, the push path and time are optimized. When the user enters the predefined point of interest, the generated personalized content is pushed to the user through multiple channels;

[0032] Collect user feedback on pushed content and send it back to the historical behavior analysis module and real-time behavior prediction module. Use machine learning algorithms to continuously update the user behavior model and optimize future content recommendations and push strategies.

[0033] Differential privacy and encryption technologies are used to protect the privacy of user data during both data transmission and reception.

[0034] After adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art. Of course, any product implementing the present invention does not necessarily need to achieve all of the following advantages at the same time:

[0035] Through the dynamic analysis capability of the real-time behavior prediction module, the system of the present invention can generate highly personalized content recommendations based on obtaining the user's current behavior and predicting future behavior, so that the recommended content not only meets the user's immediate needs, but also can foresee the user's potential needs, thereby improving the accuracy of the recommendation and user experience; through this combination of real-time and prediction, the content recommendation and generation module can comprehensively analyze the user's interests and behavior patterns, and effectively generate personalized content that is highly relevant to the user's current situation and future expectations, while the information push module optimizes the push path and time to ensure that these personalized contents are accurately delivered to the user at the most appropriate time and in the most appropriate manner, and finally further optimizes the behavior model through the feedback learning module to form a closed-loop feedback mechanism, and continuously improve the intelligence of the system and user satisfaction.

[0036] The specific implementation modes of the present invention are further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The drawings described below are only some embodiments. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0038] Figure 1 A block diagram of an intelligent information push system based on user behavior analysis according to an embodiment of the present invention;

[0039] Figure 2 The figure is a flow chart of an information push method according to an embodiment of the present invention.

[0040] It should be noted that these drawings and textual descriptions are not intended to limit the conceptual scope of the present invention in any way, but are intended to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0041] The present invention will now be described in further detail with reference to the accompanying drawings.

[0042] See also Figure 1 As shown, in this embodiment, an intelligent information push system based on user behavior analysis is provided, including a user behavior data collection module, which is used to collect behavior data in real time from multiple devices and channels of users (including mobile devices, location service systems, application usage records, social network activities, etc.), and the data includes but is not limited to the user's location information, operation records, browsing history, interaction frequency and related environmental data;

[0043] The historical behavior analysis module is used to receive and process data from the user behavior data collection module, analyze the user's historical behavior data based on the DBSCAN density clustering algorithm and association rule mining technology, identify the user's key points of interest and behavior patterns, and provide basic data for subsequent predictions and recommendations;

[0044] The real-time behavior prediction module is connected to the historical behavior analysis module. It uses the results of historical behavior analysis, long short-term memory network (LSTM) and real-time data stream processing technology to predict the user's current and future behaviors and generate a dynamic behavior model of the user;

[0045] Content recommendation and generation module: The content recommendation and generation module generates personalized content related to the user's current needs and future expectations by analyzing the user's interests and behavior patterns based on the dynamic behavior model generated by the real-time behavior prediction module and the user's key interests, including but not limited to text, images, videos and location-related service information;

[0046] The information push module is connected to the content recommendation and generation module. It optimizes the push path and time by combining the user's location information, device status and network conditions. When the user enters the predefined point of interest, the generated personalized content is pushed to the user through multiple channels (such as application content push, SMS, email, location service push);

[0047] Feedback learning module: The feedback learning module is connected to the information push module and is used to collect user feedback on the pushed content. The feedback information is then fed back to the historical behavior analysis module and the real-time behavior prediction module. The user behavior model is continuously updated through machine learning algorithms to optimize future content recommendation and push strategies, thereby improving the system's intelligence and user satisfaction.

[0048] The data security and privacy protection module runs through all modules of the entire system. The data security and privacy protection module protects the privacy of user data through differential privacy and encryption technology, ensures the security of data during collection, transmission, analysis and processing, and prevents unauthorized access and data leakage. In the financial field, the security of customer data is of vital importance. Through differential privacy and encryption technology, the system can effectively prevent data leakage and unauthorized access during the process of collecting, transmitting, analyzing and processing customer financial data. This not only meets the strict requirements of financial supervision, but also enhances customers' trust in financial institutions and promotes the establishment of deeper customer relationships.

[0049] Through the dynamic analysis capability of the real-time behavior prediction module, the system of the present invention can generate highly personalized content recommendations based on obtaining the user's current behavior and predicting future behavior, so that the recommended content not only meets the user's immediate needs, but also can foresee the user's potential needs, thereby improving the accuracy of the recommendation and user experience; through this combination of real-time and prediction, the content recommendation and generation module can comprehensively analyze the user's interests and behavior patterns, and effectively generate personalized content that is highly relevant to the user's current situation and future expectations, while the information push module optimizes the push path and time to ensure that these personalized contents are accurately delivered to the user at the most appropriate time and in the most appropriate manner, and finally further optimizes the behavior model through the feedback learning module to form a closed-loop feedback mechanism, and continuously improve the intelligence of the system and user satisfaction.

[0050] In the process of DBSCAN cluster analysis in this embodiment, first, the neighborhood density of each data point is calculated and the core point in the data is identified based on the density estimation result. Then, the core point and the points in its neighborhood are aggregated together to form a density cluster, thereby identifying the key points of interest POI of the user at different time periods and locations. The expression is: Finally, the points that do not belong to any cluster are marked as noise points, where POI represents the i-th key interest point, which is a point set, where X represents the entire data set, and N ∈ (x) represents the set of neighboring points of point x in the ∈ neighborhood, δ represents the density threshold, which is usually set to a specific value to distinguish core points from non-core points, |N ∈ (x) represents the number of neighboring points of point x in the ∈ neighborhood, and |X| represents the total number of points in the entire data set;

[0051] Then, the user's behavior frequency at each point of interest POI is analyzed based on the obtained key points of interest POI Identify high-frequency behavior patterns and use the Apriori algorithm to extract the behavioral correlation between users at different points of interest; then, screen the generated association rules according to the confidence and support of the behavior pattern, retain the rules with high confidence and high support to form the final user behavior pattern, and finally output the key points of interest POIs and behavior patterns identified by DBSCAN clustering analysis and association rule mining, where: represents the user's behavior frequency at the i-th key interest point, T represents the total length of observation time, and N i represents the number of user behavior events at key interest point i, w j represents the weight of the jth behavior event, σ j represents the normalized frequency of the jth behavioral event, represents the total weight of all behaviors at this key point of interest i. Through DBSCAN clustering analysis technology, financial institutions can accurately identify customers' key behavior patterns and points of interest (POIs), such as frequently visited bank branches, common transaction locations, or specific investment behaviors. Combined with association rule mining technology, the system can discover the potential correlation between customers' different financial behaviors, thereby providing each customer with more personalized financial product recommendations and investment portfolio suggestions. This technology not only improves customer satisfaction, but also helps financial institutions explore new market opportunities.

[0052] In this embodiment, the steps of generating the user's dynamic behavior model are:

[0053] Obtain the user's historical behavior data and real-time data stream from the user behavior data collection module, then organize the user's historical behavior data and real-time behavior data into sequence input in chronological order, and convert the original data into a feature vector suitable for model processing as the input of the LSTM model;

[0054] When building an LSTM model, you need to first set the structure of the input layer, hidden layer, and output layer and use historical behavior data to train the LSTM model;

[0055] The user behavior data received in real time is input into the trained LSTM model, and the model will update its internal state based on the input data. Finally, the LSTM model combines historical behavior data with current real-time data to predict the user's possible future behavior and generate a dynamic behavior model of the user. In the financial field, user behavior patterns often change dynamically. By analyzing historical financial transaction data and real-time behavior data, the LSTM model can effectively predict customers' future financial needs, such as loan needs, investment intentions, or account liquidity needs. Such predictive capabilities enable financial institutions to identify customers' potential needs in advance and provide more timely and appropriate financial products or services, thereby improving customer experience and enhancing customer loyalty.

[0056] In the present embodiment, when designing the structure of the LSTM model, first, the size of the feature vector in the input layer corresponding to each time step is set and the number of time steps processed by the LSTM model in a time series is set; the hidden layer is set and a 3-layer LSTM structure is set in the hidden layer, and different functions are used in the output layer according to the situation, wherein a softmax activation function is used if the output is a classification problem; and a linear activation function is used if the output is a regression problem.

[0057] In this embodiment, when generating personalized content related to the user's current needs and future expectations, first, the user's dynamic behavior model is obtained from the real-time behavior prediction module, and the user's key points of interest POIs and related behavior patterns are obtained from the historical behavior analysis module. By analyzing the user's dynamic behavior model, the user's current needs are identified. Then, combined with the user's future behavior predicted by the dynamic behavior model, the user's possible future needs are analyzed, and the expression is: Among them, D future represents the future needs of users. m represents the number of features involved in predicting future user behavior. j It represents the weight coefficient related to the jth future behavior prediction feature, reflecting the influence of the feature on future demand. j Represents the predicted feature vector (e.g. expected position, future activity, etc.) in the dynamic behavior model associated with the jth feature. future represents the predicted future time point. ψ(h j , t future ) represents the feature h j At future time t future The response function under the condition of outputting the predicted impact value related to future demand. θ represents the weight coefficient of historical behavior pattern on future demand prediction. ω(H,P j , Δt) represents the historical behavior pattern H and key interest points P j The impact function on future demand under time change Δt is used to predict the impact of historical behavior on future demand. By combining the future characteristics predicted by the dynamic behavior model (such as future location, expected activities, etc.) and historical behavior patterns, the user's future demand is predicted. Function ψ(h j , t future ) and ω(H, P j , Δt) is used to quantify the contribution of each feature to future demand. By combining real-time behavior prediction models with historical behavior analysis, the content recommendation module is able to generate highly personalized financial product recommendations. For example, based on the customer's risk preferences, historical investment behavior, and current market trends, the system can recommend the most suitable financial products or insurance plans. This not only helps customers make more informed financial decisions, but also improves the conversion rate of financial products and customers' trust in institutions.

[0058] In this embodiment, when future needs are obtained, it is also necessary to match the user's current needs and future expectations with the content library in the system and personalize the matched content according to the user's needs, interests and behavior patterns. Then, the generated personalized content is prioritized according to the relevance of the content to the user's needs, the urgency and the user's preferences. Finally, different types of personalized content are integrated into one or more information packages.

[0059] In this embodiment, when the generated personalized content is pushed to the user, the user's current geographic location is first obtained in real time through GPS, Wi-Fi positioning, and Bluetooth beacons, and compared with the pre-defined key points of interest range. When the user's current location information matches a certain point of interest range, it is recognized that the user has entered the point of interest area. Then, according to the user's current location information and network conditions, the optimal data transmission path is selected and the most appropriate push time is selected in combination with the user's device status and behavior pattern. Finally, when the user enters the pre-defined point of interest range and the device status and network conditions are suitable, the system immediately pushes the personalized content to the user through the selected channel. By optimizing the push path and time, it is ensured that personalized financial information and suggestions are accurately pushed to customers when they need it most. For example, before the customer is about to reach the credit card bill payment date, the system can intelligently push repayment reminders or installment payment suggestions; when the customer's location is close to a bank branch, push financial promotions related to the customer. This precise content push not only improves customer interaction rate, but also effectively reduces customers' financial pressure and overdue risks.

[0060] In this embodiment, first, the interactive behavior of users and pushed content is monitored and recorded in real time and the collected feedback information is classified by type. Then, the processed user feedback data is sent back to the historical behavior analysis module, the user's historical behavior record is updated to reflect the user's latest behavior changes and interest point adjustments, and the feedback data is simultaneously transmitted to the real-time behavior prediction module to adjust the prediction results of the dynamic behavior model. Subsequently, based on the updated user behavior data set, the user behavior model is retrained and optimized using a machine learning algorithm. During the model training process, the parameters of the model are automatically adjusted to improve the accuracy of the prediction and the relevance of the content recommendation. Finally, based on the updated behavior model, the content recommendation strategy is re-optimized and the content push path is optimized based on the latest model results. The interactive behavior includes user clicks on pushed content, opens links, purchases recommended products, shares content, ignores push, cancels subscriptions, etc. The sensitivity of the financial industry to customer needs requires the system to respond quickly to customer feedback. Through the feedback learning module, the system can continuously optimize the financial product recommendation strategy and behavior prediction model based on customer feedback on pushed financial products (such as clicks, purchases, ignores, etc.). This dynamic adjustment capability ensures the continued relevance and effectiveness of recommended content, improves customer satisfaction and the overall effect of financial services.

[0061] See also Figure 2 As shown, in this embodiment, a method for intelligent information push based on user behavior analysis is provided, comprising the following steps:

[0062] Collect behavioral data from users across multiple devices and channels in real time;

[0063] Receive and process user behavior data, analyze the user's historical behavior data based on the DBSCAN density clustering algorithm and association rule mining technology, and identify the user's key points of interest and behavior patterns;

[0064] Using the results of historical behavior analysis, long short-term memory network (LSTM) and real-time data stream processing technology are used to predict the user's current and future behavior and generate a dynamic behavior model of the user;

[0065] Based on the dynamic behavior model generated by the real-time behavior prediction module and the user's key interests, the module generates personalized content related to the user's current needs and future expectations by analyzing the user's interests and behavior patterns;

[0066] By combining the user's location information, device status and network conditions, the push path and time are optimized. When the user enters the predefined point of interest, the generated personalized content is pushed to the user through multiple channels;

[0067] Collect user feedback on pushed content and send it back to the historical behavior analysis module and real-time behavior prediction module. Use machine learning algorithms to continuously update the user behavior model and optimize future content recommendations and push strategies.

[0068] Differential privacy and encryption technologies are used to protect the privacy of user data during both data transmission and reception.

[0069] The present invention is not limited to the above-mentioned embodiments. Anyone should be aware that any structural changes made under the enlightenment of the present invention, and any technical solutions that are the same or similar to the present invention, fall within the protection scope of the present invention. The technology, shape, and structural parts not described in detail in the present invention are all well-known technologies.

Claims

1. An intelligent information push system based on user behavior analysis, characterized in that: include: User behavior data collection module, used to collect behavior data from multiple devices and channels of users in real time; The historical behavior analysis module is used to receive and process the data from the user behavior data collection module, analyze the user's historical behavior data based on the DBSCAN density clustering algorithm and association rule mining technology, and identify the user's key points of interest POI and behavior patterns; The real-time behavior prediction module is connected with the historical behavior analysis module. It uses the results of historical behavior analysis, long short-term memory network and real-time data stream processing technology to predict the user's current and future behaviors and generate a dynamic behavior model of the user; Content recommendation and generation module: The content recommendation and generation module generates personalized content related to the user's current needs and future expectations by analyzing the user's interests and behavior patterns based on the dynamic behavior model generated by the real-time behavior prediction module and the user's key interests; The information push module is connected to the content recommendation and generation module. It optimizes the push path and time by combining the user's location information, device status and network conditions. When the user enters the predefined point of interest, the generated personalized content is pushed to the user through multiple channels; Feedback learning module: The feedback learning module is connected to the information push module and is used to collect user feedback on the pushed content. The feedback information is then fed back to the historical behavior analysis module and the real-time behavior prediction module. The user behavior model is continuously updated through machine learning algorithms to optimize future content recommendation and push strategies. The data security and privacy protection module runs through all modules of the entire system. The data security and privacy protection module protects the privacy of user data through differential privacy and encryption technology.

2. The intelligent information push system based on user behavior analysis according to claim 1, characterized in that: In the process of DBSCAN cluster analysis, first, the neighborhood density of each data point is calculated and the core points in the data are identified based on the density estimation results. Then, the core points and the points in their neighborhood are aggregated together to form density clusters, thereby identifying the key points of interest (POIs) of users at different time periods and locations. The expression is: Finally, the points that do not belong to any cluster are marked as noise points, where POI represents the i-th key interest point, which is a point set, where X represents the entire data set, and N ∈ (x) represents the neighborhood point set of point x in the ∈ neighborhood, δ represents the density threshold, |N ∈ (x)| represents the number of neighboring points of point x in the ∈ neighborhood, and |X| represents the total number of points in the entire data set.

3. The intelligent information push system based on user behavior analysis according to claim 2 is characterized in that: After obtaining the key points of interest (POIs), analyze the user's behavior frequency at each point of interest (POI) based on the obtained key points of interest (POIs). Identify high-frequency behavior patterns and use the Apriori algorithm to extract the behavioral correlation between users at different points of interest; then, screen the generated association rules according to the confidence and support of the behavior pattern, retain the rules with high confidence and high support to form the final user behavior pattern, and finally output the key points of interest POIs and behavior patterns identified by DBSCAN clustering analysis and association rule mining, where: represents the user's behavior frequency at the i-th key interest point, T represents the total length of observation time, and N i represents the number of user behavior events at key interest point i, w j represents the weight of the jth behavior event, σ j represents the normalized frequency of the jth behavioral event, Represents the total weight of all behaviors at the key interest point i.

4. The intelligent information push system based on user behavior analysis according to claim 1, characterized in that: The steps to generate a user's dynamic behavior model are: Obtain the user's historical behavior data and real-time data stream from the user behavior data collection module, then organize the user's historical behavior data and real-time behavior data into sequence input in chronological order, and convert the original data into a feature vector suitable for model processing as the input of the LSTM model; When building an LSTM model, you need to first set the structure of the input layer, hidden layer, and output layer and use historical behavior data to train the LSTM model; The user behavior data received in real time is input into the trained LSTM model, and the model updates its internal state based on the input data. Finally, the LSTM model combines historical behavior data with current real-time data to predict the user's future behavior and generate a dynamic behavior model of the user.

5. The intelligent information push system based on user behavior analysis according to claim 4 is characterized in that: When designing the structure of the LSTM model, first, set the size of the feature vector in the input layer corresponding to each time step and set the number of time steps processed by the LSTM model in a time series; set the hidden layer and set a 3-layer LSTM structure in the hidden layer, and use different functions in the output layer according to the situation. If the output is a classification problem, use the softmax activation function; if it is a regression problem, use the linear activation function.

6. The intelligent information push system based on user behavior analysis according to claim 1, characterized in that: When generating personalized content related to the user's current needs and future expectations, first, obtain the user's dynamic behavior model from the real-time behavior prediction module and obtain the user's key points of interest POIs and related behavior patterns from the historical behavior analysis module. By analyzing the user's dynamic behavior model, the user's current needs are identified. Then, combined with the user's future behavior predicted by the dynamic behavior model, the user's future needs are analyzed. The expression is: Among them, D future represents the user's future needs, m represents the number of features involved in predicting future user behavior, and λ j represents the weight coefficient associated with the jth future behavior prediction feature, reflecting the degree of influence of the feature on future demand, h j represents the predicted feature vector in the dynamic behavior model associated with the jth feature, t future represents the predicted future time point, ψ(h j , t future ) represents the feature h j At future time t future The response function under the condition of outputting the predicted impact value related to future demand, θ represents the weight coefficient of historical behavior pattern on future demand prediction, ω(H, P j , Δt) represents the historical behavior pattern H and key interest points P j The impact function on future demand under time change Δt is used to predict the impact of historical behavior on future demand.

7. The intelligent information push system based on user behavior analysis according to claim 6, characterized in that: When obtaining future needs, it is also necessary to match the user's current needs and future expectations with the content library in the system and personalize the matching content according to the user's needs, interests and behavior patterns. Then, based on the relevance of the content to user needs, urgency and user preferences, prioritize the generated personalized content. Finally, integrate different types of personalized content into one or more information packages.

8. The intelligent information push system based on user behavior analysis according to claim 1, characterized in that: When the generated personalized content is pushed to the user, the user's current geographic location is first obtained in real time through GPS, Wi-Fi positioning, and Bluetooth beacons and compared with the pre-defined key points of interest range. When the user's current location information matches a certain point of interest range, it is recognized that the user has entered the point of interest area. Then, based on the user's current location information and network conditions, the optimal data transmission path is selected and the push time is selected in combination with the user's device status and behavior patterns. Finally, when the user enters the pre-defined point of interest range and the device status and network conditions are suitable, the system immediately pushes the personalized content to the user through the selected channel.

9. The intelligent information push system based on user behavior analysis according to claim 1, characterized in that: When continuously updating the user's behavior model through machine learning algorithms and optimizing future content recommendation and push strategies, first, monitor and record the user's interactive behavior with the pushed content in real time and classify the collected feedback information by type. Then, the processed user feedback data is sent back to the historical behavior analysis module to update the user's historical behavior record to reflect the user's latest behavior changes and interest point adjustments. The feedback data is simultaneously transmitted to the real-time behavior prediction module to adjust the prediction results of the dynamic behavior model. Subsequently, based on the updated user behavior data set, the user behavior model is retrained and optimized using a machine learning algorithm. During the model training process, the model parameters are automatically adjusted to improve the accuracy of the prediction and the relevance of the content recommendation. Finally, based on the updated behavior model, the content recommendation strategy is re-optimized and the content push path is optimized based on the latest model results.

10. An intelligent information push method based on user behavior analysis, characterized in that: The following steps are involved: Collect behavioral data from users across multiple devices and channels in real time; Receive and process user behavior data, analyze the user's historical behavior data based on the DBSCAN density clustering algorithm and association rule mining technology, and identify the user's key points of interest and behavior patterns; Using the results of historical behavior analysis, long short-term memory networks and real-time data stream processing technology are used to predict users' current and future behaviors and generate dynamic behavior models of users; Based on the dynamic behavior model generated by the real-time behavior prediction module and the user's key interests, the module generates personalized content related to the user's current needs and future expectations by analyzing the user's interests and behavior patterns; By combining the user's location information, device status and network conditions, the push path and time are optimized. When the user enters the predefined point of interest, the generated personalized content is pushed to the user through multiple channels; Collect user feedback on pushed content and send it back to the historical behavior analysis module and real-time behavior prediction module. Use machine learning algorithms to continuously update the user behavior model and optimize future content recommendations and push strategies. Differential privacy and encryption technologies are used to protect the privacy of user data during both data transmission and reception.

Images