A user behavior analysis method
By combining RNN and ARIMA models to analyze user behavior and page performance data, this approach solves the problem that traditional methods cannot fully capture the dynamic changes in user behavior and page performance, achieving high-precision performance evaluation and optimization, and improving user experience and system performance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING BAILONG MAYUN TECH CO LTD
- Filing Date
- 2024-08-02
- Publication Date
- 2026-06-23
AI Technical Summary
Traditional page performance monitoring methods cannot fully capture the complexity of user behavior and the dynamic changes in page performance, making it difficult to provide accurate and comprehensive performance assessments and optimization suggestions, resulting in unsatisfactory optimization effects.
By employing sequence models (RNN) and time-series models (ARIMA), and combining user behavior data and page performance data, an ARIMA-RNN hybrid model is used for analysis to predict future optimization schemes and formulate specific optimization measures.
It enables comprehensive analysis of user behavior and page performance, improves the model's prediction accuracy and practicality, can anticipate potential performance issues, dynamically adapts to changes in user behavior and page performance, and continuously optimizes user experience and system performance.
Smart Images
Figure CN119128252B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of substation technology, and in particular to a user behavior analysis method. Background Technology
[0002] With the rapid development of the internet and mobile applications, users have increasingly higher demands for the responsiveness and interactive experience of websites and applications. Users not only expect pages to load within seconds, but also expect a smooth, lag-free user experience. These high expectations put enormous pressure on developers, forcing them to constantly seek more efficient performance optimization methods. However, traditional page performance monitoring methods are often too simplistic to fully capture the complexity of user behavior and the dynamic changes in page performance, thus failing to provide effective decision support and performance optimization solutions.
[0003] Traditional page performance monitoring methods primarily rely on basic statistical data, such as page load time and resource load time. These methods typically just record and analyze basic performance metrics like page load time, blank screen time, and resource request time. While they provide some basic performance information, they lack in-depth analysis of user behavior patterns and the time-series characteristics of page performance. These simple statistical data cannot reflect the real user experience in different interaction scenarios, nor can they capture the dynamic characteristics of page performance changes over time.
[0004] Furthermore, traditional methods often overlook the diversity and complexity of user behavior. Users may perform various actions while using a website or application, such as clicking buttons, scrolling pages, and typing text, all of which impact page performance. However, traditional methods typically focus only on a single performance metric, failing to comprehensively consider the overall user behavior patterns and the combined impact of these actions on performance.
[0005] This limitation makes it difficult for traditional performance monitoring methods to provide accurate and comprehensive performance assessments and optimization suggestions when faced with complex user behaviors and dynamic page performance changes. Developers often have to rely on experience and intuition for performance optimization, lacking scientific data support and precise optimization directions, ultimately resulting in unsatisfactory optimization effects and limited improvement in user experience.
[0006] Therefore, this invention proposes a user behavior analysis method. Summary of the Invention
[0007] In view of this, embodiments of the present invention aim to provide a user behavior analysis method, including a sequence model and a time series model, implemented through the following steps:
[0008] S1, Data Preparation: Collect user behavior data and page performance data on the website or application;
[0009] S2, Sequence Model Construction: Using an RNN model to model user behavior sequences;
[0010] S3, Time Series Model Construction: Using the ARIMA model to model page performance data;
[0011] S4 introduces the RNN model into the ARIMA model, incorporates the constructed time series model into the sequence model, and uses page performance data as an input to the sequence model for analysis together with user behavior data;
[0012] S5. Use the trained model parameters to predict the optimization scheme and optimization cost for future time points. Based on the analysis results, use the obtained time-domain adjustment factors and factors to formulate specific optimization schemes.
[0013] S6 incorporates new optimization costs and schemes to adjust the ARIMA and RNN models, resulting in more accurate analysis of performance bottlenecks.
[0014] Optionally, the logic for collecting user behavior data on a website or application in S1 includes:
[0015] Get the start and end times of page loading;
[0016] By using JavaScript event listeners, user actions such as clicks, scrolling, and input on the page are captured, and the timestamp of each interaction is recorded, and the corresponding response latency is calculated.
[0017] Generate and record a unique user identifier (UUID), which is stored in the user's browser via cookie or localStorage to ensure user identification across sessions.
[0018] Optionally, the logic for collecting user behavior data on a website or application in S1 includes:
[0019] Get the loading time of each resource and record the start loading time and end loading time of each resource;
[0020] Obtain the times of First Render (FP) and First Content Render (FCP) and record the time points of the first page rendering and the first content rendering.
[0021] The time it takes for the page to go blank is obtained by calculating the difference between the time it takes for the browser to start initiating an HTTP request and the time it takes for the DOM tree to be fully constructed.
[0022] The collected user behavior data and page performance data are reported to the backend log server via HTTP requests;
[0023] On the server side, a log library is created to store the reported data and record the raw logs of user behavior data and page performance data.
[0024] Calculate page performance metrics over a specific time period, such as average white screen time and average resource loading time.
[0025] Optionally, the steps for using an RNN model to model user behavior sequences are as follows:
[0026] Collect user action sequence data from user behavior logs and convert user actions into training and test sets;
[0027] Train the RNN model using the training set data, set an appropriate batch size and number of training rounds, and evaluate the model's performance using the test set data, calculating metrics such as accuracy, precision, and recall.
[0028] Deploy the trained RNN model to the production environment, acquire user behavior data in real time, and input it into the RNN model for prediction;
[0029] Based on the prediction results, the page content, recommendation system, and other aspects are dynamically adjusted to optimize the user experience.
[0030] Optionally, the steps for using an ARIMA model to model page performance data are as follows:
[0031] Collect key performance indicators from websites or applications and store the data in a database, recording them in chronological order to form time-series data;
[0032] Determine the difference order, autoregressive term, and moving average term of the model, and use the determined difference order, autoregressive term, and moving average term to construct an ARIMA model;
[0033] The model makes predictions on the training set and compares them with the actual data to evaluate the model's predictive performance.
[0034] Optionally, the steps of introducing an RNN model into an ARIMA model, incorporating the existing time-series model into the sequence model, and using page performance data as input to the sequence model along with user behavior data for analysis are as follows:
[0035] User behavior data is concatenated with ARIMA residuals to form new multidimensional time series data, and the multidimensional time series data is then used to generate fixed-length subsequences, which are suitable for input to RNN models.
[0036] The trained ARIMA-RNN hybrid model was deployed to the production environment for real-time prediction of page performance and user behavior.
[0037] Regularly acquire new page performance data and user behavior data, input them into the ARIMA-RNN model for prediction, and predict performance and behavior in the future period.
[0038] The model is regularly updated based on new page performance data and user behavior data, and then retrained and optimized.
[0039] Optionally, the trained model parameters are used to predict the optimization scheme and optimization cost at future time points. Based on the analysis results, the specific optimization scheme is formulated using the obtained time-domain adjustment factors and other factors.
[0040] Collect expected page performance data and expected user behavior data at future points in time, as well as data on external factors that can easily affect page performance and user behavior;
[0041] The expected data at future time points are input into the trained ARIMA-RNN hybrid model to obtain page performance indicators and user behavior prediction results at future time points;
[0042] Analyze and predict metrics such as page load time, white screen time, and interaction latency to identify potential performance bottlenecks. Based on the analysis results, set specific optimization goals, including reducing page load time, reducing white screen time, and improving interaction speed. Determine optimization priorities based on the impact and urgency of the goals.
[0043] Based on the predicted changes in performance indicators, determine the time period for adjustment and the relevant factors.
[0044] Optionally, the steps to adjust the ARIMA and RNN models by combining new optimization costs and schemes to produce more accurate analysis results on performance bottlenecks are as follows:
[0045] Record and collect implemented optimization measures and their corresponding costs; integrate optimization plans, optimization costs, page performance data, and user behavior data to form a new multidimensional time series dataset;
[0046] New user behavior data and residual data from the ARIMA model are integrated to generate new sequence data;
[0047] The new optimization scheme and user behavior data are input into the updated model for prediction, and the time period and related factors that need to be optimized are adjusted according to the new prediction results.
[0048] Compared with the prior art, the beneficial effects of the present invention are:
[0049] First, by using RNN models to model user behavior sequences, we can gain a deeper understanding of user operation patterns and behavioral characteristics, thereby optimizing the user interaction experience. Second, by using ARIMA models to model page performance data, we can accurately capture the time-series characteristics of page performance and identify potential performance bottlenecks and trends. Third, by introducing RNN models into ARIMA models and using page performance data as input to the sequence model, along with user behavior data, we can achieve a comprehensive analysis of user behavior and page performance, improving the model's predictive accuracy and practicality. Fourth, by using trained model parameters to predict optimization schemes and costs at future time points, we can anticipate potential performance problems and optimization needs, allowing us to formulate and implement optimization schemes in advance and avoid potential performance bottlenecks. Fifth, based on the analysis results, we can use the obtained time-domain adjustment factors and factors to formulate specific optimization schemes, ensuring the accuracy and effectiveness of optimization measures.
[0050] Second, this invention adjusts the ARIMA and RNN models by combining new optimization costs and optimization schemes to generate more accurate analysis results on performance bottlenecks. It can dynamically adapt to constantly changing user behavior and page performance, and continuously optimize user experience and system performance. Attached Figure Description
[0051] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0052] Figure 1 This is a schematic diagram of the present invention. Detailed Implementation
[0053] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below;
[0054] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.
[0055] This can be achieved through the following steps:
[0056] S1, Data Preparation: Collect user behavior data and page performance data on the website or application;
[0057] S2, Sequence Model Construction: Using an RNN model to model user behavior sequences;
[0058] S3, Time Series Model Construction: Using the ARIMA model to model page performance data;
[0059] S4 introduces the RNN model into the ARIMA model, incorporates the constructed time series model into the sequence model, and uses page performance data as an input to the sequence model for analysis together with user behavior data;
[0060] S5. Use the trained model parameters to predict the optimization scheme and optimization cost for future time points. Based on the analysis results, use the obtained time-domain adjustment factors and factors to formulate specific optimization schemes.
[0061] S6 incorporates new optimization costs and schemes to adjust the ARIMA and RNN models, resulting in more accurate analysis of performance bottlenecks.
[0062] It should be noted that: through a systematic and gradual approach, user behavior data and page performance data are comprehensively integrated to achieve accurate prediction and optimization of user experience and page performance. Firstly, the data preparation phase ensures the comprehensiveness and accuracy of the basic data, laying a solid foundation for subsequent modeling. By using an RNN model to model user behavior sequences, a deeper understanding of user operation patterns and behavioral characteristics can be achieved, thereby optimizing the user interaction experience. Using an ARIMA model to model page performance data can accurately capture the time-series characteristics of page performance and identify potential performance bottlenecks and trends. Introducing the RNN model into the ARIMA model, with page performance data as an input to the sequence model, along with user behavior data... The analysis enables a comprehensive analysis of user behavior and page performance, improving the model's predictive accuracy and practicality. By using trained model parameters to predict optimization schemes and costs at future points in time, it is possible to anticipate potential performance issues and optimization needs, allowing for the early development and implementation of optimization plans to avoid potential performance bottlenecks. Based on the analysis results, specific optimization schemes are formulated using the obtained time-domain adjustment factors and other factors, ensuring the accuracy and effectiveness of optimization measures. Finally, by combining the new optimization costs and schemes, the ARIMA and RNN models are adjusted to generate more accurate analysis results regarding performance bottlenecks, dynamically adapting to constantly changing user behavior and page performance, and continuously optimizing user experience and system performance.
[0063] In this embodiment, the logic for collecting user behavior data on a website or application in step S1 includes:
[0064] Get the start and end times of page loading;
[0065] By using JavaScript event listeners, user actions such as clicks, scrolling, and input on the page are captured, and the timestamp of each interaction is recorded, and the corresponding response latency is calculated.
[0066] Generate and record a unique user identifier (UUID), which is stored in the user's browser via cookie or localStorage to ensure user identification across sessions.
[0067] It's important to note that: by obtaining the start and end times of page loading, accurate measurement of page loading speed and performance can be achieved, providing precise data support for optimizing page loading time; by using JavaScript event listeners to capture user actions such as clicks, scrolling, and input on the page and recording the timestamp of each interaction, real-time tracking of user behavior can be achieved, providing detailed insights into user operating habits and preferences on the page, thereby calculating corresponding response latency, evaluating page interaction performance, and providing a scientific basis for optimizing user experience; generating and recording a user's unique identifier (UUID) and storing it in the user's browser via cookies or localStorage ensures cross-session user identification, allowing continuous tracking of user behavior and preferences across different sessions to form a complete user behavior profile, which helps provide personalized services and recommendations, improving user satisfaction; the overall advantage of this approach is that through the systematic collection and recording of performance data and user behavior data, not only can a comprehensive understanding of page loading and interaction performance be obtained, identifying performance bottlenecks and optimization opportunities, but also, through cross-session user identification, long-term tracking and analysis of user behavior can be achieved, providing more accurate and personalized services, thereby significantly improving user experience and satisfaction, helping enterprises maintain a technological lead and user loyalty advantage in a highly competitive market, and improving operational efficiency and business success rate.
[0068] In this embodiment, the logic for collecting user behavior data on a website or application in step S1 includes:
[0069] Get the loading time of each resource and record the start loading time and end loading time of each resource;
[0070] Obtain the times of First Render (FP) and First Content Render (FCP) and record the time points of the first page rendering and the first content rendering.
[0071] The time it takes for the page to go blank is obtained by calculating the difference between the time it takes for the browser to start initiating an HTTP request and the time it takes for the DOM tree to be fully constructed.
[0072] The collected user behavior data and page performance data are reported to the backend log server via HTTP requests;
[0073] On the server side, a log library is created to store the reported data and record the raw logs of user behavior data and page performance data.
[0074] Calculate page performance metrics over a specific time period, such as average white screen time and average resource loading time.
[0075] It's important to note that: by acquiring the loading time of each resource and recording its start and end times, we can comprehensively track and analyze the loading efficiency and bottlenecks of each page resource, providing precise data support for subsequent performance optimization; acquiring and recording the times of the first render (FP) and first content render (FCP) allows us to specifically measure the initial display and content visibility of the page, helping to improve the user's perceived loading speed; calculating the difference between the time the browser initiates an HTTP request and the time the DOM tree is fully constructed yields the page blank time, accurately assessing the time users see a blank screen during page loading, further optimizing the user experience; and reporting the collected user behavior data and page performance data to the backend log server via HTTP requests ensures timely data updates. It ensures data integrity and supports real-time monitoring and analysis. On the server side, a log library is created to store reported data, recording raw logs of user behavior data and page performance data. This allows for the systematic management and storage of large amounts of data, supporting multi-dimensional analysis and mining. It calculates page performance metrics over specific time periods, such as average scrolling time and average resource loading time, comprehensively reflecting page performance at different times and helping to identify and resolve performance issues. The overall advantage of this approach is that through systematic data collection, meticulous recording, and scientific data storage, page performance can be comprehensively and accurately tracked and analyzed. This effectively identifies performance bottlenecks and optimization opportunities, providing a scientific basis for decision-making, ultimately achieving continuous optimization of page performance and a significant improvement in user experience. This helps enterprises maintain technological leadership and user satisfaction in fierce market competition, improving operational efficiency and business success rates.
[0076] In this embodiment, the steps for modeling user behavior sequences using an RNN model are as follows:
[0077] Collect user action sequence data from user behavior logs and convert user actions into training and test sets;
[0078] Train the RNN model using the training set data, set an appropriate batch size and number of training rounds, and evaluate the model's performance using the test set data, calculating metrics such as accuracy, precision, and recall.
[0079] Deploy the trained RNN model to the production environment, acquire user behavior data in real time, and input it into the RNN model for prediction;
[0080] Based on the prediction results, the page content, recommendation system, and other aspects are dynamically adjusted to optimize the user experience.
[0081] It's important to note that: systematically collecting user action sequence data from user behavior logs and converting this data into training and test sets comprehensively covers the diversity and complexity of user behavior, ensuring the representativeness and accuracy of the training and testing data. Using the training set data to train the RNN model, setting appropriate batch size and training epochs ensures the model fully understands the temporal characteristics and patterns of user behavior during the learning process, improving learning efficiency and generalization ability. Evaluating model performance using the test set data involves calculating metrics such as accuracy, precision, and recall, providing a comprehensive and objective assessment of the model's predictive ability and performance, allowing for timely identification and improvement of shortcomings, and ensuring the model's effectiveness and reliability in practical applications. Deploying the trained RNN model to a production environment enables real-time monitoring and prediction of user behavior. By acquiring user behavior data in real time and inputting it into the RNN model for prediction, user actions can be dynamically tracked and analyzed, capturing changes in user needs and preferences in a timely manner. Based on the prediction results, dynamic adjustments can be made to page content, recommendation systems, etc., to optimize user experience, achieving personalized and intelligent user services, improving user satisfaction and stickiness, and enhancing the user experience.
[0082] In this embodiment, the steps for modeling page performance data using the ARIMA model are as follows:
[0083] Collect key performance indicators from websites or applications and store the data in a database, recording them in chronological order to form time-series data;
[0084] Determine the difference order, autoregressive term, and moving average term of the model, and use the determined difference order, autoregressive term, and moving average term to construct an ARIMA model;
[0085] The model makes predictions on the training set and compares them with the actual data to evaluate the model's predictive performance.
[0086] It should be noted that systematically collecting key performance indicators from websites or applications and storing them in a database, recording them chronologically to form time series data, ensures the integrity, continuity, and accuracy of the data, providing a solid data foundation for subsequent analysis and modeling. Determining the difference order, autoregressive term, and moving average term of the model allows for the accurate capture of trends, seasonality, and random fluctuations in the time series based on the characteristics and patterns of data. Using these parameters to build an ARIMA model helps to effectively analyze and predict the future trends of time series data. Performing model predictions on the training set and comparing them with actual data to evaluate the model's predictive performance not only verifies the model's accuracy and reliability but also identifies its strengths and weaknesses, providing a reference for further optimization and adjustment.
[0087] The overall advantage of this approach lies in its ability to collect and process performance index data systematically, ensuring data quality and integrity. By determining parameters such as differencing, autoregression, and moving average terms, an ARIMA model suitable for specific data characteristics can be constructed. This allows for the accurate capture of inherent patterns in time series data, improving prediction accuracy and reliability. Through model prediction and performance evaluation on the training set, problems and areas for improvement can be identified promptly, continuously optimizing and enhancing the model's predictive capabilities. This not only helps improve the performance monitoring and prediction capabilities of websites or applications but also provides a scientific basis for performance optimization and decision-making. Ultimately, it enhances user experience, optimizes resource allocation, reduces operating costs, strengthens system stability and sustainable development capabilities, and helps enterprises maintain a leading position in technological advantage and operational efficiency in fierce market competition.
[0088] 6. The user behavior analysis method according to claim 1, characterized in that: the steps of introducing an RNN model into an ARIMA model, introducing the constructed time-series model into the sequence model, and using page performance data as an input to the sequence model, and analyzing it together with user behavior data, are as follows:
[0089] User behavior data is concatenated with ARIMA residuals to form new multidimensional time series data, and the multidimensional time series data is then used to generate fixed-length subsequences, which are suitable for input to RNN models.
[0090] The trained ARIMA-RNN hybrid model was deployed to the production environment for real-time prediction of page performance and user behavior.
[0091] Regularly acquire new page performance data and user behavior data, input them into the ARIMA-RNN model for prediction, and predict performance and behavior in the future period.
[0092] The model is regularly updated based on new page performance data and user behavior data, and then retrained and optimized.
[0093] It's important to note that by concatenating user behavior data with ARIMA residuals to create new multidimensional time series data and generating fixed-length subsequences suitable for RNN model input, the advantages of time series analysis and deep learning can be effectively combined to improve data processing accuracy and model predictive power. Deploying a pre-trained ARIMA-RNN hybrid model to a production environment provides an efficient and accurate solution for real-time prediction of page performance and user behavior, helping to respond promptly to user needs and market changes. Regularly acquiring new page performance and user behavior data and inputting it into the ARIMA-RNN model allows for continuous updates to future performance and behavior predictions, ensuring real-time accuracy. Regularly updating the model with new page performance and user behavior data allows for retraining and optimization, maintaining its up-to-dateness and effectiveness while continuously improving performance and prediction accuracy, ensuring high adaptability in constantly changing real-world environments.
[0094] The overall advantage of this approach lies in combining the powerful analytical capabilities of the ARIMA model for time-series data with the deep learning capabilities of the RNN model for sequence data. This enables comprehensive, dynamic, and accurate prediction and analysis of page performance and user behavior, thereby effectively identifying potential problems and optimization opportunities, formulating timely and effective optimization measures, continuously improving user experience and system performance, reducing operating costs, and enhancing the scientific and agile nature of business decisions. Ultimately, this helps enterprises maintain technological advantages and user satisfaction in market competition, and achieve significant improvements in operational efficiency and resource utilization.
[0095] In this embodiment, the steps of using trained model parameters to predict optimization schemes and costs at future time points, and formulating specific optimization schemes based on the analysis results and the obtained time-domain adjustment factors and factors are as follows:
[0096] Collect expected page performance data and expected user behavior data at future points in time, as well as data on external factors that can easily affect page performance and user behavior;
[0097] The expected data at future time points are input into the trained ARIMA-RNN hybrid model to obtain page performance indicators and user behavior prediction results at future time points;
[0098] Analyze and predict metrics such as page load time, white screen time, and interaction latency to identify potential performance bottlenecks. Based on the analysis results, set specific optimization goals, including reducing page load time, reducing white screen time, and improving interaction speed. Determine optimization priorities based on the impact and urgency of the goals.
[0099] Based on the predicted changes in performance indicators, determine the time period for adjustment and the relevant factors.
[0100] It should be noted that this comprehensive step systematically integrates and utilizes existing model parameters and data prediction results, improving the accuracy and efficiency of page performance optimization at future time points. By collecting expected page performance data, user behavior data, and data on external factors that easily affect page performance and user behavior at future time points, it ensures the comprehensiveness and relevance of the input data to the prediction model, thereby improving the accuracy of model predictions. This expected data is then input into a trained ARIMA-RNN hybrid model, fully leveraging the advantages of the ARIMA model in time series analysis and the powerful performance of the RNN model in processing sequential data. This yields predicted page performance metrics and user behavior results for future time points. Based on these prediction results, in-depth analysis of page load time, white page load time, and other factors can be conducted. By accurately identifying potential performance bottlenecks based on key performance indicators such as screen time and interaction latency, the system provides comprehensive performance insights. Based on these analyses, specific optimization goals are set, including reducing page load time, minimizing white screen time, and improving interaction speed. Optimization priorities are rationally determined according to the impact and urgency of these goals, ensuring efficient use of resources and time. Simultaneously, based on predicted changes in performance indicators, the system identifies the time periods and relevant factors requiring adjustment, dynamically adapting to changes in performance and behavior. This provides precise time-domain adjustment suggestions, enabling the development of more specific and actionable optimization solutions. Ultimately, this helps businesses improve page performance, optimize user experience, reduce operating costs, achieve continuous improvement and stable operation of system performance, and maintain a technological advantage and user satisfaction in market competition.
[0101] In this embodiment, the steps for adjusting the ARIMA and RNN models by combining new optimization costs and optimization schemes to generate more accurate analysis results regarding performance bottlenecks are as follows:
[0102] Record and collect implemented optimization measures and their corresponding costs; integrate optimization plans, optimization costs, page performance data, and user behavior data to form a new multidimensional time series dataset;
[0103] New user behavior data and residual data from the ARIMA model are integrated to generate new sequence data;
[0104] The new optimization scheme and user behavior data are input into the updated model for prediction, and the time period and related factors to be optimized are adjusted based on the new prediction results. This not only effectively utilizes existing optimization results and related cost information to improve the accuracy and adaptability of model predictions, but also achieves a comprehensive insight into user behavior and page performance through the integration and analysis of multi-dimensional data. This allows for the identification and localization of more granular performance bottlenecks, leading to the development of more precise and effective optimization measures. Furthermore, by combining the time series analysis capabilities of the ARIMA model with the powerful performance of the RNN model in processing sequence data, this method can dynamically adapt to changes in page performance and user behavior, providing real-time performance optimization suggestions and decision support. Ultimately, this helps enterprises improve user experience, optimize resource allocation, reduce operating costs, and enhance the overall performance and stability of the system in complex network environments.
Claims
1. A user behavior analysis method, comprising a sequence model and a time-series model, characterized in that: This can be achieved through the following steps: S1, Data Preparation: Collect user behavior data and page performance data on the website or application; S2, Sequence Model Construction: Using an RNN model to model user behavior sequences; S3, Time Series Model Construction: Using the ARIMA model to model page performance data; S4 introduces the RNN model into the ARIMA model, incorporates the constructed time series model into the sequence model, and uses page performance data as an input to the sequence model for analysis together with user behavior data; S5. Use the trained model parameters to predict the optimization scheme and optimization cost for future time points. Based on the analysis results, use the obtained time-domain adjustment factors and factors to formulate specific optimization schemes. S6, combining new optimization costs and optimization schemes, adjusts the ARIMA and RNN models to produce more accurate analysis results on performance bottlenecks; The steps involve using the trained model parameters to predict optimization schemes and costs at future time points, and then, based on the analysis results and the obtained time-domain adjustment factors, to formulate specific optimization schemes. Collect expected page performance data and expected user behavior data at future points in time, as well as data on external factors that can easily affect page performance and user behavior; The expected data at future time points are input into the trained ARIMA-RNN hybrid model to obtain page performance indicators and user behavior prediction results at future time points; Analyze and predict page load time, white screen time, and interaction latency metrics to identify potential performance bottlenecks. Based on the analysis results, set specific optimization goals, including reducing page load time, reducing white screen time, and improving interaction speed. Determine optimization priorities based on the impact and urgency of the goals. Based on the predicted changes in performance indicators, determine the time period for adjustment and the relevant factors.
2. The user behavior analysis method according to claim 1, characterized in that: The logic for collecting user behavior data on a website or application in S1 includes: Get the start and end times of page loading; By using JavaScript event listeners, we can capture user actions such as clicking, scrolling, and typing on the page, record the timestamp of each interaction, and calculate the corresponding response latency. Generate and record a unique user identifier, which is stored in the user's browser via cookie or localStorage to ensure user identification across sessions.
3. The user behavior analysis method according to claim 1, characterized in that: The logic for collecting user behavior data on a website or application in S1 includes: Get the loading time of each resource and record the start loading time and end loading time of each resource; Get the times of the first rendering and the first content rendering, and record the time points of the first page rendering and the first content rendering. The time it takes for the page to go blank is obtained by calculating the difference between the time it takes for the browser to start initiating an HTTP request and the time it takes for the DOM tree to be fully constructed. The collected user behavior data and page performance data are reported to the backend log server via HTTP requests; On the server side, a log library is created to store the reported data and record the raw logs of user behavior data and page performance data. Calculate page performance metrics, including average white screen time and average resource loading time, over a specific time period.
4. The user behavior analysis method according to claim 1, characterized in that: The steps for using an RNN model to model user behavior sequences are as follows: Collect user action sequence data from user behavior logs and convert user actions into training and test sets; Train the RNN model using the training set data, set an appropriate batch size and number of training rounds, and evaluate the model's performance using the test set data, calculating the model's accuracy, precision, and recall metrics. Deploy the trained RNN model to the production environment, acquire user behavior data in real time, and input it into the RNN model for prediction; Based on the prediction results, the page content and recommendation system are dynamically adjusted to optimize the user experience.
5. The user behavior analysis method according to claim 1, characterized in that: The steps for using the ARIMA model to model page performance data are as follows: Collect key performance indicators from websites or applications and store the data in a database, recording them in chronological order to form time-series data; Determine the difference order, autoregressive term, and moving average term of the model, and use the determined difference order, autoregressive term, and moving average term to construct an ARIMA model; The model makes predictions on the training set and compares them with the actual data to evaluate the model's predictive performance.
6. The user behavior analysis method according to claim 1, characterized in that: The steps for introducing an RNN model into an ARIMA model, incorporating an existing time-series model into a sequence model, and using page performance data as input to the sequence model along with user behavior data are as follows: User behavior data is concatenated with ARIMA residuals to form new multidimensional time series data, and the multidimensional time series data is then used to generate fixed-length subsequences, which are suitable for input to RNN models. The trained ARIMA-RNN hybrid model was deployed to the production environment for real-time prediction of page performance and user behavior. Regularly acquire new page performance data and user behavior data, input them into the ARIMA-RNN model for prediction, and predict performance and behavior in the future period. The model is regularly updated based on new page performance data and user behavior data, and then retrained and optimized.
7. The user behavior analysis method according to claim 1, characterized in that: The steps to adjust the ARIMA and RNN models by combining new optimization costs and schemes to produce more accurate analysis results on performance bottlenecks are as follows: Record and collect implemented optimization measures and their corresponding costs; integrate optimization plans, optimization costs, page performance data, and user behavior data to form a new multidimensional time series dataset; New user behavior data and residual data from the ARIMA model are integrated to generate new sequence data; The new optimization scheme and user behavior data are input into the updated model for prediction, and the time period and related factors that need to be optimized are adjusted according to the new prediction results.