Internet user behavior accurate analysis method based on artificial intelligence

Abstract

The invention discloses an accurate analysis method for Internet user behaviors based on artificial intelligence, and the method belongs to the technical field of Internet, and comprises the steps: collecting user behaviors and basic attribute information through a multi-source data collection technology, and integrating data through a data fusion algorithm after data cleaning and preprocessing; designing and extracting a plurality of behavior features, screening key features by using a feature selection algorithm, and performing feature transformation processing at the same time; constructing a deep learning model fusing a recurrent neural network (such as LSTM and GRU), a convolutional neural network and the like, training the model by using a large-scale data set, and optimizing parameters; establishing a real-time analysis system, inputting new data into the model in real time, providing personalized service and recommendation according to an analysis result, and collecting feedback data to update the model online; and finally, evaluating the model by adopting multiple evaluation indexes, verifying the generalization ability through methods such as cross validation and the like, and further optimizing and adjusting the model.

Description

Technical Field

[0001] The present invention belongs to the technical fields related to Internet technology and artificial intelligence, and particularly relates to a method for accurately analyzing Internet user behavior based on artificial intelligence. Background Art

[0002] In today's digital age, the Internet has generated a vast amount of user behavior data. By analyzing this data, enterprises and platforms can deeply understand user needs, preferences, and behavior patterns, so as to provide personalized services for users, optimize product designs, formulate accurate marketing strategies, etc. However, traditional user behavior analysis methods have many limitations.

[0003] On the one hand, traditional methods often rely on simple statistical analysis and rule matching, and it is difficult to process large-scale, high-dimensional, complex and changeable data. For example, when analyzing the behavior of users browsing web pages, only basic information such as the number of views and the stay time can be counted, and it is impossible to deeply explore the potential interests and behavior motives of users. This results in inaccurate analysis results and cannot provide strong support for decision-making.

[0004] On the other hand, with the continuous expansion of Internet services and the increasing diversification of user needs, new user behavior patterns and data types are emerging continuously. Traditional analysis methods lack the ability to adapt to these new situations and are difficult to quickly and effectively identify and analyze new types of user behavior. For example, in emerging business scenarios such as short videos and live broadcasts, the interaction behaviors of users are more complex and diverse, and traditional methods are difficult to accurately capture and analyze the user intentions behind these behaviors.

[0005] In addition, some existing preliminary analysis methods based on artificial intelligence, although improving the analysis ability to a certain extent, still have deficiencies in aspects such as the accuracy, real-time performance of the model, and the ability to fuse and process multi-source data. For example, when processing user data from different sources, some models cannot fully explore the correlation relationships between data, resulting in the integrity and reliability of the analysis results being affected. Therefore, there is an urgent need for a new, accurate analysis method based on artificial intelligence to solve the problems existing in the prior art and meet the growing demand for Internet user behavior analysis. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for accurately analyzing Internet user behavior based on artificial intelligence to solve the problems raised in the above background art.

[0007] To achieve the above purpose, the present invention provides the following technical solutions:

[0008] A method for accurately analyzing Internet user behavior based on artificial intelligence, comprising:

[0009] Multi-source data collection and integration steps: Collect user behavior data and basic attribute information from various channels of the Internet platform through a variety of data collection technologies, use data cleaning and preprocessing technologies to remove noise, duplicates, and missing values ​​from the collected data, and use data fusion algorithms to integrate data from different data sources and formats into a unified user behavior data set;

[0010] Feature engineering construction steps: design and extract a variety of user behavior features, use feature selection algorithms to screen key features, and use feature transformation technology to normalize and standardize some features;

[0011] Deep learning model construction and training steps: Build a user behavior analysis model based on deep learning, including using recurrent neural networks and their variants or convolutional neural networks, integrating multiple deep learning models, using large-scale user behavior data sets to train the model, and adjusting model parameters through optimization algorithms;

[0012] Real-time analysis and feedback mechanism steps: Establish a real-time data processing and analysis system, input the newly generated user behavior data into the trained model in a timely manner for real-time analysis, provide personalized services and recommendations to users based on the analysis results, and collect user feedback data and feed it back to the model in real time for online update and optimization;

[0013] Model evaluation and optimization steps: Use a variety of evaluation indicators to evaluate the trained model, verify the model's generalization ability through cross-validation, and optimize the model based on the evaluation results.

[0014] Preferably, in the multi-source data collection and integration step, user behavior data from public pages on the Internet are collected through web crawler technology, and user operation behavior data on specific devices are collected through sensor technology; and for different types of data sources, adapted data interface protocols are used for data access to ensure the comprehensiveness and real-time nature of data collection, and a data denoising method based on an outlier detection algorithm is used to identify and remove outliers in the data; a missing value filling algorithm based on a probability statistical model is used to accurately fill in missing data to improve data quality, and a data fusion algorithm based on a graph model is used to construct a correlation relationship graph between data, and based on the weights and connection relationships between nodes, deep fusion of data from different data sources is achieved, and potential data association patterns are mined, and the collected user behavior data includes but is not limited to user browsing records, click behaviors, search histories, purchase transaction records, and social interaction information.

[0015] Preferably, in the feature engineering construction step, the extracted user behavior features include user activity features, interest preference features, and behavior sequence features. The interaction behavior features of users in the virtual reality (VR) / augmented reality (AR) scenario are extracted, including gesture operation frequency, perspective switching mode, etc.; and the behavior features of users in blockchain applications, such as transaction verification and information encryption. A feature selection network based on deep learning is used to automatically screen out the features that have a key impact on user behavior analysis by constructing a non-linear mapping relationship between features and target variables. Compared with traditional algorithms, the accuracy and efficiency of feature selection are significantly improved. The feature transformation method based on the generative adversarial network (GAN) is used to transform the unstructured features in user behavior data into structured features that meet the model input requirements, enhancing the model's processing ability for complex data.

[0016] Preferably, in the deep learning model construction and training step, the optimization algorithms include stochastic gradient descent and Adam optimizer. A hybrid deep learning model that combines Transformer with LSTM with an attention mechanism is constructed. Transformer is used to capture the global dependencies in user behavior data, and LSTM is used to process the time dependencies of sequence data. The two work together to improve the modeling ability for complex user behaviors. Data augmentation techniques are used to perform operations such as random cropping, flipping, and adding noise to the training data to expand the diversity of the training data and improve the generalization ability of the model. The distributed training framework is used to parallelly allocate the training tasks to multiple computing nodes to accelerate the model training process and shorten the training time. The adaptive learning rate adjustment algorithm is used to dynamically adjust the learning rate according to the performance metrics of the model during the training process to ensure that the model converges quickly during training and avoids falling into local optimal solutions.

[0017] Preferably, in the real-time analysis and feedback mechanism step, a real-time data processing system based on the stream computing architecture is constructed, which has a data processing delay of milliseconds and can process large-scale user behavior data in real time. The message queue technology is used for data buffering and asynchronous processing to ensure the high availability and stability of the system. Based on the user's real-time behavior data and historical behavior portraits, a recommendation algorithm that combines collaborative filtering and deep learning is used to provide personalized recommendations for users on products, content, or services. The feedback data of users on the recommended content is collected in real time, and the online learning algorithm is used to incrementally update the recommendation model to continuously optimize the recommendation effect. The feedback data of users on different terminals (such as mobile terminals and PC terminals) and different application scenarios is integrated, and techniques such as sentiment analysis and behavior sequence analysis are used to deeply explore the intentions and needs behind the user feedback, providing a more comprehensive basis for model optimization.

[0018] Preferably, in the model evaluation and optimization step, the evaluation metrics include accuracy, recall, F1-score, and mean squared error. Evaluation metrics based on user experience, such as user satisfaction and retention rate, are adopted and combined with traditional metrics such as accuracy and recall to comprehensively evaluate the performance of the model in actual application scenarios. For different business scenarios (such as e-commerce promotion activities, social media hot events, etc.), the adaptability and stability of the model are customized. The model verification method based on reinforcement learning is used to dynamically verify the model by simulating the interaction process between users and the system, and evaluate the decision-making ability and adaptability of the model in complex environments. Cross-domain verification technology is adopted to verify the model by applying it to user behavior data in different fields, expanding the generalization ability boundary of the model. A model automatic optimization system based on meta-learning is constructed, which can automatically adjust the model structure, parameter settings, etc. according to the model evaluation results to achieve intelligent optimization of the model. An adaptive model update strategy is adopted to optimize the model in real time and dynamically according to the changes in data distribution and the adjustment of business requirements, ensuring that the model always maintains the best performance.

[0019] Preferably, an Internet user behavior precise analysis system based on artificial intelligence includes:

[0020] A data collection and integration module for performing multi-source data collection and integration steps, that is, collecting user behavior data and basic attribute information from various channels of the Internet platform through various data collection technologies, performing operations such as denoising, duplicate removal, and missing value processing on the collected data using data cleaning and preprocessing technologies, and integrating data from different data sources and formats into a unified user behavior data set using data fusion algorithms;

[0021] A feature engineering module for performing feature engineering construction steps, that is, designing and extracting various user behavior features, screening key features using feature selection algorithms, and performing normalization, standardization, etc. on some features using feature transformation technologies;

[0022] A model construction and training module for performing deep learning model construction and training steps.

[0023] Compared with the prior art, the present invention provides an Internet user behavior precise analysis method based on artificial intelligence, which has the following beneficial effects:

[0024] Improve analysis accuracy: By deeply mining and analyzing multi-source and complex user behavior data through a deep learning model, it can more accurately capture the intentions and preferences behind user behavior, and greatly improve the accuracy of analysis results compared with traditional methods.

[0025] Enhance adaptability to complex data and new business scenarios: The method of the present invention can effectively process various types and scales of user behavior data, including complex behavior data in emerging business scenarios, and has strong adaptability and scalability, enabling it to promptly address new challenges brought about by the development of Internet businesses.

[0026] Achieve real-time analysis and feedback: The establishment of a real-time data processing and analysis system and a feedback mechanism enables the provision of personalized services in a timely manner based on users' real-time behaviors, and continuously optimizes the model through user feedback, improving the user experience and service quality.

[0027] Optimize multi-source data fusion processing: Advanced data collection, integration, and feature engineering technologies can fully explore the correlation relationships between multi-source data, achieve a comprehensive and integrated analysis of user behaviors, and provide richer and more accurate information for decision-making. Description of the Drawings

[0028] Figure 1 It is a schematic structural diagram of the present invention. Detailed Embodiment

[0029] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0030] The present invention provides a method for accurate analysis of Internet user behaviors based on artificial intelligence as shown in Figure 1 and includes:

[0031] Multi-source data collection and integration step: Collect user behavior data and basic attribute information from various channels of the Internet platform through various data collection technologies, perform operations such as noise removal, duplicate removal, and missing value processing on the collected data using data cleaning and preprocessing technologies, and use data fusion algorithms to integrate data from different data sources and formats into a unified user behavior data set;

[0032] Feature engineering construction step: Design and extract various user behavior features, use feature selection algorithms to screen key features, and perform normalization and standardization processing on some features using feature transformation technologies;

[0033] Deep learning model construction and training step: Construct a user behavior analysis model based on deep learning, including using recurrent neural networks and their variants or convolutional neural networks, fuse multiple deep learning models, train the model using a large-scale user behavior data set, and adjust the model parameters through optimization algorithms;

[0034] Real-time analysis and feedback mechanism steps: Establish a real-time data processing and analysis system, input the newly generated user behavior data into the trained model in a timely manner for real-time analysis, provide personalized services and recommendations to users based on the analysis results, and collect user feedback data and feed it back to the model in real time for online update and optimization;

[0035] Model evaluation and optimization steps: Use a variety of evaluation indicators to evaluate the trained model, verify the model's generalization ability through cross-validation, and optimize the model based on the evaluation results.

[0036] Example 1: User behavior analysis of a comprehensive e-commerce platform

[0037] 1. Data collection and preprocessing

[0038] Data collection: This e-commerce platform has a huge user base and a variety of business types. In order to comprehensively collect user behavior data, it has adopted a variety of technical means. On the website and mobile application side, JavaScript scripts and SDK (software development kit) are used to collect users' real-time operation data, including page browsing, product clicks, search keyword input, adding to shopping carts, placing orders, and payment. At the same time, the server log is used to record basic data such as user access time, IP address, and device type. In addition, it also cooperates with third-party data platforms to obtain external data such as the geographic location of some users and social media account association information to enrich user portraits.

[0039] Data cleaning and preprocessing: The collected data has problems such as noise, duplication and missing, and needs to be cleaned and preprocessed. Use data cleaning tools to identify and delete abnormal data caused by network fluctuations, system errors, etc., such as records with negative browsing time or far beyond the normal range. Through hash algorithms and data comparison, duplicate user behavior records are removed to ensure the uniqueness of each data. For missing values, different filling methods are used according to data types and business logic. For example, for missing values ​​of user age, the average age of users of the same age group is used to fill in; for missing values ​​of product prices, the average price of the category to which the product belongs is used to fill in. At the same time, the text data is denoised to remove special characters, typos and stop words to make the text more standardized and easier to analyze.

[0040] 2. Feature Engineering Construction

[0041] Feature Extraction: Considering the characteristics of e-commerce business, a rich set of user behavior features are designed and extracted. In terms of activity features, the number of logins, total number of page views, and total number of operations within a week or a month are calculated. Interest preference features are determined by analyzing information such as the categories, brands, and keywords of the products browsed, searched, and purchased by users. For example, if a user frequently browses 3C digital products and the search keywords include "mobile phone", "computer", etc., it indicates that the user is interested in 3C digital products. Behavioral sequence features record the operation sequence during the shopping process, such as the process of first browsing product details, then adding to the shopping cart, and finally placing an order, as well as the time intervals between different operations.

[0042] Feature Selection: From the large number of extracted features, feature selection algorithms such as chi-square test and information gain are used to screen out the key features that have a significant impact on user purchase behavior. After screening, 15 key features are determined, including the number of times a user browses products, the number of products added to the shopping cart, the user's purchase history, and the correlation between browsed products and purchased products. These key features not only reduce the feature dimension and improve the model training efficiency but also can more accurately reflect the user's purchase intention.

[0043] Feature Transformation: Some features are normalized and standardized to make different features have the same scale. For example, for the feature of the user's consumption amount, the Z-score standardization method is used to transform it into a standard normal distribution with a mean of 0 and a standard deviation of 1, which is convenient for the model to better learn and process. At the same time, for some categorical features, such as product categories and user genders, one-hot encoding is used to transform them into numerical features to meet the input requirements of the model.

[0044] 3. Deep Learning Model Construction and Training

[0045] Model Construction: A comprehensive analysis model integrating multiple deep learning models is constructed. Among them, the long short-term memory network (LSTM) is used to process user behavior sequence data to capture the time dependence and long-term trends of user behavior. For example, through LSTM, the changes in a user's shopping preferences over a period of time and the sequential relationship of purchase behaviors can be learned. At the same time, the convolutional neural network (CNN) is used to extract and analyze features from unstructured data such as product pictures and evaluation texts uploaded by users. The outputs of LSTM and CNN are fused, and then classified and predicted through a fully connected layer to determine the user's purchase intention, degree of preference for products, etc.

[0046] Model Training: Divide the preprocessed data into a training set (70%), a validation set (20%), and a test set (10%). Use the training set to train the model, setting the number of training epochs to 80 and the learning rate to 0.0005. During training, use the Adam optimizer to adjust the model parameters and minimize the cross-entropy loss function to improve the model's accuracy. At the same time, to prevent overfitting, use L1 and L2 regularization techniques and add a Dropout layer to the model to randomly discard some neurons to enhance the model's generalization ability. After multiple rounds of training, the model achieves an accuracy of 90% and a recall rate of 85% on the validation set.

[0047] 4. Real-time Analysis and Feedback Mechanism

[0048] Real-time Analysis: Deploy a real-time analysis system on the server side of the e-commerce platform. When a user operates on the platform, the system uses a message queue (such as Kafka) to collect the user's behavior data in real time and quickly transmits it to the analysis model for processing. For example, when a user views a product details page, the system immediately extracts the user's behavior characteristics, including browsing history, recent search records, etc., and inputs them into the trained model for real-time analysis to predict the user's purchase probability for the product and other related products that the user may be interested in.

[0049] Personalized Recommendation: Provide personalized product recommendation services based on the real-time analysis results. If the model predicts that the user has a high purchase probability for the currently viewed product, the system will recommend the product's complementary products, similar styles, or other related products that users often purchase after buying this product at a prominent position on the page. At the same time, display the recommendation results on the user's personalized recommendation page, APP push messages, etc., to increase the probability that the user discovers products of interest.

[0050] Feedback Mechanism: The user's actions such as clicking, purchasing, favoriting, and ignoring recommended products are collected by the system in real time as feedback data. The system inputs this feedback data into the model to perform online updates and optimizations. For example, if a user frequently clicks on a certain type of recommended product, the model will increase the weight of the relevant features of this type of product to improve the accuracy of subsequent recommendations; if a user does not perform any operations on the recommended product, the model will adjust the recommendation strategy to reduce the recommendation of similar products. Through this real-time feedback mechanism, the model can continuously adapt to changes in user behavior and continuously improve the recommendation effect.

[0051] 5. Model Evaluation and Optimization

[0052] Model Evaluation: The model is regularly evaluated using a variety of evaluation metrics, including accuracy, recall, F1-score, mean absolute error (MAE), root mean square error (RMSE), etc. On the test set, the model achieved an accuracy of 88%, a recall of 83%, an F1-score of 85.5%, an MAE of 0.05, and an RMSE of 0.08, indicating that the model has high accuracy and reliability in predicting user purchase behavior. At the same time, through user satisfaction surveys, feedback from users on the personalized recommendation service is collected to further evaluate the improvement effect of the model on the user experience.

[0053] Model Optimization: Based on the evaluation results, the model is optimized and adjusted. If it is found that the model has a low accuracy in predicting certain categories of goods, the reason may be that the data volume of this category of goods is small or the features are not obvious enough. Therefore, the data collection volume of this category of goods is increased, and relevant features are redesigned and extracted. In addition, attempts are also made to adjust the model structure, such as increasing the number of layers of the LSTM layer or CNN layer, and adjusting the number of neurons in the fully connected layer. By comparing the performance of different model structures on the validation set, the optimal model structure is selected. At the same time, the training data is updated regularly to reflect the latest changes in user behavior and keep the model in good performance all the time.

[0054] Example Two: User Behavior Analysis of an Online Education Platform

[0055] 1. Data Collection and Preprocessing

[0056] Data Collection: The online education platform collects user behavior data through various methods. On the learning course page, behavior data such as the user's course viewing progress, pause times, fast forward and rewind times, note-taking, questions, and answers are recorded. Basic attribute information of users, such as age, education level, occupation, learning goals, etc., is obtained from the user registration information and learning profiles. At the same time, using the social functions of the platform, interactive data between users, such as likes, comments, and sharing of course content, is collected. In addition, by cooperating with educational industry data platforms, external data such as relevant educational market data and industry trend data are obtained to provide a more comprehensive perspective for user behavior analysis.

[0057] Data Cleaning and Preprocessing: Clean and preprocess the collected data to remove invalid and noisy data. For example, records with extremely short course viewing time (less than 1 minute) and no operations are considered invalid data and deleted. Through data comparison and duplicate checking algorithms, remove duplicate user behavior records. For missing values, handle them according to the importance and relevance of the data. For missing values of important user attribute information, such as education level, learning goals, etc., supplement them through user questionnaires or communication with users; for missing values of some secondary behavior data, such as missing note records, fill them with default values or according to the behavior of similar users. For text data, such as user questions, comments, etc., perform word segmentation, stemming, stop word removal, etc. to convert it into a format suitable for analysis.

[0058] 2. Feature Engineering Construction

[0059] Feature Extraction: According to the characteristics of the online education platform, a series of features related to user learning behavior and learning effect are extracted. The learning activity features include the number of times the user logs in to the platform per week, the daily learning duration, the total course viewing duration, etc. The learning interest preference features are determined by analyzing information such as the course categories, topics, and lecturers watched by the user. For example, if a user frequently watches programming language courses, it indicates that they are interested in programming learning. The learning behavior sequence features record the operation sequence of the user during the learning process, such as the process of first watching the course video, then doing the after-class exercises, and finally taking the exam, as well as the time intervals between each link. In addition, the learning achievement features of the user are also extracted, such as exam scores, homework completion, obtained certificates, etc.

[0060] Feature Selection: Use feature selection methods such as random forest algorithm and gradient boosting tree algorithm to screen out the key features that have an important impact on user learning effect and learning behavior prediction from the numerous extracted features. After screening, 12 key features such as the user's learning duration, course completion rate, exam scores, number of questions asked, and interaction frequency with other users are determined. These key features can effectively reflect the user's learning attitude, learning ability, and learning needs, providing strong support for subsequent model training.

[0061] Feature Transformation: For some continuous features, such as learning duration, test scores, etc., perform normalization processing to scale them to the [0, 1] interval to eliminate the dimensional differences between different features. For categorical features, such as course categories, user occupations, etc., use Label Encoding or One-Hot Encoding methods to convert them into numerical features. At the same time, in order to explore the potential relationships between features, some features are combined and crossed to generate new features. For example, cross the user's learning duration and course difficulty to obtain the learning time distribution feature of the user on courses with different difficulties, further enriching the feature information.

[0062] 3. Deep Learning Model Construction and Training

[0063] Model Construction: A deep learning model based on the Transformer architecture is constructed to analyze the user's learning behavior and predict the learning effect. The Transformer model can effectively handle long-distance dependencies in sequence data and has achieved good results in fields such as natural language processing and time series analysis. In this embodiment, the user's learning behavior sequence data (such as course viewing records, exercise completion records, etc.) is used as input, and through the encoder and decoder structures of the Transformer model, feature extraction and modeling of the data are performed. At the same time, combined with the attention mechanism, the model can automatically focus on the behavioral features that have a greater impact on the learning effect, improving the prediction accuracy of the model. In addition, a fully connected layer and a classifier are added to the model to output the prediction results of the user's learning effect, such as whether the course learning can be completed, whether the expected learning goals can be achieved, etc.

[0064] Model Training: The preprocessed data is divided according to the ratio of 60% training set, 25% validation set, and 15% test set. The training set is used to train the model, setting the number of training epochs to 100 and the learning rate to 0.001. During the training process, the Adagrad optimizer is used to adjust the model parameters, and the mean squared error loss function is minimized to improve the prediction accuracy of the model. To prevent overfitting, the early stopping method (Early Stopping) is adopted, that is, when the loss function value on the validation set no longer decreases for several consecutive rounds, the training is stopped to avoid the model overfitting the training data. At the same time, an L2 regularization term is added to the model to constrain the model parameters and improve the generalization ability of the model. After training, the mean squared error of the model on the validation set is 0.12, and the correlation coefficient (R 2 ) reaches 0.85, indicating that the model can better fit the user's learning behavior data and accurately predict the learning effect.

[0065] 4. Real-Time Analysis and Feedback Mechanism

[0066] Real-time analysis: Establish a real-time data transmission and analysis system on the learning client and server side of the online education platform. When users perform learning operations on the platform, the client sends the user's behavior data to the server side in real time. The server side quickly transmits the data to the analysis model through the message queue for real-time analysis. For example, when users are watching course videos, the system analyzes the user's viewing progress, pause, and fast-forward behaviors in real time to judge the user's learning status and understanding of the course content. If it is found that the user frequently pauses or fast-forwards, it may indicate that the user does not understand some content or feels that the course rhythm is inappropriate. The system will record relevant information in a timely manner and provide a basis for subsequent personalized learning recommendations.

[0067] Personalized learning recommendations: Provide personalized learning recommendation services based on the real-time analysis results. If the system determines that the user has difficulty understanding the current course content being studied, it will recommend relevant basic knowledge courses, explanatory videos, or learning materials to help the user consolidate knowledge. At the same time, according to the user's learning interest preferences and learning progress, recommend advanced courses, extended learning resources, or learning activities suitable for the user. For example, for users who are learning the Python basic course and have a good learning progress, recommend relevant courses such as Python advanced programming and data analysis practice, as well as online programming competition activities to stimulate the user's learning interest and enthusiasm.

[0068] Feedback mechanism: Collect feedback data from users on the learning recommendation content, including whether they click to view the recommended content, whether they study according to the recommendation, and their evaluations after learning. Input these feedback data into the model in real time to update and optimize the model online. For example, if users give good feedback on the recommended courses, the model will increase the weight of the relevant features of such courses and improve the recommendation probability of similar courses; if users are not interested in the recommended content or give poor feedback, the model will analyze the reasons and adjust the recommendation strategy to avoid recommending similar irrelevant content again. Through this real-time feedback mechanism, continuously improve the quality of personalized learning recommendations and the user's learning experience.

[0069] 5. Model evaluation and optimization

[0070] Model evaluation: Comprehensively evaluate the model using a variety of evaluation metrics. In addition to metrics such as mean squared error, R 2 and others, metrics such as mean absolute percentage error (MAPE) and coefficient of determination are also used to measure the prediction accuracy and stability of the model. On the test set, the MAPE of the model is 8%, and the coefficient of determination is 0.82, indicating that the model has high accuracy and reliability in predicting the user's learning effect. At the same time, through the user learning effect tracking survey, compare the model prediction results with the user's actual learning achievements to further verify the effectiveness of the model.

[0071] Model optimization: Optimize and adjust the model based on the evaluation results. If it is found that the model has deviations in the predictions of certain specific user groups (such as users of different age groups and different educational backgrounds), the reason for analysis may be that the model has not fully learned the characteristics of these user groups. Therefore, the data sample size of these user groups is increased in a targeted manner, and the model is retrained and optimized. In addition, attempts are made to adjust the hyperparameters of the Transformer model, such as adjusting the number of attention heads, the dimension of the hidden layer, etc., and by comparing the performance of the model on the validation set under different hyperparameter settings, the optimal hyperparameter combination is selected. At the same time, pay attention to the latest developments in the education industry and changes in user needs, and update training data and model structures in a timely manner to adapt to changing education scenarios and user behaviors.

[0072] It can be seen from the above two specific embodiments that the Internet user behavior precision analysis method based on artificial intelligence of the present invention can be effectively applied on different types of Internet platforms, and through the precise analysis of user behavior data, it can provide valuable decision support for the platform and improve the service quality and user experience of the platform. In practical applications, each link of the method can be appropriately adjusted and optimized according to the characteristics and needs of different platforms to give full play to its advantages.

[0073] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A precise analysis method of Internet user behavior based on artificial intelligence, characterized in that: include: Multi-source data collection and integration steps: Collect user behavior data and basic attribute information from various channels of the Internet platform through a variety of data collection technologies, use data cleaning and preprocessing technologies to remove noise, duplicates, and missing values ​​from the collected data, and use data fusion algorithms to integrate data from different data sources and formats into a unified user behavior data set; Feature engineering construction steps: design and extract multiple user behavior features, Use feature selection algorithms to screen key features, and use feature transformation technology to normalize and standardize some features; Deep learning model construction and training steps: Build a user behavior analysis model based on deep learning, including using recurrent neural networks and their variants or convolutional neural networks, integrating multiple deep learning models, using large-scale user behavior data sets to train the model, and adjusting model parameters through optimization algorithms; Real-time analysis and feedback mechanism steps: Establish a real-time data processing and analysis system, input the newly generated user behavior data into the trained model in a timely manner for real-time analysis, provide personalized services and recommendations to users based on the analysis results, and collect user feedback data and feed it back to the model in real time for online update and optimization; Model evaluation and optimization steps: Use a variety of evaluation indicators to evaluate the trained model, verify the model's generalization ability through cross-validation, and optimize the model based on the evaluation results.

2. According to claim 1, a method for accurately analyzing Internet user behavior based on artificial intelligence is characterized by: In the multi-source data collection and integration step, user behavior data from public pages on the Internet are collected through web crawler technology, and user operation behavior data on specific devices are collected through sensor technology; For different types of data sources, adapted data interface protocols are used for data access to ensure the comprehensiveness and real-time nature of data collection. Data denoising methods based on outlier detection algorithms are used to identify and remove outliers in the data. Missing value filling algorithms based on probability statistical models are used to accurately fill in missing data to improve data quality. Data fusion algorithms based on graph models are used to construct association graphs between data. Based on the weights and connection relationships between nodes, deep fusion of data from different data sources is achieved to explore potential data association patterns. The collected user behavior data include but are not limited to user browsing records, click behaviors, search histories, purchase transaction records, and social interaction information.

3. According to claim 1, a method for accurately analyzing Internet user behavior based on artificial intelligence is characterized by: In the feature engineering construction step, the extracted user behavior features include user activity features, interest preference features, and behavior sequence features. The interactive behavior features of users in virtual reality / augmented reality scenarios are extracted, including gesture operation frequency and perspective switching mode; as well as the behavioral features of transaction verification and information encryption of users in blockchain applications. A feature selection network based on deep learning is used to automatically screen out features that have a key impact on user behavior analysis by constructing a nonlinear mapping relationship between features and target variables. Compared with traditional algorithms, the accuracy and efficiency of feature selection are significantly improved. A feature transformation method based on a generative adversarial network is used to transform unstructured features in user behavior data to generate structured features that meet the model input requirements, thereby enhancing the model's ability to process complex data.

4. According to claim 1, a method for accurately analyzing Internet user behavior based on artificial intelligence is characterized by: In the deep learning model construction and training steps, the optimization algorithm includes stochastic gradient descent and Adam optimizer, and a hybrid deep learning model of Transformer and LSTM integrating attention mechanism is constructed. Transformer is used to capture the global dependency in user behavior data, and LSTM is used to process the time dependency of sequence data. The two work together to improve the modeling ability of complex user behavior. Data enhancement technology is used to randomly crop, flip, and add noise to the training data to expand the diversity of training data and improve the generalization ability of the model. Use a distributed training framework to distribute training tasks in parallel to multiple computing nodes, accelerate the model training process, shorten the training time, and use an adaptive learning rate adjustment algorithm to dynamically adjust the learning rate according to the performance indicators of the model during training, ensuring that the model converges quickly during training and avoids falling into local optimal solutions.

5. According to claim 1, a method for accurately analyzing Internet user behavior based on artificial intelligence is characterized by: In the real-time analysis and feedback mechanism step, a real-time data processing system based on a stream computing architecture is constructed, which has a millisecond-level data processing delay and can process large-scale user behavior data in real time; message queue technology is used for data buffering and asynchronous processing to ensure the high availability and stability of the system, and based on the user's real-time behavior data and historical behavior portraits, a recommendation algorithm combining collaborative filtering and deep learning is used to provide users with personalized product, content or service recommendations; user feedback data on recommended content is collected in real time, and the recommendation model is incrementally updated using an online learning algorithm to continuously optimize the recommendation effect, integrate user feedback data from different terminals and different application scenarios, and use sentiment analysis and behavior sequence analysis techniques to deeply explore the intentions and needs behind user feedback, providing a more comprehensive basis for model optimization.

6. The method for accurately analyzing Internet user behavior based on artificial intelligence according to claim 1, characterized in that: In the model evaluation and optimization steps, the evaluation indicators include accuracy, recall rate, F1 value, and mean square error. Evaluation indicators based on user experience, such as user satisfaction and retention rate, are used in combination with traditional accuracy and recall rate indicators to comprehensively evaluate the performance of the model in actual application scenarios. The adaptability and stability of the evaluation model are customized for different business scenarios. The model verification method based on reinforcement learning is used to dynamically verify the model by simulating the interaction process between the user and the system, and evaluate the decision-making ability and adaptability of the model in complex environments. Cross-domain verification technology is used to apply the model to user behavior data in different fields for verification, expand the generalization ability boundary of the model, and build a model automatic optimization system based on meta-learning, which can automatically adjust the model structure and parameter settings according to the model evaluation results to achieve intelligent optimization of the model. An adaptive model update strategy is used to dynamically optimize the model in real time according to changes in data distribution and adjustments to business needs to ensure that the model always maintains optimal performance.

7. An artificial intelligence-based precise analysis system for Internet user behavior, characterized in that: include: The data collection and integration module is used to perform multi-source data collection and integration steps, that is, to collect user behavior data and basic attribute information from various channels of the Internet platform through various data collection technologies, use data cleaning and preprocessing technologies to remove noise, duplicates, and missing values ​​from the collected data, and use data fusion algorithms to integrate data from different data sources and formats into a unified user behavior data set; The feature engineering module is used to perform the feature engineering construction step, that is, to design and extract a variety of user behavior features. Use feature selection algorithms to screen key features, and use feature transformation technology to normalize and standardize some features; The model building and training module is used to perform deep learning model building and training steps.

Images