CN117763230B

CN117763230B - Data analysis method and system based on neural network model

Info

Publication number: CN117763230B
Application number: CN202311768273.6A
Authority: CN
Inventors: 符珊
Original assignee: Hainan Ningningqi Technology Co ltd
Current assignee: Hainan Ningningqi Technology Co ltd
Priority date: 2023-12-21
Filing date: 2023-12-21
Publication date: 2024-08-23
Anticipated expiration: 2043-12-21
Also published as: CN117763230A

Abstract

The present invention relates to the field of data processing technologies, and in particular, to a data analysis method and system based on a neural network model. The method comprises the following steps: collecting user information data of the network information platform to generate user information data; abnormal user information data is removed from the user information data, and safe user information data is generated; collecting user browsing data according to the safe user information data to generate user browsing data; performing word frequency data clustering processing of the user browsing data to generate user browsing word frequency clustering data; user browsing interest type analysis is carried out on the user browsing data according to the user browsing word frequency clustering data, and user browsing interest type data is generated; and generating optimized browsing type interest scoring data based on the interest scoring optimization analysis of the browsing type of the user browsing interest type data row. The invention eliminates the analyzed user abnormal data and realizes more accurate user interest type data analysis.

Description

Data analysis method and system based on neural network model

Technical Field

The present invention relates to the field of data processing technologies, and in particular, to a data analysis method and system based on a neural network model.

Background

The neural network can effectively capture complex nonlinear data relationships, can automatically extract useful features in data without manual intervention, has remarkable advantages especially when processing high-scale and large-scale data, can process multi-modal data, integrates different data types, is suitable for multiple fields such as natural language processing, computer vision, voice recognition and the like, can cope with changeable data in the aspect of large-scale data processing, also shows high efficiency, can extract valuable information, find modes and make predictions from the data based on the neural network model, helps make evidence-based decisions, and is beneficial to solving problems and optimizing processes. However, the conventional neural network model-based data analysis method does not analyze data anomalies through interactive data in terms of users, programs, and the like, so that the analysis efficiency is poor, and task events are not analyzed through analysis of user behavior data, so that the data analysis result is inaccurate.

Disclosure of Invention

Based on the above, the present invention provides a data analysis method and system based on a neural network model, so as to solve at least one of the above technical problems.

In order to achieve the above object, a data analysis method based on a neural network model includes the following steps:

Step S1, user information data acquisition is carried out on a network information platform by utilizing a preset web crawler program, and user information data are generated; collecting user interaction frequency data according to the user information data to generate user interaction frequency data; carrying out abnormal interaction frequency data analysis on the user interaction frequency data to generate abnormal interaction frequency data;

S2, carrying out interaction frequency fluctuation analysis of a time window on abnormal interaction frequency data according to a preset time window to generate window fluctuation data; extracting abnormal interaction frequency data with window fluctuation deviation according to the window fluctuation data to generate dangerous interaction frequency data; abnormal user information data is removed from the user information data according to the dangerous interaction frequency data, and safe user information data is generated;

S3, user browsing data acquisition is carried out according to the safe user information data, and user browsing data are generated; performing word frequency data clustering processing of user browsing data on the user browsing data based on a pre-trained K-Means clustering model, and generating user browsing word frequency clustering data; performing clustering tag design according to word frequency clustering data browsed by a user to generate clustering tag data;

S4, analyzing the user browsing interest types of the user browsing data according to the clustering label data to generate user browsing interest type data; user pushing data acquisition is carried out based on user browsing interest type data, user pushing data are generated, and the user pushing data are fed back to the terminal;

S5, when a user browses the user push data through the terminal, the built-in monitoring equipment of the terminal is utilized to collect the pupil image data of the user, and the pupil image data of the user is generated; performing pupil scaling degree real-time calculation on the pupil image data of the user to generate real-time pupil scaling data; integrating the real-time pupil scaling data with real-time user pushing data in a corresponding sequence to generate real-time pupil-pushing data;

Step S6, pupil interest score calculation is carried out on the real-time pupil image data, and real-time pupil interest score data are generated; and performing interest score optimization analysis of the browsing type based on the real-time pupil score data and the real-time pupil-pushing data, and generating optimized browsing type interest score data.

The invention further analyzes the behavior of the user on the network platform through the user information data collected by the web crawler program, and the invention comprises interests, click modes and the like of the user, thereby being beneficial to knowing the behavior mode of the user. The user interaction frequency data is acquired according to the user information data, so that the behavior of the user on the network information platform, including interests, interests and interaction modes of the user, can be known in depth, and the user requirements and behavior characteristics can be better understood. By analyzing the abnormal interaction frequency data of the user interaction frequency data, abnormal behaviors such as abnormal high-frequency interaction or abnormal mode interaction can be identified, potential fraud, malicious behaviors or abnormal activities can be found early, and therefore safety of the platform is enhanced. the abnormal interaction frequency data is analyzed through a preset time window, the time correlation of the user behavior is identified, and abnormal activities in a specific time period are found, so that potential risks are more accurately located. The window fluctuation data generated by the interaction frequency fluctuation analysis of the time window provides information about the variation trend of the user behavior, so that the dynamic characteristics of the user behavior can be better understood, and a basis is provided for subsequent risk assessment and prediction. By analyzing the window fluctuation data, abnormal interaction frequency data, namely dangerous interaction frequency data, of window fluctuation deviation can be extracted, possible safety risks can be rapidly identified, and sensitivity of the system to potential threats is improved. According to the dangerous interaction frequency data, abnormal information corresponding to the user information data can be removed, abnormal user information is removed, and therefore safer and more reliable user information data are generated, and the overall safety of the platform is guaranteed. And acquiring specific browsing behaviors of the user on the platform by acquiring user browsing data according to the safe user information data. The interests and preferences of the user are more fully understood, providing basic data for personalized recommendations. The word frequency data clustering processing is carried out on the user browsing data by utilizing a pre-trained K-Means clustering model, so that similar user browsing behaviors can be classified into the same cluster, the commonality in the user group can be found, and guidance is provided for subsequent personalized recommendation and content optimization. The clustering labels are designed based on the word frequency clustering data browsed by the user, namely each cluster is assigned with a representative label. This helps to simplify understanding of user behavior, making the user population easier to resolve and understand. By analyzing the user browsing data according to the clustering label data, the browsing interest types of the users are identified and understood, the interest field of the users is deeply known, and more accurate guidance is provided for personalized recommendation. The collection of user push data is performed based on user browsing interest type data to provide content that meets the user's interests. This helps to improve the user experience, making it more likely that the user is interested in pushing content, thereby increasing user engagement and retention. the generated user push data is fed back to the terminal, which means that the user can directly receive personalized push content on the terminal equipment, so that the satisfaction degree of the user is improved, and the user obtains customized information instead of general push. The pupil image data acquisition is carried out on the user by using the built-in monitoring equipment of the terminal, so that a physiological signal can be provided, the physiological state of the user when browsing the push content is reflected, and the attention and the excitation degree of the user are more comprehensively known. Calculating pupil compression in real-time for user pupil image data can reflect the user's level of attention to different pushed content, with a larger pupil compression possibly indicating that the user is interested in a particular content and a smaller pupil compression possibly indicating a lower attention. And integrating the real-time pupil scaling data with the real-time user push data to generate real-time pupil-push data, and establishing the association between the user pupil scaling and the push content, so that the response of the user when browsing the push content is better understood. And (3) performing pupil interest score calculation on the real-time pupil image data, so that the interest degree of the user on different push contents can be quantified. This provides an objective indicator reflecting the actual degree of attention of the user to the particular content. Based on the real-time pupil score data and the real-time pupil-pushing data, optimization analysis of browsing interest types can be performed, and the preference and interest types of the user can be better understood by analyzing the real-time response of the user, so that the pushing content is adjusted, and the accuracy of personalized recommendation is improved. Therefore, the data analysis method based on the neural network model can analyze data anomalies through interaction data in the aspects of users, programs and the like, so that the analysis efficiency is improved, task events can be analyzed according to analysis of user behavior data, and the data analysis result is accurate.

The present specification provides a data analysis system based on a neural network model, for performing the data analysis method based on a neural network model as described above, the data analysis system based on a neural network model comprising:

the user information acquisition module is used for acquiring user information data of the network information platform by utilizing a preset web crawler program to generate user information data; collecting user interaction frequency data according to the user information data to generate user interaction frequency data; carrying out abnormal interaction frequency data analysis on the user interaction frequency data to generate abnormal interaction frequency data;

The dangerous user information analysis module is used for carrying out interaction frequency fluctuation analysis of a time window on the abnormal interaction frequency data according to a preset time window to generate window fluctuation data; extracting abnormal interaction frequency data with window fluctuation deviation according to the window fluctuation data to generate dangerous interaction frequency data; abnormal user information data is removed from the user information data according to the dangerous interaction frequency data, and safe user information data is generated;

The user browsing information clustering module is used for collecting user browsing data according to the safe user information data and generating user browsing data; performing word frequency data clustering processing of user browsing data on the user browsing data based on a pre-trained K-Means clustering model, and generating user browsing word frequency clustering data; performing clustering tag design according to word frequency clustering data browsed by a user to generate clustering tag data;

The user browsing data pushing module is used for carrying out user browsing interest type analysis on the user browsing data according to the clustering tag data to generate user browsing interest type data; user pushing data acquisition is carried out based on user browsing interest type data, user pushing data are generated, and the user pushing data are fed back to the terminal;

The user pupil image acquisition module is used for acquiring user pupil image data by using the built-in monitoring equipment of the terminal when the user browses the user push data through the terminal, so as to generate user pupil image data; performing pupil scaling degree real-time calculation on the pupil image data of the user to generate real-time pupil scaling data; integrating the real-time pupil scaling data with real-time user pushing data in a corresponding sequence to generate real-time pupil-pushing data;

The user browsing interest type analysis module is used for carrying out pupil interest score calculation on the real-time pupil image data to generate real-time pupil interest score data; and performing interest score optimization analysis of the browsing type based on the real-time pupil score data and the real-time pupil-pushing data, and generating optimized browsing type interest score data.

The application has the advantages that the network information platform is subjected to user information data acquisition by utilizing the web crawler program to form a comprehensive user information database, the user interaction data acquisition is carried out to obtain detailed interaction behavior records of the user, the interaction strength of the user and the platform is effectively quantized through interaction frequency calculation, the abnormal analysis of the user interaction frequency is carried out to generate accurate abnormal interaction frequency data, potential risks and abnormal behaviors can be sharply captured, a comprehensive user information file is built, the deep insight of the user behaviors is realized, and a solid foundation is provided for subsequent analysis and optimization. The accurate division of the time window and the deep analysis of the abnormal interaction frequency data realize the careful monitoring of the user behavior, the abnormal interaction frequency data are divided by the time window, the abnormal interaction frequency data are generated, the window fluctuation analysis is carried out, the window fluctuation data are formed, the comparison standard with the conventional behavior is established by designing the conventional window fluctuation interval, and the analysis accuracy is further improved. When the window fluctuation data deviate from a conventional window fluctuation interval, the corresponding abnormal interaction frequency data are marked as dangerous interaction frequency data, potential risk behaviors can be captured timely, user information data are screened through the dangerous interaction frequency data, safe user information data are generated, a reliable data basis is provided for subsequent analysis and recommendation, monitoring and processing capabilities of the abnormal behaviors are enhanced, and the level of guaranteeing the safety of a platform to users is improved. The safe user information data is utilized to collect user browsing data, thus forming detailed user browsing data, providing rich information basis for subsequent analysis, converting word frequency data of the user browsing data by natural language technology, converting text information into a machine processable form, better understanding the browsing content of the user, clustering the user browsing word frequency data by a pre-trained K-Means clustering model, forming user browsing word frequency clustering data, effectively classifying similar browsing behaviors together, extracting characteristics of the clustering data by a principal component analysis method, generating clustering characteristic data, the method is beneficial to dimension reduction and extraction of key information, performs clustering label design according to the clustering feature data, endows meaningful labels for user browsing behaviors, improves the interpretability and interpretability of user interests, realizes fine analysis of the user interests by deep mining of the user browsing behaviors, and provides more accurate and deep guidance for personalized recommendation. The user browsing data is classified according to the clustering label data, the user behaviors are finely divided, and the user browsing interest type data is formed by carrying out user browsing interest type analysis on the classified user browsing data, so that the interest preference of the user is deeply revealed. Through the push type interest scoring design, interest scores of users are given to different types of push contents, the individuation degree of the push contents is improved, user push data acquisition is carried out based on the push type interest scoring data, user push data are generated and fed back to the terminal, recommended contents which are more in line with the interests are provided for the users, the interests of the users are deeply analyzed, the push strategy is optimized, the relevance and the user satisfaction degree of the push contents are improved, and the interactive experience of the users and the platform is further enhanced. by performing personalized pupil scaling limit analysis according to the pupil image data of the user, personalized pupil scaling limit data is established for different users, physiological differences of the users are considered, and analysis accuracy is improved. And when the pupil image data of the user is collected in real time, pupil scaling degree calculation is carried out on the real-time pupil image data through personalized pupil scaling limit data, so that real-time pupil scaling data is generated, and physiological objective indexes are provided for real-time user feedback. The real-time pupil scaling data and the real-time user pushing data are subjected to data integration of a corresponding sequence, so that the real-time pupil-pushing data are generated, the interests and the focus points of the user are more comprehensively understood to provide beneficial information, the perception of the real-time interests of the user is realized through comprehensive analysis of pupil image data, and scientific basis is provided for individuation of pushing content and improvement of user experience. The pupil interest score calculation is carried out on the real-time pupil image data through the pupil interest score algorithm, the system can accurately quantify the interest degree of the user on the current content, provide physiological objective basis for personalized pushing, carry out real-time browsing type interest score weighting on the real-time pupil-pushing data through the real-time pupil score data, balance the interest degree of different pushing contents according to the actual physiological reaction of the user, improve the personalized degree of pushing, comprehensively consider the real-time browsing type interest score data and the pushing type interest score data, realize real-time optimization on browsing interest types, generate optimized browsing interest type data, The real-time adjustment and the user experience of the push content are improved, and the perception of the real-time interest of the user and the real-time optimization of the push strategy are realized through the comprehensive analysis of the physiological response data, so that the individuation degree of the push content and the satisfaction degree of the user are improved.

Drawings

FIG. 1 is a schematic flow chart of steps of a data analysis method based on a neural network model according to the present invention;

FIG. 2 is a flowchart illustrating the detailed implementation of step S2 in FIG. 1;

FIG. 3 is a flowchart illustrating the detailed implementation of step S3 in FIG. 1;

FIG. 4 is a flowchart illustrating the detailed implementation of step S4 in FIG. 1;

FIG. 5 is a flowchart illustrating the detailed implementation of step S5 in FIG. 1;

FIG. 6 is a flowchart illustrating the detailed implementation of step S6 in FIG. 1;

The achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.

Detailed Description

The following is a clear and complete description of the technical method of the present patent in conjunction with the accompanying drawings, and it is evident that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, are intended to fall within the scope of the present invention.

Furthermore, the drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus a repetitive description thereof will be omitted. Some of the block diagrams shown in the figures are functional entities and do not necessarily correspond to physically or logically separate entities. The functional entities may be implemented in software or in one or more hardware modules or integrated circuits or in different networks and/or processor methods and/or microcontroller methods.

It will be understood that, although the terms "first," "second," etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another element. For example, a first element could be termed a second element, and, similarly, a second element could be termed a first element, without departing from the scope of example embodiments. The term "and/or" as used herein includes any and all combinations of one or more of the associated listed items.

In order to achieve the above objective, referring to fig. 1 to 4, the present invention provides a data analysis method based on a neural network model, comprising the following steps:

In the embodiment of the present invention, as described with reference to fig. 1, a step flow diagram of a data analysis method based on a neural network model of the present invention is provided, and in the embodiment, the data analysis method based on the neural network model includes the following steps:

In the embodiment of the invention, a web crawler program, such as Beautiful Soup library in Python, is adopted to collect user information data of the network information platform, including obtaining basic information, release content and interaction condition with other users. Through this process, a piece of user information data containing rich data is obtained. Based on the collected user information data, data acquisition of user interaction frequency is carried out, and interaction behaviors of users such as praise, comment, sharing and the like on a platform are monitored, so that user interaction frequency data is established, and interaction conditions between the users and the platform and between the users and other users are recorded in detail. After obtaining the user interaction frequency data, performing analysis of abnormal interaction frequency data, wherein the analysis stage involves statistical processing of the user interaction frequency, such as calculating average value, standard deviation and the like, so as to identify abnormal frequency deviating from a normal range, by setting a threshold value, we can determine which user interaction frequency belongs to abnormal conditions, and generate abnormal interaction frequency data, a Z score (Z-score) method is used to convert the interaction frequency of each user into a standardized score, and then the abnormal interaction frequency data is screened out by setting a threshold value (such as Z score is greater than 2 or less than-2). This in-depth and detailed embodiment ensures our sensitivity to user behavior, enables accurate capture of anomalies, and provides reliable underlying data for subsequent analysis.

In the embodiment of the invention, the interaction frequency fluctuation analysis of the time window is carried out on the abnormal interaction frequency data according to the preset time window, the abnormal interaction frequency data is grouped according to the daily interval by taking the daily as an example, the frequency fluctuation in each time window is calculated, and the method is realized by using the techniques of moving average, sliding window and the like, so that the time change trend is ensured to be captured. And extracting abnormal interaction frequency data with deviated window fluctuation from the abnormal interaction frequency data according to the window fluctuation data, and determining which abnormal interaction frequency data deviate from a normal fluctuation range in a specific time window by setting a threshold value of the window fluctuation, so as to generate dangerous interaction frequency data. After dangerous interaction frequency data are obtained, the processing of the user information data is turned to, and the user information related to the abnormal frequency data is removed by correlating the dangerous interaction frequency data, so that safe user information data are generated, the health and reliability of a data set are ensured, and the accuracy of subsequent analysis is further improved.

In the embodiment of the invention, the user browsing data acquisition is performed based on the safe user information data. By monitoring the browsing behavior of the user on the platform, including clicking links, viewing content and the like, the browsing data of the user is established, and the interests and activities of the user on different contents on the platform are recorded in detail. And performing word frequency data clustering processing on the user browsing data by using a pre-trained K-Means clustering model, converting the user browsing data into text data, and then clustering by using a K-Means algorithm. The K value here may be determined by cross-validation or the like, the best clustering effect. After obtaining word frequency clusters of user browsing data, we have conducted cluster tag design, and by analyzing keywords in each cluster, we have designed a representative tag for each cluster, reflecting the user browsing interests represented by the cluster. This step ensures a more intuitive and specific understanding of the user's interests in the subsequent analysis. For example, a scikit-learn library in Python is used for K-Means clustering, user browsing data are converted into a TF-IDF (Term Frequency-Inverse Document Frequency) matrix, then a K-Means algorithm is applied for clustering, and clustering labels are designed by analyzing keywords of each clustering cluster, so that the category of browsing interest content of a user is carefully known, and a foundation is laid for subsequent personalized recommendation and analysis.

in the embodiment of the invention, the user browsing interest type analysis is carried out on the user browsing data according to the clustering tag data, and the user browsing data and the clustering tag data are associated, for example, the clustering tag is an astronomical content clustering tag, the corresponding astronomical content is classified into astronomical browsing data, and the browsing interest type of each user is determined. And collecting user pushing data based on the user browsing interest type data. By associating the user browsing interest types with the related push content, user push data is generated, which includes various types of information pushed for the user browsing interest types according to the user interest types, and the user push data is fed back to the user terminal so as to provide personalized content which accords with the user interests. For example, by using tools such as SQL or Pandas, the user browsing data and the clustering tag data are associated to obtain browsing interest types corresponding to each user, and push contents related to the user browsing interest types are collected through a push system or an API of the monitoring platform to form user push data.

In the embodiment of the invention, when a user browses the user to push data through the terminal, pupil image data acquisition is carried out on the user by using the built-in monitoring equipment of the terminal, and the application of the terminal camera is involved, and pupil data is acquired by acquiring the image of eyes of the user. The pupil image data of the user is subjected to real-time calculation of the pupil scaling degree, the diameter of the pupil can be measured by using a computer vision technology, the pupil scaling degree is calculated according to the diameter of the pupil, and the real-time calculation ensures that specific states of the pupil of the user at different time points can be captured. And integrating the real-time pupil scaling data with the real-time user pushing data in a corresponding sequence to generate real-time pupil-pushing data, and corresponding the pupil data with the pushing content browsed by the user through a time stamp or other identifiers to form a comprehensive real-time data set which reflects the dynamic change of the pupil when the user browses the pushing content. For example, an image processing library such as OpenCV may be used, an eye image of the user is acquired through a terminal camera, the pupil scaling degree is calculated by using an image processing technology, and pupil scaling data and user push data are integrated according to a timestamp to form real-time pupil-push data.

In the embodiment of the invention, the pupil interest score calculation is performed on the real-time pupil image data to generate the real-time pupil interest score data, which reflects the intensity and frequency of pupil change when the user browses the push content, and is a key index for measuring the interest degree of the user, for example, the pupil zoom degree of the content of interest can be greatly changed by the user, and the content of no interest is vice versa. Real-time pupil image data is processed by using a computer vision algorithm, such as indexes of pupil area change, and real-time pupil interest scoring data can be obtained by quantifying the indexes, so that the interest level of a user in the current content can be measured. The method comprises the steps of carrying out optimization analysis on browsing interest types based on real-time pupil score data and real-time pupil-pushing data, generating optimized browsing interest type data, optimizing the classification of pushing content by analyzing pupil interest scores of users under different interest types, ensuring pushing more in line with real-time interests of the users, for example, calculating pupil interest scores by utilizing an image processing library in Python such as OpenCV, analyzing pupil changes, and analyzing pupil interest scores of the users under different interest types by correlating with the real-time pupil-pushing data, thereby ensuring that real-time interest changes of the pushing content by the users can be accurately captured, and continuously optimizing interest analysis of personalized browsing content of the users, so as to generate optimized browsing interest type data.

Preferably, step S1 comprises the steps of:

step S11, user information data acquisition is carried out on the network information platform by utilizing a preset web crawler program, and user information data are generated;

Step S12, user interaction data acquisition is carried out on the user information data to generate user interaction data;

Step S13, performing interaction frequency calculation according to the user interaction data to generate user interaction frequency data;

and S14, carrying out abnormal interaction frequency data analysis on the user interaction frequency data to generate abnormal interaction frequency data.

The invention collects the user information data of the network information platform through the web crawler program, and can acquire a large amount of user related information including personal data, hobbies and interests, interaction frequency using the network information platform and the like. This helps to build a user representation, improving the overall knowledge of the user population. The interactive data acquisition is carried out on the user information data, so that the actual operation behaviors of the user on the platform, such as clicking, browsing and the like, can be captured, and the analysis of whether the user is a real person or not is facilitated, and the network platform behavior interaction is carried out, so that a foundation is provided for subsequent analysis. And carrying out interaction frequency calculation according to the user interaction data to generate the user interaction frequency data, providing quantitative evaluation on the user liveness and the interest degree of specific content, and providing important basis for personalized recommendation. And carrying out abnormal interaction frequency data analysis on the user interaction frequency data to generate abnormal interaction frequency data, and detecting and analyzing abnormal interaction behaviors to improve the perception of potential risks and enhance the safety of the platform.

In the embodiment of the invention, a preset web crawler program, such as Beautiful Soup library in Python, is used to collect user information data from the network information platform, and through accessing the user page, personal data, browsing data, interactive data and the like of the platform, rich user information including user ID, geographic position, age, gender and the like is obtained, so that an exhaustive user information data set is formed. And collecting user interaction data of the collected user information data. This includes monitoring various interactions of the user on the platform, such as posting content, praise, comments, etc., to form comprehensive user interaction data. And calculating the interaction frequency based on the obtained user interaction data, and generating the interaction frequency data of the user by counting the interaction times of the user in a certain time. And carrying out data analysis of abnormal interaction frequency on the user interaction frequency data. By using statistical methods, such as Z-scores, it is identified which users' interaction frequencies deviate from the normal range, and too high interaction frequencies are identified as robot operations, generating abnormal interaction frequency data.

Preferably, step S14 comprises the steps of:

And carrying out data comparison on the user interaction frequency data by utilizing a preset safe interaction frequency threshold value, marking the user interaction frequency data as abnormal interaction frequency data when the user interaction frequency data is larger than the safe interaction frequency threshold value, and marking the user interaction frequency data as conventional interaction frequency data when the user interaction frequency data is not larger than the safe interaction frequency threshold value.

According to the invention, a standard can be set by using the preset safe interaction frequency threshold value, the user interaction frequency data is divided into two types of abnormality and routine, and the comparison is helpful for determining what interaction frequency is considered to be abnormality, so that the sensitivity to potential risks is improved. When the user interaction frequency data is larger than the safety interaction frequency threshold value, the user interaction frequency data is marked as abnormal interaction frequency data, so that user behaviors possibly at risk can be rapidly identified, and the detection accuracy of abnormal interaction is improved. When the user interaction frequency data is not greater than the safety interaction frequency threshold value, the user interaction frequency data is marked as conventional interaction frequency data, so that the record of normal user behaviors is reserved, the possibility of false alarm is reduced, and the system is ensured to judge the user behaviors more accurately.

In the embodiment of the invention, a preset safe interaction frequency threshold is set, and the threshold can be adjusted according to historical data, service requirements and characteristics of a specific platform, for example, the safe interaction frequency threshold can be set to be a standard deviation compared with the average interaction times per hour. And (3) carrying out data comparison on the user interaction frequency data, and calculating whether the interaction frequency of each user is larger than a set safe interaction frequency threshold value in each time window, wherein the calculation can be realized by using a statistical method such as Z score (Z score), namely, the interaction frequency of each user is standardized and compared with the set threshold value. When the user interaction frequency data is greater than the safety interaction frequency threshold, marking the user interaction frequency data as abnormal interaction frequency data, wherein the abnormal interaction frequency data corresponds to the robot operation which is possibly not artificial. Otherwise, when the user interaction frequency data is not greater than the safety interaction frequency threshold value, the data is marked as conventional interaction frequency data, and the conventional interaction frequency data corresponds to conventional operation of a user. For example, using pandas and numpy libraries in Python for data processing and calculation, calculating the Z-score for each user, and then comparing with a set safe interaction frequency threshold to determine whether to mark it as abnormal interaction frequency data.

Preferably, step S2 comprises the steps of:

S21, carrying out time window division processing on abnormal interaction frequency data according to a preset time window to generate divided abnormal interaction frequency data, and carrying out interaction frequency fluctuation analysis on the divided abnormal interaction frequency data to generate window fluctuation data;

s22, designing a conventional window fluctuation interval according to conventional interaction frequency data to generate a conventional window fluctuation interval;

Step S23, window fluctuation interval comparison is carried out on window fluctuation data according to a conventional window fluctuation interval, and when the window fluctuation data deviate from the conventional window fluctuation interval, abnormal interaction frequency data corresponding to the window fluctuation data are marked as dangerous interaction frequency data;

and step S24, eliminating abnormal user information data from the user information data according to the dangerous interaction frequency data to generate safe user information data.

According to the method, the time window division processing and the fluctuation analysis are carried out on the abnormal interaction frequency data, so that the time dynamics of the abnormal interaction behavior can be known more carefully. This helps to capture transient fluctuations in abnormal behavior and improve the perceptibility of potential risks. The conventional window fluctuation interval is designed according to conventional interaction frequency data, so that the fluctuation range of normal user behaviors can be more flexibly adapted, the false alarm rate can be reduced, and the normal user behaviors are ensured not to be misjudged as abnormal. And comparing the window fluctuation data with a conventional window fluctuation interval, and when the data deviate from the conventional interval, marking the corresponding abnormal interaction frequency data as dangerous interaction frequency data, so that the system can be helped to quickly identify and mark user behaviors possibly at risk, such as irregular interaction behaviors such as a large number of automatic clicks and the like in a short time through a robot, and the sensitivity of the system to potential threats is improved. Abnormal user information rejection is carried out on the user information data according to the dangerous interaction frequency data, and safe user information data is generated, so that the overall safety of the platform is guaranteed, and the influence of potential threats on users and systems is reduced.

As an example of the present invention, referring to fig. 2, a detailed implementation step flow diagram of step S2 in fig. 1 is shown, where step S2 includes:

In the embodiment of the invention, the abnormal interaction frequency data is divided according to the preset time window, the abnormal interaction frequency data is divided according to the time window per hour, the abnormal interaction frequency in each time window is counted, the interaction frequency fluctuation analysis of the time window is carried out, and the window fluctuation data is generated by calculating the fluctuation index, for example, the interaction frequency of a user in each hour is a data point, and the fluctuation index is calculated according to the interaction frequency data point.

In the embodiment of the invention, the design of the conventional window fluctuation interval is carried out according to the conventional interaction frequency data, for example, the interaction frequency of the conventional interaction frequency data in each hour is data point, the interaction frequencies of different users in each hour are different, and a conventional multiple which accords with the conventional is enlarged according to the range to construct a conventional window fluctuation interval.

In the embodiment of the invention, window fluctuation data is compared with a conventional window fluctuation interval. When the window fluctuation data deviates from the conventional window fluctuation interval, the abnormal interaction frequency data corresponding to the window fluctuation data is marked as dangerous interaction frequency data, and by considering the relative deviation of the fluctuation in the window, for example, the interaction frequency of one window of the window fluctuation data exceeds the limit range of the conventional window fluctuation interval, the action can be non-manual operation, the operation is marked as mechanical operation, and the corresponding abnormal interaction frequency data is marked as dangerous interaction frequency data.

In the embodiment of the invention, abnormal user information data is removed from the user information data according to the dangerous interaction frequency data, so as to generate safe user information data, and the dangerous interaction frequency data is associated to the user information data, so that the user information data can be performed by non-manual operation, and the corresponding user information data performed by the non-manual operation is provided, thereby improving the filtering accuracy of the system on potential risks and generating the safe user information data.

Preferably, step S3 comprises the steps of:

S31, collecting user browsing data according to the safe user information data to generate user browsing data;

s32, performing word frequency data conversion on the user browsing data by using a natural language technology to generate user browsing word frequency data;

s33, performing word frequency data clustering processing on the user browsing word frequency data based on a pre-trained K-Means clustering model to generate user browsing word frequency clustering data;

Step S34, feature extraction is carried out on the word frequency cluster data browsed by the user by utilizing a principal component analysis method, and cluster feature data is generated;

And step S35, carrying out cluster tag design according to the cluster feature data to generate cluster tag data.

According to the invention, the user browsing data is acquired according to the safe user information data, so that the actual browsing behavior of the user on the platform can be obtained, the analysis of more accurate user interest directions is facilitated, and the recommendation accuracy is improved. The word frequency data conversion is carried out on the user browsing data by utilizing the natural language technology, the text information can be converted into a machine processable form, more structured data is provided for subsequent clustering processing, and the clustering accuracy is improved. Based on a pre-trained K-Means clustering model, clustering processing is carried out on the user browsing word frequency data, and similar user browsing behaviors are divided into the same cluster, so that the user interest group can be found, and more targeted information can be provided for personalized recommendation. And feature extraction is performed on the word frequency clustering data browsed by the user by using a principal component analysis method, so that the dimensional efficiency of the data is improved, key information is reserved, the data dimension is reduced, model training is accelerated, and the clustering effect is improved. And (3) carrying out cluster label design according to the cluster characteristic data, endowing each cluster with a representative label, simplifying the understanding of user behaviors through the design of the cluster characteristic data jin and g label, and providing more visual information for a subsequent recommendation system.

As an example of the present invention, referring to fig. 3, a detailed implementation step flow diagram of step S3 in fig. 1 is shown, where step S3 includes:

In the embodiment of the invention, the acquisition of the user browsing data is performed according to the safe user information data. By monitoring the browsing behavior of the user on the platform, including clicking links, viewing content, etc., we have established user browsing data detailing the user's interests and activities for different content on the platform.

In the embodiment of the invention, the word frequency data conversion is carried out on the user browsing data by utilizing the natural language technology, the word frequency statistics is carried out by converting the user browsing data into text data and utilizing the natural language processing technology such as NLTK or Spacy, the word frequency data browsed by the user is generated, and the keyword information of the content browsed by the user is captured.

In the embodiment of the invention, a Elbow method, a Silhouette score or other suitable cluster evaluation method is utilized to select a proper cluster number K, which is a key decision in K-Means clustering and affects the quality of a final cluster result, wherein the cluster number K is obtained by carrying out model training on a K-Means cluster model through a training sample corresponding to user browsing word frequency data, the training sample is obtained by analyzing user historical data, carrying out cluster analysis on the user browsing word frequency data through the trained K-Means cluster model, analyzing the cluster result, and dividing the content browsed by a user into different clusters to form user browsing word frequency cluster data, which is helpful for better understanding the focus point of user interest, and checking keywords or topics in each cluster so as to allocate a proper label for each cluster. This can be done by looking at the center point of each cluster (cluster center) or other statistical information of the clustering result.

In the embodiment of the invention, the feature extraction is carried out on the word frequency clustering data browsed by the user by utilizing a principal component analysis method, the principal component analysis helps us find key features in the word frequency data browsed by the user, the dimension of the data is reduced, the abstract representation of the browsing interest of the user is improved, and the text with the key features of each clustering result is obtained to generate clustering feature data.

In the embodiment of the invention, the clustering label design is carried out according to the clustering characteristic data, and the representative label is designed for each cluster by analyzing the key characteristic in each cluster, for example, the cluster characteristic of one cluster is related to astronomy, then the cluster is marked as astronomy label, the browsing interest of the user represented by the cluster is reflected, and the subsequent analysis of the user interest is more visual and specific.

Preferably, step S4 comprises the steps of:

S41, performing browsing data classification processing on the user browsing data according to the clustering label data to generate classified user browsing data;

Step S42, analyzing the type of the user browsing interest according to the classified user browsing data to generate user browsing interest type data;

Step S43, carrying out push type interest scoring design according to user browsing interest type data to generate push type interest scoring data;

And S44, collecting user push data based on the push type interest scoring data, generating user push data, and feeding the user push data back to the terminal.

According to the method and the device for classifying the user browsing data, the user browsing data are classified according to the clustering tag data, so that the classified user browsing data are generated, the user browsing behavior can be more accurately classified into the specific interest type, and the detailed understanding of the user interests is improved. And analyzing the user browsing interest types according to the classified user browsing data, generating the user browsing interest type data, helping to understand the interest field of the user deeply, and providing more personalized guidance for subsequent recommendation. And designing push type interest scores according to the user browsing interest type data, and giving interest scores to different types of push contents to the user, so that the interest degree of the user in different push types can be estimated more accurately. User push data acquisition is carried out based on push type interest scoring data, user push data are generated and fed back to the terminal, personalized push content which better accords with user interests can be provided, and user participation and satisfaction are improved.

As an example of the present invention, referring to fig. 4, a detailed implementation step flow diagram of step S4 in fig. 1 is shown, where step S4 includes:

in the embodiment of the invention, the browsing data of the user is classified according to the clustering tag data, the browsing data of the user is classified into different browsing categories by associating the user browsing data with the clustering tag data generated before, the browsing categories are preset by a network platform, and the browsing categories corresponding to the clustering tag data are classified, so that each category can represent one browsing interest type of the user on the platform.

In the embodiment of the invention, the user browsing interest type analysis is performed according to the classified user browsing data, and statistics and analysis are performed on each browsing classified data, so that the user browsing interest type data is generated, for example, the browsing preference of the user under different interest classifications is described based on the access type times corresponding to the user browsing data.

In the embodiment of the invention, the push type interest scoring design is carried out according to the user browsing interest type data, and a set of scoring mechanism is designed to quantitatively evaluate the push interests of each user under different browsing interest types, so that the push type interest scoring data is generated, and the interest degree of the user on different push types is reflected.

In the embodiment of the invention, the user push data acquisition is performed based on the push type interest scoring data. The designed scoring mechanism selects push content meeting the user interests from the platform, generates user push data and feeds the data back to the user terminal, so that individuation of the push content and meeting the user interests are ensured, and user experience is improved.

Preferably, step S5 comprises the steps of:

Step S51, when a user browses the user push data through the terminal, the built-in monitoring equipment of the terminal is utilized to collect the pupil image data of the user, and the pupil image data of the user is generated;

step S52, personalized pupil constriction limit analysis is carried out according to the pupil image data of the user, and personalized pupil constriction limit data is generated;

step S53, collecting pupil image data of a user in real time to generate real-time pupil image data;

Step S54, calculating the pupil scaling degree of the real-time pupil image data according to the personalized pupil scaling limit data to generate real-time pupil scaling data;

And S55, integrating the real-time pupil scaling data with the real-time user pushing data in a corresponding sequence to generate real-time pupil-pushing data.

According to the invention, pupil image data acquisition is carried out on the user through the built-in monitoring equipment of the terminal, so that physiological response of the user can be obtained in real time, and attention and interest degree of the user when browsing push content can be more comprehensively known. The personalized pupil scaling limit analysis is carried out according to the pupil image data of the user, and the pupil scaling judgment standard can be adjusted according to the individual difference, so that the method is beneficial to capturing the attention change of the user more accurately and improving the adaptability to the individual difference of the user. The pupil image data of the user is collected in real time, pupil changes of the user at different time points can be captured, the attention level of the user is monitored in real time, and more timely feedback is provided. The pupil scaling degree calculation is carried out on the real-time pupil image data according to the personalized pupil scaling limit data, so that the attention level of the user to the push content can be quantified, and the interest degree of the user to different contents can be estimated more accurately. And integrating the real-time pupil scaling data with the real-time user pushing data in a corresponding sequence to generate real-time pupil-pushing data, establishing the association between the user pupil scaling and the pushing content, and providing more comprehensive user feedback.

As an example of the present invention, referring to fig. 5, a detailed implementation step flow diagram of step S5 in fig. 1 is shown, where step S5 includes:

In the embodiment of the invention, when the user browses the user push data through the terminal, the user acquires the pupil image data of the user by using the built-in monitoring equipment of the terminal, and the pupil image data of the user is acquired through the camera or other related sensors, so that a basis is provided for subsequent personalized analysis.

In the embodiment of the invention, personalized pupil scaling limit analysis is carried out according to the pupil image data of the user, and the personalized pupil scaling limit data is generated by analyzing the reference size and the variation range of the pupil of the user, so that the physiological differences among different users are considered.

In the embodiment of the invention, the pupil image data of the user is acquired in real time, so that the pupil dynamic change of the user when browsing the push content can be captured.

In the embodiment of the invention, the pupil scaling degree calculation is carried out on the real-time pupil image data according to the personalized pupil scaling limit data, the relation between the interest degree of the user on the push content and the pupil scaling degree is reflected by comparing the real-time pupil size with the personalized scaling limit, the real-time pupil scaling data is generated, and the relative size of the personalized pupil scaling degree of each user is considered by the real-time pupil scaling data.

In the embodiment of the invention, the real-time pupil-pushing data and the real-time user pushing data are subjected to data integration of corresponding sequences, and the real-time pupil-pushing data are generated by integrating the pushing content browsed by the user under the specific pupil-scaling degree into the sequence data, so that a key basis is provided for the real-time optimization of personalized pushing.

Preferably, step S6 comprises the steps of:

Step S61, performing pupil interest score calculation on the real-time pupil image data by using a pupil interest score algorithm to generate real-time pupil interest score data;

step S62, real-time browsing type interest scoring weighting is carried out on the real-time pupil-pushing data according to the real-time pupil score data, and real-time browsing type interest scoring data is generated;

And step S63, performing real-time optimization on the interest scores of the browsing types according to the real-time browsing type interest score data and the pushing type interest score data, and generating optimized browsing type interest score data.

The invention utilizes the pupil interest score algorithm to calculate the pupil interest score of the real-time pupil image data, can quantify the interest level of the user on the push content, provides a physiological objective measure, and is helpful for knowing the real-time interest of the user more accurately. The real-time pupil-pushing data is subjected to real-time browsing type interest scoring and weighting according to the real-time pupil score data, the interest degree of different pushing contents can be weighted according to the physiological response of the user, and the personalized degree of the pushing contents can be improved. The browsing interest type is optimized in real time according to the real-time browsing interest scoring data and the pushing interest scoring data, so that the understanding of the system to the user interest can be dynamically adjusted, the pushing strategy is optimized in real time, and the user experience is improved.

As an example of the present invention, referring to fig. 6, a flowchart illustrating a detailed implementation step of step S6 in fig. 1 is shown, where step S6 includes:

In the embodiment of the invention, the pupil interest score algorithm is utilized to calculate the pupil interest score of the real-time pupil image data, the information such as the pupil scaling degree, the scaling speed and the like in the real-time pupil image data is calculated through the pupil interest score algorithm, the real-time pupil interest score data is generated, the attention degree of a user to the current content is reflected, and the pupil interest score can be calculated through the conventional pupil score algorithm, but the accuracy of the result of calculating the pupil interest degree of the user and the browsing content is not as high as that of the pupil interest score algorithm.

In the embodiment of the invention, the real-time pupil-pushing data is weighted according to the real-time pupil score data, the real-time pupil score is used as the weight, the pupil interest score is combined with the pushing content, and the weighted calculation is performed, so that the real-time browsing type interest score data is generated, and the interest of the user to the pushing content is more accurately reflected.

In the embodiment of the invention, the real-time optimization of the browsing interest types is performed according to the real-time browsing interest scoring data and the pushing interest scoring data, the pushing interest scoring data has the initial score of pupil scoring optimization, and the optimized browsing interest type data is generated by comprehensively considering the real-time browsing interest scoring data and the pushing interest scoring data, and reflects the content score of the latest interest of the user in each type, so that the browsing data pushed to the user can be optimized in real time according to the optimized browsing interest type data, and a more personalized direction is provided for subsequent pushing and content presentation.

Preferably, the pupil interest score algorithm in step S61 is as follows:

Where P (T) is represented as real-time pupil interest score data, T is represented as browsing time, alpha is represented as a scaling factor of the real-time pupil interest score data, beta is represented as personalized pupil scaling limit data, gamma is represented as a current pupil diameter, delta is represented as an index adjustment value of the pupil diameter, e is represented as a natural constant, eta is represented as a rate of decay of the change in pupil diameter, T is represented as a browsing time period, zeta is represented as the number of scaling changes in pupil diameter, theta is represented as an unchanged time of pupil diameter scaling, An outlier represented as real-time pupil interest score data.

The invention utilizes a pupil interest scoring algorithm which comprehensively considers interaction relations among browsing time T, scaling factor alpha of real-time pupil interest scoring data, personalized pupil scaling limit data beta, current pupil diameter gamma, index adjustment value delta of pupil diameter, natural constant e, change attenuation rate eta of pupil diameter, browsing time period T, scaling change times zeta of pupil diameter, unchanged time theta of pupil diameter scaling and functions to form a functional relation:

That is to say, The functional relation is used for calculating whether the user is interested in browsing the push content or not by analyzing the relative degree of pupil change of each user when browsing the push content, and the attention degree of the user in browsing the push content can be judged by the frequency and the speed of the change times of pupil scaling of the user, so that the pupil interest score of the user on the type of the push content is obtained. Browsing time, namely the time when a user browses a certain type of content, is correspondingly changed along with the passage of the browsing time, and a longer browsing time reflects that the user has higher interest in the type; the scaling factor of the real-time pupil interest score data is used for adjusting the integral amplitude of the interest score and is used for normalizing the data result; personalized pupil scaling limit data, which affects the personalized degree of interest scores, for personalized adjustment of the influence of pupil changes of each different user; the current pupil diameter represents a real-time physiological index considered in an algorithm, the change of the pupil diameter influences the interest scoring trend, the larger pupil diameter probably takes into account the attention reduction, and the interest scoring result correspondingly reduces; the index adjustment value of the pupil diameter is used for adjusting the nonlinear influence of the pupil diameter on the interest score and adjusting the nonlinear effect of the pupil diameter change; the rate of decay of the pupil diameter variation determines the rate of pupil diameter variation; the browsing time period reflects the variation range of the integration; the number of scaling changes in pupil diameter reflects the frequency of pupil changes, and the faster pupil change frequency reflects that the user's attention mechanism is more concentrated and the higher the score of the generated result is; the unchanged time of pupil diameter scaling participates in parameters of Sigmoid function, and influences the change amplitude of interest scores. By introducing a plurality of parameters of pupil diameter and nonlinear elements, the algorithm can dynamically respond to pupil interest changes of the user in different browsing time periods, and the dynamic property enables the algorithm to capture transient changes of the browsing behaviors of the user more sensitively, so that accuracy of real-time interest scoring is improved. Abnormal adjustment value using real-time pupil interest scoring dataThe functional relation is adjusted and corrected, so that error influence caused by abnormal data or error items is reduced, real-time pupil interest score data P (T) are generated more accurately, and accuracy and reliability of pupil interest score calculation on real-time pupil image data are improved. Meanwhile, the scaling factors and the adjustment values in the formula can be adjusted according to actual conditions and are applied to different real-time pupil image data, so that the flexibility and applicability of the algorithm are improved.

The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.

The foregoing is only a specific embodiment of the invention to enable those skilled in the art to understand or practice the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. The data analysis method based on the neural network model is characterized by comprising the following steps of:

step S6, pupil interest score calculation is carried out on the real-time pupil image data, and real-time pupil interest score data are generated; performing interest score optimization analysis of the browsing type based on the real-time pupil score data and the real-time pupil-pushing data, and generating optimized browsing type interest score data;

Wherein, step S6 includes:

The pupil interest score algorithm in step S61 is as follows:

Where P (T) is represented as real-time pupil interest score data, T is represented as browsing time, alpha is represented as a scaling factor of the real-time pupil interest score data, beta is represented as personalized pupil scaling limit data, gamma is represented as a current pupil diameter, delta is represented as an index adjustment value of the pupil diameter, e is represented as a natural constant, eta is represented as a rate of decay of the change in pupil diameter, T is represented as a browsing time period, zeta is represented as the number of scaling changes in pupil diameter, theta is represented as an unchanged time of pupil diameter scaling, An outlier adjustment represented as real-time pupil interest score data;

2. The data analysis method based on the neural network model according to claim 1, wherein the step S1 includes the steps of:

3. The data analysis method based on the neural network model according to claim 2, wherein the step S14 includes the steps of:

4. A data analysis method based on a neural network model according to claim 3, wherein step S2 comprises the steps of:

5. The data analysis method based on the neural network model according to claim 1, wherein the step S3 includes the steps of:

6. The data analysis method based on the neural network model according to claim 1, wherein the step S4 includes the steps of:

7. The data analysis method based on the neural network model according to claim 1, wherein the step S5 includes the steps of:

8. A neural network model-based data analysis system for performing the neural network model-based data analysis method of claim 1, the neural network model-based data analysis system comprising: