Method and system for extracting features of user access behavior data

Abstract

The present invention discloses a method and system for extracting features of user access behavior data, and relates to the technical field of user access behavior data analysis. The user access behavior data of the present invention focuses on the current page dwell time data and mouse sliding trajectory data; if it is greater than a minimum dwell time threshold, it is necessary to collect user access behavior data, and the user access behavior data is mainly data related to mouse movement, and the mouse sliding trajectory data can reflect the user's main focus area on the current page. Through the analysis of the mouse sliding trajectory data, the user's focus in the process of browsing the current web page can be accurately obtained, thereby realizing targeted pop-up window-guided recommendations. For the same user, recording the pop-up window-guided recommendation data of each web page browsed by the same user can form the user's historical interest portrait data; greatly improving the accuracy of user access behavior analysis, and avoiding flood-style recommendations that cause user disgust.

Description

Technical Field

[0001] The present invention relates to the technical field of user access behavior data analysis, and in particular to a method and system for extracting features from user access behavior data. Background Art

[0002] User access behavior data refers to the data traces left by users when they visit websites, use applications or other online services. This data is very valuable to enterprises and can help them better understand user needs, optimize product design, improve user experience, and formulate marketing strategies.

[0003] Existing user access behavior data management is generally based on the server's automatic recording of access logs, such as Apache's access.log or Nginx's access.log. The above management methods cannot accurately analyze user intentions. Some even analyze user interests by capturing user shooting tools, analyzing points of interest by shooting user portraits and eye direction, or obtaining user voice information through recording tools, and performing keyword analysis on voice information to obtain user points of interest. These methods are suspected of infringing user privacy and are not in compliance with GDPR, CCPA and other regulations.

[0004] Moreover, log files may be very large and difficult to analyze; it is impossible to track the specific behavior of users, and only the page visits can be seen. Summary of the invention

[0005] In order to solve the above technical problems of analyzing user access behavior data, the present invention provides a method and system for extracting features of user access behavior data. The following technical solutions are adopted:

[0006] The feature extraction method of user access behavior data includes the following steps:

[0007] Step 1: Obtain the data of the current web page browsed by the user and the user's access behavior data;

[0008] Step 2: extract key feature data of user access behavior data, including current page dwell time data, mouse sliding trajectory data, mouse pause time data, and mouse click data;

[0009] Step 3: Analyze whether the current page dwell time data is greater than the minimum dwell time threshold. If so, perform user access behavior analysis to obtain current interest profile data based on the current web page data and key feature data, and make pop-up window-guided recommendations on the current page based on the current interest profile data.

[0010] Step 4: If it is determined that the current page dwell time data is less than or equal to the minimum dwell time threshold, it is determined that the user is not interested in the current web page the user is browsing, and a pop-up window guide recommendation is made on the current page based on the user's historical interest portrait. The pop-up window guide recommendation content is generated based on the pop-up window guide recommendation database.

[0011] By adopting the above technical solution, due to the restrictions of laws and regulations, the user's picture and sound data cannot be directly obtained. The user access behavior data here focuses on the current page dwell time data and mouse sliding trajectory data. If the current page dwell time data is less than or equal to the minimum dwell time threshold, for example, the minimum dwell time threshold can be designed to be 1 second, and the user closes or enters other pages within 1 second, it is considered that the user does not pay attention to the current web page information at all;

[0012] If it is greater than the minimum stay time threshold, it is necessary to collect user access behavior data. User access behavior data is mainly data related to mouse movement. Mouse sliding trajectory data can reflect the user's main focus area on the current page. Under normal circumstances, the mouse sliding back and forth under or to the side of a line of words or an image represents the user's main focus, that is, the point of interest. Therefore, through the analysis of mouse sliding trajectory data, the user's focus in the process of browsing the current web page can be accurately obtained. Combined with the current web page data corresponding to the mouse sliding trajectory data, interest-related keywords can be obtained, thereby realizing targeted pop-up window-guided recommendations. For the same user, recording the pop-up window-guided recommendation data of each web page browsed can form the user's historical interest portrait data;

[0013] For pop-up window recommendations that are not of interest to the current web page, recommendations can be guided based on the user's historical interest profile data;

[0014] Greatly improve the accuracy of user access behavior analysis to avoid flooding recommendations that may cause user disgust.

[0015] Optionally, in step 2, the timing starts from when the user opens the current web page. If the user closes the current web page within the minimum stay time threshold, step 4 is directly executed.

[0016] Optionally, in step 1, if the current web page is still open after exceeding the minimum stay time threshold, the following steps are used to obtain the current web page data:

[0017] Step 11, using JavaScript tools to access and operate the document object model of the web page;

[0018] Step 12, using JavaScript tools to obtain the query parameters of the current URL, the scroll position of the page, and the viewport size information;

[0019] Step 13. Use the Performance API tool to get detailed metrics on page navigation and load times.

[0020] Step 14: Use the browser's developer tools to obtain network requests, page resources, and DOM element data.

[0021] By adopting the above technical solution, obtaining the current page data usually involves front-end development technology, and JavaScript tools can be used to obtain the document object model of the web page, thereby obtaining the query parameters of the current URL, the scroll position of the page and the viewport size information, providing basic data for the subsequent positioning of the mouse track on the page;

[0022] Use the Performance API tool to get detailed metrics on page navigation and loading times. This allows you to factor in loading times when calculating minimum dwell time thresholds to avoid false positives caused by long loading times.

[0023] Use the browser's developer tools to obtain network requests, page resources, DOM element data, etc., which can provide data for subsequent keyword extraction.

[0024] Optionally, in step 1, a JavaScript tool is used to track the mouse sliding trajectory data, and points where the dwell time exceeds 0.2 mm are recorded as pause points to form a pause point set, and the pause points are connected in chronological order to form a trajectory curve, which is the mouse sliding trajectory data; after the pause point set is embedded in the pause time, it becomes the mouse pause time data.

[0025] Optionally, step 3 of analyzing user access behavior includes the following steps:

[0026] Step 31, analyzing the mouse sliding trajectory data, if the number of pauses between the mouse starting point and the mouse click point is less than or equal to 2, directly extracting the mouse click data of the mouse click point, obtaining the corresponding page resource of the mouse click data, and performing a pop-up window guidance recommendation operation based on the corresponding page resource;

[0027] Step 32, if the number of pauses is greater than 2, use Fourier descriptor, Zernike moment or HOG to describe the shape of the trajectory, determine whether there is a reciprocating fold line, if so, extract the reciprocating fold line, and determine whether the reciprocating fold line is a horizontal line or a vertical line. If it is a horizontal line, extract the page resource corresponding to the rectangular area above the horizontal line. If it is a vertical line, process the reciprocating fold line into a thick vertical line, take the center of the thick vertical line as the dot, extract the corresponding page resource in the circular area with a radius of R, and perform a pop-up window-guided recommendation operation based on the corresponding page resource.

[0028] Optionally, in step 32, the method for calculating the rectangular area above the horizontal line is:

[0029] Assume that the starting point of the bottom edge of the rectangular area is the left end point of the top horizontal line of the reciprocating fold line, and the right end point of the top horizontal line is the end point of the bottom edge of the rectangular area;

[0030] Let the length of the top horizontal line be A, then the length of the bottom side of the rectangular area is A, and the length of the side is ;

[0031] The method to calculate the radius R of the circular area is:

[0032] Process the reciprocating broken line into a thick vertical line, take the center point of the thick vertical line as the circle point, calculate the length of the thick vertical line as C, and the width as B, then , where n is the range adjustment factor, which ranges from 1 to 5.

[0033] By adopting the above technical solution, when the user operates the mouse to browse the current web page, pauses will be formed at some points. These pauses usually stay for a short time, but will be greater than 0.2mm. These pause data are formed into a pause set. In specific applications, the pause set is embedded in the pause time to obtain the mouse pause time data. The pause time threshold can also be set. By detecting if the mouse pause time exceeds the pause time threshold, the corresponding keyword information of the nearby web page can be obtained for pop-up window guidance recommendation;

[0034] The pause points are connected in chronological order to form trajectory curves, and the shape of these trajectory curves can be analyzed. If there are reciprocating broken lines, which may be horizontal lines or broken lines, it is highly likely that users are interested in the web page information in these related areas, thereby accurately obtaining the user's focus from the user's access behavior;

[0035] For reciprocating horizontal lines, users usually focus on the area above the horizontal lines. When processing reciprocating horizontal lines, the horizontal lines can be standardized. For example, the reciprocating horizontal lines can be standardized into horizontal thick horizontal lines. The top edge of the horizontal thick horizontal line is used to replace the top horizontal line. The bottom length of the rectangular area is A, and the side length is , which can basically cover the content of the concerned area;

[0036] For a reciprocating vertical fold line, the user's focus may be on both sides of the vertical fold line, so a circular area is used to obtain web page information in the associated area.

[0037] Optionally, the pop-up guided recommendation operation includes the following steps: extracting keywords of the corresponding page resources, matching keywords in the pop-up guided recommendation database, making surrounding pop-up recommendations for the successfully matched interest-oriented keywords, first determining the number of successfully matched interest-oriented keywords, using JavaScript tools to create and display multiple recommendation blocks, and displaying the multiple recommendation blocks around the current mouse cursor.

[0038] Optionally, in the pop-up window guiding the recommended operation, the block appears by fading in or rotating to create a dynamic visual feedback effect. The duration, type and complexity of the animation can be adjusted as needed.

[0039] By adopting the above technical solution, the pop-up window guided recommendation operation is formally restricted during the specific application process. The pop-up window guided recommendation area surrounds the current mouse and can appear in a fade-in or rotation manner. The transparency can be adjusted in specific application. When the recommendation block is formed, it is fixed in position and the user can click it by moving the mouse, so as to avoid interfering with the user's current operation as much as possible.

[0040] The feature extraction system of user access behavior data includes multiple user intelligent terminals, a server computer and a database storage. The server computer is respectively connected to the multiple user intelligent terminals through network communication. The multiple user intelligent terminals respectively access the web pages of the server computer through the network. The server computer is installed with a program designed using the feature extraction method of user access behavior data, and the current web page data browsed by the user of the running program or the user intelligent terminal and the user access behavior data, and pop-up guidance recommendation is performed based on the pop-up guidance recommendation data stored in the pop-up guidance recommendation database.

[0041] Optionally, the pop-up window-guided recommendation data includes recommended keyword text, keyword supporting graphic data and keyword link data.

[0042] By adopting the above technical solution, the recommendation database also stores each pop-up window recommendation data corresponding to the user name data, and each pop-up window recommendation data can form the user's historical interest portrait data.

[0043] In summary, the present invention includes at least one of the following beneficial technical effects:

[0044] The present invention can provide a feature extraction method and system for user access behavior data. The user access behavior data focuses on the current page dwell time data and the mouse sliding trajectory data; if it is greater than the minimum dwell time threshold, the user access behavior data needs to be collected. The user access behavior data is mainly data related to mouse movement. The mouse sliding trajectory data can reflect the user's main focus area on the current page. The user's focus in the process of browsing the current web page can be accurately obtained through analysis of the mouse sliding trajectory data. In combination with the current web page data corresponding to the mouse sliding trajectory data, interest-related keywords can be obtained, thereby realizing targeted pop-up guided recommendations. For the same user, the pop-up guided recommendation data of each web page browsed by the user can be recorded to form the user's historical interest portrait data; the pop-up guided recommendation for the current web page that is not of interest can be recommended based on the user's historical interest portrait data; the accuracy of user access behavior analysis is greatly improved to avoid flooding recommendations that cause user disgust. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flow chart of the feature extraction method of user access behavior data of the present invention;

[0046] Figure 2 It is a schematic diagram of the effect of mouse sliding trajectory data in a specific embodiment of the method for extracting features of user access behavior data of the present invention;

[0047] Figure 3 It is a schematic diagram of the effect of processing a vertical broken line into a standard vertical line in a specific embodiment of the method for extracting features of user access behavior data of the present invention, with the left side being the original mouse sliding track image and the right side being a schematic diagram of the effect of a standard vertical line;

[0048] Figure 4 It is a schematic diagram of the pop-up window guided recommendation effect in a specific embodiment of the feature extraction method of user access behavior data of the present invention. DETAILED DESCRIPTION

[0049] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0050] The embodiment of the present invention discloses a method and system for extracting features of user access behavior data.

[0051] Reference Figure 1-Figure 4 ,The feature extraction method of user access behavior data includes the following steps:

[0052] Step 1: Obtain the data of the current web page browsed by the user and the user's access behavior data;

[0053] Step 2: extract key feature data of user access behavior data, including current page dwell time data, mouse sliding trajectory data, mouse pause time data, and mouse click data;

[0054] Step 3: Analyze whether the current page dwell time data is greater than the minimum dwell time threshold. If so, perform user access behavior analysis to obtain current interest profile data based on the current web page data and key feature data, and make pop-up window-guided recommendations on the current page based on the current interest profile data.

[0055] Step 4: If it is determined that the current page dwell time data is less than or equal to the minimum dwell time threshold, it is determined that the user is not interested in the current web page the user is browsing, and a pop-up window guide recommendation is made on the current page based on the user's historical interest portrait. The pop-up window guide recommendation content is generated based on the pop-up window guide recommendation database.

[0056] Due to legal restrictions, it is not possible to directly obtain user screen and sound data. The user access behavior data here focuses on the current page dwell time data and mouse sliding trajectory data. If the current page dwell time data is less than or equal to the minimum dwell time threshold, for example, the minimum dwell time threshold can be set to 1 second, and the user closes the page or enters another page within 1 second, it is considered that the user does not pay attention to the current web page information at all;

[0057] If it is greater than the minimum stay time threshold, it is necessary to collect user access behavior data. User access behavior data is mainly data related to mouse movement. Mouse sliding trajectory data can reflect the user's main focus area on the current page. Under normal circumstances, the mouse sliding back and forth under or to the side of a line of words or an image represents the user's main focus, that is, the point of interest. Therefore, through the analysis of mouse sliding trajectory data, the user's focus in the process of browsing the current web page can be accurately obtained. Combined with the current web page data corresponding to the mouse sliding trajectory data, interest-related keywords can be obtained, thereby realizing targeted pop-up window-guided recommendations. For the same user, recording the pop-up window-guided recommendation data of each web page browsed can form the user's historical interest portrait data;

[0058] For pop-up window recommendations that are not of interest to the current web page, recommendations can be guided based on the user's historical interest profile data;

[0059] Greatly improve the accuracy of user access behavior analysis to avoid flooding recommendations that may cause user disgust.

[0060] In step 2, the time starts from when the user opens the current web page. If the user closes the current web page within the minimum stay time threshold, step 4 is directly executed.

[0061] In step 1, if the current web page is still open after exceeding the minimum stay time threshold, the following steps are used to obtain the current web page data:

[0062] Step 11, using JavaScript tools to access and operate the document object model of the web page;

[0063] Step 12, using JavaScript tools to obtain the query parameters of the current URL, the scroll position of the page, and the viewport size information;

[0064] Step 13. Use the Performance API tool to get detailed metrics on page navigation and load times.

[0065] Step 14: Use the browser's developer tools to obtain network requests, page resources, and DOM element data.

[0066] Obtaining the current page data usually involves front-end development technology. JavaScript tools can be used to obtain the document object model of the web page, thereby obtaining the query parameters of the current URL, the scroll position of the page, and the viewport size information, providing basic data for the subsequent positioning of the mouse track on the page;

[0067] Use the Performance API tool to get detailed metrics on page navigation and loading times. This allows you to factor in loading times when calculating minimum dwell time thresholds to avoid false positives caused by long loading times.

[0068] Use the browser's developer tools to obtain network requests, page resources, DOM element data, etc., which can provide data for subsequent keyword extraction.

[0069] In step 1, a JavaScript tool is used to track the mouse sliding trajectory data, and points with a dwell time of more than 0.2 mm are recorded as pause points to form a pause point set, and the pause points are connected in chronological order to form a trajectory curve, which is the mouse sliding trajectory data; after the pause point set is embedded in the pause time, it becomes the mouse pause time data.

[0070] Step 3: User access behavior analysis includes the following steps:

[0071] Step 31, analyzing the mouse sliding trajectory data, if the number of pauses between the mouse starting point and the mouse click point is less than or equal to 2, directly extracting the mouse click data of the mouse click point, obtaining the corresponding page resource of the mouse click data, and performing a pop-up window guidance recommendation operation based on the corresponding page resource;

[0072] Step 32, if the number of pauses is greater than 2, use Fourier descriptor, Zernike moment or HOG to describe the shape of the trajectory, determine whether there is a reciprocating fold line, if so, extract the reciprocating fold line, and determine whether the reciprocating fold line is a horizontal line or a vertical line. If it is a horizontal line, extract the page resource corresponding to the rectangular area above the horizontal line. If it is a vertical line, process the reciprocating fold line into a thick vertical line, take the center of the thick vertical line as the dot, extract the corresponding page resource in the circular area with a radius of R, and perform a pop-up window-guided recommendation operation based on the corresponding page resource.

[0073] In step 32, the method for calculating the rectangular area above the horizontal line is:

[0074] Assume that the starting point of the bottom edge of the rectangular area is the left end point of the top horizontal line of the reciprocating fold line, and the right end point of the top horizontal line is the end point of the bottom edge of the rectangular area;

[0075] Let the length of the top horizontal line be A, then the length of the bottom side of the rectangular area is A, and the length of the side is ;

[0076] The method to calculate the radius R of the circular area is:

[0077] Process the reciprocating broken line into a thick vertical line, take the center point of the thick vertical line as the circle point, calculate the length of the thick vertical line as C, and the width as B, then , where n is the range adjustment factor, which ranges from 1 to 5.

[0078] When the user uses the mouse to browse the current web page, pauses will be formed at some points. These pauses usually stay for a short time, but will be greater than 0.2mm. These pause data are formed into a pause set. In specific applications, the pause set is embedded in the pause time to obtain the mouse pause time data. The pause time threshold can also be set. By detecting if the mouse pause time exceeds the pause time threshold, the corresponding keyword information of the nearby web page can be obtained for pop-up window guidance recommendation;

[0079] The pause points are connected in chronological order to form trajectory curves, and the shape of these trajectory curves can be analyzed. If there are reciprocating broken lines, which may be horizontal lines or broken lines, it is highly likely that users are interested in the web page information in these related areas, thereby accurately obtaining the user's focus from the user's access behavior;

[0080] For reciprocating horizontal lines, users usually focus on the area above the horizontal lines. When processing reciprocating horizontal lines, the horizontal lines can be standardized. For example, the reciprocating horizontal lines can be standardized into horizontal thick horizontal lines. The top edge of the horizontal thick horizontal line is used to replace the top horizontal line. The bottom length of the rectangular area is A, and the side length is , which can basically cover the content of the concerned area;

[0081] For a reciprocating vertical fold line, the user's focus may be on both sides of the vertical fold line, so a circular area is used to obtain web page information in the associated area.

[0082] The pop-up window recommendation operation includes the following steps: extracting keywords of the corresponding page resources, matching keywords in the pop-up window recommendation database, making surrounding pop-up window recommendations for the successfully matched interest-oriented keywords, first determining the number of successfully matched interest-oriented keywords, using JavaScript tools to create and display multiple recommendation blocks, and displaying multiple recommendation blocks around the current mouse cursor.

[0083] In the pop-up window guiding the recommended operation, the block appears by fading in or rotating to create a dynamic visual feedback effect. The duration, type and complexity of the animation can be adjusted as needed.

[0084] The pop-up window-guided recommendation operation is formally restricted in its specific application process. The pop-up window-guided recommendation area surrounds the current mouse and can appear in a fade-in or rotation manner. The transparency can be adjusted in specific application. When the recommendation block is formed, it is fixed in position and the user can click it by moving the mouse, so as to avoid interfering with the user's current operation as much as possible.

[0085] A feature extraction system for user access behavior data includes multiple user smart terminals, a server computer and a database storage device. The server computer is connected to the multiple user smart terminals through a network communication. The multiple user smart terminals access the web pages of the server computer through the network. The server computer is installed with a program designed using a feature extraction method for user access behavior data. The server computer runs the program or the user smart terminal to collect the current web page data browsed by the user and the user access behavior data, and performs pop-up guidance recommendation based on the pop-up guidance recommendation data stored in the pop-up guidance recommendation database.

[0086] The pop-up window recommendation data includes the recommended keyword text, keyword supporting graphic data and keyword link data.

[0087] The recommendation database also stores each pop-up window recommendation data corresponding to the user name data, and each pop-up window recommendation data can form the user's historical interest portrait data.

[0088] The following specific embodiments are used to illustrate the implementation principle of the feature extraction method of user access behavior data of the present invention:

[0089] When a user visits a web page, the user's browsing data and user access behavior data are obtained;

[0090] Extract key feature data of user access behavior data, including current page dwell time data, mouse sliding trajectory data, mouse pause time data and mouse click data;

[0091] Analyze the current page dwell time data to see if it is greater than the minimum dwell time threshold, perform user access behavior analysis, obtain current interest profile data based on the current web page data and key feature data, and make pop-up window-guided recommendations on the current page based on the current interest profile data;

[0092] Use JavaScript tools to track the mouse sliding trajectory data, record the points where the dwell time exceeds 0.2mm as pause points, form a pause point set, and connect the pause points in chronological order to form a trajectory curve, see Figure 2 , which is the mouse sliding trajectory data; after the pause point set is embedded in the pause time, it is the mouse pause time data.

[0093] Analyze the mouse sliding trajectory data. If the number of pauses is 10 and greater than 2, use Fourier descriptor, Zernike moment or HOG to describe the shape of the trajectory to determine whether there is a reciprocating broken line. Through analysis, a vertical broken line that goes back and forth three times is obtained. The vertical broken line is processed into a standard vertical line. See Figure 3 , take the center of the thick vertical line as the dot, extract the corresponding page resources in the circular area with a radius of R, and obtain the keyword men's jackets through the corresponding page resource data analysis. Then, four brands of men's jackets are matched in the pop-up window guidance recommendation database, four recommendation blocks are generated, and corresponding web page links are generated. The four recommendation blocks surround the current mouse cursor. See Figure 4 .

[0094] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for extracting features from user access behavior data, characterized in that: The following steps are involved: Step 1: Obtain the data of the current web page browsed by the user and the user's access behavior data; Step 2: extract key feature data of user access behavior data, including current page dwell time data, mouse sliding trajectory data, mouse pause time data, and mouse click data; Step 3: Analyze whether the current page dwell time data is greater than the minimum dwell time threshold. If so, perform user access behavior analysis to obtain current interest profile data based on the current web page data and key feature data, and make pop-up window-guided recommendations on the current page based on the current interest profile data. Step 4: If it is determined that the current page dwell time data is less than or equal to the minimum dwell time threshold, it is determined that the user is not interested in the current web page browsed by the user, and a pop-up window guidance recommendation is performed on the current page based on the user's historical interest profile, and the pop-up window guidance recommendation content is generated based on the pop-up window guidance recommendation database; Step 3: User access behavior analysis includes the following steps: Step 31, analyzing the mouse sliding trajectory data, if the number of pauses between the mouse starting point and the mouse click point is less than or equal to 2, directly extracting the mouse click data of the mouse click point, obtaining the corresponding page resource of the mouse click data, and performing a pop-up window guidance recommendation operation based on the corresponding page resource; Step 32, if the number of pauses is greater than 2, use Fourier descriptor, Zernike moment or HOG to describe the shape of the trajectory, determine whether there is a reciprocating fold line, if so, extract the reciprocating fold line, determine whether the reciprocating fold line is a horizontal line or a vertical line, if it is a horizontal line, extract the page resource corresponding to the rectangular area above the horizontal line, if it is a vertical line, process the reciprocating fold line into a thick vertical line, take the center of the thick vertical line as the dot, extract the corresponding page resource in the circular area with a radius of R, and perform a pop-up window guidance recommendation operation based on the corresponding page resource; In step 32, the method for calculating the rectangular area above the horizontal line is: Assume that the starting point of the bottom edge of the rectangular area is the left end point of the top horizontal line of the reciprocating fold line, and the right end point of the top horizontal line is the end point of the bottom edge of the rectangular area; Let the length of the top horizontal line be A, then the length of the bottom side of the rectangular area is A, and the length of the side is ; The method to calculate the radius R of the circular area is: Process the reciprocating broken line into a thick vertical line, take the center point of the thick vertical line as the circle point, calculate the length of the thick vertical line as C, and the width as B, then , where n is the range adjustment factor, which ranges from 1 to 5.

2. The feature extraction method of user access behavior data according to claim 1 is characterized in that: In step 2, the time starts from when the user opens the current web page. If the user closes the current web page within the minimum stay time threshold, step 4 is directly executed.

3. The feature extraction method of user access behavior data according to claim 2 is characterized in that: In step 1, if the current web page is still open after exceeding the minimum stay time threshold, the following steps are used to obtain the current web page data: Step 11, using JavaScript tools to access and operate the document object model of the web page; Step 12, using JavaScript tools to obtain the query parameters of the current URL, the scroll position of the page, and the viewport size information; Step 13. Use the Performance API tool to get detailed metrics on page navigation and load times. Step 14: Use the browser's developer tools to obtain network requests, page resources, and DOM element data.

4. The feature extraction method of user access behavior data according to claim 3 is characterized in that: In step 1, a JavaScript tool is used to track the mouse sliding trajectory data, and points with a dwell time of more than 0.2 mm are recorded as pause points to form a pause point set, and the pause points are connected in chronological order to form a trajectory curve, which is the mouse sliding trajectory data; after the pause point set is embedded in the pause time, it becomes the mouse pause time data.

5. The method for extracting features from user access behavior data according to claim 4, characterized in that: The pop-up window recommendation operation includes the following steps: extracting keywords of the corresponding page resources, matching keywords in the pop-up window recommendation database, making surrounding pop-up window recommendations for the successfully matched interest-oriented keywords, first determining the number of successfully matched interest-oriented keywords, using JavaScript tools to create and display multiple recommendation blocks, and displaying multiple recommendation blocks around the current mouse cursor.

6. The method for extracting features of user access behavior data according to claim 5, characterized in that: In the pop-up window guiding the recommended operation, the block appears by fading in or rotating to create a dynamic visual feedback effect. The duration, type and complexity of the animation can be adjusted as needed.

7. A feature extraction system for user access behavior data, characterized by: The invention comprises a plurality of user intelligent terminals, a server computer and a database storage device. The server computer is respectively connected to the network of the plurality of user intelligent terminals. The plurality of user intelligent terminals respectively access the web pages of the server computer through the network. The server computer is installed with a program designed by the feature extraction method of user access behavior data described in any one of claims 1 to 6. The program or the user intelligent terminal runs the current web page data browsed by the user and the user access behavior data, and performs pop-up guidance recommendation based on the pop-up guidance recommendation data stored in the pop-up guidance recommendation database.

8. The feature extraction system for user access behavior data according to claim 7, characterized in that: The pop-up window recommendation data includes the recommended keyword text, keyword supporting graphic data and keyword link data.

Images