An e-commerce analysis and diagnosis method based on artificial intelligence

Abstract

The application is specifically an e-commerce analysis and diagnosis method based on artificial intelligence, comprising: behavior data collection; behavior analysis; comprehensive judgment; intention diagnosis and evaluation.In the application, a multi-dimensional analysis system is constructed, the behavior data of a user from the whole link of contacting a commodity to browsing details, comments and the like is comprehensively collected, a foundation is laid for accurate analysis, two core indexes of a matching coefficient and an interest coefficient are introduced, the former quantifies the matching degree of the user and the commodity information from the angles of browsing fluency, information acquisition efficiency, key information attention and the like, the latter combines three dimensions of comment interaction, historical portrait matching and picture browsing to mine the potential interest of the user, finally, the intention evaluation coefficient is calculated and graded by comprehensively adding the purchase behavior and the collection behavior, the purchase intention of the user can be more accurately judged, the waste of marketing resources is avoided, the conversion rate of marketing activities and the input-output ratio are improved, and a scientific basis is provided for accurate marketing of e-commerce.

Description

Technical Field

[0001] This invention relates to the field of e-commerce technology, and in particular to an e-commerce analysis and diagnosis method based on artificial intelligence. Background Technology

[0002] In the e-commerce sector, with the explosive growth of user behavior data, traditional manual analysis or simple statistical methods are no longer sufficient to meet the needs of real-time and accurate operations. The pain points include low data processing efficiency, insufficient depth of user insights (only focusing on surface indicators such as click-through rate and add-to-cart rate, lacking in-depth exploration of browsing rhythm, interest preferences, emotional tendencies, and other deep behavioral patterns), vague marketing audience positioning and poor content adaptability, fragmented technical tools, and rising user retention costs due to intensified industry competition.

[0003] From a user insight perspective, existing technologies lack the ability to deeply analyze user behavior. Traditional analysis focuses only on surface-level metrics such as click-through rate and conversion rate, neglecting the subtle behavioral characteristics of users during the browsing process:

[0004] Differences in browsing pace: The length of time a user pauses while scrolling on a product details page implies the depth of their digestion of information (e.g., a long pause may indicate a careful reading of specifications and parameters);

[0005] Interest focus: The number of times users zoom in on specific areas of product images and the frequency of keywords in the comments section directly reflect their focus (e.g., users who zoom in on fabric detail images multiple times are more concerned about the material).

[0006] Sentiment bias: The polarity (positive / negative) of users' feelings toward comments and their interactive behaviors (likes, replies) reveal their underlying attitudes toward the product (e.g., frequent likes of negative comments may indicate interest in competitors).

[0007] The limitations of existing technologies are mainly reflected in: inefficient data processing, superficial user insights, extensive marketing matching, fragmented system architecture, and lagging model iteration.

[0008] Therefore, there is an urgent need for an AI-based e-commerce analysis and diagnostic method to address the aforementioned problems. Summary of the Invention

[0009] The purpose of this invention is to propose an artificial intelligence-based e-commerce analysis and diagnosis method to solve the above-mentioned problems.

[0010] To achieve the above objectives, the present invention adopts the following technical solution:

[0011] An AI-based e-commerce analysis and diagnostic method includes:

[0012] Behavioral data collection: Obtaining user interaction behavior data after marketing information is displayed on users' mobile phones;

[0013] Behavioral analysis: Matching coefficients and interest coefficients are obtained by analyzing user interaction behavior data;

[0014] Comprehensive judgment: The basic evaluation coefficient is obtained by weighting the matching coefficient and the interest coefficient. The intention evaluation coefficient is obtained by combining the basic evaluation coefficient with the coefficients corresponding to the actions of adding to cart and favorites.

[0015] Intention Diagnosis and Assessment: Determine the degree of user's purchase intention for the marketed product based on the intention assessment coefficient.

[0016] Preferably, the collection of behavioral data specifically includes:

[0017] When a shopping app on a user's mobile phone receives a product marketing image display, it obtains corresponding user interaction behavior information data, which includes:

[0018] Data information includes clicks on product marketing images, interaction time after entering the product page, add-to-cart and favorite data, and viewing of product details and reviews.

[0019] Preferably, the process of obtaining the matching coefficient includes:

[0020] The pause time between each page swipe when the user is browsing the product details page is obtained, and a pause time threshold for page swipe is preset. The pause time between each page swipe is compared with the pause time threshold. The pause time between page swipes that is less than the pause time threshold is recorded as normal pause time, and the pause time between page swipes that is greater than the pause time threshold is recorded as abnormal pause time.

[0021] After counting the number of all normal pause times and the number of pause times between all page scrolling, the normal browsing ratio is obtained by dividing the number of all normal pause times by the number of pause times between all page scrolling.

[0022] The total normal pause time is obtained by summing up all normal pause times;

[0023] The total abnormal pause time is obtained by summing up all abnormal pause times.

[0024] Get the time when the user clicks on a product and the time when the user exits browsing the product, and record the difference between the two times as the total browsing time;

[0025] Divide the total normal pause time by the total browsing time to obtain the effective time ratio;

[0026] Subtract the total abnormal pause time from the total browsing time to obtain the effective browsing time;

[0027] Identify all keywords for the product, including specifications, after-sales policy, and related recommendations;

[0028] Obtain the times when keywords appear and disappear on the user's mobile screen while browsing product details; subtract the appearance time from the disappearance time to obtain the keyword appearance duration;

[0029] The occurrence duration of each keyword is obtained sequentially, and these durations are accumulated to obtain the total occurrence duration of the keywords; the total occurrence duration of the keywords is divided by the effective browsing time to obtain the keyword ratio;

[0030] The matching coefficient is obtained by weighting the normal browsing ratio, effective time ratio, and keyword ratio.

[0031] Preferably, the process of obtaining the interest coefficient includes:

[0032] The comment interaction value is obtained by analyzing the interaction data in the comment section;

[0033] The user profile matching value is obtained by analyzing the data of similar products browsed by the user within a preset time period.

[0034] Image values ​​are obtained by analyzing user browsing behavior on product detail pages and in comment sections.

[0035] Preferably, the process of obtaining the comment interaction value includes:

[0036] Get the time when a user opens the comment section and the time when they exit the comment section, and record the time difference between the two times as the comment dwell time;

[0037] Get the number of comments that users clicked to view, as well as the corresponding interactive behaviors for each comment, including clicking "view more comments", "reply", and "like".

[0038] The interaction rate is obtained by summing up the interaction behaviors corresponding to each comment, dividing the total number of interactions by the comment dwell time.

[0039] Using natural language processing technology, the comments viewed by users are matched with a pre-set keyword database, and the number of hits is counted.

[0040] The content of each comment viewed by the user is matched against a preset keyword database in turn to obtain the number of times the keyword is hit for each comment;

[0041] Using a sentiment analysis model from natural language processing, the comments viewed by users are scored with sentiment, with a score range of -1 to 1, where -1 is negative and 1 is positive.

[0042] Sort the hit counts of each keyword in descending order of numerical value, and extract the three keywords with the highest hit counts;

[0043] Multiply the number of times the three keywords hit the most by the corresponding sentiment score of the comment to obtain the sentiment score of the number of times the three keywords hit the most.

[0044] The three sentiment scores are averaged. If the average value is greater than 0, it is marked as the sentiment average.

[0045] The comment interaction value is obtained by combining the interaction rate and the sentiment score.

[0046] Preferably, the process of obtaining the profile matching value includes:

[0047] Get the user's search or favorites of similar products and their corresponding purchase prices within a preset historical time range;

[0048] The product details pages of each similar product purchased are retrieved sequentially, and the corresponding product keywords are extracted from them. The keywords of all similar products purchased are then unified to obtain a similar product keyword library.

[0049] Match each keyword of the marketing product with a keyword database of similar products, count the number of successful matches, and record it as a success value;

[0050] Divide the total number of keywords for the marketing product by the total number of keywords in the keyword library for similar products to obtain the overlap rate;

[0051] The average price of similar products is obtained by averaging the prices of all similar products.

[0052] The price difference is calculated by comparing the price of the marketed product with the average price of similar products. The allowable fluctuation range of the price difference is preset, and the price difference is matched with the allowable fluctuation range. If the price difference is within the allowable fluctuation range, it is recorded as the reasonable price difference.

[0053] The profile matching value is obtained by comprehensively processing the success value, overlap, and reasonable price difference.

[0054] Preferably, the process of obtaining the image value includes:

[0055] Monitor changes in the properties of the image's parent container; if the size exceeds a preset threshold, it is determined to be an image enlargement operation.

[0056] Obtain the marketing product area corresponding to the user's zoomed-in image, and extract the zoom-in dwell time of that area;

[0057] The overall appearance of the marketing product is divided into multiple areas, and each area is distinguished based on its own characteristics;

[0058] The system acquires images displayed on the user's mobile phone screen, extracts relevant marketing product features from them, and identifies the marketing product area corresponding to the image.

[0059] The system sequentially retrieves the marketing product area corresponding to each time a user zooms in on an image, and counts the number of zooms in on each area of ​​the marketing product; it then extracts the two highest zoom counts and the image zoom dwell time corresponding to the two highest zoom counts.

[0060] Multiply the two largest magnification counts by their respective magnification dwell times to obtain the quantization values;

[0061] The two quantized values ​​are used as the major and minor axes of the ellipse, respectively, to construct an ellipse model, and the area of ​​the ellipse model is recorded as the image value.

[0062] Preferably, the step of obtaining the basic evaluation coefficient by weighting the matching coefficient and the interest coefficient includes:

[0063] After normalizing the matching coefficient and interest coefficient, a circle is constructed using the matching coefficient as the radius and the interest coefficient is constructed using the height of the circle to establish a cone model. The volume of the cone model is calculated to obtain the basic evaluation coefficient.

[0064] If a user adds a marketing product to their shopping cart, the action value for the "add to cart" action is recorded as 1; otherwise, it is recorded as -1.

[0065] Similarly, if a user adds a marketing product to their favorites, the action value for the "favorite" action is recorded as 1, and vice versa, it is -1.

[0066] Preset the weighting factors corresponding to the actions of adding to cart and adding to favorites. Then, multiply the action value of adding to cart and the action value of adding to favorites with their corresponding weighting factors and sum them to obtain the coefficient of adding to cart and adding to favorites.

[0067] After normalizing the basic evaluation coefficient and the add-to-cart / favorite coefficient, determine whether the add-to-cart / favorite coefficient is greater than 0:

[0068] If the add-to-cart and add-to-collect coefficient is greater than 0, then the basic evaluation coefficient and the add-to-cart and add-to-collect coefficient are respectively used as the two legs of a right triangle, and the remaining leg is connected to form a complete right triangle. The area of ​​the right triangle is recorded as the intention evaluation coefficient.

[0069] If the add-to-cart and add-to-collect coefficient is less than 0, then the basic evaluation coefficient and the add-to-cart and add-to-collect coefficient are respectively used as two legs of a right triangle. Connect the remaining leg to form a complete right triangle. Then, take the center of the leg corresponding to the add-to-collect coefficient as the origin and half of the add-to-collect coefficient as the radius to build a circle. This circle will divide the right triangle into areas. Calculate the area of ​​the remaining right triangle and record it as the intention evaluation coefficient.

[0070] Preferably, determining the user's purchase intention for the marketed product based on the intention assessment coefficient specifically includes:

[0071] Three threshold ranges are preset, and each threshold range corresponds to a purchase intention level. The intention evaluation coefficient is compared with the three threshold ranges to obtain the purchase intention level corresponding to the intention evaluation coefficient. The purchase intention levels include low intention, medium intention and high intention.

[0072] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0073] 1. This invention constructs a multi-dimensional analysis system, comprehensively collecting user behavior data from the entire process of product contact to browsing details and reviews, laying the foundation for accurate analysis. It introduces two core indicators: a matching coefficient and an interest coefficient. The former quantifies the degree of matching between users and product information from the perspectives of browsing smoothness, information acquisition efficiency, and attention to key information; the latter combines comment interaction, historical profile matching, and image browsing to uncover users' potential interests. Finally, it calculates an intention evaluation coefficient based on add-to-cart and favorite behaviors and classifies them into levels, enabling more accurate judgment of user purchase intentions, avoiding waste of marketing resources, improving marketing activity conversion rates and return on investment, and providing a scientific basis for e-commerce precision marketing.

[0074] 2. This invention utilizes SDKs and event tracking technologies to collect user behavior data in real time. Based on dynamically calculated matching coefficients, interest coefficients, and intention evaluation coefficients, it promptly reflects changes in user behavior and market feedback. Once user behavior data is updated, the system immediately recalculates the relevant coefficients and updates the purchase intention evaluation results. E-commerce companies can quickly adjust their marketing strategies accordingly, such as optimizing product information display and improving comment section design for low-intent groups; increasing resource investment and pushing personalized promotions to high-intent users; and monitoring the differences in effectiveness across different marketing channels and time periods to optimize resource allocation. Through dynamic evaluation and real-time strategy adjustments, companies can better adapt to market changes, enhance user stickiness, and improve market competitiveness. Attached Figure Description

[0075] Further details, features, and advantages of this application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which:

[0076] Figure 1 This is a flowchart of the present invention; Detailed Implementation

[0077] Several embodiments of this application will now be described in more detail with reference to the accompanying drawings to enable those skilled in the art to implement this application. This application may be embodied in many different forms and for various purposes and should not be limited to the embodiments set forth herein. These embodiments are provided to make this application thorough and complete, and to fully convey the scope of this application to those skilled in the art. The embodiments described do not limit this application.

[0078] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It will be further understood that terms such as those defined in commonly used dictionaries shall be interpreted as having a meaning consistent with their meaning in the relevant field and / or the context of this specification, and shall not be interpreted in an idealized or overly formal sense unless expressly defined herein.

[0079] Please see Figure 1 As shown, the present invention provides a technical solution:

[0080] An AI-based e-commerce analysis and diagnostic method includes:

[0081] Behavioral data collection: Obtaining user interaction behavior data after marketing information is displayed on users' mobile phones;

[0082] Specifically, it includes:

[0083] When a shopping app on a user's mobile phone receives a product marketing image display, it obtains corresponding user interaction behavior information data, which includes:

[0084] Data information includes clicks on product marketing images, interaction time after entering the product page, add-to-cart and favorites data, and product details and review viewing data.

[0085] Click data information for product marketing images: When a shopping app on a user's mobile phone displays product marketing images, the software development kit (SDK) or JavaScript tracking technology is used to accurately capture data such as the time, coordinates, and device model of the user clicking on the product marketing images;

[0086] Interaction time data after entering the product page: After a user enters the product page, the time difference between entering the page and leaving the page is continuously recorded using a front-end timer (such as setInterval) as interaction time information. At the same time, page switching events (such as visibilitychange) are combined to handle situations such as the user switching out of the application.

[0087] Add to cart and favorite data: Bind event listeners to the "Add to cart" and "Favorite" buttons in the shopping app. When a user triggers the action, immediately obtain information such as the product ID, action time, and user account.

[0088] Product details and review viewing data: Embed tracking code or integrate SDK into the front-end page of the shopping software to monitor user behavior in real time: On the product details page, use scroll events (such as scroll) to calculate the user's browsing depth (page scroll percentage), record the dwell time through lifecycle hooks (such as onLoad / onUnload), and bind interactive events (such as play / click) to elements such as videos and parameter switching to obtain content interaction data; Set load listener events in the product buyer review area, and when the page loads the review content, record the trigger time for the user to view the review, the number of reviews viewed, whether all reviews were expanded, and other information.

[0089] Behavioral analysis: Matching coefficients and interest coefficients are obtained by analyzing user interaction behavior data;

[0090] The process of obtaining the matching coefficient includes:

[0091] The pause time between each page swipe when the user is browsing the product details page is obtained, and a pause time threshold for page swipe is preset. The pause time between each page swipe is compared with the pause time threshold. The pause time between page swipes that is less than the pause time threshold is recorded as normal pause time, and the pause time between page swipes that is greater than the pause time threshold is recorded as abnormal pause time.

[0092] After counting the number of all normal pause times and the number of pause times between all page scrolling, the normal browsing ratio is obtained by dividing the number of all normal pause times by the number of pause times between all page scrolling.

[0093] The normal browsing ratio reflects the percentage of pauses that follow a normal rhythm during a user's browsing process. The higher the normal browsing ratio, the smoother and more natural the user's browsing behavior is. For example, if the total number of pauses is 10 and the number of normal pauses is 8, then the normal browsing ratio is 0.8.

[0094] The total normal pause time is obtained by summing up all normal pause times;

[0095] The total abnormal pause time is obtained by summing up all abnormal pause times.

[0096] Get the time when the user clicks on a product and the time when the user exits browsing the product, and record the difference between the two times as the total browsing time;

[0097] Divide the total normal pause time by the total browsing time to obtain the effective time ratio;

[0098] Subtract the total abnormal pause time from the total browsing time to obtain the effective browsing time;

[0099] Identify all keywords for the product, including specifications, after-sales policy, and related recommendations;

[0100] Obtain the times when keywords appear and disappear on the user's mobile screen while browsing product details; subtract the appearance time from the disappearance time to obtain the keyword appearance duration;

[0101] The occurrence duration of each keyword is obtained sequentially, and these durations are accumulated to obtain the total occurrence duration of the keywords; the total occurrence duration of the keywords is divided by the effective browsing time to obtain the keyword ratio;

[0102] The keyword ratio reflects the degree of attention users pay to the key information of a product. The higher the keyword ratio, the more users pay attention to the core information of the product.

[0103] The matching coefficient is obtained by weighting the normal browsing ratio, effective time ratio, and keyword ratio.

[0104] After setting the weighting factors for normal browsing ratio, effective time ratio, and keyword ratio, the matching coefficient is obtained by multiplying the normal browsing ratio, effective time ratio, and keyword ratio with their corresponding weighting factors.

[0105] The process of obtaining the interest coefficient includes:

[0106] The comment interaction value is obtained by analyzing the interaction data in the comment section;

[0107] The user profile matching value is obtained by analyzing the data of similar products browsed by the user within a preset time period.

[0108] Image values ​​are obtained by analyzing user browsing behavior on product detail pages and in comment sections.

[0109] The interest coefficient is obtained by weighting the comment interaction value, profile matching value, and image value. The interest coefficient is obtained by multiplying the interaction value, profile matching value, and image value with their corresponding weight factors and summing them.

[0110] The process of obtaining comment interaction value includes:

[0111] Get the time when a user opens the comment section and the time when they exit the comment section, and record the time difference between the two times as the comment dwell time;

[0112] Get the number of comments that users clicked to view, as well as the corresponding interactive behaviors for each comment, including clicking "view more comments", "reply", and "like".

[0113] The interaction rate is obtained by summing up the interaction behaviors corresponding to each comment, dividing the total number of interactions by the comment dwell time.

[0114] The interaction rate reflects the frequency with which users interact with the content in the comment section within a unit of time. The higher the interaction rate, the higher the user's participation in the comment section, and the higher their interest or attention to the product may be.

[0115] Using natural language processing technology, the comments viewed by users are matched with a pre-set keyword database (such as words related to the core selling points of the product and common concerns), and the number of hits is counted.

[0116] Suppose the keyword library contains words such as "durable" and "battery life". If the user sees a review that mentions "durable", the hit count is 1.

[0117] The content of each comment viewed by the user is matched against a preset keyword database in turn to obtain the number of times the keyword is hit for each comment;

[0118] Using sentiment analysis models in natural language processing (such as BERT-based sentiment classification models), the comments viewed by users are scored with sentiment, with scores ranging from -1 to 1, where -1 is negative and 1 is positive.

[0119] Sort the hit counts of each keyword in descending order of numerical value, and extract the three keywords with the highest hit counts;

[0120] Multiply the number of times the three keywords hit the most by the corresponding sentiment score of the comment to obtain the sentiment score of the number of times the three keywords hit the most.

[0121] The three sentiment scores are averaged. If the average value is greater than 0, it is marked as the sentiment average.

[0122] A sentiment score greater than 0 indicates that users' overall sentiment attitude towards key information in the comments is biased towards a positive one;

[0123] The comment interaction value is obtained by comprehensively calculating the interaction rate and the sentiment score.

[0124] Substitute the interaction rate and sentiment equalization into the formula as SI and EM respectively: The comment interaction value ζ is obtained; where EM′ is the reference sentiment average score; a1 and a2 are the weighting factors corresponding to the interaction rate and sentiment average score, respectively;

[0125] The process of obtaining profile matching values ​​includes:

[0126] Get the user's search or favorites of similar products and their corresponding purchase prices within a preset historical time range;

[0127] Within a preset historical timeframe (e.g., the last 30 days, 90 days), information on similar products searched, favorited, or purchased by users, along with their corresponding purchase prices, is obtained. This step utilizes the e-commerce platform's user behavior database to filter similar products based on product category tags (e.g., "laptops" or "sports shoes"). For example, if a user purchased three different brands of laptops in the past 30 days, the names and purchase prices of these three products are recorded. This data provides a foundation for subsequent analysis, reflecting the user's historical consumption preferences.

[0128] The product details pages of each similar product purchased are retrieved sequentially, and the corresponding product keywords are extracted from them. The keywords of all similar products purchased are then unified to obtain a similar product keyword library.

[0129] The system sequentially visits the product detail pages of similar products purchased by the user, using natural language processing techniques (such as text extraction and keyword extraction algorithms) to extract keywords from fields such as product titles, descriptions, and specifications. For example, keywords such as "Core i7," "16GB RAM," and "512GB SSD" are extracted from a laptop's detail page. All extracted keywords from similar products are then deduplicated, merged, and processed to create a keyword library for similar products. This keyword library represents the core characteristics of the user's historical product purchases.

[0130] Match each keyword of the marketing product with a keyword database of similar products, count the number of successful matches, and record it as a success value;

[0131] Divide the total number of keywords for the marketing product by the total number of keywords in the keyword library for similar products to obtain the overlap rate;

[0132] The overlap ratio is calculated by dividing the total number of keywords for the marketing product by the total number of keywords in the keyword library for similar products. For example, if the marketing product has 5 keywords and the keyword library for similar products has 20 keywords, the overlap ratio is 0.25. The overlap ratio reflects the overall correlation between the marketing product's keywords and the keywords based on users' historical preferences.

[0133] The average price of similar products is obtained by averaging the prices of all similar products.

[0134] The price difference is calculated by comparing the price of the marketed product with the average price of similar products. The allowable fluctuation range of the price difference is preset, and the price difference is matched with the allowable fluctuation range. If the price difference is within the allowable fluctuation range, it is recorded as the reasonable price difference.

[0135] The profile matching value is obtained by comprehensively processing the success value, overlap, and reasonable price difference.

[0136] The success value, overlap, and reasonable price difference are preset as weighting factors. The success value, overlap, and reasonable price difference are multiplied by their corresponding weighting factors to calculate the profile matching value.

[0137] The process of obtaining image values ​​includes:

[0138] Monitor changes in the width / height properties of the image's parent container. If the size exceeds a preset threshold (e.g., exceeding 50% of the screen's visible area), it is determined to be an image enlargement operation.

[0139] Obtain the marketing product area corresponding to the user's zoomed-in image, and extract the zoom-in dwell time of that area;

[0140] The moment when the image displayed on the user's mobile phone screen is magnified is recorded as the first time, and the time when the image displayed on the screen changes is recorded as the second time. The time difference between the first time and the second time is calculated to obtain the magnification dwell time; the unit of magnification dwell time is seconds.

[0141] The overall appearance of the marketing product is divided into multiple areas, and each area is distinguished based on its own characteristics;

[0142] The system acquires images displayed on the user's mobile phone screen, extracts relevant marketing product features from them, and identifies the marketing product area corresponding to the image.

[0143] The system sequentially retrieves the marketing product area corresponding to each time a user zooms in on an image, and counts the number of zooms in on each area of ​​the marketing product; it then extracts the two highest zoom counts and the image zoom dwell time corresponding to the two highest zoom counts.

[0144] Multiply the two largest magnification counts by their respective magnification dwell times to obtain the quantization values;

[0145] The two quantized values ​​are used as the major and minor axes of the ellipse, respectively, to construct an ellipse model, and the area of ​​the ellipse model is recorded as the image value.

[0146] Comprehensive judgment: The basic evaluation coefficient is obtained by weighting the matching coefficient and the interest coefficient. The intention evaluation coefficient is obtained by combining the basic evaluation coefficient with the coefficients corresponding to the actions of adding to cart and favorites.

[0147] After normalizing the matching coefficient and interest coefficient, a circle is constructed using the matching coefficient as the radius and the interest coefficient is constructed using the height of the circle to establish a cone model. The volume of the cone model is calculated to obtain the basic evaluation coefficient.

[0148] If a user adds a marketing product to their shopping cart, the action value for the "add to cart" action is recorded as 1; otherwise, it is recorded as -1.

[0149] Similarly, if a user adds a marketing product to their favorites, the action value for the "favorite" action is recorded as 1, and vice versa, it is -1.

[0150] Preset the weighting factors corresponding to the actions of adding to cart and adding to favorites. Then, multiply the action value of adding to cart and the action value of adding to favorites with their corresponding weighting factors and sum them to obtain the coefficient of adding to cart and adding to favorites.

[0151] After normalizing the basic evaluation coefficient and the add-to-cart / favorite coefficient, determine whether the add-to-cart / favorite coefficient is greater than 0:

[0152] If the add-to-cart and add-to-collect coefficient is greater than 0, then the basic evaluation coefficient and the add-to-cart and add-to-collect coefficient are respectively used as the two legs of a right triangle, and the remaining leg is connected to form a complete right triangle. The area of ​​the right triangle is recorded as the intention evaluation coefficient.

[0153] If the add-to-cart and add-to-collect coefficient is less than 0, then the basic evaluation coefficient and the add-to-cart and add-to-collect coefficient are respectively used as two legs of a right triangle. Connect the remaining leg to form a complete right triangle. Then, take the center of the leg corresponding to the add-to-collect coefficient as the origin and half of the add-to-collect coefficient as the radius to build a circle. This circle will cut the right triangle into areas. Calculate the area of ​​the remaining right triangle and record it as the intention evaluation coefficient.

[0154] Intention Diagnosis and Assessment: Determine the degree of user's purchase intention for the marketed product based on the intention assessment coefficient;

[0155] Three threshold ranges are preset, and each threshold range corresponds to a purchase intention level. The intention evaluation coefficient is compared with the three threshold ranges to obtain the purchase intention level corresponding to the intention evaluation coefficient. The purchase intention levels include low intention, medium intention and high intention.

[0156] The above formulas are derived from software simulations using a large amount of data and are selected to be close to the actual values. The influence weight factors and specific coefficient values ​​in the formulas are set by those skilled in the art based on the actual situation and can be adjusted and modified in the future.

[0157] The above description of the embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not 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. An e-commerce analysis and diagnostic method based on artificial intelligence, characterized in that, include: Behavioral data collection: Obtaining user interaction behavior data after marketing information is displayed on users' mobile phones; Behavioral analysis: Matching coefficients and interest coefficients are obtained by analyzing user interaction behavior data; The process of obtaining the interest coefficient includes analyzing users' browsing behavior on the product details page and in the comment section to obtain image values; The process of obtaining image values ​​includes: Monitor changes in the properties of the image's parent container; if the size exceeds a preset threshold, it is determined to be an image enlargement operation. Obtain the marketing product area corresponding to the user's zoomed-in image, and extract the zoom-in dwell time of that area; The overall appearance of the marketing product is divided into multiple areas, and each area is distinguished based on its own characteristics; The system acquires images displayed on the user's mobile phone screen, extracts relevant marketing product features from them, and identifies the marketing product area corresponding to the image. The system sequentially retrieves the marketing product area corresponding to each time a user zooms in on an image, and counts the number of zooms in on each area of ​​the marketing product; it then extracts the two highest zoom counts and the image zoom dwell time corresponding to the two highest zoom counts. Multiply the two largest magnification counts by their respective magnification dwell times to obtain the quantization values; The two quantized values ​​are used as the major and minor axes of the ellipse, respectively, to construct an ellipse model, and the area of ​​the ellipse model is recorded as the image value. The basic evaluation coefficient is obtained by weighting the matching coefficient and the interest coefficient, including: After normalizing the matching coefficient and interest coefficient, a circle is constructed using the matching coefficient as the radius and the interest coefficient is constructed using the height of the circle to establish a cone model. The volume of the cone model is calculated to obtain the basic evaluation coefficient. If a user adds a marketing product to their shopping cart, the action value for the "add to cart" action is recorded as 1; otherwise, it is recorded as -1. Similarly, if a user adds a marketing product to their favorites, the action value for the "favorite" action is recorded as 1, and vice versa, it is -1. Preset the weighting factors corresponding to the actions of adding to cart and adding to favorites. Then, multiply the action value of adding to cart and the action value of adding to favorites with their corresponding weighting factors and sum them to obtain the coefficient of adding to cart and adding to favorites. After normalizing the basic evaluation coefficient and the add-to-cart / favorite coefficient, determine whether the add-to-cart / favorite coefficient is greater than 0: If the add-to-cart and add-to-collect coefficient is greater than 0, then the basic evaluation coefficient and the add-to-cart and add-to-collect coefficient are respectively used as the two legs of a right triangle, and the remaining leg is connected to form a complete right triangle. The area of ​​the right triangle is recorded as the intention evaluation coefficient. If the add-to-cart and add-to-collect coefficient is less than 0, then the basic evaluation coefficient and the add-to-cart and add-to-collect coefficient are respectively used as two legs of a right triangle. Connect the remaining leg to form a complete right triangle. Then, take the center of the leg corresponding to the add-to-collect coefficient as the origin and half of the add-to-collect coefficient as the radius to build a circle. This circle will cut the right triangle into areas. Calculate the area of ​​the remaining right triangle and record it as the intention evaluation coefficient. Comprehensive judgment: The basic evaluation coefficient is obtained by weighting the matching coefficient and the interest coefficient. The intention evaluation coefficient is obtained by combining the basic evaluation coefficient with the coefficients corresponding to the actions of adding to cart and favorites. Intention Diagnosis and Assessment: Determine the degree of user's purchase intention for the marketed product based on the intention assessment coefficient.

2. The e-commerce analysis and diagnosis method based on artificial intelligence according to claim 1, characterized in that, Behavioral data collection specifically includes: When a shopping app on a user's mobile phone receives a product marketing image display, it obtains corresponding user interaction behavior information data, which includes: Data information includes clicks on product marketing images, interaction time after entering the product page, add-to-cart and favorite data, and viewing of product details and reviews.

3. The e-commerce analysis and diagnosis method based on artificial intelligence according to claim 2, characterized in that, The process of obtaining the matching coefficient includes: The pause time between each page swipe when the user is browsing the product details page is obtained, and a pause time threshold for page swipe is preset. The pause time between each page swipe is compared with the pause time threshold. The pause time between page swipes that is less than the pause time threshold is recorded as normal pause time, and the pause time between page swipes that is greater than the pause time threshold is recorded as abnormal pause time. After counting the number of all normal pause times and the number of pause times between all page scrolling, the normal browsing ratio is obtained by dividing the number of all normal pause times by the number of pause times between all page scrolling. The total normal pause time is obtained by summing up all normal pause times; The total abnormal pause time is obtained by summing up all abnormal pause times. Get the time when the user clicks on a product and the time when the user exits browsing the product, and record the difference between the two times as the total browsing time; Divide the total normal pause time by the total browsing time to obtain the effective time ratio; Subtract the total abnormal pause time from the total browsing time to obtain the effective browsing time; Identify all keywords for the product, including specifications, after-sales policy, and related recommendations; Obtain the times when keywords appear and disappear on the user's mobile screen while browsing product details; subtract the appearance time from the disappearance time to obtain the keyword appearance duration; The occurrence duration of each keyword is obtained sequentially, and these durations are accumulated to obtain the total occurrence duration of the keywords; the total occurrence duration of the keywords is divided by the effective browsing time to obtain the keyword ratio; The matching coefficient is obtained by weighting the normal browsing ratio, effective time ratio, and keyword ratio.

4. The e-commerce analysis and diagnosis method based on artificial intelligence according to claim 3, characterized in that, The process of obtaining the interest coefficient includes: The comment interaction value is obtained by analyzing the interaction data in the comment section; The user profile matching value is obtained by analyzing the data of similar products viewed by the user within a preset time period. The interest coefficient is obtained by weighting the comment interaction value, profile matching value, and image value.

5. The e-commerce analysis and diagnosis method based on artificial intelligence according to claim 4, characterized in that, The process of obtaining comment interaction value includes: Get the time when a user opens the comment section and the time when they exit the comment section, and record the time difference between the two times as the comment dwell time; Get the number of comments that users clicked to view, as well as the corresponding interactive behaviors for each comment, including clicking "view more comments", "reply", and "like". The interaction rate is obtained by summing up the interaction behaviors corresponding to each comment, dividing the total number of interactions by the comment dwell time. Using natural language processing technology, the comments viewed by users are matched with a pre-set keyword database, and the number of hits is counted. The content of each comment viewed by the user is matched against a preset keyword database in turn to obtain the number of times the keyword is hit for each comment; Using a sentiment analysis model from natural language processing, the comments viewed by users are scored with sentiment, with a score range of -1 to 1, where -1 is negative and 1 is positive. Sort the hit counts of each keyword in descending order of numerical value, and extract the three keywords with the highest hit counts; Multiply the number of times the three keywords hit the most by the corresponding sentiment score of the comment to obtain the sentiment score of the number of times the three keywords hit the most. The three sentiment scores are averaged. If the average value is greater than 0, it is marked as the sentiment average. The comment interaction value is obtained by combining the interaction rate and the sentiment score.

6. The e-commerce analysis and diagnosis method based on artificial intelligence according to claim 5, characterized in that, The process of obtaining profile matching values ​​includes: Get the user's search or favorites of similar products and their corresponding purchase prices within a preset historical time range; The product details pages of each similar product purchased are retrieved sequentially, and the corresponding product keywords are extracted from them. The keywords of all similar products purchased are then unified to obtain a similar product keyword library. Match each keyword of the marketing product with a keyword database of similar products, count the number of successful matches, and record it as a success value; Divide the total number of keywords for the marketing product by the total number of keywords in the keyword library for similar products to obtain the overlap rate; The average price of similar products is obtained by averaging the prices of all similar products. The price difference is calculated by comparing the price of the marketed product with the average price of similar products. The allowable fluctuation range of the price difference is preset, and the price difference is matched with the allowable fluctuation range. If the price difference is within the allowable fluctuation range, it is recorded as the reasonable price difference. The profile matching value is obtained by comprehensively processing the success value, overlap, and reasonable price difference.

7. The e-commerce analysis and diagnosis method based on artificial intelligence according to claim 1, characterized in that, The intention assessment coefficient is used to determine the degree of user's purchase intention for the marketed products, specifically including: Three threshold ranges are preset, and each threshold range corresponds to a purchase intention level. The intention evaluation coefficient is compared with the three threshold ranges to obtain the purchase intention level corresponding to the intention evaluation coefficient. The purchase intention levels include low intention, medium intention and high intention.

Images