User feature extraction method of e-commerce recommendation model

Abstract

The invention relates to the technical field of electronic commerce data processing, in particular to a user feature extraction method of an electronic commerce recommendation model. The method comprises the following steps: acquiring an electronic commerce data set; performing data preprocessing on the electronic commerce data set to obtain a standard electronic commerce data set; commodity order information of the standard electronic commerce data set is extracted, and commodity order data is generated; order processing time recording is carried out on the commodity order data to generate order processing time data; performing order completion rate statistics on the commodity order data according to the order processing time data to obtain order completion rate data; and carrying out commodity type identification on the order completion rate data to obtain commodity order type data. Through a data processing technology, an image mapping technology, an intelligent control technology and a cloud platform technology, commodity demand characteristics of a user are accurately and objectively reflected, a feedback mechanism is perfected, and a recommendation result is intelligently adjusted, so that the accuracy of an electronic commerce recommendation model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of e-commerce data processing, and in particular to a method for extracting user features of an e-commerce recommendation model. Background Art

[0002] The development of e-commerce recommendation systems has gone from simple content-based recommendations to complex collaborative filtering recommendations, and then to the current hybrid recommendations and deep learning-based recommendation models. Initially, recommendation systems mainly relied on commodity attribute information, such as category, brand, and price, for simple similarity matching. Over time, the development of technology has enabled recommendation systems to use users' historical behavior data, such as purchase records and browsing history, to provide personalized recommendations. Furthermore, in order to improve the accuracy of recommendations and user satisfaction, recommendation systems have begun to integrate multiple data sources, including user reviews, social network information, and search queries. In addition, with the development of machine learning technology, recommendation systems have begun to adopt more complex algorithms, such as matrix decomposition and deep neural networks, to explore users' potential preferences and needs. However, the user feature extraction methods of existing e-commerce recommendation models tend to analyze users' subjective behavior, lack technical objectivity, and are difficult to accurately reflect users' commodity demand characteristics. In addition, the feedback mechanism is imperfect, resulting in the optimization and adjustment of recommendation results being not intelligent and accurate enough. Summary of the invention

[0003] Based on this, it is necessary to provide a user feature extraction method for an e-commerce recommendation model to solve at least one of the above technical problems.

[0004] To achieve the above purpose, a method for extracting user features of an e-commerce recommendation model is provided, the method comprising the following steps:

[0005] Step S1: obtaining an e-commerce data set; performing data preprocessing on the e-commerce data set to obtain a standard e-commerce data set;

[0006] Step S2: extracting commodity order information from a standard e-commerce data set to generate commodity order data; recording order processing time for the commodity order data to generate order processing time data; performing order completion rate statistics on the commodity order data according to the order processing time data to obtain order completion rate data; performing commodity type identification on the order completion rate data to obtain commodity order type data;

[0007] Step S3: extracting product page browsing information of the standard e-commerce data set to generate product page browsing data; performing page interaction depth evaluation on the product page browsing data to obtain page interaction evaluation data; identifying product detail page access conditions on the product page browsing data according to the page interaction evaluation data to obtain product detail page access data;

[0008] Step S4: extracting the product shopping cart information of the standard e-commerce data set to generate product shopping cart data; calculating the product storage time of the product shopping cart data to generate product storage time data; performing product deletion analysis on the product shopping cart data according to the product storage time data to obtain deleted product data;

[0009] Step S5: dividing the product order type data, product detail page access data and deleted product data by influence weight to obtain product influence weight data; constructing an e-commerce recommendation model based on the product influence weight data to obtain an e-commerce recommendation model; recommending products through the e-commerce recommendation model to generate product recommendation data;

[0010] Step S6: monitor the click rate of the product recommendation data to obtain the recommended product click rate data; analyze the interactive characteristics of the product recommendation data to obtain the recommended product interactive data; record the skipping of the product recommendation data to obtain the recommended product skipping data; integrate the recommended product click rate data, the recommended product interactive data and the recommended product skipping data to obtain the recommended product feedback data;

[0011] Step S7: Based on the recommended product feedback data, intelligently adjust the product recommendation data to generate product recommendation adjustment data; perform real-time adjustment feedback on the e-commerce recommendation model according to the product recommendation adjustment data to obtain a real-time adjustment feedback report; construct a cloud platform according to the real-time adjustment feedback report to obtain a real-time adjustment feedback cloud platform; execute the user feature extraction method operation of the e-commerce recommendation model based on the real-time adjustment feedback cloud platform.

[0012] The present invention ensures the quality and consistency of data by acquiring an e-commerce data set and performing data preprocessing, and provides a standardized e-commerce data set for subsequent analysis. By extracting commodity order information and generating order data, the order performance of different commodity types can be accurately identified by recording the order processing time and statistically calculating the order completion rate, providing data support for understanding the user's purchase tendency for various commodities. By analyzing the commodity page browsing data and evaluating the page interaction depth, the user's access to the commodity details page can be identified, thereby gaining insight into the user's detailed interest in specific commodities. By measuring the commodity storage time and analyzing the deleted commodity data, the user's retention time in the shopping cart and the commodities that are eventually deleted can be understood, which helps to understand the user's purchase decision process. The commodity order type data, the commodity details page access data and the deleted commodity data are weighted, and the commodity impact weight data is constructed, which provides key user behavior features for the e-commerce recommendation model and enhances the personalization and accuracy of the recommendation model; by monitoring the click-through rate, interactive features and skipping of the recommended commodities, the feedback data of the recommended commodities is collected, providing direct quantitative indicators for evaluating the recommendation effect and user acceptance. Based on the feedback data of recommended products, the recommended data is intelligently adjusted to achieve the optimization adjustment and real-time update of the recommendation model. The constructed real-time adjustment feedback cloud platform further strengthens the dynamic adjustment capability of the recommendation model and ensures the timeliness and user satisfaction of the recommended content. The present invention uses data processing technology, image mapping technology, intelligent control technology and cloud platform technology to accurately and objectively reflect the user's product demand characteristics, improve the feedback mechanism, and intelligently adjust the recommendation results, thereby improving the accuracy of the e-commerce recommendation model.

[0013] Preferably, step S2 comprises the following steps:

[0014] Step S21: extracting commodity order information from a standard e-commerce data set to generate commodity order data;

[0015] Step S22: Analyze the processing status of the commodity order data to obtain commodity order processing status data; identify the timestamp of the commodity order processing status data to obtain order processing timestamp data; record the order processing time of the order processing timestamp data to generate order processing time data;

[0016] Step S23: performing payment feature recognition on the commodity order data to obtain order payment feature data; performing payment attribute classification on the order payment feature data to obtain order payment attribute data, extracting paid order attributes from the order payment attribute data to obtain paid order attribute data;

[0017] Step S24: Calculate the ratio of the paid order attribute data to the commodity order data to generate order percentage data; perform time-based statistics on the order percentage data according to the order processing time data to obtain order completion rate data;

[0018] Step S25: extracting commodity type features of the standard e-commerce data set to obtain commodity type feature data; numbering the commodity type feature data to obtain commodity type number data;

[0019] Step S26: Map the order completion rate data to commodity types based on the commodity type number data to obtain commodity order type data.

[0020] The present invention generates commodity order data by extracting commodity order information from standard e-commerce data sets, which provides basic data support for subsequent order feature analysis. The processing status of commodity order data is analyzed, the order processing timestamp is identified, the order processing time is recorded, and an accurate time benchmark is provided for analysis and user order processing. The order payment characteristics are identified and the payment attributes are divided, and the paid order attributes are extracted, which helps to distinguish the user's payment behavior and provides data support for understanding the user's willingness and ability to pay. By calculating the ratio of the paid order attribute data to the commodity order data, the order percentage data is generated, and then the order processing time is combined with the time period statistics to obtain the order completion rate data, which provides a quantitative indicator for evaluating the efficiency of the order process and user satisfaction. The commodity type characteristics are extracted and numbered to generate commodity type number data, which provides a standardized method for commodity classification and feature management. The order completion rate data is mapped based on the commodity type number data to obtain the commodity order type data, which helps to analyze the order completion of different types of commodities and provides a basis for optimizing commodity management and recommendation strategies.

[0021] Preferably, step S3 comprises the following steps:

[0022] Step S31: extracting product page browsing information of a standard e-commerce data set to generate product page browsing data;

[0023] Step S32: Record user click information on the product page browsing data to obtain user click data; calculate page dwell time on the user click data to obtain page dwell time calculation data; sort the page dwell time calculation data in descending order to obtain page dwell time descending order data;

[0024] Step S33: Perform heat map color mapping on the descending data of page dwell time to generate a page dwell time heat map; extract the dark coordinate data of the page dwell time heat map to obtain long-stay page data;

[0025] Step S34: performing page scroll depth analysis on the product page browsing data based on the long-stay page data to obtain page scroll depth data; performing page interaction depth evaluation on the product page browsing data based on the page scroll depth data to obtain page interaction evaluation data; identifying product detail page access situations on the product page browsing data based on the page interaction evaluation data to obtain product detail page access data.

[0026] The present invention generates product page browsing data by extracting product page browsing information from a standard e-commerce data set, which provides basic data for understanding user browsing behavior; records user click information and measures page dwell time, and obtains descending data of page dwell time by arranging in descending order, which helps to identify that users pay more attention to specific pages. Through heat map color mapping, a page dwell time heat map is generated, and the dark coordinate data in the heat map is extracted as long-stay page data, which provides an intuitive representation for visualizing the hot spots of user stay on the page. Based on the long-stay page data, a page scroll depth analysis is performed to obtain page scroll depth data, which helps to evaluate the user's in-depth exploration of the page content; further, the product page browsing data is evaluated for page interaction depth based on the page scroll depth data to obtain page interaction evaluation data, which provides a basis for identifying the user's access to the product details page, and finally generates product details page access data.

[0027] Preferably, step S34 includes the following steps:

[0028] Step S341: Record the page depth of the product page browsing data to obtain product browsing page depth data; analyze the page scrolling situation of the product browsing page depth data according to the long-stay page data to obtain page scrolling situation data;

[0029] Step S342: extracting the numerical information of the product browsing page depth data to obtain the product browsing page depth value; performing a preset depth calculation on the product browsing page depth value; and generating page preset depth data;

[0030] Step S343: extracting the depth value of the page scrolling situation data to obtain page scrolling depth data; comparing the page scrolling depth data with the page preset depth data, if the page scrolling depth data is greater than the page preset depth data, marking the page scrolling depth data as page high scrolling data; if the page scrolling depth data is less than or equal to the page preset depth data, marking the page scrolling depth data as page low scrolling data; integrating the page high scrolling data and the page low scrolling data to obtain page interaction evaluation data;

[0031] Step S344: perform product detail page feature recognition on the product page browsing data to obtain product detail page feature data; perform access situation analysis on the product detail page feature data based on the page interaction evaluation data to obtain product detail page access data, wherein the product detail page access data includes detail page non-visited data and detail page visited data.

[0032] The present invention records the page depth through the product page browsing data, generates the product browsing page depth data, and helps to quantify the user's browsing depth of each page; by analyzing the page scrolling situation through the long-term stay page data, the page scrolling situation data is obtained, which provides important information for understanding the degree of user interaction on the page. The numerical information of the product browsing page depth data is extracted and the preset depth calculation is performed to generate the page preset depth data; a quantitative standard is provided for evaluating the user browsing behavior, which is helpful for the subsequent page scrolling depth evaluation. The page scrolling depth data is obtained by extracting the depth value of the page scrolling situation data; the page scrolling depth data is compared with the page preset depth data, and the page scrolling depth data is classified into page high scrolling data and page low scrolling data, and the page interaction evaluation data is obtained by integration, which can identify the user's in-depth exploration of the page content and provide a basis for evaluating the user's interest points. The product details page features are identified on the product page browsing data to obtain product details page feature data. The access situation of the product details page feature data is analyzed based on the page interaction evaluation data to obtain product details page access data, wherein the product details page access data includes details page non-visited data and details page visited data, providing a clear indicator for understanding whether the user has conducted an in-depth browsing of the product details page.

[0033] Preferably, step S4 comprises the following steps:

[0034] Step S41: extracting the commodity shopping cart information of the standard e-commerce data set to generate commodity shopping cart data;

[0035] Step S42: Record the time of adding the product to the shopping cart data to generate product purchase time point data;

[0036] Step S43: performing payment feature recognition on the shopping cart data to obtain shopping cart settlement data; extracting the time point data of the shopping cart settlement data to obtain the shopping cart settlement time point data; calculating the settlement commodity storage time of the shopping cart data according to the shopping cart settlement time point data and the commodity purchase time point data to generate the settlement commodity storage time data;

[0037] Step S44: performing removal feature recognition on the shopping cart data to obtain shopping cart removal data; extracting the time point data of the shopping cart removal data to obtain the shopping cart removal time point data; calculating the storage time of the removed goods on the shopping cart data according to the shopping cart removal time point data and the product purchase time point data to generate the removed goods storage time data;

[0038] Step S45: Integrate the settled commodity storage time data and the removed commodity storage time data to generate commodity storage time data; perform a deleted commodity analysis on the commodity shopping cart data according to the commodity storage time data to obtain deleted commodity data, wherein the deleted commodity data includes the settled commodity data and the removed commodity data.

[0039] The present invention generates shopping cart data by extracting the information of the commodity shopping cart in the standard e-commerce data set, providing basic data for analyzing the user's shopping behavior; recording the time point when the user adds the shopping cart, generating the commodity purchase time point data, and clarifying the starting point of the user's interest in the commodity; identifying the payment characteristics of the shopping cart and extracting the settlement time point data, and calculating the settlement commodity storage time by combining with the commodity purchase time point data, and generating the settlement commodity storage time data, which can clearly analyze the user's decision cycle from adding the shopping cart to settlement. Identify the removal characteristics of the commodity in the shopping cart and extract the removal time point data, and calculate the removal commodity storage time by combining with the commodity purchase time point data, and generate the removal commodity storage time data; identify the user's reconsideration and final abandonment of the commodity in the shopping cart; integrate the settlement commodity storage time and the removal commodity storage time data, and generate the commodity storage time data, and perform a commodity deletion analysis on the commodity storage time data to obtain the deleted commodity data, including the settlement commodity data and the removal commodity data, so as to provide key information for clarifying the user's final purchase decision, including the commodities that the user chooses to purchase and abandon.

[0040] Preferably, step S5 comprises the following steps:

[0041] Step S51: dividing the product order type data, product detail page access data and deleted product data by impact weight to obtain product impact weight data;

[0042] Step S52: constructing an e-commerce recommendation model according to the commodity influence weight data to obtain an e-commerce recommendation pre-model; and using the commodity influence weight data to perform model training on the e-commerce recommendation pre-model to obtain an e-commerce recommendation training model;

[0043] Step S53: performing model cross-validation evaluation on the e-commerce recommendation training model to obtain e-commerce recommendation evaluation data; adjusting model parameters of the e-commerce recommendation training model using the e-commerce recommendation evaluation data to obtain an e-commerce recommendation model;

[0044] Step S54: Recommend products through the e-commerce recommendation model to generate product recommendation data.

[0045] The present invention divides the influence weights through product order type data, product detail page access data and deleted product data, generates product influence weight data, assigns corresponding importance to different user behavior characteristics, and helps to more accurately reflect user preferences. The product influence weight data is used to build an e-commerce recommendation pre-model, and through the model training process, it is ensured that the recommendation model can make effective personalized recommendations based on user behavior characteristics. Through model cross-validation evaluation, e-commerce recommendation evaluation data is obtained, which helps to fully understand the performance of the recommendation model; the model parameters are adjusted based on the evaluation data, and the e-commerce recommendation training model is further optimized to improve the accuracy and robustness of the model. The e-commerce recommendation model with adjusted parameters is used to make product recommendations and generate product recommendation data; the optimized recommendation model is directly applied to the actual recommendation process to provide users with more personalized and accurate product recommendations.

[0046] Preferably, step S51 includes the following steps:

[0047] Step S511: performing order type similarity analysis on the commodity order type data to obtain commodity type similarity data;

[0048] Step S512: performing association analysis on the settlement product data according to the product detail page access data to obtain settlement product detail page association data; identifying the number of visits to the details page visited data according to the settlement product detail page association data to generate details page visit number data; extracting multiple visit information from the details page visit number data to obtain details page multiple visit data;

[0049] Step S513: Analyze the reasons why the removed product data has not been accessed through the details page not accessed data to obtain the details page not accessed reason data;

[0050] Step S514: constructing a user product type feature matrix for the product type similarity data based on the details page multiple access data and the details page non-access reason data to obtain a user product type feature matrix;

[0051] Step S515: Perform multiple linear regression analysis on the user product type feature matrix to generate multiple linear regression coefficients; perform significant impact analysis on the user product type feature matrix based on the multiple linear regression coefficients to obtain product significant impact data; perform weight analysis on the product significant impact data to obtain product impact weight data.

[0052] The present invention performs order type similarity analysis through commodity order type data to obtain commodity type similarity data, which can identify the correlation between commodity types of interest to users. According to commodity detail page access data, settlement commodity data is subjected to correlation analysis to generate settlement commodity detail page correlation data; by identifying the number of visits, detail page access number data is obtained; multiple visit information is extracted to obtain detail page multiple visit data, which can identify users' repeated attention and in-depth consideration of specific commodities. The removed commodity data is subjected to non-visit reason analysis through detail page non-visit data to obtain detail page non-visit reason data, which helps to understand the reasons why users do not have a deep understanding of specific commodities. Based on the detail page multiple visit data and detail page non-visit reason data, a user commodity type feature matrix is ​​constructed, which provides a comprehensive data framework for analyzing user preferences and behavior patterns for different commodity types. Multiple linear regression analysis is performed on the user commodity type feature matrix to generate multiple linear regression coefficients, which can quantify the influence of different commodity features on user purchase decisions. Through significant influence analysis, commodity significant influence data is obtained, and commodity features that have a significant influence on user purchase behavior are further identified. Weight analysis is performed on commodity significant influence data to obtain commodity influence weight data, which provides accurate feature weights for recommendation models.

[0053] Preferably, step S6 comprises the following steps:

[0054] Step S61: Count the number of times the product recommendation data is displayed to obtain display number statistics; monitor the number of clicks on the product recommendation data to obtain recommended product click number data; calculate the ratio of the recommended product click number data and the display number statistics to obtain recommended product click rate data;

[0055] Step S62: extracting collection features from the product recommendation data according to the recommended product click rate data to obtain recommended product collection attribute data; identifying purchase records from the recommended product collection attribute data to obtain recommended product purchase data; analyzing purchase situations from the recommended product purchase data to obtain recommended product purchase data; identifying the whole process interactive features from the recommended product purchase data to obtain recommended product interactive data;

[0056] Step S63: Analyze the skipping situation of the product recommendation data according to the recommended product click count data to obtain the recommended product skipping data;

[0057] Step S64: Integrate the recommended product click rate data, recommended product interaction data and recommended product skip data to obtain recommended product feedback data.

[0058] The present invention accurately calculates the click-through rate of recommended products by counting the number of displays and clicks on product recommendation data, directly reflecting the user's initial interest and participation in the recommended content. Based on the click-through rate data of recommended products, the user's collection behavior is further analyzed to identify products with high collection potential; through the identification of purchase records and the analysis of purchase conditions, the user's purchase intention and interaction depth for recommended products are deeply understood, thereby obtaining the full-process interaction data of recommended products. The click data of recommended products is deeply analyzed to identify product recommendations skipped by users and obtain recommended product skip data. It helps to identify the shortcomings of recommended content and provide a basis for improving recommendation strategies. The click-through rate data, interaction data and skip data are integrated to form comprehensive recommended product feedback data, which comprehensively reflects the user's acceptance and behavioral characteristics of recommended products, and provides key information for the continuous optimization of recommendation models and personalized recommendations.

[0059] Preferably, step S7 comprises the following steps:

[0060] Step S71: intelligently adjusting the product recommendation data based on the recommended product feedback data to generate product recommendation adjustment data;

[0061] Step S72: performing real-time monitoring, adjustment and feedback on the e-commerce recommendation model through the commodity recommendation adjustment data to obtain real-time adjustment data and real-time monitoring feedback data;

[0062] Step S73: Integrate the real-time adjustment data and the real-time monitoring feedback data to obtain a real-time adjustment feedback report;

[0063] Step S74: construct a cloud platform according to the real-time adjustment feedback report to obtain a real-time adjustment feedback cloud platform, wherein the real-time adjustment feedback cloud platform construction includes the construction of a real-time adjustment interface for product user features and the construction of a real-time monitoring feedback interface for product user features; execute a user feature extraction method operation of an e-commerce recommendation model based on the real-time adjustment feedback cloud platform.

[0064] The present invention uses the recommended product feedback data to make detailed intelligent adjustments to the product recommendation data and generate product recommendation adjustment data. This step ensures that the recommendation content can be optimized according to the real-time feedback of the user, thereby improving the accuracy of the recommendation and user satisfaction. The e-commerce recommendation model is monitored and adjusted in real time through the product recommendation adjustment data to obtain real-time adjustment data and real-time monitoring feedback data, providing immediate, data-driven insights for the continuous improvement of the recommendation model. The real-time adjustment data and real-time monitoring feedback data are integrated to form a real-time adjustment feedback report. The report comprehensively reflects the current status and performance of the recommendation model and provides a detailed basis for further decision-making. According to the real-time adjustment feedback report, a real-time adjustment feedback system is constructed on the cloud platform, including a real-time adjustment interface for product user features and a real-time monitoring feedback interface for product user features. The construction of this platform realizes the cloud management and operation of the recommendation model and improves the response speed and flexibility of the model. Based on the real-time adjustment feedback cloud platform, the user feature extraction method of the e-commerce recommendation model is executed to ensure the efficient operation and real-time update of the recommendation model.

[0065] Preferably, step S71 includes the following steps:

[0066] Step S711: performing click rate evaluation on the recommended product click rate data to obtain recommended product click rate evaluation data, wherein the recommended product click rate evaluation data includes high click rate data and low click rate data;

[0067] Step S712: intelligently adjust the product recommendation data based on the recommended product click rate evaluation data. If the recommended product click rate evaluation data is high click rate data, the product recommendation data is adjusted by increasing the recommended page area of ​​products of the same type to obtain high click rate product adjustment data; if the recommended product click rate evaluation data is low click rate data, the product recommendation data is adjusted by reducing the recommended page area of ​​products of the same type to obtain low click rate product adjustment data; integrate the high click rate product adjustment data and the low click rate product adjustment data to generate click rate product adjustment data;

[0068] Step S713: Perform product interaction stage analysis on the recommended product interaction data to obtain recommended product interaction stage data, wherein the recommended product interaction stage data includes product collection stage, product purchase stage and product purchase stage;

[0069] Step S714: intelligently adjust the product recommendation data based on the recommended product interaction stage data. If the recommended product interaction stage data is the product collection stage and the product purchase stage, the product recommendation data is adjusted to increase the pre-purchase push frequency of related types of products to obtain the collection purchase stage adjustment data; if the recommended product interaction stage data is the product purchase stage, the product recommendation data is adjusted to increase the post-purchase display frequency of the same type of products to obtain the purchase stage adjustment data; the collection purchase stage adjustment data and the purchase stage adjustment data are integrated to generate the product interaction stage adjustment data;

[0070] Step S715: intelligently adjusting the product recommendation data based on the recommended product skip data. If the product recommendation data is recommended product skip data, removing recommendations of products of the same type from the product recommendation data to obtain product skip adjustment data.

[0071] Step S716: Integrate the click-through rate product adjustment data, the product interaction stage adjustment data and the product skip adjustment data to obtain product recommendation adjustment data.

[0072] The present invention evaluates the click-through rate data of recommended products, distinguishes high-click-through rate and low-click-through rate products, obtains recommended product click-through rate evaluation data, and clearly identifies the hot spots of user preference and the product categories that need to be improved. Based on the click-through rate evaluation data, the product recommendation data is intelligently adjusted; for high-click-through rate products, the recommendation page area of ​​the same type of products is increased to meet the interests and needs of users; for low-click-through rate products, the recommendation page area is reduced to optimize the user interface and improve the relevance of recommendations; and the click-through rate product adjustment data is generated by integration and adjustment. The interactive data of recommended products is analyzed to identify the user's interactive behaviors at different stages of product collection, additional purchase and purchase, and the recommended product interactive stage data is obtained, which is helpful to understand the user's behavior pattern during the purchase process. Based on the product interactive stage data, the recommended data is intelligently adjusted. In the product collection and additional purchase stages, the push frequency of related types of products is increased to promote users' further purchase decisions; in the product purchase stage, the display frequency of the same type of products is increased to meet users' post-purchase needs and potential interests; and the product interactive stage adjustment data is generated by integration and adjustment. Analyze the recommended product skip data, identify the types of products that users are not interested in, and remove these products from the recommendations to obtain product skip adjustment data, which helps reduce users' negative experience and improve the relevance of recommendations. Integrate the click-through rate product adjustment data, product interaction stage adjustment data, and product skip adjustment data to obtain comprehensive recommendation adjustment data, ensuring that the adjustment of the recommendation model can comprehensively consider various user feedback and realize personalized product recommendations.

[0073] The method of the present invention realizes deep mining and intelligent analysis of user behavior data in an e-commerce recommendation system to improve the relevance and accuracy of recommendations. The e-commerce data set is thoroughly preprocessed to ensure the accuracy and consistency of the data set used for analysis. On this basis, the product order information is carefully extracted and analyzed, the order processing time is recorded, and the order completion rate is counted, so as to deeply understand the user's purchase behavior of different types of products. Further, by tracking the user's browsing behavior on the product details page, the page interaction depth is evaluated, and the user's detailed interest points for specific products are identified. At the same time, by analyzing the user's storage time and deletion behavior of products in the shopping cart, the user's purchase intention and preference are insightful. In addition, the order type, page access and shopping cart behavior data are integrated to construct the product influence weight data, which provides refined user behavior parameters for the e-commerce recommendation model. By real-time monitoring of user feedback on recommended products, including click-through rate, interactive behavior and neglect, rich user feedback information is collected, providing an empirical basis for evaluating the recommendation effect. Based on the feedback information, the recommended content is intelligently adjusted, and the recommendation strategy is continuously optimized to ensure that the recommendation system can respond to changes in user behavior in a timely manner. By building a cloud platform, real-time monitoring and dynamic adjustment of the recommendation system are achieved, which improves the real-time nature of the recommended content and user satisfaction, thereby bringing higher user loyalty and market competitiveness to the e-commerce platform. Therefore, the present invention uses data processing technology, image mapping technology, intelligent control technology and cloud platform technology to accurately and objectively reflect the user's commodity demand characteristics, improve the feedback mechanism, and intelligently adjust the recommendation results, thereby improving the accuracy of the e-commerce recommendation model. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 A flowchart of the steps of the user feature extraction method for an e-commerce recommendation model;

[0075] Figure 2 for Figure 1 Detailed implementation steps of step S3 in FIG.

[0076] Figure 3 for Figure 1 Detailed implementation steps of step S7;

[0077] Figure 4 for Figure 3 Detailed implementation steps of step S71;

[0078] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0079] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0080] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0081] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0082] To achieve this, please refer to Figures 1 to 4 , a user feature extraction method for an e-commerce recommendation model, the method comprising the following steps:

[0083] Step S1: obtaining an e-commerce data set; performing data preprocessing on the e-commerce data set to obtain a standard e-commerce data set;

[0084] Step S2: extracting commodity order information from a standard e-commerce data set to generate commodity order data; recording order processing time for the commodity order data to generate order processing time data; performing order completion rate statistics on the commodity order data according to the order processing time data to obtain order completion rate data; performing commodity type identification on the order completion rate data to obtain commodity order type data;

[0085] Step S3: extracting product page browsing information of the standard e-commerce data set to generate product page browsing data; performing page interaction depth evaluation on the product page browsing data to obtain page interaction evaluation data; identifying product detail page access conditions on the product page browsing data according to the page interaction evaluation data to obtain product detail page access data;

[0086] Step S4: extracting the product shopping cart information of the standard e-commerce data set to generate product shopping cart data; calculating the product storage time of the product shopping cart data to generate product storage time data; performing product deletion analysis on the product shopping cart data according to the product storage time data to obtain deleted product data;

[0087] Step S5: dividing the product order type data, product detail page access data and deleted product data by influence weight to obtain product influence weight data; constructing an e-commerce recommendation model based on the product influence weight data to obtain an e-commerce recommendation model; recommending products through the e-commerce recommendation model to generate product recommendation data;

[0088] Step S6: monitor the click rate of the product recommendation data to obtain the recommended product click rate data; analyze the interactive characteristics of the product recommendation data to obtain the recommended product interactive data; record the skipping of the product recommendation data to obtain the recommended product skipping data; integrate the recommended product click rate data, the recommended product interactive data and the recommended product skipping data to obtain the recommended product feedback data;

[0089] Step S7: Based on the recommended product feedback data, intelligently adjust the product recommendation data to generate product recommendation adjustment data; perform real-time adjustment feedback on the e-commerce recommendation model according to the product recommendation adjustment data to obtain a real-time adjustment feedback report; construct a cloud platform according to the real-time adjustment feedback report to obtain a real-time adjustment feedback cloud platform; execute the user feature extraction method operation of the e-commerce recommendation model based on the real-time adjustment feedback cloud platform.

[0090] The present invention ensures the quality and consistency of data by acquiring an e-commerce data set and performing data preprocessing, and provides a standardized e-commerce data set for subsequent analysis. By extracting commodity order information and generating order data, the order performance of different commodity types can be accurately identified by recording the order processing time and statistically calculating the order completion rate, providing data support for understanding the user's purchase tendency for various commodities. By analyzing the commodity page browsing data and evaluating the page interaction depth, the user's access to the commodity details page can be identified, thereby gaining insight into the user's detailed interest in specific commodities. By measuring the commodity storage time and analyzing the deleted commodity data, the user's retention time in the shopping cart and the commodities that are eventually deleted can be understood, which helps to understand the user's purchase decision process. The commodity order type data, the commodity details page access data and the deleted commodity data are weighted, and the commodity impact weight data is constructed, which provides key user behavior features for the e-commerce recommendation model and enhances the personalization and accuracy of the recommendation model; by monitoring the click-through rate, interactive features and skipping of the recommended commodities, the feedback data of the recommended commodities is collected, providing direct quantitative indicators for evaluating the recommendation effect and user acceptance. Based on the feedback data of recommended products, the recommended data is intelligently adjusted to achieve the optimization adjustment and real-time update of the recommendation model. The constructed real-time adjustment feedback cloud platform further strengthens the dynamic adjustment capability of the recommendation model and ensures the timeliness and user satisfaction of the recommended content. Therefore, the present invention uses data processing technology, image mapping technology, intelligent control technology and cloud platform technology to accurately and objectively reflect the user's product demand characteristics, improve the feedback mechanism, intelligently adjust the recommendation results, and improve the accuracy of the e-commerce recommendation model.

[0091] In the embodiment of the present invention, reference Figure 1 The above is a schematic flow chart of the steps of the method for extracting user features of an e-commerce recommendation model of the present invention. In this example, the method for extracting user features of an e-commerce recommendation model includes the following steps:

[0092] Step S1: obtaining an e-commerce data set; performing data preprocessing on the e-commerce data set to obtain a standard e-commerce data set;

[0093] In an embodiment of the present invention, a data set containing user behavior records is first imported from the database of an e-commerce platform, which contains order information, user browsing history, and multi-dimensional information of product detail page interactions. Next, the e-commerce data set is preprocessed, wherein the data preprocessing includes data cleaning and format unification. Data cleaning removes duplicate records in the data set, corrects obvious errors and inconsistent data items, and handles missing values; for example, for the problem of inconsistent date formats in order information, all date fields are converted to a standard format, and invalid and abnormal order records are identified and deleted. Format unification is to adjust the various fields in the data set to a format suitable for analysis; for example, convert the product classification from text description to numerical code, and convert the user's browsing behavior record into a structured log format. After data preprocessing, the standard e-commerce data set obtained will be a clean, consistent, and uniformly formatted data set, which will be used for further data analysis.

[0094] Step S2: extracting commodity order information from a standard e-commerce data set to generate commodity order data; recording order processing time for the commodity order data to generate order processing time data; performing order completion rate statistics on the commodity order data according to the order processing time data to obtain order completion rate data; performing commodity type identification on the order completion rate data to obtain commodity order type data;

[0095] In the embodiment of the present invention, query statements and data extraction technology are used to extract all records related to commodity orders. The commodity order information includes order number, order status, order generation time and order completion time fields; the order generation time and order completion time in each order record are analyzed by time measurement tools, and the processing time of each order is recorded; the order processing time of each order is calculated by using timestamp data, and the completion rate of each order is calculated based on the order processing time to identify the order completion status of each commodity type; the order completion rate data is used to identify the commodity type according to the order completion rate data and commodity type information to obtain commodity order type data.

[0096] Step S3: extracting product page browsing information of the standard e-commerce data set to generate product page browsing data; performing page interaction depth evaluation on the product page browsing data to obtain page interaction evaluation data; identifying product detail page access conditions on the product page browsing data according to the page interaction evaluation data to obtain product detail page access data;

[0097] In the embodiment of the present invention, a data extraction operation is performed using an SQL query statement to capture records of product page browsing from a standard e-commerce data set, wherein the records include user ID, product ID, browsing timestamp, and browsing duration key fields. The product page browsing data is evaluated for page interaction depth, and the user's residence time, click count, and scrolling behavior interaction behavior on each product page are quantitatively analyzed. The user's interaction depth is evaluated by calculating the user's average residence time on each page, and the user's scrolling percentage on the page is monitored to determine whether the user has viewed the page content in detail. The results of the page interaction depth evaluation are used to further identify the access to the product details page, and a specific interaction depth threshold is set to identify whether the user has actually visited the product details page. Only when the user's residence time on the page exceeds the average value and scrolls to the bottom of the page, the user is considered to have visited the product details page. The generated product details page access data will record in detail when the user visits the details page of a specific product, as well as the interaction depth, providing important information for understanding the user's interest in and purchase intention for a specific product, and providing key user behavior features for the subsequent recommendation system.

[0098] Step S4: extracting the product shopping cart information of the standard e-commerce data set to generate product shopping cart data; calculating the product storage time of the product shopping cart data to generate product storage time data; performing product deletion analysis on the product shopping cart data according to the product storage time data to obtain deleted product data;

[0099] In the embodiment of the present invention, information related to the shopping cart of goods is extracted from the standard e-commerce data set, including user ID, product ID, timestamp of adding to the shopping cart and timestamp of removing the goods from the shopping cart, to generate shopping cart data of goods, wherein the shopping cart data of goods clearly records the purchase and deletion behavior of each user on each product; the shopping cart data of goods is analyzed to measure the storage time of each product in the shopping cart; this is achieved by calculating the difference between the two timestamps from when the goods are added to the shopping cart to when they are finally purchased and deleted; the measurement of the storage time provides a quantitative indicator for understanding the user's consideration and decision-making process. Using the storage time data of goods, the shopping cart data of goods is deeply analyzed to identify the goods that have not been purchased within a certain time threshold and are finally deleted from the shopping cart by the user, including the detailed information of the deleted goods and the storage time in the shopping cart.

[0100] Step S5: dividing the product order type data, product detail page access data and deleted product data by influence weight to obtain product influence weight data; constructing an e-commerce recommendation model based on the product influence weight data to obtain an e-commerce recommendation model; recommending products through the e-commerce recommendation model to generate product recommendation data;

[0101] In an embodiment of the present invention, the influence factor analysis technology is used to comprehensively consider the product order type data, product detail page access data and deleted product data to determine their relative importance in the recommendation model; weights are divided based on the data analysis results. For example, if the data shows that the user's visit to the product detail page is strongly correlated with the purchase behavior, the data set is given a higher weight. The e-commerce recommendation model is constructed using the influence weight data; machine learning algorithms such as collaborative filtering, content recommendation algorithms and deep learning models are used to analyze user behavior and product features to predict the products that the user is interested in. In the model construction process, the influence weight data is used as an input feature to learn user preferences through the training algorithm. The e-commerce recommendation model will be used to generate product recommendation data. The e-commerce recommendation model processes the current user's behavior data and combines it with the patterns learned during the training process to output a series of recommended products.

[0102] Step S6: monitor the click rate of the product recommendation data to obtain the recommended product click rate data; analyze the interactive characteristics of the product recommendation data to obtain the recommended product interactive data; record the skipping of the product recommendation data to obtain the recommended product skipping data; integrate the recommended product click rate data, the recommended product interactive data and the recommended product skipping data to obtain the recommended product feedback data;

[0103] In the embodiment of the present invention, firstly, the commodity recommendation data is monitored in real time to capture the user's click behavior on the recommended commodities; by tracking the event of the user clicking on the recommended commodities, the timestamp and user ID of each click are recorded to generate the recommended commodity click-through rate data; the click-through rate is calculated by dividing the number of clicks by the number of recommendation displays to reflect the degree to which the recommended commodities attract the user's attention; the commodity recommendation data is analyzed for interactive features, wherein the interactive features include monitoring the user's collection, add-to-cart and settlement interactive behaviors to generate the recommended commodity interactive data; the interactive data is obtained by analyzing various interactive behaviors between the user and the recommended commodities. By recording the user's skipping of the recommended commodities and tracking the user's scrolling and skipping behaviors when browsing the recommended commodities, the commodity information that does not arouse the user's interest is captured to generate the recommended commodity skipping data. The recommended commodity click-through rate data, recommended commodity interactive data and recommended commodity skipping data are integrated to form a comprehensive recommended commodity feedback data, which provides a perspective for comprehensive evaluation of the recommendation effect.

[0104] Step S7: Based on the recommended product feedback data, intelligently adjust the product recommendation data to generate product recommendation adjustment data; perform real-time adjustment feedback on the e-commerce recommendation model according to the product recommendation adjustment data to obtain a real-time adjustment feedback report; construct a cloud platform according to the real-time adjustment feedback report to obtain a real-time adjustment feedback cloud platform; execute the user feature extraction method operation of the e-commerce recommendation model based on the real-time adjustment feedback cloud platform.

[0105] In the embodiment of the present invention, data mining technology is used to conduct a comprehensive analysis of existing product recommendation data through recommended product feedback data, and the recommendation logic is intelligently adjusted and automatically updated to generate product recommendation adjustment data. The e-commerce recommendation model is adjusted in real time according to the product recommendation adjustment data; new user feedback is introduced during the model training process to reflect the latest user behavior and preference changes; through the machine learning framework, the model is retrained or incrementally learned according to real-time data to generate a real-time adjustment feedback report. Furthermore, based on the real-time adjustment feedback report, cloud computing technology is used to develop a cloud platform service, which integrates the real-time monitoring and adjustment functions of the recommendation model, and executes the user feature extraction method of the e-commerce recommendation model on the real-time adjustment feedback cloud platform.

[0106] Preferably, step S2 comprises the following steps:

[0107] Step S21: extracting commodity order information from a standard e-commerce data set to generate commodity order data;

[0108] Step S22: Analyze the processing status of the commodity order data to obtain commodity order processing status data; identify the timestamp of the commodity order processing status data to obtain order processing timestamp data; record the order processing time of the order processing timestamp data to generate order processing time data;

[0109] Step S23: performing payment feature recognition on the commodity order data to obtain order payment feature data; performing payment attribute classification on the order payment feature data to obtain order payment attribute data, extracting paid order attributes from the order payment attribute data to obtain paid order attribute data;

[0110] Step S24: Calculate the ratio of the paid order attribute data to the commodity order data to generate order percentage data; perform time-based statistics on the order percentage data according to the order processing time data to obtain order completion rate data;

[0111] Step S25: extracting commodity type features of the standard e-commerce data set to obtain commodity type feature data; numbering the commodity type feature data to obtain commodity type number data;

[0112] Step S26: Map the order completion rate data to commodity types based on the commodity type number data to obtain commodity order type data.

[0113] In the embodiment of the present invention, data extraction technology is used to extract commodity order information of a standard e-commerce data set to generate commodity order data; a data processing tool is used to analyze the order status in the commodity order data set to identify the current processing status of each order, such as "pending payment" and "completed", to obtain commodity order processing status data; the timestamps of order creation and update are extracted from the order record by a timestamp extraction function to obtain order processing timestamp data; the difference between the timestamps is calculated to record the processing time of the order to generate order processing time data; the payment method and payment time payment characteristics are identified from the commodity order data by regular expression matching or field value mapping technology to obtain order payment characteristic data; the order is classified according to the payment characteristics and the attributes indicating that the order has been paid are extracted to obtain paid order attribute data. The paid order attribute data is calculated as a ratio with the commodity order data, and the Excel formula calculation function is used to generate order percentage data; the order percentage data is statistically divided into time periods according to the order processing time data to obtain order completion rate data. The commodity type characteristics of the standard e-commerce data set are extracted to obtain commodity type characteristic data; the commodity type characteristic data are numbered to obtain commodity type number data. Through data association mapping technology, the commodity type number data is matched with the order completion rate data to obtain the commodity order type data.

[0114] As an example of the present invention, refer to Figure 2 As shown, in this example, step S3 includes:

[0115] Step S31: extracting product page browsing information of a standard e-commerce data set to generate product page browsing data;

[0116] Step S32: Record user click information on the product page browsing data to obtain user click data; calculate page dwell time on the user click data to obtain page dwell time calculation data; sort the page dwell time calculation data in descending order to obtain page dwell time descending order data;

[0117] Step S33: Perform heat map color mapping on the descending data of page dwell time to generate a page dwell time heat map; extract the dark coordinate data of the page dwell time heat map to obtain long-stay page data;

[0118] Step S34: performing page scroll depth analysis on the product page browsing data based on the long-stay page data to obtain page scroll depth data; performing page interaction depth evaluation on the product page browsing data based on the page scroll depth data to obtain page interaction evaluation data; identifying product detail page access situations on the product page browsing data based on the page interaction evaluation data to obtain product detail page access data.

[0119] In an embodiment of the present invention, a data extraction tool is used to extract the commodity page browsing information of a standard e-commerce data set to generate commodity page browsing data. First, the detailed information of each time a user clicks on a commodity page is recorded, including the click time and the click position, to obtain the user click data; the user's stay time on each page is calculated by calculating the difference between the click time and the page loading time; the page stay time measurement data is arranged in descending order by a sorting algorithm to obtain the page stay time descending data. Using data visualization technology, the page stay time descending data is represented by the color depth to represent the user's stay time on the page, and a page stay time heat map is generated; the darkest color coordinate point is extracted from the heat map to represent the area where the user stays the longest, and long-term stay page data is obtained. Based on the long-term stay page data, the commodity page browsing data is analyzed, and the page scrolling depth data is obtained by calculating the user's scrolling depth of the page, such as the percentage of scrolling to the page. Then, the user's interaction depth with the page is evaluated according to the page scrolling depth data to obtain the page interaction evaluation data. Finally, the commodity page browsing data is analyzed using the evaluation data to identify the user's actual visit to the commodity details page, and the commodity details page access data is obtained.

[0120] Preferably, step S34 includes the following steps:

[0121] Step S341: Record the page depth of the product page browsing data to obtain product browsing page depth data; analyze the page scrolling situation of the product browsing page depth data according to the long-stay page data to obtain page scrolling situation data;

[0122] Step S342: extracting the numerical information of the product browsing page depth data to obtain the product browsing page depth value; performing a preset depth calculation on the product browsing page depth value; and generating page preset depth data;

[0123] Step S343: extracting the depth value of the page scrolling situation data to obtain page scrolling depth data; comparing the page scrolling depth data with the page preset depth data, if the page scrolling depth data is greater than the page preset depth data, marking the page scrolling depth data as page high scrolling data; if the page scrolling depth data is less than or equal to the page preset depth data, marking the page scrolling depth data as page low scrolling data; integrating the page high scrolling data and the page low scrolling data to obtain page interaction evaluation data;

[0124] Step S344: perform product detail page feature recognition on the product page browsing data to obtain product detail page feature data; perform access situation analysis on the product detail page feature data based on the page interaction evaluation data to obtain product detail page access data, wherein the product detail page access data includes detail page non-visited data and detail page visited data.

[0125] In an embodiment of the present invention, firstly, by tracking the user's behavior path on the product page, the depth of page browsing, such as the page access sequence and the access time of each page, is recorded to obtain product browsing page depth data. Then, the page scrolling situation is analyzed in combination with the long-stay page data, including the scrolling distance and the scrolling speed, to obtain page scrolling situation data. The numerical information in the product browsing page depth data is extracted, such as the depth percentage of the page access and the length of the stay time, to obtain the product browsing page depth value; according to the preset page depth threshold, such as the percentage of the page scrolling to the bottom, the page depth value is calculated to generate the page preset depth data. The depth value is extracted from the page scrolling situation data to obtain the page scrolling depth data; the page scrolling depth data is compared with the page preset depth data, if the page scrolling depth data exceeds the preset value, it is classified as page high scrolling data; if the page scrolling depth data does not exceed the preset value, it is classified as page low scrolling data; by writing conditional judgment logic to achieve, the page interaction evaluation data is obtained. The product page browsing data is used to identify product details page features, such as the loading status of page elements and user interaction behavior, to obtain product details page feature data; the page interaction evaluation data is used to analyze the access status of the product details page feature data to distinguish whether the user visits the product details page, and obtain product details page access data, which includes details page non-visited data and details page visited data.

[0126] Preferably, step S4 comprises the following steps:

[0127] Step S41: extracting the commodity shopping cart information of the standard e-commerce data set to generate commodity shopping cart data;

[0128] Step S42: Record the time of adding the product to the shopping cart data to generate product purchase time point data;

[0129] Step S43: performing payment feature recognition on the shopping cart data to obtain shopping cart settlement data; extracting the time point data of the shopping cart settlement data to obtain the shopping cart settlement time point data; calculating the settlement commodity storage time of the shopping cart data according to the shopping cart settlement time point data and the commodity purchase time point data to generate the settlement commodity storage time data;

[0130] Step S44: performing removal feature recognition on the shopping cart data to obtain shopping cart removal data; extracting the time point data of the shopping cart removal data to obtain the shopping cart removal time point data; calculating the storage time of the removed goods on the shopping cart data according to the shopping cart removal time point data and the product purchase time point data to generate the removed goods storage time data;

[0131] Step S45: Integrate the settled commodity storage time data and the removed commodity storage time data to generate commodity storage time data; perform a deleted commodity analysis on the commodity shopping cart data according to the commodity storage time data to obtain deleted commodity data, wherein the deleted commodity data includes the settled commodity data and the removed commodity data.

[0132] In an embodiment of the present invention, the commodity shopping cart information of the standard e-commerce data set is extracted through database query technology to generate commodity shopping cart data; the specific time point when each commodity is added to the shopping cart is recorded for the commodity shopping cart data to generate commodity purchase time point data; the characteristics of the completed payment of the commodity shopping cart data are identified to obtain shopping cart settlement data; the shopping cart settlement time point data is generated by extracting the payment time point in the shopping cart settlement data; then, the storage time of each settlement commodity in the shopping cart is calculated by combining the shopping cart settlement time point data and the commodity purchase time point data to generate settlement commodity storage time data. The removal feature of the commodity shopping cart data is identified to obtain shopping cart removal data. The time point information in the shopping cart removal data is extracted to generate shopping cart removal time point data, and the storage time of each removed commodity in the shopping cart is calculated according to the shopping cart removal time point data and the commodity purchase time point data to generate removal commodity storage time data. The storage time data of settled goods and the storage time data of removed goods are integrated to generate the storage time data of goods. The shopping cart data of goods is analyzed according to the storage time data of goods using data analysis technology to identify goods whose storage time exceeds a preset threshold for deletion analysis to obtain the deleted goods data.

[0133] Preferably, step S5 comprises the following steps:

[0134] Step S51: dividing the product order type data, product detail page access data and deleted product data by impact weight to obtain product impact weight data;

[0135] Step S52: constructing an e-commerce recommendation model according to the commodity influence weight data to obtain an e-commerce recommendation pre-model; and using the commodity influence weight data to perform model training on the e-commerce recommendation pre-model to obtain an e-commerce recommendation training model;

[0136] Step S53: performing model cross-validation evaluation on the e-commerce recommendation training model to obtain e-commerce recommendation evaluation data; adjusting model parameters of the e-commerce recommendation training model using the e-commerce recommendation evaluation data to obtain an e-commerce recommendation model;

[0137] Step S54: Recommend products through the e-commerce recommendation model to generate product recommendation data.

[0138] In the embodiment of the present invention, the weight distribution algorithm in machine learning, such as decision tree classifier and principal component analysis method, is used to evaluate the feature importance of commodity order type data, commodity detail page access data and deleted commodity data, determine the contribution of the recommendation model, and obtain commodity influence weight data. According to the commodity influence weight data, the e-commerce recommendation model is constructed. First, the hyperparameters of the e-commerce recommendation system are set, and the neural collaborative filtering model based on deep learning is determined. The number of neurons in the hidden layer is set to 128, the learning rate is set to 0.001, the batch size is 256, and the early stopping method is used as a regularization method to avoid overfitting. Next, the model is trained using the commodity influence weight data. During the training process, the model learns the mapping relationship between user preferences and commodity characteristics by receiving the user's historical interaction data and the commodity feature data; using the TensorFlow framework, a neural network including an embedding layer, multiple fully connected layers and a nonlinear activation function is constructed; binary cross-entropy is used as the loss function of the model, and the optimizer uses Adam. Cross-validation is implemented to evaluate the generalization ability of the model; the dataset is divided into 5 parts, the model is trained on 4 parts, and tested on the remaining part. This process is repeated 5 times, and a different part is selected as the test set each time. In each iteration, the performance indicators of the model, such as precision, recall, and F1 score, are recorded. The hyperparameters of the model are adjusted according to the e-commerce recommendation evaluation data obtained by cross-validation. If the accuracy of the model is low, it is necessary to increase the dimension of the embedding layer and adjust the learning rate; after adjustment, the model is retrained and evaluated until the optimal hyperparameter combination is found. Finally, the e-commerce recommendation model is obtained that can generate product recommendation data based on given user behavior data.

[0139] Preferably, step S51 includes the following steps:

[0140] Step S511: performing order type similarity analysis on the commodity order type data to obtain commodity type similarity data;

[0141] Step S512: performing association analysis on the settlement product data according to the product detail page access data to obtain settlement product detail page association data; identifying the number of visits to the details page visited data according to the settlement product detail page association data to generate details page visit number data; extracting multiple visit information from the details page visit number data to obtain details page multiple visit data;

[0142] Step S513: Analyze the reasons why the removed product data has not been accessed through the details page not accessed data to obtain the details page not accessed reason data;

[0143] Step S514: constructing a user product type feature matrix for the product type similarity data based on the details page multiple access data and the details page non-access reason data to obtain a user product type feature matrix;

[0144] Step S515: Perform multiple linear regression analysis on the user product type feature matrix to generate multiple linear regression coefficients; perform significant impact analysis on the user product type feature matrix based on the multiple linear regression coefficients to obtain product significant impact data; perform weight analysis on the product significant impact data to obtain product impact weight data.

[0145] In the embodiment of the present invention, cluster analysis in data mining technology, such as K-means and hierarchical clustering algorithms, is used to process commodity order type data to identify the similarity between different commodity types; by calculating the attribute similarity of the commodities in the order, the cosine similarity index is used to obtain commodity type similarity data. Association rule learning is performed on the commodity detail page access data and the settlement commodity data, and the frequent item sets and association patterns between commodities are used by the Apriori algorithm to obtain the settlement commodity detail page association data; the number of visits in the detail page visited data is identified by counting analysis to generate the detail page visit number data; for records with a number of visits exceeding a specific threshold, multiple visit information is extracted to obtain the detail page multiple visit data. The detail page non-visit data is analyzed, and the reasons why the user did not visit the commodity detail page are analyzed by logistic regression technology to obtain the details page non-visit reason data. Based on the details page multiple visit data and the details page non-visit reason data, the user commodity type feature matrix is ​​constructed for the commodity type similarity data by using the t-distribution random neighborhood embedding technology to obtain the user commodity type feature matrix. Perform multiple linear regression analysis on the user product type feature matrix, use the least squares method to estimate the model parameters, and generate multiple linear regression coefficients; perform an F test to evaluate the significance of the regression coefficients and obtain product significance impact data; assign weights to product features based on the significance analysis results and the size of the regression coefficients to obtain product impact weight data.

[0146] Preferably, step S6 comprises the following steps:

[0147] Step S61: Count the number of times the product recommendation data is displayed to obtain display number statistics; monitor the number of clicks on the product recommendation data to obtain recommended product click number data; calculate the ratio of the recommended product click number data and the display number statistics to obtain recommended product click rate data;

[0148] Step S62: extracting collection features from the product recommendation data according to the recommended product click rate data to obtain recommended product collection attribute data; identifying purchase records from the recommended product collection attribute data to obtain recommended product purchase data; analyzing purchase situations from the recommended product purchase data to obtain recommended product purchase data; identifying the whole process interactive features from the recommended product purchase data to obtain recommended product interactive data;

[0149] Step S63: Analyze the skipping situation of the product recommendation data according to the recommended product click count data to obtain the recommended product skipping data;

[0150] Step S64: Integrate the recommended product click rate data, recommended product interaction data and recommended product skip data to obtain recommended product feedback data.

[0151] In the embodiment of the present invention, Logstash is used to collect user interaction logs through log analysis technology, and the number of times each recommended product is displayed is counted to obtain display statistics; the user's click behavior on the recommended products is monitored and recorded to obtain the recommended product click data; arithmetic operations are used to calculate the ratio of the number of clicks to the number of displays to obtain the recommended product click rate data. The recommended products with high click rates are extracted using data analysis tools, and their collection features are identified to obtain the recommended product collection attribute data; further, the user's purchase records of these products are identified through the retrieval function of the database to obtain the recommended product purchase data; the final purchased products are identified through association analysis or sequence pattern mining to obtain the recommended product purchase data; the user's full-process interactive features of the recommended products are identified through Google Analytics user behavior tracking technology, including page dwell time and scrolling behavior, to obtain the recommended product interaction data. The product recommendation data is skipped according to the recommended product click data to obtain the recommended product skip data. The recommended product click rate data, recommended product interaction data and recommended product skip data are integrated using data fusion technology to obtain the recommended product feedback data.

[0152] As an example of the present invention, refer to Figure 3 As shown, in this example, step S7 includes:

[0153] Step S71: intelligently adjusting the product recommendation data based on the recommended product feedback data to generate product recommendation adjustment data;

[0154] Step S72: performing real-time monitoring, adjustment and feedback on the e-commerce recommendation model through the commodity recommendation adjustment data to obtain real-time adjustment data and real-time monitoring feedback data;

[0155] Step S73: Integrate the real-time adjustment data and the real-time monitoring feedback data to obtain a real-time adjustment feedback report;

[0156] Step S74: construct a cloud platform according to the real-time adjustment feedback report to obtain a real-time adjustment feedback cloud platform, wherein the real-time adjustment feedback cloud platform construction includes the construction of a real-time adjustment interface for product user features and the construction of a real-time monitoring feedback interface for product user features; execute a user feature extraction method operation of an e-commerce recommendation model based on the real-time adjustment feedback cloud platform.

[0157] In an embodiment of the present invention, the product recommendation data is intelligently adjusted based on the recommended product feedback data to generate product recommendation adjustment data. The recommendation model is deployed to realize real-time monitoring of the product recommendation adjustment data; the real-time performance of the recommendation model is monitored by stream processing technology, and the real-time adjustment data and real-time monitoring feedback data are collected. Through the ETL process integration technology, the real-time adjustment data and the real-time monitoring feedback data are integrated to generate a real-time adjustment feedback report, which records in detail the effect of the recommendation adjustment, including the changes in key performance indicators (KPIs). According to the real-time adjustment feedback report, a real-time adjustment feedback cloud platform is constructed using cloud platform services, such as Amazon Web Services technology; the platform includes two main parts: a real-time adjustment interface for product user features, which can dynamically adjust the recommendation strategy according to feedback data; a real-time monitoring feedback interface for product user features, which displays the real-time performance of the recommendation model and the results of user behavior analysis; based on the real-time adjustment feedback cloud platform, the user feature extraction method of the e-commerce recommendation model is executed.

[0158] As an example of the present invention, refer to Figure 4 As shown, in this example, step S71 includes:

[0159] Step S711: performing click rate evaluation on the recommended product click rate data to obtain recommended product click rate evaluation data, wherein the recommended product click rate evaluation data includes high click rate data and low click rate data;

[0160] Step S712: intelligently adjust the product recommendation data based on the recommended product click rate evaluation data. If the recommended product click rate evaluation data is high click rate data, the product recommendation data is adjusted by increasing the recommended page area of ​​products of the same type to obtain high click rate product adjustment data; if the recommended product click rate evaluation data is low click rate data, the product recommendation data is adjusted by reducing the recommended page area of ​​products of the same type to obtain low click rate product adjustment data; integrate the high click rate product adjustment data and the low click rate product adjustment data to generate click rate product adjustment data;

[0161] Step S713: Perform product interaction stage analysis on the recommended product interaction data to obtain recommended product interaction stage data, wherein the recommended product interaction stage data includes product collection stage, product purchase stage and product purchase stage;

[0162] Step S714: intelligently adjust the product recommendation data based on the recommended product interaction stage data. If the recommended product interaction stage data is the product collection stage and the product purchase stage, the product recommendation data is adjusted to increase the pre-purchase push frequency of related types of products to obtain the collection purchase stage adjustment data; if the recommended product interaction stage data is the product purchase stage, the product recommendation data is adjusted to increase the post-purchase display frequency of the same type of products to obtain the purchase stage adjustment data; the collection purchase stage adjustment data and the purchase stage adjustment data are integrated to generate the product interaction stage adjustment data;

[0163] Step S715: intelligently adjusting the product recommendation data based on the recommended product skip data. If the product recommendation data is recommended product skip data, removing recommendations of products of the same type from the product recommendation data to obtain product skip adjustment data.

[0164] Step S716: Integrate the click-through rate product adjustment data, the product interaction stage adjustment data and the product skip adjustment data to obtain product recommendation adjustment data.

[0165] In an embodiment of the present invention, a data analysis tool is used to quantitatively analyze the click-through rate data of recommended products, determine the threshold of the click-through rate, and distinguish between high click-through rate data and low click-through rate data. Based on the click-through rate evaluation results, a machine learning model and a rule engine are used to adjust the parameters of the recommendation model; for high click-through rate products, by increasing the weight of the product in the recommendation model or improving its position in the recommendation list, and increasing the recommended page area for the same type of products, high click-through rate product adjustment data is obtained; on the contrary, for low click-through rate products, their weight is reduced or their frequency of occurrence is reduced, and the recommended page area for the same type of products is reduced, and low click-through rate product adjustment data is obtained; then, these two types of adjustment data are integrated to generate click-through rate product adjustment data. The recommended product interaction data is analyzed in stages, and the user's collection, add-to-cart and purchase behaviors of the recommended products are collected through user behavior tracking systems, such as web page tracking codes and mobile application analysis APIs, to obtain the recommended product interaction stage data. According to the data of the product interaction stage, the recommendation system is adjusted using machine learning models and rule engines; for products in the collection and add-to-cart stage, the push frequency of related types of products before purchase is increased, and the number of their appearances in the recommendation model is increased to obtain the collection and add-to-cart stage adjustment data; for products that have reached the purchase stage, the post-purchase display frequency of the same type of products is increased to promote cross-selling or increase the attention of the same type of products, and the purchase stage adjustment data is obtained. After that, these two types of data are integrated to generate the product interaction stage adjustment data. If the user skips a specific recommended product, the skipping behavior is identified through the user feedback collection mechanism, and the recommendation model is adjusted to remove the recommendation of the same type of products to obtain the product skip adjustment data. The click-through rate product adjustment data, product interaction stage adjustment data, and product skip adjustment data are integrated to form a product recommendation adjustment data set.

[0166] The method of the present invention realizes deep mining and intelligent analysis of user behavior data in an e-commerce recommendation system to improve the relevance and accuracy of recommendations. The e-commerce data set is thoroughly preprocessed to ensure the accuracy and consistency of the data set used for analysis. On this basis, the product order information is carefully extracted and analyzed, the order processing time is recorded, and the order completion rate is counted, so as to deeply understand the user's purchase behavior of different types of products. Further, by tracking the user's browsing behavior on the product details page, the page interaction depth is evaluated, and the user's detailed interest points for specific products are identified. At the same time, by analyzing the user's storage time and deletion behavior of products in the shopping cart, the user's purchase intention and preference are insightful. In addition, the order type, page access and shopping cart behavior data are integrated to construct the product influence weight data, which provides refined user behavior parameters for the e-commerce recommendation model. By real-time monitoring of user feedback on recommended products, including click-through rate, interactive behavior and neglect, rich user feedback information is collected, providing an empirical basis for evaluating the recommendation effect. Based on the feedback information, the recommended content is intelligently adjusted, and the recommendation strategy is continuously optimized to ensure that the recommendation system can respond to changes in user behavior in a timely manner. By building a cloud platform, real-time monitoring and dynamic adjustment of the recommendation system are achieved, which improves the real-time nature of the recommended content and user satisfaction, thereby bringing higher user loyalty and market competitiveness to the e-commerce platform. Therefore, the present invention uses data processing technology, image mapping technology, intelligent control technology and cloud platform technology to accurately and objectively reflect the user's commodity demand characteristics, improve the feedback mechanism, and intelligently adjust the recommendation results, thereby improving the accuracy of the e-commerce recommendation model.

[0167] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0168] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be 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 present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A user feature extraction method for an e-commerce recommendation model, characterized in that: The following steps are involved: Step S1: obtaining an e-commerce data set; performing data preprocessing on the e-commerce data set to obtain a standard e-commerce data set; Step S2: extracting commodity order information from a standard e-commerce data set to generate commodity order data; recording order processing time for the commodity order data to generate order processing time data; Perform order completion rate statistics on the commodity order data according to the order processing time data to obtain order completion rate data; perform commodity type identification on the order completion rate data to obtain commodity order type data; Step S3: extracting product page browsing information of the standard e-commerce data set to generate product page browsing data; performing page interaction depth evaluation on the product page browsing data to obtain page interaction evaluation data; identifying product detail page access conditions on the product page browsing data according to the page interaction evaluation data to obtain product detail page access data; Step S4: extracting the product shopping cart information of the standard e-commerce data set to generate product shopping cart data; calculating the product storage time of the product shopping cart data to generate product storage time data; performing product deletion analysis on the product shopping cart data according to the product storage time data to obtain deleted product data; Step S5: dividing the product order type data, product detail page access data and deleted product data by influence weight to obtain product influence weight data; constructing an e-commerce recommendation model based on the product influence weight data to obtain an e-commerce recommendation model; recommending products through the e-commerce recommendation model to generate product recommendation data; Step S6: monitor the click rate of the product recommendation data to obtain the recommended product click rate data; analyze the interactive characteristics of the product recommendation data to obtain the recommended product interactive data; record the skipping of the product recommendation data to obtain the recommended product skipping data; integrate the recommended product click rate data, the recommended product interactive data and the recommended product skipping data to obtain the recommended product feedback data; Step S7: Based on the recommended product feedback data, intelligently adjust the product recommendation data to generate product recommendation adjustment data; perform real-time adjustment feedback on the e-commerce recommendation model according to the product recommendation adjustment data to obtain a real-time adjustment feedback report; construct a cloud platform according to the real-time adjustment feedback report to obtain a real-time adjustment feedback cloud platform; execute the user feature extraction method operation of the e-commerce recommendation model based on the real-time adjustment feedback cloud platform.

2. The user feature extraction method of the e-commerce recommendation model according to claim 1 is characterized in that: Step S2 includes the following steps: Step S21: extracting commodity order information from a standard e-commerce data set to generate commodity order data; Step S22: Analyze the processing status of the commodity order data to obtain commodity order processing status data; identify the timestamp of the commodity order processing status data to obtain order processing timestamp data; record the order processing time of the order processing timestamp data to generate order processing time data; Step S23: performing payment feature recognition on the commodity order data to obtain order payment feature data; performing payment attribute classification on the order payment feature data to obtain order payment attribute data, extracting paid order attributes from the order payment attribute data to obtain paid order attribute data; Step S24: Calculate the ratio of the paid order attribute data to the commodity order data to generate order percentage data; perform time-based statistics on the order percentage data according to the order processing time data to obtain order completion rate data; Step S25: extracting commodity type features of the standard e-commerce data set to obtain commodity type feature data; numbering the commodity type feature data to obtain commodity type number data; Step S26: Map the order completion rate data to commodity types based on the commodity type number data to obtain commodity order type data.

3. The user feature extraction method of the e-commerce recommendation model according to claim 1 is characterized in that: Step S3 includes the following steps: Step S31: extracting product page browsing information of a standard e-commerce data set to generate product page browsing data; Step S32: Record user click information on the product page browsing data to obtain user click data; calculate page dwell time on the user click data to obtain page dwell time calculation data; sort the page dwell time calculation data in descending order to obtain page dwell time descending order data; Step S33: Perform heat map color mapping on the descending data of page dwell time to generate a page dwell time heat map; extract the dark coordinate data of the page dwell time heat map to obtain long-stay page data; Step S34: performing page scroll depth analysis on the product page browsing data based on the long-stay page data to obtain page scroll depth data; performing page interaction depth evaluation on the product page browsing data based on the page scroll depth data to obtain page interaction evaluation data; identifying product detail page access situations on the product page browsing data based on the page interaction evaluation data to obtain product detail page access data.

4. The user feature extraction method of the e-commerce recommendation model according to claim 3 is characterized in that: Step S34 includes the following steps: Step S341: Record the page depth of the product page browsing data to obtain product browsing page depth data; analyze the page scrolling situation of the product browsing page depth data according to the long-stay page data to obtain page scrolling situation data; Step S342: extracting the numerical information of the product browsing page depth data to obtain the product browsing page depth value; performing a preset depth calculation on the product browsing page depth value; and generating page preset depth data; Step S343: extracting the depth value of the page scrolling situation data to obtain page scrolling depth data; comparing the page scrolling depth data with the page preset depth data, if the page scrolling depth data is greater than the page preset depth data, marking the page scrolling depth data as page high scrolling data; if the page scrolling depth data is less than or equal to the page preset depth data, marking the page scrolling depth data as page low scrolling data; integrating the page high scrolling data and the page low scrolling data to obtain page interaction evaluation data; Step S344: perform product detail page feature recognition on the product page browsing data to obtain product detail page feature data; perform access situation analysis on the product detail page feature data based on the page interaction evaluation data to obtain product detail page access data, wherein the product detail page access data includes detail page non-visited data and detail page visited data.

5. The user feature extraction method of the e-commerce recommendation model according to claim 1 is characterized in that: Step S4 includes the following steps: Step S41: extracting the commodity shopping cart information of the standard e-commerce data set to generate commodity shopping cart data; Step S42: Record the time of adding the product to the shopping cart data to generate product purchase time point data; Step S43: performing payment feature recognition on the shopping cart data to obtain shopping cart settlement data; extracting the time point data of the shopping cart settlement data to obtain the shopping cart settlement time point data; calculating the settlement commodity storage time of the shopping cart data according to the shopping cart settlement time point data and the commodity purchase time point data to generate the settlement commodity storage time data; Step S44: performing removal feature recognition on the shopping cart data to obtain shopping cart removal data; extracting the time point data of the shopping cart removal data to obtain the shopping cart removal time point data; calculating the storage time of the removed goods on the shopping cart data according to the shopping cart removal time point data and the product purchase time point data to generate the removed goods storage time data; Step S45: Integrate the settled commodity storage time data and the removed commodity storage time data to generate commodity storage time data; perform a deleted commodity analysis on the commodity shopping cart data according to the commodity storage time data to obtain deleted commodity data, wherein the deleted commodity data includes the settled commodity data and the removed commodity data.

6. The user feature extraction method of the e-commerce recommendation model according to claim 1, characterized in that: Step S5 includes the following steps: Step S51: dividing the product order type data, product detail page access data and deleted product data by impact weight to obtain product impact weight data; Step S52: constructing an e-commerce recommendation model according to the commodity influence weight data to obtain an e-commerce recommendation pre-model; and using the commodity influence weight data to perform model training on the e-commerce recommendation pre-model to obtain an e-commerce recommendation training model; Step S53: performing model cross-validation evaluation on the e-commerce recommendation training model to obtain e-commerce recommendation evaluation data; adjusting model parameters of the e-commerce recommendation training model using the e-commerce recommendation evaluation data to obtain an e-commerce recommendation model; Step S54: Recommend products through the e-commerce recommendation model to generate product recommendation data.

7. The user feature extraction method of the e-commerce recommendation model according to claim 6 is characterized in that: Step S51 includes the following steps: Step S511: performing order type similarity analysis on the commodity order type data to obtain commodity type similarity data; Step S512: performing association analysis on the settlement product data according to the product detail page access data to obtain settlement product detail page association data; identifying the number of visits to the details page visited data according to the settlement product detail page association data to generate details page visit number data; extracting multiple visit information from the details page visit number data to obtain details page multiple visit data; Step S513: Analyze the reasons why the removed product data has not been accessed through the details page not accessed data to obtain the details page not accessed reason data; Step S514: constructing a user product type feature matrix for the product type similarity data based on the details page multiple access data and the details page non-access reason data to obtain a user product type feature matrix; Step S515: Perform multiple linear regression analysis on the user product type feature matrix to generate multiple linear regression coefficients; perform significant impact analysis on the user product type feature matrix based on the multiple linear regression coefficients to obtain product significant impact data; perform weight analysis on the product significant impact data to obtain product impact weight data.

8. The method for extracting user features of an e-commerce recommendation model according to claim 1, characterized in that: Step S6 includes the following steps: Step S61: Count the number of times the product recommendation data is displayed to obtain display number statistics; monitor the number of clicks on the product recommendation data to obtain recommended product click number data; calculate the ratio of the recommended product click number data and the display number statistics to obtain recommended product click rate data; Step S62: extracting collection features from the product recommendation data according to the recommended product click rate data to obtain recommended product collection attribute data; identifying purchase records from the recommended product collection attribute data to obtain recommended product purchase data; analyzing purchase situations from the recommended product purchase data to obtain recommended product purchase data; identifying the whole process interactive features from the recommended product purchase data to obtain recommended product interactive data; Step S63: Analyze the skipping situation of the product recommendation data according to the recommended product click count data to obtain the recommended product skipping data; Step S64: Integrate the recommended product click rate data, recommended product interaction data and recommended product skip data to obtain recommended product feedback data.

9. The user feature extraction method of the e-commerce recommendation model according to claim 1, characterized in that: Step S7 includes the following steps: Step S71: intelligently adjusting the product recommendation data based on the recommended product feedback data to generate product recommendation adjustment data; Step S72: performing real-time monitoring, adjustment and feedback on the e-commerce recommendation model through the commodity recommendation adjustment data to obtain real-time adjustment data and real-time monitoring feedback data; Step S73: Integrate the real-time adjustment data and the real-time monitoring feedback data to obtain a real-time adjustment feedback report; Step S74: construct a cloud platform according to the real-time adjustment feedback report to obtain a real-time adjustment feedback cloud platform, wherein the real-time adjustment feedback cloud platform construction includes the construction of a real-time adjustment interface for product user features and the construction of a real-time monitoring feedback interface for product user features; execute a user feature extraction method operation of an e-commerce recommendation model based on the real-time adjustment feedback cloud platform.

10. The user feature extraction method of the e-commerce recommendation model according to claim 9, characterized in that: Step S71 includes the following steps: Step S711: performing click rate evaluation on the recommended product click rate data to obtain recommended product click rate evaluation data, wherein the recommended product click rate evaluation data includes high click rate data and low click rate data; Step S712: intelligently adjust the product recommendation data based on the recommended product click rate evaluation data. If the recommended product click rate evaluation data is high click rate data, the product recommendation data is adjusted by increasing the recommended page area of ​​products of the same type to obtain high click rate product adjustment data; if the recommended product click rate evaluation data is low click rate data, the product recommendation data is adjusted by reducing the recommended page area of ​​products of the same type to obtain low click rate product adjustment data; integrate the high click rate product adjustment data and the low click rate product adjustment data to generate click rate product adjustment data; Step S713: Perform product interaction stage analysis on the recommended product interaction data to obtain recommended product interaction stage data, wherein the recommended product interaction stage data includes product collection stage, product purchase stage and product purchase stage; Step S714: intelligently adjust the product recommendation data based on the recommended product interaction stage data. If the recommended product interaction stage data is the product collection stage and the product purchase stage, the product recommendation data is adjusted to increase the pre-purchase push frequency of related types of products to obtain the collection purchase stage adjustment data; if the recommended product interaction stage data is the product purchase stage, the product recommendation data is adjusted to increase the post-purchase display frequency of the same type of products to obtain the purchase stage adjustment data; the collection purchase stage adjustment data and the purchase stage adjustment data are integrated to generate the product interaction stage adjustment data; Step S715: intelligently adjusting the product recommendation data based on the recommended product skip data. If the product recommendation data is recommended product skip data, removing recommendations of products of the same type from the product recommendation data to obtain product skip adjustment data. Step S716: Integrate the click-through rate product adjustment data, the product interaction stage adjustment data and the product skip adjustment data to obtain product recommendation adjustment data.

Images