A commodity search method based on big data and electronic shopping mall

Abstract

The present application discloses a commodity search method based on big data and electronic shopping malls. The user terminal collects payment order information and scene interaction data generated by the intelligent payment device, and obtains negative interaction behaviors related to user commodities at the same time; obtains commodity categories or features that the user is not interested in through the user's historical purchase records; generates anti-intention tags based on negative interaction behaviors; generates anti-object tags based on commodity categories that the user is not interested in, and integrates anti-intention tags and anti-object tags into a reverse screening group; parses the payment order list to obtain the label features of the commodities in the payment order; summarizes the scene interaction data to form a scene interaction list, and obtains the commodity matching results of the payment order through the classification unit. Through reverse screening, the accuracy of recommendations is improved, and multi-path computing further enhances the accuracy of recommendations, ensuring that the recommended commodities are highly matched with the actual needs of users, and the multi-path computing push scheme improves the accuracy of scene determination.

Description

Technical Field

[0001] The present invention relates to the technical field of big data and electronic shopping malls, and in particular to a commodity search method based on big data and electronic shopping malls. Background Art

[0002] With the development of Internet technology, e-commerce has developed rapidly with its advantages of convenience, speed, low price and complete product variety. Among them, online shopping has gradually replaced offline shopping and become a shopping mode favored by consumers. In the prior art, the product recommendation model for consumers to shop online is mainly based on the user's search content. When consumers search for a certain product in an online shopping mall, the big data cloud service platform in the background obtains the user's interactive data through data mining, and recommends products to consumers based on this.

[0003] For example, Chinese patent application number CN202011067597.3 discloses a product search method, system and cloud service platform based on big data and e-commerce malls. The method is based on an order scenario list, a scenario interaction feature list, and an object label feature distribution list. The method obtains the product matching results corresponding to the payment order object through the classification unit included in the product push model, and then determines the product recommendation list corresponding to the payment order object list based on the product matching results. This design conducts a specific analysis based on the payment order information, and conducts a comprehensive analysis by extracting multi-dimensional feature information such as order scenarios and product types. It can perform a correlation analysis with the products that consumers expect to purchase, thereby carefully considering the association information between consumers and products, and then accurately analyzing the needs of consumers, and recommending qualified products to consumers, so that the recommended products have higher accuracy and can be recommended to consumers more accurately.

[0004] However, this specific analysis based on payment order information extracts multi-dimensional feature information such as order scenarios and product types for comprehensive analysis. If the user's purchasing behavior changes for some reason, it is impossible to respond to the user's behavior changes in real time, resulting in inaccurate recommendation lists. Summary of the invention

[0005] The present application provides a product search method based on big data and e-commerce malls. By dynamically testing products and calculating intention paths, the system can respond to changes in user behavior in real time and provide personalized product recommendations that meet current needs.

[0006] This application provides a product search method based on big data and electronic shopping malls, including:

[0007] S101, the user terminal collects payment order information and scene interaction data generated by the smart payment device, and obtains user feedback on the product and interaction behavior related to the product;

[0008] Obtain product categories or features that users are not interested in through user purchase history, browsing behavior, and search keywords;

[0009] S102, based on the feedback information and interactive behavior obtained in step S101, the obtained feedback information and interactive behavior are compared and analyzed with the sentiment concept network to generate an anti-intention label; based on the product categories or features that the user is not interested in, an anti-object label is generated, and the anti-intention label and the anti-object label are integrated into a reverse screening group;

[0010] S103, extracting a payment order list according to the collected payment order information, wherein the payment order list includes multiple payment orders, and the first product push parsing unit in the product push model parses the payment order list to obtain an order scenario list;

[0011] S104, the second product push parsing unit in the product push model parses the payment order list, obtains the label features of the products in the payment order, and aggregates the label features of each order to form a label feature distribution list;

[0012] S105, aggregating the scene interaction data to form a scene interaction list, and obtaining a product matching result of the payment order through a classification unit based on the order scene list, the scene interaction list and the tag feature distribution list;

[0013] S106, use the reverse screening group to filter and screen the product matching results to form a possible intended product pool, and organize the products and scenes in the possible intended product pool into a set, which is a possible space.

[0014] Preferably, based on step S101, the user's historical purchase records, browsing behaviors and search keywords are collected, and a collaborative filtering algorithm is used to calculate products similar to the target user's historical purchase records. The screened products are arranged according to the degree of similarity, and a recommendation list is generated. The generated recommendation list is recommended to the user, and the remaining dissimilar products or products that the user is not interested in are sorted into a collection.

[0015] Preferably, the product push model includes a first product push parsing unit, a second product push parsing unit and a classification unit. The first product push parsing unit analyzes product information, purchase time and purchase frequency in an order to identify different order scenarios. The second product push parsing unit analyzes the attributes, categories, prices and other characteristics of each product in the order to form an object label feature distribution. The classification unit obtains the product matching result corresponding to the payment order through a classification algorithm.

[0016] Preferably, the specific steps for refining the possible space are:

[0017] S201, extracting feature parameters from the order scenario list, the scenario interaction list, and the tag feature distribution list, and selecting the middle value of the feature parameter as a reference value;

[0018] S202, selecting multiple groups of products from the possible space, each group of products selected has different features, to test the impact of changes in different features on user interaction;

[0019] S203, selecting test users according to their purchase history, delivering the test product to the test users, and collecting the test users' clicks, browsing time, and purchase interaction behavior data;

[0020] S204, analyzing the preferences of the test users based on the collected interactive behavior data, adjusting the characteristic parameters of the test product based on the analysis results, and optimizing the possible space.

[0021] Preferably, the user purchases the product after selecting the product multiple times, and the nodes generated when the user selects the product are recorded to form an intention path.

[0022] Preferably, the intention path forming step includes:

[0023] S301, quantifying each tag in the tag feature distribution list, wherein the quantification processing means that each tag has a numerical range or a scoring system, and based on the user's historical transaction records and real-time browsing data, the historical data is sorted into a training set, and the training set is input into the model to train the model, and the expected quantitative tag range of the recommended node is calculated by the model;

[0024] S302, using a reverse screening group to screen the products within the expected quantitative label range, removing products beyond the expected calculation range, and narrowing the expected quantitative label range based on the user's interactive information;

[0025] S303, according to the questions preset when the user first enters the system, determine the initial intention tag set, preset the tag priority, explore each intention tag in the preset tag priority order, and stop exploring the intention tag once the range of an intention tag is determined, and turn to the next tag with an undetermined range, record the range path of the user's intention tag from the initial to the final determination, and form an intention path;

[0026] S304, real-time monitoring and collection of user behavior. If the user behavior changes, recalculate the expected quantitative label range, update the intention path, and dynamically adjust the recommended products.

[0027] Preferably, the intention path formed in step S303 is combined with the intersecting purchase scenarios to form multiple intention paths.

[0028] Preferably, the steps of forming the multiple intention paths include:

[0029] S401, based on steps S301 to S304, the user's historical purchase behavior and current scene interaction data, generate an initial intention path;

[0030] S402, based on the user's actual needs and changes in purchasing behavior, find multiple cross-scenarios related to the user's current intended scenario, design multiple cross-paths based on the cross-scenarios, recommend products under different paths to the user, observe the user's clicks, browsing, and dwell time, and record the characteristics of products that the user has not ordered;

[0031] S403, based on the analysis results of the features of the products that the user has not ordered, extracting boundary indicators that affect the user's purchase result, the boundary indicators include price, brand, function and style; searching for products that meet the boundary indicator requirements in the cross-scenario;

[0032] S404, when the user shows interactive behavior of browsing or staying on the newly generated cross-scenario product, the intention measurement of the path is strengthened, and the priority and weight of each path are adjusted according to the user's interactive behavior.

[0033] Preferably, the boundary indicator is a key indicator that causes a directional change in the intention path during the user interaction process, and the boundary indicator appears at the critical point where the user transfers from one purchase intention to another.

[0034] Preferably, a special tag for the user is generated based on multiple boundary indicators, and the special tag integrates the user's purchase history, current preferences, and potential purchase intention transfer trends.

[0035] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0036] Through quantitative processing, expected calculation and reverse screening, the system can accurately capture the user's purchase intention and improve the accuracy of recommendations. The multi-path computing push solution and the definition of boundary indicators and special tags further enhance the accuracy of recommendations and ensure that the recommended products are highly matched with the user's actual needs. The multi-path computing push solution improves the accuracy of scenario determination and comprehensively covers the user's potential purchase intention by designing multiple cross-intention paths.

[0037] Dynamic testing of products and intention path calculations enable the system to respond to changes in user behavior in real time and provide personalized product recommendations that meet current needs. Accurate push notifications reduce the interference of invalid information and improve user acceptance and satisfaction with recommended products, thereby optimizing the overall user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of a commodity search method based on big data and electronic shopping malls of the present invention;

[0039] Figure 2 A schematic diagram of a flow chart of optimizing a possible space according to an embodiment of the present invention;

[0040] Figure 3 A schematic diagram of a process for forming an intention path according to an embodiment of the present invention;

[0041] Figure 4 The figure is a schematic diagram of a process of forming multiple intention paths according to an embodiment of the present invention. DETAILED DESCRIPTION

[0042] To facilitate the understanding of the present invention, the present application will be described more comprehensively below with reference to the relevant drawings; the drawings show preferred embodiments of the present invention, but the present invention can be implemented in many different forms and is not limited to the embodiments described herein; on the contrary, the purpose of providing these embodiments is to enable a more thorough and comprehensive understanding of the disclosed content of the present invention.

[0043] It should be noted that the terms “vertical”, “horizontal”, “up”, “down”, “left”, “right” and similar expressions used in this document are only for illustrative purposes and do not represent the only implementation method.

[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which the present invention belongs; the terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention; the term "and / or" used herein includes any and all combinations of one or more related listed items.

[0045] Embodiment 1: Figure 1 The flowchart of a commodity search method based on big data and an electronic mall according to an embodiment of the present invention includes the following steps:

[0046] S101, the user terminal collects payment order information and scenario interaction data generated by the smart payment device, and obtains user feedback on the product and interaction behavior related to the product; obtains product categories or features that the user is not interested in through the user's historical purchase records, browsing behavior and search keywords.

[0047] Specifically, a smartphone is used to connect to a mobile payment terminal through an application programming (API) interface to collect payment order information in real time, including transaction amount, time, type, status, and product details. At the same time, the smartphone records the user's scene interaction data during the payment process through MouseFlow, including click behavior, dwell time, and page jump path before and after payment. A user feedback channel for in-APP evaluation, customer service complaints, and return applications is established, and the reasons for in-APP evaluation, customer service complaints, and return applications are set as mandatory items. Negative feedback from users on products is collected through the reasons for user feedback, and negative interaction behaviors such as users frequently viewing products but not purchasing them, deleting them after adding them to the shopping cart, and exiting the page after a quick browse are monitored and recorded.

[0048] Collect users' historical purchase records, browsing behaviors, and search keywords, use collaborative filtering algorithms to calculate products that are similar to the target users' historical purchase records, sort the selected products according to their similarity, generate a recommendation list, recommend the generated recommendation list to the user, and organize the remaining dissimilar products or products that the user is not interested in into a collection.

[0049] S102, generating anti-intention tags through sentiment analysis according to the feedback information and interactive behaviors obtained in step S101; generating anti-object tags according to the product categories or features that the user is not interested in, and integrating the anti-intention tags and anti-object tags into a reverse screening group.

[0050] Specifically, extract words and phrases from the collected feedback information, establish a sentiment concept network (SenticNet) containing negative words and their synonyms and antonyms, input the words and phrases into the established sentiment concept network (SenticNet), match the input words and phrases with the words in the sentiment concept network (SenticNet), compare and identify all negative words, output the level of negative emotion according to the identified negative words, and the negative emotion level is divided into indifference, disappointment and disgust, and generate a series of anti-intention labels according to the negative emotion level. Count the categories to which the products marked by the user as uninterested belong, find out the categories that the user is obviously not interested in or has never paid attention to, and generate anti-object labels according to the products that the user is not interested in.

[0051] In some embodiments, after purchasing a dress, a user leaves a comment: Very disappointed! The size of this dress is completely inaccurate. It is marked as L size but is as small as M size, and it is tight when worn. Moreover, the fabric does not match the description. It is said to be pure cotton, but the one received is blended, which is not breathable and easy to pilling. I will never buy clothes from this store again. Based on the above comments, sentiment analysis is performed to extract the words "inaccurate size", "tight", "fabric inconsistent" and "easy to pilling". Based on these words, the sentiment level of disappointment is output, and the anti-intention labels of "size inconsistent", "fabric inconsistent with description" and "easy to pilling" are generated according to the above extracted words and sentiment levels. When browsing products, when the user browses to products with lace trims and tight labels, he clicks the "not interested" button, and by analyzing the characteristics of the products that he is not interested in, the anti-object labels of "lace trims" and "tight" are generated.

[0052] S103, extracting a payment order list according to the collected payment order information, wherein the payment order list includes multiple payment orders, and parsing the payment order list through the first product push parsing unit in the product push model to obtain an order scenario list.

[0053] Specifically, the product push parsing model includes a first product push parsing unit, a second product push parsing unit and a classification unit. The first product push parsing unit can analyze the product information, purchase time and purchase frequency in the order to identify different order scenarios. The second product push parsing unit analyzes the attributes, categories, prices and other characteristics of each product in the order to form an object label feature distribution. The classification unit obtains the product matching result corresponding to the payment order through a complex classification algorithm. The classification unit is the core of the product push model and can utilize multi-dimensional feature information to accurately match products that consumers may be interested in through methods such as label multi-dimensional feature clustering processing and scene pointing trajectory simulation.

[0054] The first product push parsing unit in the product push model traverses each order in the payment order list and identifies the purchase scenarios in each order, which include daily shopping, holiday gifts and promotional activities. According to the above purchase scenarios, the first product push parsing unit classifies the identified purchase scenarios. Each order purchase scenario represents a similar purchase intention. Finally, the first product push unit generates an order scenario list, and each order scenario corresponds to one or more payment orders.

[0055] S104, the second product push parsing unit parses the payment order list, obtains label features of the products in the payment order, and aggregates the label features of each order to form a label feature distribution list.

[0056] The second product push parsing unit in the product push model traverses each order in the payment order list, parses the product information of each order, identifies the multidimensional features of the products in each order, the multidimensional features include brand, price range, size specifications, color preferences and material characteristics, constructs corresponding label features for each payment order based on the multidimensional features, and summarizes and analyzes the label features in each order to form a label feature distribution list.

[0057] The product push model is an existing technology, and the specific working method will not be repeated in this embodiment.

[0058] S105, the scene interaction data is aggregated to form a scene interaction list, and based on the order scene list, the scene interaction list and the label feature distribution list, the commodity matching result of the payment order is obtained through the classification unit.

[0059] Among them, based on the orders in the order scenario list, scenario interaction list and tag feature distribution list, a group of filtering tags are preset, and the filtering tags are confirmed according to historical browsing data and order information, and the orders in the order scenario list, scenario interaction list and tag feature distribution list are compared with the preset filtering tags. When the similarity between the order and the filtering tag exceeds a preset threshold, it is considered that the order matches the tag successfully, that is, it is a tag with high similarity. The successfully matched tag with high similarity is used as the product matching result, and the threshold is preset according to historical information and expert experience.

[0060] S106, using a reverse screening group to filter and screen the product matching results to form a pool of possible intended products.

[0061] Specifically, the product matching results are matched with the anti-intention tags. If a match is found, the result is marked as not recommended and removed from the product matching results. Products containing anti-object tags in the product matching results are directly removed, and the remaining product matching results form a possible intention product pool. The products in the possible intention product pool are more in line with the user's potential purchasing needs. The anti-intention tags and anti-object tags are updated in time according to the user's real-time feedback.

[0062] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0063] By introducing anti-intention tags and anti-object tags, anti-intention tags are used to exclude products and scenarios that match users' negative preferences, reducing products that users don't like in the recommendation list, and anti-object tags are used to directly eliminate products that users clearly will not buy, avoiding resource waste and a decline in user experience.

[0064] Embodiment 2: Based on the possible intended product pool in Embodiment 1, the products and scenarios in the possible intended product pool are sorted into a set, and the set is a possible space. With the continuous changes in user behavior and the update of market trends, the content in the possible space is also constantly adjusted.

[0065] like Figure 2 As shown, the specific steps to optimize the possible space are:

[0066] S201, extracting feature parameters from the order scenario list, the scenario interaction list and the label feature distribution list, and selecting the middle value of the feature parameters as the reference value.

[0067] Specifically, in the order scenario list, orders are classified according to the type of goods in the scenario, and each type of scenario is assigned an identification code according to the frequency of occurrence from large to small. If the frequency of occurrence is the same, they are sorted by the first letter of the product name, and the number of clicks and average dwell time are extracted from the scenario interaction list. The extracted number of clicks and average dwell time are sorted from large to small, and the median, that is, the middle value, is selected. Similarly, prices are extracted from the label feature list, sorted, and the middle value is taken. For brands and colors, they are encoded according to the frequency of occurrence. If the frequency of occurrence is the same, they are sorted by the first letter of the product name. The codes are sorted and the middle value is taken. The middle value is taken as the benchmark value to make the product universal.

[0068] S202, selecting multiple groups of products from the possible space, each group of products selected has differences in features, and is used to test the impact of changes in different features on user interaction.

[0069] Among them, keep the non-test characteristics consistent to ensure the accuracy and comparability of the test, select three groups of products from the possible space for testing, and use the control variable method to keep the non-test characteristics consistent, that is, except for the characteristics being tested, control other variables unchanged.

[0070] S203, selecting users to participate in the test according to their purchase history, delivering the test product to the users to be tested, and collecting the test users' clicks, browsing time, and purchase interaction behavior data.

[0071] Traverse the user's purchase history, build user portraits based on the user's historical purchase records, and then match these portraits with user portraits in the possible space. Select the top 10 users as the test group by sorting, and select users who participate in the test based on their purchase history. This makes the selected users potential users in the possible space, making the test results more accurate. In addition to collecting user interaction behavior data, the user feedback system collects test users' evaluations and suggestions.

[0072] S204, analyzing the preferences of the test users based on the collected interactive behavior data, adjusting the characteristic parameters of the test product based on the analysis results, and optimizing the possible space.

[0073] Specifically, the interactive behavior data are analyzed using SPSS statistical software, mainly for hypothesis testing and correlation analysis of click-through rate, browsing time, and conversion rate. Based on the analysis results, specific optimization measures are formulated for different characteristics, such as increasing product inventory in popular colors, adjusting marketing strategies for unpopular styles, and setting more attractive price discounts.

[0074] In some embodiments, the key feature parameters extracted from the comprehensive scene pointing tag include: price, style and color. The price is arranged in order from small to large, and the frequency of style and color is arranged according to size. The median value of the selected price is 400 yuan, the median value of the style frequency is a dress, and the median value of the color is black. Three groups of test products are designed, each group contains two dresses. The first group tests the color, one is a black dress, and the other is a beige dress. The second group tests the style, one is an A-shaped dress (black in the first group), and the other is a slim dress of the same brand. The third group tests price sensitivity, one keeps the original price of 400 yuan, and the other is a limited-time discount price of 300 yuan. Select user groups that meet the possible space user portraits as test objects, display the test products to these users, and collect their clicks, browsing time, adding to shopping carts, final purchases and other behavioral data. At the same time, collect users' direct evaluation and suggestions on the test products through questionnaires or user feedback systems. Through analysis, it is found that the click rate and the proportion of adding to shopping carts of beige dresses are significantly higher than those of black dresses, indicating that users may be more inclined to choose light-colored products. Although the number of views for slim-fit dresses is not low, the purchase conversion rate is relatively low, which may be related to the high comfort requirements of its target user group. The purchase conversion rate of dresses at a limited-time discount is significantly higher than that of products at the original price, which verifies the important influence of price sensitivity on user purchase decisions. Based on the test results, the platform decided to increase the inventory of light-colored dresses and optimize the recommendation algorithm to increase the exposure of such products. For slim-fit dresses, the platform is considering adjusting its marketing strategy to emphasize its wearing occasions and characteristics to attract more potential buyers.

[0075] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0076] By introducing the test group as mentioned above, the specific performance of user preferences in the possible space is verified. At the same time, the test group optimizes the data in the possible space, further improving the accuracy of recommendations and user satisfaction, and enhancing the user experience through personalized recommendations.

[0077] Embodiment 3: Based on the optimization of the possible space in Embodiment 2, Embodiment 3 further screens the commodities, so that the scope of the possible space is further narrowed, an intention path is formed, and the accuracy of the recommendation is further improved.

[0078] like Figure 3 As shown in the figure, the specific steps to form the intention path are:

[0079] S301, quantify each tag in the tag feature distribution list, where the quantization means that each tag has a numerical range or a scoring system. Based on the user's historical transaction records and real-time browsing data, the historical data is sorted into a training set, the training set is input into the model, the model is trained using a logistic regression algorithm, and the expected quantitative tag range of the recommended node is calculated by the model.

[0080] In some embodiments, price tags are quantified into 100-500 yuan, colors are converted into quantitative scores based on color psychology, and for non-numerical style tags, user preference data on styles are collected through user surveys, combined with expert evaluations, to divide styles into multiple levels, and assign corresponding quantitative scores to each level.

[0081] S302, using a reverse screening group to screen the products within the expected quantitative label range, removing the products beyond the expected calculation range, and narrowing the expected quantitative label range according to the user's interactive information.

[0082] Specifically, the products within the expected quantitative label range are matched with the labels in the reverse screening group. If a match is found, the result is marked as not recommended and removed from the expected quantitative label range.

[0083] S303, based on the questions preset when the user first enters the system, determine the initial set of intention tags, preset the priority of the tags, and explore each intention tag in the order of the preset tag priority. Whenever the scope of an intention tag is determined, stop exploring the intention tag and turn to the next tag with an undetermined scope. Record the user's range path from the initial to the final intention tag to form an intention path.

[0084] Among them, the preset questions are set based on in-depth market research and user portraits, and the principle of tag priority is based on historical records, that is, the more active the user is on a certain tag in the past, the higher the priority of the tag.

[0085] S304, real-time monitoring and collection of user behavior. If the user behavior changes, recalculate the expected quantitative label range, update the intention path, and dynamically adjust the recommended products.

[0086] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0087] Through quantitative processing, expectation calculation and reverse screening, the system can more accurately grasp the user's purchasing intention, thereby improving the accuracy of recommendations. Dynamic testing of products and intention path calculations enable the recommendation system to respond to changes in user behavior in real time and provide product recommendations that are more in line with the user's current needs, thereby optimizing the user's shopping experience. Accurate recommendations can increase the possibility of users clicking, browsing and purchasing products, thereby improving the conversion rate of e-commerce platforms.

[0088] Embodiment 4: An intention path based on Embodiment 3 is very limited for the purchase scenario. Therefore, multiple intention paths are designed to determine the user's purchase intention range. In the process of forming the intention path, after testing the products that have not been ordered for the previous N times, products that meet the boundary indicator quantity in the cross-scenario are selected on the boundary indicator to form a multi-path computing and push solution.

[0089] like Figure 4 As shown, the steps of multi-path push are:

[0090] S401, based on the user's historical purchase records, current browsing behavior and scenario data collected in steps S301 to S304, an initial intention path reflecting the user's current purchase intention is generated through a data analysis algorithm, such as cluster analysis or a decision tree model.

[0091] S402, based on the user's real-time feedback and changes in purchasing behavior, use data mining technology to identify multiple cross-scenarios associated with the user's current intention scenario, design multiple cross-paths for each cross-scenario, and display product sets under different paths to the user through A / B testing, record and analyze the user's interactive behavior under different paths, especially the specific characteristics of the products that the user has not ordered.

[0092] S403, based on the analysis results of the features of the products that the user has not ordered, extract boundary indicators that affect the user's purchase results, the boundary indicators include price, brand, function and style; search for products that meet the boundary indicator requirements in the cross-scenario.

[0093] Specifically, boundary indicators are key indicators that cause directional changes in the intention path during user interaction. These indicators often appear at the critical point where users shift from one purchase intention to another. By recording the product indicator at which the user ends browsing as the end indicator, when the next push begins, the last end indicator is compared with the new boundary indicator to form a calculation node, which contains information about changes in the user's preference for the product. Based on the information in the calculation node, the indicator direction and value range of the next push are predicted. In particular, when there are multiple boundary indicators, the user's purchase intention can be more accurately predicted based on the common characteristics and change trends of the multiple boundary indicators. Detailed characterization; based on multiple boundary indicators, a special label is generated for the user. The special label integrates the user's purchase history, current preferences and potential purchase intention transfer trends. The special label is further subdivided into two sub-labels: scenario demand orientation and product demand orientation. Scenario demand orientation describes the user's purchase preferences in different scenarios, while product demand orientation is specific to the functions, styles and other detailed features of the goods. According to the range of these two labels, a more accurate push strategy is formulated. For products that meet the user's scenario demand orientation, the system will give priority to push; at the same time, products that meet the user's scenario demand orientation and product demand orientation are extracted in the cross-scenario to form multiple intention paths.

[0094] S404, when the user shows interactive behavior of browsing or staying on the newly generated cross-scenario product, the intention measurement of the path is strengthened, and the priority and weight of each path are adjusted according to the user's interactive behavior.

[0095] Among them, for paths that users are more interested in, the recommendation frequency and exposure are increased; at the same time, for paths that users are less interested in, the system also needs to maintain a certain level of attention so that the recommendation strategy can be adjusted in time when user needs change.

[0096] In some embodiments, the system finds that the user has recently searched for relevant information about basketball shoes and browsed multiple basketball shoe product pages. Based on the search and browsing records, the system determines that the user's initial intention path is to purchase basketball shoes. The system identifies purchase scenarios related to purchasing basketball shoes, such as purchasing running shoes, casual shoes, and sports equipment. If the user does not purchase a certain pair of basketball shoes due to price reasons, the system should use the price of the basketball shoes as the upper limit of the price boundary quantity; if the user does not purchase high-priced basketball shoes due to price reasons, but the system finds that there are moderately priced products with similar functions in the running shoe category, these running shoes can be included in the new intention path.

[0097] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0098] By designing multiple cross-intention paths, we can more comprehensively cover users' potential purchasing intentions and improve the accuracy of scenario determination. At the same time, the ability to dynamically adjust paths makes the system more flexible and able to respond to changes in user needs in real time. The multi-path computing push solution can better match users' actual needs, reduce invalid recommendations, and thus improve user satisfaction.

[0099] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A commodity search method based on big data and electronic shopping mall, characterized in that: include: S101, the user terminal collects payment order information and scene interaction data generated by the smart payment device, and obtains user feedback on the product and interaction behavior related to the product; Obtain product categories or features that users are not interested in through user purchase history, browsing behavior, and search keywords; S102, based on the feedback information and interactive behavior obtained in step S101, the obtained feedback information and interactive behavior are compared and analyzed with the sentiment concept network to generate an anti-intention label; based on the product categories or features that the user is not interested in, an anti-object label is generated, and the anti-intention label and the anti-object label are integrated into a reverse screening group; S103, extracting a payment order list according to the collected payment order information, wherein the payment order list includes multiple payment orders, and the first product push parsing unit in the product push model parses the payment order list to obtain an order scenario list; S104, the second product push parsing unit in the product push model parses the payment order list, obtains the label features of the products in the payment order, and aggregates the label features of each order to form a label feature distribution list; S105, aggregating the scene interaction data to form a scene interaction list, and obtaining a product matching result of the payment order through a classification unit based on the order scene list, the scene interaction list and the tag feature distribution list; S106, filtering and screening the product matching results with a reverse screening group to form a possible intended product pool, and sorting the products and scenes in the possible intended product pool into a set, which is a possible space; Specific steps to refine the possible space: S201, extracting feature parameters from the order scenario list, the scenario interaction list, and the tag feature distribution list, and selecting the middle value of the feature parameter as a reference value; S202, selecting multiple groups of products from the possible space, each group of products selected has different features, so as to test the impact of changes in different features on user interaction; S203, selecting test users according to their purchase history, delivering the test product to the test users, and collecting the test users' clicks, browsing time, and purchase interaction behavior data; S204, analyzing the preferences of the test users according to the collected interactive behavior data, adjusting the characteristic parameters of the test product according to the analysis results, and optimizing the possible space; The method further includes: the user purchases the product after selecting the product multiple times, and forms an intention path by recording the nodes generated when the user selects the product; The intention path formation steps are: S301, quantifying each tag in the tag feature distribution list, wherein the quantification processing means that each tag has a numerical range or a scoring system, and based on the user's historical transaction records and real-time browsing data, the historical data is sorted into a training set, and the training set is input into the model to train the model, and the expected quantitative tag range of the recommended node is calculated by the model; S302, using a reverse screening group to screen the products within the expected quantitative label range, removing products beyond the expected calculation range, and narrowing the expected quantitative label range based on the user's interactive information; S303, according to the questions preset when the user first enters the system, determine the initial intention tag set, preset the tag priority, explore each intention tag in the preset tag priority order, and stop exploring the intention tag once the range of an intention tag is determined, and turn to the next tag with an undetermined range, record the range path of the user's intention tag from the initial to the final determination, and form an intention path; S304, real-time monitoring and collection of user behavior. If the user behavior changes, recalculate the expected quantitative tag range, update the intention path, and dynamically adjust the recommended products; The method further includes: combining the intention path formed in step S303 with the intersecting purchase scenarios to form multiple intention paths; Steps for forming the multiple intention paths: S401, based on steps S301 to S304, the user's historical purchase behavior and current scene interaction data, generate an initial intention path; S402, based on the user's actual needs and changes in purchasing behavior, find multiple cross-scenarios related to the user's current intended scenario, design multiple cross-paths based on the cross-scenarios, recommend products under different paths to the user, observe the user's clicks, browsing, and dwell time, and record the characteristics of products that the user has not ordered; S403, based on the analysis results of the features of the products that the user has not ordered, extracting boundary indicators that affect the user's purchase result, the boundary indicators include price, brand, function and style; searching for products that meet the boundary indicator requirements in the cross-scenario; S404, when the user shows interactive behavior of browsing or staying on the newly generated cross-scenario product, the intention measurement of the path is strengthened, and the priority and weight of each path are adjusted according to the user's interactive behavior.

2. A commodity search method based on big data and electronic shopping mall as claimed in claim 1, characterized in that: Based on step S101, the user's historical purchase records, browsing behaviors and search keywords are collected, and a collaborative filtering algorithm is used to calculate products similar to the target user's historical purchase records. The screened products are arranged according to the degree of similarity, and a recommendation list is generated. The generated recommendation list is recommended to the user, and the remaining dissimilar products or products that the user is not interested in are sorted into a collection.

3. A commodity search method based on big data and electronic shopping mall as claimed in claim 1, characterized in that: The product push model includes a first product push parsing unit, a second product push parsing unit and a classification unit. The first product push parsing unit analyzes product information, purchase time and purchase frequency in an order to identify different order scenarios. The second product push parsing unit analyzes the attributes, categories, prices and other features of each product in the order to form an object label feature distribution. The classification unit obtains the product matching result corresponding to the payment order through a classification algorithm.

4. A commodity search method based on big data and electronic shopping mall as claimed in claim 1, characterized in that: The boundary indicator is a key indicator that causes a directional change in the intention path during the user interaction process. The boundary indicator appears at the critical point where the user transfers from one purchase intention to another.

5. A commodity search method based on big data and electronic shopping mall as claimed in claim 4, characterized in that: Based on multiple boundary indicators, a special label for the user is generated. The special label integrates the user's purchase history, current preferences, and potential purchase intention transfer trends.

Images