E-commerce data intelligent classification recommendation method and system based on artificial intelligence
By calculating the user's platform activity indicators, product attraction indicators and behavioral interference indicators in the e-commerce platform, and combining the collaborative filtering algorithm, the products are divided into different categories, which solves the problem of ignoring the dynamic characteristics of users based on static attributes in the existing technology. The problem achieves more accurate user interests and preferences reflection and recommendation results.
Patent Information
- Application Number
- CN202411304238.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-19
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-09-19
AI Technical Summary
The product recommendation algorithm in the existing e-commerce platform is classified based on static product attributes, ignoring the dynamic characteristics of user behavior, resulting in low accuracy of recommendation results and unable to accurately reflect the changes in users' real interests and preferences.
By obtaining the historical shopping information of users on the e-commerce platform, calculating platform activity indicators, product attraction indicators and behavioral interference indicators, combining collaborative filtering algorithms, products are divided into different categories, and the improved item co-occurrence matrix and user co-occurrence matrix are used. Make recommendations for users.
It improves the accuracy of product recommendations, can better adapt to users' personalized needs and dynamic behaviors, reflects users' real interests and preferences, and improves the accuracy of recommendation results.
Smart Images

Figure CN118822698B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of commodity recommendation, and in particular to an e-commerce data intelligent classification recommendation method and system based on artificial intelligence. Background Art
[0002] The product recommendation system is a vital component of modern e-commerce platforms. The recommendation system uses user behavior, product information and historical interaction data to predict products that users may be interested in through algorithm models and displays them in a personalized manner. It aims to increase user purchasing intention, increase product sales and sales, provide accurate market analysis and forecasts, improve the company's operational efficiency and service quality, and enhance the company's competitiveness.
[0003] In actual application, the product recommendation table is generated by analyzing the user's product purchase information using the recommendation algorithm. However, there are many products on the e-commerce platform, which leads to a high cost for calculating the similarity between products in the recommendation algorithm. The existing technology classifies products according to product attributes, and then recommends products to users based on the similarity between products in different categories. However, product attributes are static features, which can easily make the product classification more rigid and ignore the dynamic features of user behavior. As a result, the recommendation results may not accurately reflect the user's real interests and preference changes, and thus the accuracy of product recommendation results is low. Summary of the invention
[0004] In order to solve the technical problem that commodities are classified based on commodity attributes and the dynamic characteristics of user behavior are ignored, resulting in low accuracy of commodity recommendations to users, the purpose of the present invention is to provide an e-commerce data intelligent classification recommendation method and system based on artificial intelligence. The technical solutions adopted are as follows:
[0005] In a first aspect, an embodiment of the present invention provides an e-commerce data intelligent classification recommendation method based on artificial intelligence, the method comprising:
[0006] Obtaining different types of shopping information of a number of users in the process of purchasing goods on an e-commerce platform within a historical period of time; the types include: platform type and user type;
[0007] Obtain each user's platform activity index based on each user's platform-based shopping information in a historical period;
[0008] According to the difference between the shopping information of the user class of the same commodity purchased by each user and all users, a local shopping feature vector of each user for each commodity is obtained; according to the similarity between the local shopping feature vectors of each commodity purchased by each user and all commodities, and the shopping information of the user class of each commodity purchased by each user, a commodity attraction index of each commodity for each user is obtained;
[0009] According to the similarity between each commodity purchased by each user and the platform-type shopping information of other commodities, obtain the behavior interference index of each user purchasing each commodity;
[0010] The commodities are divided into different categories according to the platform activity index, the commodity attraction index and the behavior interference index; and commodities are recommended to users by using a recommendation algorithm according to the number of times commodities of different categories are purchased by users.
[0011] Furthermore, obtaining the platform activity index of each user includes:
[0012] The platform-based shopping information includes: the time each user logs into the e-commerce platform each time, the number of purchases each user makes on the e-commerce platform, and the number of reviews each user makes on the purchased goods;
[0013] The platform activity index of each user is obtained based on the number of product evaluations by each user during the historical period, the number of shopping times during the number of times the user logs into the e-commerce platform, and the time interval between two adjacent log-ins to the e-commerce platform.
[0014] Furthermore, obtaining the commodity attraction index of each commodity to each user includes:
[0015] The average of the local shopping feature vectors of all commodities purchased by each user on the e-commerce platform within the historical time period is used as the comprehensive shopping feature vector of each user;
[0016] The shopping information of the user class includes: the total browsing time of each user on the product details page of each product purchased; obtaining the first similarity between the local shopping feature vector and the comprehensive shopping feature vector for each product purchased by each user; obtaining the product attraction index of each product for each user based on the first similarity and the total browsing time of the product details page of each product purchased by each user; the first similarity and the total browsing time are negatively correlated with the product attraction index.
[0017] Furthermore, obtaining the behavior interference index of each user purchasing each commodity includes:
[0018] Obtain product details of each product, convert keywords in the product details into word vectors, and obtain a product details vector for each product;
[0019] The platform-based shopping information also includes: the number of purchases by each user in each store on the e-commerce platform, and the product details vector of each product;
[0020] Obtaining a second similarity between each commodity purchased by each user and the commodity detail vectors of other commodities; and obtaining similar commodities to each commodity purchased by each user based on the second similarity;
[0021] According to the shopping frequency of each user at the store where each commodity purchased belongs and the shopping frequency of similar commodities in the historical time period, the behavior interference index of each user purchasing each commodity is obtained.
[0022] Furthermore, the commodities are classified into different categories according to the platform activity index, the commodity attraction index and the behavior interference index, including:
[0023] The commodity attraction index and the behavior interference index are weighted by using the platform activity index to obtain the shopping behavior index of each user for purchasing each commodity;
[0024] The set consisting of the shopping behavior indicators of all users who purchase the same product is recorded as the user shopping behavior set of each product; the third similarity between the user shopping behavior sets of different products is obtained, and the products are divided into different clusters according to the third similarity; all products in a cluster are one category.
[0025] Furthermore, the method of recommending products to users by using a recommendation algorithm based on the number of times that products of different categories are purchased by users comprises:
[0026] An improved item co-occurrence matrix is constructed according to the number of times different categories of items are purchased by users; and a first predicted interest level of each user for each category of items is obtained using an item-based collaborative filtering algorithm according to the improved item co-occurrence matrix;
[0027] Constructing an improved user co-occurrence matrix based on the number of categories of commodities commonly purchased by different users in the same category; obtaining the second predicted interest of each user for each category of commodities using a user-based collaborative filtering algorithm based on the improved user co-occurrence matrix;
[0028] The first predicted interest level and the second predicted interest level are weightedly summed to obtain a comprehensive predicted interest level of each user for each category of goods; and goods are recommended to each user based on the comprehensive predicted interest level.
[0029] Furthermore, the obtaining of the local shopping feature vector of each user for each commodity includes:
[0030] The user-class shopping information also includes: the browsing time of each user on different detail description parts of the product detail page of each product;
[0031] According to the browsing time of each detail description part of all users who purchased the same product, the overall browsing time of each detail description part of each product is obtained;
[0032] A vector formed by the difference between the browsing time of the detail description part of the product detail page of each product purchased by each user and the overall browsing time is recorded as the local shopping feature vector of each user for each product.
[0033] Furthermore, the acquiring similar commodities of each commodity purchased by each user based on the second similarity includes:
[0034] Arrange each product purchased by each user in order of the second similarity between the product detail vectors of other products, to obtain a detail similarity sequence of each product purchased by each user; obtain the difference between two adjacent elements in the detail similarity sequence, and select the largest element among the elements corresponding to the largest difference as a demarcation element; and take the products corresponding to the demarcation element in the detail similarity sequence that are greater than or equal to the demarcation element as similar products of each product purchased by each user.
[0035] Furthermore, the overall browsing time of each detail description part of each commodity is equal to the average browsing time of each detail description part of all users who purchased the commodity.
[0036] In a second aspect, another embodiment of the present invention provides an e-commerce data intelligent classification recommendation system based on artificial intelligence, the system comprising:
[0037] The data collection module is used to obtain different types of shopping information of several users in the process of purchasing goods on the e-commerce platform within a historical period of time; the types include: platform type and user type;
[0038] Use the activity analysis module to obtain each user's platform activity index based on each user's platform-based shopping information in a historical period;
[0039] The commodity attraction analysis module is used to obtain a local shopping feature vector of each user for each commodity based on the difference between the shopping information of the user class of the same commodity purchased by each user and all users; and obtain a commodity attraction index of each commodity for each user based on the similarity between the local shopping feature vectors of each commodity purchased by each user and all commodities, and the shopping information of the user class of each commodity purchased by each user;
[0040] A behavior interference module is used to obtain a behavior interference index for each user purchasing each product based on the similarity between each product purchased by each user and the platform-type shopping information of other products;
[0041] The product recommendation module is used to classify the products into different categories according to the platform activity index, the product attraction index and the behavior interference index; and recommend products to users using a recommendation algorithm according to the number of times that products of different categories are purchased by users.
[0042] The present invention has the following beneficial effects:
[0043] In an embodiment of the present invention, there are many users with different usage habits, that is, different levels of activity, in the e-commerce platform. The usage habits can be presented through the platform-type shopping information of the users in the historical time period, and the platform activity index of the users can be obtained; there are differences in the purchase process of different users in the actual purchase of goods, which is mainly reflected in the degree of users' understanding of the product information, which can be measured by the user-type shopping information of the goods. In order to highlight the differences in users' purchase processes, the difference between the user-type shopping information of each user and the same product purchased by all users is analyzed, so that the obtained local shopping feature vector reflects the shopping process characteristics of the user's purchase of a single product, that is, the shopping habits; the similarity between the local shopping feature vectors of each product purchased by the user and all products reflects the gap between the user's understanding of the information of the purchase of a single product and the usual purchase process, and is analyzed in combination with the user-type shopping information of each product purchased by the user. Both of the above can reflect the degree of attraction of the product to the user, thereby obtaining the product attraction index; the similarity between the platform-type shopping information of each product purchased by the user and other products shows the degree of interference of the user's historical purchase behavior on the user's purchase of the product. The user's purchasing behavior of goods is affected by both the degree of attraction of the goods' own information to the user, i.e., the goods attraction index, and the influence of the user's historical purchasing behavior, i.e., the behavior interference index. The platform activity index affects the credibility of the goods attraction index and the behavior interference index. The platform activity index, the goods attraction index and the behavior interference index jointly determine the user's shopping behavior characteristics for goods. Commodities are divided into different categories based on the above three indicators, and then the recommendation algorithm is used to recommend goods to users, which can better adapt to the user's personalized needs and dynamic behaviors, so that the recommendation results can accurately reflect the user's real interests and preference changes, and improve the accuracy of product recommendations to users. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0045] Figure 1 A flowchart of a method for intelligent classification and recommendation of e-commerce data based on artificial intelligence provided by an embodiment of the present invention;
[0046] Figure 2 A structural diagram of an e-commerce data intelligent classification recommendation system based on artificial intelligence provided by an embodiment of the present invention;
[0047] Figure 3 A schematic diagram of a computer device for an artificial intelligence-based e-commerce data intelligent classification and recommendation device provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0048] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of the e-commerce data intelligent classification recommendation method and system based on artificial intelligence proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.
[0049] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
[0050] The following is a detailed description of a specific scheme of an e-commerce data intelligent classification recommendation method and system based on artificial intelligence provided by the present invention in conjunction with the accompanying drawings.
[0051] Embodiment 1:
[0052] This invention proposes an e-commerce data intelligent classification recommendation method based on artificial intelligence, please refer to Figure 1 , which shows a flowchart of a method for intelligent classification and recommendation of e-commerce data based on artificial intelligence provided by an embodiment of the present invention, the method comprising:
[0053] Step S1: Acquire different types of shopping information of a number of users in the process of purchasing goods on an e-commerce platform within a historical period of time; the types include: platform type and user type.
[0054] Collect each user's purchase record, browsing record, login record and all text information contained in the product details description page on the e-commerce platform within the historical time period; the purchase record includes: the time when the user places an order for each product, the products evaluated by the user and the store from which the products purchased by the user belong; the browsing record includes: the length of time the user browses all the details description parts of the product details page of the product; the login record includes: the time each time the user logs into the e-commerce platform.
[0055] For each user, during the historical time period, the number of times the user shopped on the e-commerce platform, the number of times the user logged into the e-commerce platform, the number of times the user shopped at each store on the e-commerce platform, and the number of times the user evaluated the purchased products were counted. All the content in the product details description page of each product is used as the product details content, and the term frequency-inverse document frequency (TF-IDF) algorithm is used to extract keywords from all the text information contained in the product details description page of each product, and the Word2Vec algorithm is used to convert the keywords into word vectors, which are used as the product details vectors of each product.
[0056] In this embodiment, the platform-type shopping information includes: the time each user logs into the e-commerce platform each time, the number of purchases each user makes on the e-commerce platform, the number of reviews each user gives to purchased goods, and the number of purchases each user makes in each store on the e-commerce platform; the user-type shopping information includes: the browsing time each user spends on several detail description sections of the product details page of each purchased product, and the total browsing time each user spends on the product details page of each product.
[0057] It should be noted that, in this embodiment, the historical time period is the past year, and the implementer can set it according to the specific situation; the product details description page refers to the detailed display page of the product on the e-commerce platform, and the product details page includes three parts: product introduction, product price and product review area.
[0058] Step S2: Obtain the platform activity index of each user based on the platform-based shopping information of each user in the historical time period.
[0059] When using the collaborative filtering algorithm based on items to obtain the user's product recommendation table, the similarity between products is obtained by analyzing the user's behavior records on the e-commerce platform, and then the recommendation table is generated. However, when there are a large number of products, the cost of directly calculating the similarity matrix of the products is high. In order to improve the user experience, the recommendation algorithm needs to be optimized.
[0060] There are many users with different usage habits on e-commerce platforms. When using recommendation algorithms to generate product recommendation tables for users, there are differences in the degree of demand for products among users with different usage habits on e-commerce platforms. The user's usage habit characteristics are affected by the user's activeness in using the e-commerce platform. The user's platform activity index is obtained based on the user's platform shopping information.
[0061] Preferably, in some possible implementation modes of the embodiments of the present invention, the method for obtaining the platform activity index includes: the platform shopping information includes: the time each user logs into the e-commerce platform each time, the number of shopping times each user makes on the e-commerce platform and the number of evaluations each user gives to the purchased goods; the platform activity index of each user is obtained according to the number of evaluations of the goods by each user in the number of shopping times on the e-commerce platform within the historical time period, the number of shopping times in the number of logins to the e-commerce platform and the time interval between two adjacent logins to the e-commerce platform.
[0062] The user's activity level on the e-commerce platform can be comprehensively reflected by the completeness of the shopping process and the shopping frequency. The online evaluation of the purchased goods can measure the completeness of the shopping process. The more times the user evaluates the goods in the number of times he / she shopped on the e-commerce platform, the more complete the user's shopping experience on the e-commerce platform is, and the richer the user's experience in using the e-commerce platform is. The shopping frequency is presented by the frequency of logging into the e-commerce platform and the frequency of purchasing goods. The time interval between two consecutive logins of the user to the e-commerce platform reflects the frequency of logging into the e-commerce platform. If the time interval between two consecutive logins to the e-commerce platform is shorter, it means that the user uses the e-commerce platform more frequently, and the user is more active on the e-commerce platform; the number of shopping times in the number of logins to the e-commerce platform directly reflects the shopping frequency. The more times the user purchases goods, the more frequently the user uses the e-commerce platform, and the more active the user is on the e-commerce platform.
[0063] In this embodiment, the ratio of the number of times each user evaluates products on the e-commerce platform to the number of times they shop on the e-commerce platform in the historical time period is recorded as the evaluation frequency of each user; the average of the time intervals between two consecutive logins to the e-commerce platform of each user in the historical time period is taken as the average login interval of each user; the ratio of the number of shopping times to the number of platform logins of each user in the historical time period is taken as the purchase frequency of each user; based on the evaluation frequency, the average login interval and the purchase frequency, the platform activity index of each user is obtained.
[0064] The larger the platform activity index is, the more experience the user has in using the e-commerce platform and the more frequently the user uses it, the higher the user's trust in the e-commerce platform. Therefore, the evaluation frequency and purchase frequency are both positively correlated with the platform activity index, and the average login interval is negatively correlated with the platform activity index.
[0065] In this embodiment, the average login interval is negatively correlated and the product of the evaluation frequency, purchase frequency and the mapping result is normalized to obtain the platform activity index of each user. This embodiment can also construct the correlation between the evaluation frequency, the average login interval and the purchase frequency and the platform activity index through other basic mathematical operations, which will not be limited or elaborated here.
[0066] It should be noted that this embodiment uses the Norm function for normalization, and takes the inverse to achieve negative correlation mapping. This embodiment can also select other normalization methods, such as function conversion, maximum and minimum normalization, etc., and use methods such as exponential functions with natural constants as the base for negative correlation mapping, which are not limited here. Only users who purchase a certain product can evaluate the product once, and some users may not evaluate the product. In this case, the number of shopping times for each user on the e-commerce platform is less than or equal to the number of product evaluations; if a user logs into the e-commerce platform once and there are two situations: purchasing a product or not purchasing a product, then the number of times the user logs into the e-commerce platform is less than or equal to the number of times the product is purchased.
[0067] The total amount of consumption of each user on the e-commerce platform, the number of times the product is browsed, and the number of times the product is added to the shopping cart, etc., can also reflect the activity level of each user on the e-commerce platform. The platform shopping information can also include: the total amount of consumption of each user on the e-commerce platform, the number of times the product is browsed, and the number of times the product is added to the shopping cart, etc.
[0068] In a specific implementation of the embodiment of the present invention, the platform activity index is expressed by the formula:
[0069]
[0070] In the formula, H is the platform activity index of each user; P is the evaluation frequency of each user; G is the purchase frequency of each user; T is the average login interval of each user; Norm is the normalization function. It should be noted that the larger the platform activity index H, the more experience users have in using the e-commerce platform and the higher their trust in the e-commerce platform.
[0071] Step S3: Based on the difference between the shopping information of the user class of the same commodity purchased by each user and all users, obtain the local shopping feature vector of each user for each commodity; based on the similarity between the local shopping feature vectors of each commodity purchased by each user and all commodities, as well as the shopping information of the user class of each commodity purchased by each user, obtain the commodity attraction index of each commodity to each user.
[0072] There are differences in the purchase process of different users in the actual purchase process. This difference is mainly reflected in the degree of user's understanding of product information during the purchase process. The user-class shopping information of a product can reflect the degree of user's understanding of product information during the purchase process. In order to highlight the differences in users' purchase processes, the difference between the user-class shopping information of each user and all users for the same product is analyzed, so that the local shopping feature vector obtained can reflect the shopping process characteristics of users purchasing a single product.
[0073] Preferably, in some possible implementation modes of the embodiments of the present invention, the method for obtaining the local shopping feature vector includes: the shopping information of the user class also includes: the browsing time of each user on different detail description parts of the product details page of each product; according to the browsing time of each detail description part of all users purchasing the same product, the overall browsing time of each detail description part of each product is obtained; the vector formed by the difference between the browsing time of the detail description part of the product details page of each user purchasing each product and the overall browsing time is recorded as the local shopping feature vector of each user for each product.
[0074] This embodiment shows the user's understanding of the product information through the browsing time of different detailed description parts of the product details page during the user's purchase process. The longer the browsing time, the more the user understands the corresponding product information. The number of times the user searches for product keywords, the number of questions asked to the customer service of the product, and the number of times the product details page is viewed can also reflect the user's understanding of the product information. The user-related shopping information can also include: the number of times the user searches for product keywords, the number of questions asked to the customer service of the product, and the number of times the product details page is viewed.
[0075] In this embodiment, the proportion of each user's browsing time of each detail description part of the product details page of each product purchased to the total browsing time of the product details page is taken as the proportion of each user's browsing time of each detail description part of each product; the average of the proportion of the browsing time of each detail description part of all users purchasing the same product is taken as the overall browsing time of each detail description part of each product.
[0076] The browsing time of users in the detailed description of the product reflects the decisive influence of the information in the detailed description on the user's shopping behavior; the overall browsing time reflects the overall level of browsing time of each detailed description of all users who purchase the same product. Other embodiments can replace the mean with indicators such as mode and median to measure the overall level of data. The larger the value of different dimensions in the local shopping feature vector, the greater the degree of reference of the user to the information in the detailed description corresponding to the dimension during the shopping process.
[0077] It should be noted that the product details page of each product includes three parts: product introduction, product price and product review area; the sum of the browsing time of the user in the product introduction, product price and product review area of each product detail page is equal to the total browsing time. The values of the same dimension of different local shopping feature vectors represent the browsing time of the user in the same detail description part of the product detail pages of different products.
[0078] The local shopping feature vector reflects the shopping habits of users in the process of purchasing goods, that is, the characteristics of the overall shopping process. The similarity between the local shopping feature vectors of each product purchased by the user and all the products purchased by the user reflects the gap between the user's understanding of the information of a single product and the usual purchasing process. It is combined with the total browsing time of the product details page of each product purchased by the user for analysis. Both can reflect the degree of attractiveness of the product to the user. The product attractiveness index is obtained by combining the above factors.
[0079] Preferably, in some possible implementation modes of the embodiments of the present invention, the method for obtaining the product attraction index includes: taking the average of the local shopping feature vectors of all products purchased by each user on the e-commerce platform within the historical time period as the comprehensive shopping feature vector of each user; the user-class shopping information includes: the total browsing time of each user on the product details page of each purchased product; obtaining the first similarity between the local shopping feature vector and the comprehensive shopping feature vector of each purchased product by each user; obtaining the product attraction index of each product for each user based on the first similarity and the total browsing time of each user on the product details page of each purchased product; the first similarity and the total browsing time are both negatively correlated with the product attraction index.
[0080] The local shopping feature vector reflects the shopping habits of users in the process of purchasing a single product. By analyzing the shopping habits of users in the process of purchasing all products, a comprehensive shopping feature vector can be obtained. The comprehensive shopping feature vector reflects the shopping habits of individual users.
[0081] It should be noted that the similarity between two vectors can be analyzed by the cosine similarity and the Euclidean distance between the two vectors. In this embodiment, the cosine similarity between the local shopping feature vector and the comprehensive shopping feature vector of each commodity purchased by each user is normalized to obtain the first similarity; other embodiments can negatively correlate and normalize the Euclidean distance between the local shopping feature vector and the comprehensive shopping feature vector of each commodity purchased by each user to obtain the first similarity.
[0082] The attractiveness of a product to a user is reflected in the time the user spends on purchasing a product and the purchase decision structure, specifically: the shorter the total browsing time of the product details page of each product purchased by the user, the shorter the user's hesitation time in purchasing the product, and the more attractive the product is to the user; when purchasing a product, the greater the gap between the understanding of the product information and the usual purchasing process, that is, the greater the difference between the user's shopping habits for a product and the user's individual shopping habits, the greater the attractiveness of the product to the user; if the difference between the user's shopping habits for a product and the user's individual shopping habits is greater, the smaller the first similarity is.
[0083] Therefore, the first similarity between the local shopping feature vector and the comprehensive shopping feature vector of each commodity purchased by each user, and the total browsing time of the commodity details page of each commodity by each user are negatively correlated with the commodity attraction index. In this embodiment, the product of the first similarity between the local shopping feature vector and the comprehensive shopping feature vector of each commodity purchased by each user and the total browsing time of the commodity details page is negatively correlated and normalized to obtain the commodity attraction index of each commodity for each user. The larger the commodity attraction index, the greater the degree of attraction of each commodity to each user.
[0084] In other embodiments, other basic mathematical operations may be used to construct the first similarity between the local shopping feature vector and the comprehensive shopping feature vector of each product purchased by the user, and the correlation between the total browsing time of the product details page and the product attractiveness index, which is not limited or elaborated here.
[0085] Step S4: According to the similarity between each commodity purchased by each user and the platform-type shopping information of other commodities, the behavior interference index of each user purchasing each commodity is obtained.
[0086] The user's purchase behavior of goods is not only affected by the attractiveness of the goods' own information to the user, but also by the user's historical purchase behavior. The greater the similarity between each product purchased by the user and the platform-type shopping information of other products, the greater the interference of historical purchase behavior on the user's purchase of goods, thereby obtaining a behavior interference index.
[0087] Preferably, in some possible implementation modes of the embodiments of the present invention, the method for obtaining the behavior interference index includes: obtaining the product details content of each product, converting the keywords in the product details content into word vectors, and obtaining the product details vector of each product; the platform shopping information also includes: the number of shopping times of each user in each store on the e-commerce platform, and the product details vector of each product; obtaining the second similarity between each product purchased by each user and the product details vectors of other products; obtaining similar products of each product purchased by each user based on the second similarity; obtaining the behavior interference index of each user purchasing each product according to the number of shopping times in the store where each product purchased by each user belongs, and the number of shopping times for similar products in a historical time period.
[0088] When users buy goods, they usually tend to buy from stores where they have bought goods more often to ensure the quality of the goods. The user's historical purchase behavior is also affected by historical purchase factors. If a user has bought similar goods to a certain product more times, the user's historical purchase behavior will interfere more with the current purchase. Therefore, the number of purchases of each product purchased by the user and the number of purchases of similar products in the historical time period are positively correlated with the behavior interference index.
[0089] In this embodiment, the total number of shopping times of each user on the e-commerce platform in the historical time period is used to normalize the number of shopping times of the store to which each commodity purchased by each user belongs, so as to obtain the store frequency of each commodity purchased by each user; the ratio of the number of similar commodities of each commodity purchased by each user to the length of the historical time period is used as the similar purchase frequency of each commodity purchased by each user; the product of the store frequency and the similar purchase frequency is normalized to obtain the behavior interference index of each commodity purchased by each user.
[0090] In other embodiments, the correlation between the store frequency and the similar purchase frequency and the behavior interference index, such as the sum value, may also be constructed through other basic mathematical operations, which are not limited or elaborated herein.
[0091] The specific method for obtaining similar products for each product purchased by each user is as follows:
[0092] In this embodiment, each product purchased by each user is arranged in order of the second similarity between the product detail vectors of other products, so as to obtain a detail similarity sequence of each product purchased by each user; the difference between two adjacent elements in the detail similarity sequence is obtained, and the largest element among the elements corresponding to the largest difference is selected as the delimiting element; and the products corresponding to the delimiting element in the detail similarity sequence that are greater than or equal to the delimiting element are taken as similar products of each product purchased by each user.
[0093] It should be noted that if each user purchases more similar products in the historical time period, it means that the user's purchase behavior has a more obvious interference with the current product purchase behavior. The elements in the detail similarity sequence are the second similarity between the product detail vectors of each product and other products; the method for obtaining the similarity between two vectors has been described in detail in step S3 and will not be repeated here.
[0094] In other embodiments, for the second similarity between each product purchased by each user and the product detail vectors of other products, the products corresponding to the second similarity greater than the similarity threshold are used as similar products of each product purchased by each user. The similarity threshold is an empirical value of 0.7, which can be set by the implementer according to the specific situation.
[0095] Step S5: Divide the commodities into different categories according to the platform activity index, commodity attraction index and behavior interference index; and recommend commodities to users using a recommendation algorithm according to the number of times commodities of different categories are purchased by users.
[0096] The user's purchasing behavior of goods is affected by both the degree of attraction of the product's own information to the user, i.e., the product attraction index, and the user's historical purchasing behavior, i.e., the behavior interference index. The platform activity index determines the credibility of the product attraction index and the behavior interference index. The platform activity index, product attraction index, and behavior interference index jointly determine the user's shopping behavior characteristics for goods.
[0097] In this embodiment, the commodity attraction index and the behavior interference index are weighted by using the platform activity index to obtain the shopping behavior index of each user purchasing each commodity.
[0098] If the user's platform activity index is smaller, the user's historical usage data on the e-commerce platform is less, then the degree of interference of the user's historical purchase behavior on the user's purchase of a certain product, that is, the credibility of the behavior interference index is lower. At this time, it is necessary to use the product attraction index to analyze the user's shopping behavior characteristics for the product to ensure the accuracy of the shopping behavior characteristics. If the user's platform activity index H is larger, the user has more experience in using the e-commerce platform, the higher the trust in the e-commerce platform, there are obvious personalized differences in the purchase decisions of different products, and the user's browsing time for the product is relatively short, then the credibility of the product attraction index is lower. At this time, it is necessary to use the behavior interference index to analyze the user's shopping behavior characteristics for the product to ensure the accuracy of the shopping behavior characteristics.
[0099] In a specific implementation of the embodiment of the present invention, the shopping behavior index is expressed by the formula:
[0100]
[0101] In the formula, GX is the shopping behavior indicator of each user purchasing each product; H is the platform activity indicator of each user; Provides an indicator of behavioral interference for each product on each user; Product attraction metrics for each product purchased by each user.
[0102] The set of shopping behavior indicators of all users who purchase the same product is recorded as the user shopping behavior set of each product; the third similarity between the user shopping behavior sets of different products is obtained, and the products are divided into different clusters according to the third similarity; all products in a cluster are one category. Classifying products based on shopping behavior indicators not only adapts to users' personalized needs and dynamic behaviors, but also makes the recommendation results accurately reflect users' real interests and preference changes, thereby improving the accuracy of product recommendations for users.
[0103] It should be noted that, in this embodiment, the Jaccard distance between the user shopping behavior sets of two commodities is used as the third similarity between the two sets; the K-means clustering algorithm is selected to cluster the third similarity between the shopping behavior sets of two commodities, to obtain K clusters, and the K value is determined by the elbow method. Among them, the Jaccard distance, the K-means clustering algorithm and the elbow method are all well-known technologies for those skilled in the art, and will not be described in detail here.
[0104] In order to reduce the amount of calculation of similarity between products and ensure the accuracy of product recommendations, the co-occurrence matrix is simplified by dividing the products into different categories, thereby reducing the calculation cost and realizing product recommendations for users. The specific process is as follows:
[0105] An improved item co-occurrence matrix is constructed based on the number of times different categories of goods are purchased together by users; based on the improved item co-occurrence matrix, an item-based collaborative filtering algorithm is used to obtain each user's first predicted interest in each category of goods; an improved user co-occurrence matrix is constructed based on the number of categories of goods in the same category purchased by different users; based on the improved user co-occurrence matrix, a user-based collaborative filtering algorithm is used to obtain each user's second predicted interest in each category of goods.
[0106] As an example, if A, B, C, D, E represent users, a1, a2, b, c1, c2, d, e represent products, a1 and a2 are the first category LB1, product b is the second category LB2, c1 and c2 are the third category LB3, product d is the fourth category LB4, and product e is the fifth category LB5. Assuming that the reverse order from user to product is A: a1, a2, b, d, B: b, c1, c2, e, C: c1, c2, d, D: b, c1, c2, d, E: a1, a2, d, the improved product co-occurrence matrix X1 is constructed according to the number of times different categories of products are purchased by users; the improved user co-occurrence matrix X2 is constructed according to the number of categories of products purchased by different users in the same category.
[0107] The improved item co-occurrence matrix X1 is shown in Table 1:
[0108] Table 1
[0109]
[0110] If all products of the first category LB1 and the second category LB2 are purchased by user A at the same time, the final co-occurrence number of the first category LB1 and the second category LB2 is 1; if all products of the second category LB2 and the third category LB3 are purchased by users B and D at the same time, the final co-occurrence number of the second category LB2 and the third category LB3 is 2.
[0111] It should be noted that only when all the products of two categories are purchased by a user at the same time, the co-occurrence times of the two categories will be increased by 1; the number of rows and columns of the improved item co-occurrence matrix are equal to the number of categories into which the products are divided.
[0112] The improved user co-occurrence matrix X2 is shown in Table 2:
[0113] Table 2
[0114]
[0115] If user A and user E both purchase all products in the first category LB1 and the fourth category LB4, the final co-occurrence count of user A and user E is 2; if user B and user C purchase the third category LB3 at the same time, the final co-occurrence count of user B and user C is 1. It should be noted that only when two users purchase all products in the same category at the same time, the co-occurrence count of the two users will be increased by 1.
[0116] It should be noted that the improved item co-occurrence matrix X1 is equivalent to the improved item-item co-occurrence matrix in the collaborative filtering algorithm, and the improved user co-occurrence matrix X2 is equivalent to the improved user-user co-occurrence matrix in the collaborative filtering algorithm. The subsequent steps in the collaborative filtering algorithm remain unchanged, and finally the first predicted interest degree and the second predicted interest degree are obtained respectively.
[0117] The two algorithms, item-based collaborative filtering algorithm and user-based collaborative filtering algorithm, are combined to predict the user's interest in the product, further improving the accuracy of product recommendations to users.
[0118] In this embodiment, the method for obtaining the comprehensive predicted interest is: weighted summing the first predicted interest and the second predicted interest to obtain the comprehensive predicted interest of each user for each category of goods; and recommending goods to each user based on the comprehensive predicted interest. The comprehensive predicted interest of each user for each category of goods is expressed by the formula:
[0119]
[0120] In the formula, ZP is the comprehensive predicted interest of each user in each category of goods; The first predicted interest level of each user for each category of goods; The second predicted interest level of each user for each category of goods; is the first interest coefficient; is the second interest coefficient.
[0121] It is known that the item-based collaborative filtering algorithm is suitable for scenarios with a large number of products, and the user-based collaborative filtering algorithm is suitable for scenarios with a large number of users. This solution recommends products to users when there are a large number of products. The first interest coefficient Greater than Second interest coefficient. In this embodiment, the first interest coefficient Take the experience value 0.7, the second interest coefficient 2 takes the empirical value of 0.3, and implementers can set it according to specific circumstances.
[0122] For each user's comprehensive predicted interest in all categories of goods, the Top-N algorithm is used to recommend goods to each user. It should be noted that when recommending goods to users, all goods in a category are recommended to the user, and N is the number of categories of goods recommended to the user. Among them, N takes an empirical value of 50, and the implementer can set it according to the specific situation; the Top-N algorithm is a well-known technology for those skilled in the art and will not be described in detail here.
[0123] So far, the present invention is completed.
[0124] Embodiment 2:
[0125] This invention proposes an e-commerce data intelligent classification recommendation system based on artificial intelligence, please refer to Figure 2 , which shows a structure diagram of an e-commerce data intelligent classification recommendation system based on artificial intelligence provided by an embodiment of the present invention, the system includes:
[0126] The data collection module 610 is used to obtain different types of shopping information of a number of users in the process of purchasing goods on the e-commerce platform within a historical period of time; the types include: platform type and user type;
[0127] Using the activity analysis module 620, for obtaining the platform activity index of each user according to the platform-type shopping information of each user in the historical time period;
[0128] The commodity attraction analysis module 630 is used to obtain a local shopping feature vector of each user for each commodity based on the difference between the shopping information of the user class of the same commodity purchased by each user and all users; obtain a commodity attraction index of each commodity for each user based on the similarity between the local shopping feature vectors of each commodity purchased by each user and all commodities, and the shopping information of the user class of each commodity purchased by each user;
[0129] The behavior interference module 640 is used to obtain a behavior interference index of each user's purchase of each commodity according to the similarity between each commodity purchased by each user and the platform-type shopping information of other commodities;
[0130] The product recommendation module 650 is used to classify products into different categories according to the platform activity index, product attraction index and behavior interference index; and recommend products to users using a recommendation algorithm according to the number of times products of different categories are purchased by users.
[0131] It should be noted that the device provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device is divided into different functional modules to complete all or part of the functions described above. In addition, the e-commerce data intelligent classification recommendation system based on artificial intelligence and the e-commerce data intelligent classification recommendation method based on artificial intelligence provided in the above embodiment belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.
[0132] Embodiment 3:
[0133] Figure 3 A computer device schematic diagram of an e-commerce data intelligent classification recommendation device based on artificial intelligence provided by an embodiment of the present invention. For example, Figure 3 As shown, the computer device includes: a memory 701, a processor 702, and a computer program 703 stored in the memory 701 and running on the processor 702, wherein when the processor 702 executes the computer program 703, the computer device can execute any one of the artificial intelligence-based e-commerce data intelligent classification recommendation methods introduced above.
[0134] In addition, an embodiment of the present application also protects a device, which may include a memory and a processor, wherein the memory stores executable program code, and the processor is used to call and execute the executable program code to execute an artificial intelligence-based e-commerce data intelligent classification recommendation method provided in an embodiment of the present application.
[0135] In this embodiment, the functional modules of the device can be divided according to the above method example. For example, each functional module can be corresponded, or two or more functions can be integrated into one processing module. The above integrated module can be implemented in the form of hardware. It should be noted that the division of modules in this embodiment is schematic and is only a logical function division. There may be other division methods in actual implementation.
[0136] It should be understood that the device provided in this embodiment is used to execute the above-mentioned artificial intelligence-based e-commerce data intelligent classification recommendation method, and therefore can achieve the same effect as the above-mentioned implementation method.
[0137] In the case of an integrated unit, the device may include a processing module and a storage module. When the device is applied to a device, the processing module may be used to control and manage the actions of the device. The storage module may be used to support the device to execute mutual program codes, etc.
[0138] The processing module may be a processor or a controller, which may implement or execute various exemplary logic blocks, modules and circuits included in the disclosure of this application. The processor may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of digital signal processing (DSP) and a microprocessor, etc. The storage module may be a memory.
[0139] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0140] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.
[0141] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.
Claims
1. An e-commerce data intelligent classification recommendation method based on artificial intelligence, characterized in that: The method includes: Obtaining different types of shopping information of a number of users in the process of purchasing goods on an e-commerce platform within a historical period of time; the types include: platform type and user type; Obtain each user's platform activity index based on each user's platform-based shopping information in a historical period; According to the difference between the shopping information of the user class of the same commodity purchased by each user and all users, a local shopping feature vector of each user for each commodity is obtained; according to the similarity between the local shopping feature vectors of each commodity purchased by each user and all commodities, and the shopping information of the user class of each commodity purchased by each user, a commodity attraction index of each commodity for each user is obtained; According to the similarity between each commodity purchased by each user and the platform-type shopping information of other commodities, obtain the behavior interference index of each user purchasing each commodity; Classifying commodities into different categories according to the platform activity index, the commodity attraction index and the behavior interference index; and recommending commodities to users by using a recommendation algorithm according to the number of times commodities of different categories are purchased by users; The obtaining of each user's platform activity indicator includes: The platform-based shopping information includes: the time each user logs into the e-commerce platform each time, the number of purchases each user makes on the e-commerce platform, and the number of reviews each user makes on the purchased goods; Obtain each user's platform activity index based on the number of product reviews by each user during the historical period, the number of purchases during the number of logins to the e-commerce platform, and the time interval between two consecutive logins to the e-commerce platform; The obtaining of the commodity attraction index of each commodity to each user includes: The average of the local shopping feature vectors of all commodities purchased by each user on the e-commerce platform within the historical time period is used as the comprehensive shopping feature vector of each user; The shopping information of the user class includes: the total browsing time of each user on the product details page of each product purchased; obtaining the first similarity between the local shopping feature vector and the comprehensive shopping feature vector of each product purchased by each user; obtaining the product attraction index of each product for each user according to the first similarity and the total browsing time of each user on the product details page of each product purchased; the first similarity and the total browsing time are both negatively correlated with the product attraction index; The obtaining of the behavior interference index of each user purchasing each commodity includes: Obtain product details of each product, convert keywords in the product details into word vectors, and obtain a product details vector for each product; The platform-based shopping information also includes: the number of purchases by each user in each store on the e-commerce platform, and the product details vector of each product; Obtaining a second similarity between each commodity purchased by each user and the commodity detail vectors of other commodities; and obtaining similar commodities to each commodity purchased by each user based on the second similarity; Obtain a behavior interference index for each user's purchase of each product based on the number of purchases at the store where each product purchased by each user belongs, and the number of purchases of similar products in a historical period; The step of obtaining the local shopping feature vector of each user for each commodity includes: The user-class shopping information also includes: the browsing time of each user on different detail description parts of the product detail page of each product; According to the browsing time of each detail description part of all users who purchased the same product, the overall browsing time of each detail description part of each product is obtained; A vector formed by the difference between the browsing time of the detail description part of the product detail page of each product purchased by each user and the overall browsing time is recorded as the local shopping feature vector of each user for each product.
2. According to the artificial intelligence-based e-commerce data intelligent classification recommendation method of claim 1, it is characterized by: The commodities are classified into different categories according to the platform activity index, the commodity attraction index and the behavior interference index, including: The commodity attraction index and the behavior interference index are weighted by using the platform activity index to obtain the shopping behavior index of each user for purchasing each commodity; The set consisting of the shopping behavior indicators of all users who purchase the same product is recorded as the user shopping behavior set of each product; the third similarity between the user shopping behavior sets of different products is obtained, and the products are divided into different clusters according to the third similarity; all products in a cluster are one category.
3. The method for intelligent classification and recommendation of e-commerce data based on artificial intelligence according to claim 1, characterized in that: The method of recommending products to users by using a recommendation algorithm based on the number of times that products of different categories are purchased by users comprises: An improved item co-occurrence matrix is constructed according to the number of times different categories of items are purchased by users; and a first predicted interest level of each user for each category of items is obtained using an item-based collaborative filtering algorithm according to the improved item co-occurrence matrix; Constructing an improved user co-occurrence matrix based on the number of categories of commodities commonly purchased by different users in the same category; obtaining the second predicted interest of each user for each category of commodities using a user-based collaborative filtering algorithm based on the improved user co-occurrence matrix; The first predicted interest level and the second predicted interest level are weightedly summed to obtain a comprehensive predicted interest level of each user for each category of goods; and goods are recommended to each user based on the comprehensive predicted interest level.
4. The method for intelligent classification and recommendation of e-commerce data based on artificial intelligence according to claim 1, characterized in that: The acquiring similar commodities of each commodity purchased by each user based on the second similarity includes: Arrange each product purchased by each user in order of the second similarity between the product detail vectors of other products, to obtain a detail similarity sequence of each product purchased by each user; obtain the difference between two adjacent elements in the detail similarity sequence, and select the largest element among the elements corresponding to the largest difference as a demarcation element; and take the products corresponding to the demarcation element in the detail similarity sequence that are greater than or equal to the demarcation element as similar products of each product purchased by each user.
5. The method for intelligent classification and recommendation of e-commerce data based on artificial intelligence according to claim 1, characterized in that: The overall browsing time of each detail description part of each commodity is equal to the average browsing time of each detail description part of all users who purchased the commodity.
6. A system for implementing the artificial intelligence-based e-commerce data intelligent classification recommendation method as claimed in claim 1, characterized in that: The system includes: The data collection module is used to obtain different types of shopping information of several users in the process of purchasing goods on the e-commerce platform within a historical period of time; the types include: platform type and user type; Use the activity analysis module to obtain each user's platform activity index based on each user's platform-based shopping information in a historical period; The commodity attraction analysis module is used to obtain a local shopping feature vector of each user for each commodity based on the difference between the shopping information of the user class of the same commodity purchased by each user and all users; and obtain a commodity attraction index of each commodity for each user based on the similarity between the local shopping feature vectors of each commodity purchased by each user and all commodities, and the shopping information of the user class of each commodity purchased by each user; A behavior interference module is used to obtain a behavior interference index for each user to purchase each product based on the similarity between each product purchased by each user and the platform-type shopping information of other products; The product recommendation module is used to classify the products into different categories according to the platform activity index, the product attraction index and the behavior interference index; and recommend products to users using a recommendation algorithm according to the number of times that products of different categories are purchased by users.
Citation Information
Patent Citations
Commodity recommendation method, device and system, and storage medium
CN113761347A
E-commerce personalized precise shopping guide method based on artificial intelligence
CN116720928A
Online network sales user potential demand analysis monitoring method and system
CN116894692A
Shopping recommendation method and system based on big data and artificial intelligence
CN117495513A
Personalized recommendation method based on user analysis
CN117575747A
