E-commerce product recommendation method and system based on big data analysis

Abstract

The invention provides an e-commerce product recommendation method and system based on big data analysis. The method comprises the following steps: determining to-be-analyzed user data of a target user, and mining and determining category data of various categories for the user data; based on the category data of the multiple categories, performing cross-category indexing from the level of the representation information of the multiple categories to obtain respective multi-category index portrait elements of the category data of the multiple categories; determining a plurality of to-be-determined portrait elements for the user data based on the respective multi-category index portrait elements of the category data of the plurality of categories; and performing portrait element reasoning based on the plurality of undetermined portrait elements and the category data of the plurality of categories to obtain user portrait elements for the user data, and further generating a target user portrait. And finally, realizing personalized product recommendation for the target user according to the user portrait. According to the method, the accuracy of the recommendation system is effectively improved by utilizing a big data analysis technology.

Description

Technical Field

[0001] The present application relates to the technical field of big data analysis, and in particular to an e-commerce product recommendation method and system based on big data analysis. Background Art

[0002] In the field of e-commerce, with the rapid development of big data technology and the increasing abundance of user behavior data, personalized product recommendations have become an important means to improve user experience, enhance user stickiness and promote sales conversion. Traditional e-commerce product recommendation methods often analyze single-dimensional data such as user purchase history, browsing history or search keywords to recommend relevant products to users. However, these methods ignore the diversity and complexity of user behavior and the potential correlation between different categories of data, resulting in limited accuracy and pertinence of recommendation results.

[0003] In order to more fully understand user needs and preferences and improve the accuracy of product recommendations, the industry has begun to explore multi-dimensional, cross-category e-commerce product recommendation methods based on big data analysis. This type of method attempts to build a more detailed and comprehensive user profile by mining and analyzing various user behavior data (such as clicks, browsing, adding to cart, purchases, etc.) and basic attribute data (such as age, gender, region, etc.) on e-commerce platforms. However, how to effectively cross-category index and merge multiple categories of data, and how to accurately infer user profile elements based on these data to generate target user profiles, are still challenges facing current technology. Summary of the invention

[0004] The purpose of the present invention is to provide an e-commerce product recommendation method and device based on big data analysis. The embodiment of the present application is implemented as follows: In a first aspect, an embodiment of the present invention provides an e-commerce product recommendation method based on big data analysis, the method comprising: Determine user data to be analyzed of a target user, and determine multiple categories of category data by mining the user data; Based on the multiple categories of category data, cross-category indexing is performed on each of the multiple categories of representation information to obtain a multi-category index portrait element of each of the multiple categories of category data; Determining a plurality of pending portrait elements for the user data based on the respective multi-category index portrait elements of the plurality of categories of category data; Performing portrait element reasoning based on the multiple undetermined portrait elements and the multiple categories of category data to obtain a user portrait element for the user data, and generating a target user portrait of the target user based on the user portrait element; Recommend products to the target user based on the target user portrait.

[0005] On the other hand, an embodiment of the present invention provides an e-commerce product recommendation system based on big data analysis, including a memory and a processor, wherein the memory stores a computer program that can be executed on the processor, and the processor implements the steps in the above method when executing the program.

[0006] The present invention improves the accuracy and personalization level of e-commerce product recommendations through a method based on big data analysis. First, through deep mining of the user data of the target user, multiple categories of category data are determined, and cross-category indexing is performed on the representation information level of these category data, so as to obtain richer and more accurate multi-category index portrait elements, which helps to understand the user's interests and preferences more comprehensively; then, multiple pending portrait elements are determined based on these index portrait elements, and specific user portrait elements are obtained through a portrait element reasoning process combined with the category data, and finally a target user portrait is generated. This process can effectively improve the accuracy of user portrait construction, thereby enabling the e-commerce platform to provide personalized product recommendation services for the target user based on a more accurate user portrait, which not only improves the user's shopping experience, but also improves the efficiency and accuracy of the recommendation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for describing the embodiments of the present invention are briefly introduced below.

[0008] Figure 1 A schematic diagram of the implementation process of an e-commerce product recommendation method based on big data analysis provided in an embodiment of the present invention.

[0009] Figure 2 It is a hardware entity schematic diagram of an e-commerce product recommendation system based on big data analysis provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0010] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention are further elaborated in detail below in conjunction with the drawings and embodiments. The described embodiments should not be regarded as limiting the present invention. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present invention.

[0011] Figure 1 A schematic diagram of an implementation flow of an e-commerce product recommendation method based on big data analysis provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown, the method includes: Step 100: determine the user data to be analyzed of the target user, and mine the user data to determine multiple categories of category data.

[0012] The recommendation system obtains the user's basic attribute data through the user registration information, such as the name, age, gender, occupation, etc. filled in by the user when registering on the e-commerce platform. The recommendation system also obtains behavior monitoring data by monitoring the user's behavior on the e-commerce platform. For example, when a user browses products on the platform, the recommendation system records the user's clicks, favorites, shopping carts, purchases, etc.

[0013] In order to analyze user data more accurately, the collected raw data can be preprocessed. The preprocessing steps include data cleaning, data transformation, and data normalization. Data cleaning refers to removing noise, missing values, and outliers from the data to improve the quality of the data. The recommendation system uses statistical methods or machine learning algorithms to detect and process missing values ​​and outliers. Data transformation refers to converting raw data into a format suitable for analysis. The recommendation system converts text data into numerical vectors for numerical calculations and analysis. Data normalization refers to scaling the data to a specific range to eliminate the dimensionality of the data. The recommendation system uses methods such as min-max normalization or z-score normalization to normalize the data.

[0014] After completing data preprocessing, the recommendation system mines the user data to determine the category data of multiple categories. This is achieved through data mining algorithms, such as cluster analysis, association rule mining, and classification algorithms. Cluster analysis refers to the division of user data into different categories or clusters, and the users in each cluster have similar characteristics and behaviors. The recommendation system uses the K-means clustering algorithm to cluster users according to characteristics such as age, gender, and consumption habits to obtain different user groups. Association rule mining refers to the discovery of associations between different variables in the data. The recommendation system uses the Apriori algorithm to discover the association rules between users' purchases of goods, such as users who buy mobile phones often also buy mobile phone cases. Classification algorithms refer to the classification of user data into different categories. The recommendation system uses the decision tree algorithm to classify users into high-value users, medium-value users, and low-value users based on their basic attribute data and behavior monitoring data.

[0015] The recommendation system adopts different strategies for mining basic attribute data and behavior monitoring data. For basic attribute data, the recommendation system constructs attribute feature vectors, taking each attribute as a dimension of the vector. For example, age is represented by a numerical value, and occupation is represented by a code. By analyzing these attribute feature vectors, the relationship and pattern between different attributes are discovered. For behavior monitoring data, the recommendation system constructs behavior sequences to record the user's behavior at different time points. For example, the user's click behavior is represented as a time series, and by analyzing the behavior sequence, the user's behavior patterns and trends are discovered.

[0016] Step 200, based on the category data of the multiple categories, cross-category indexing is performed on each of the representation information of the multiple categories to obtain multi-category index portrait elements of each of the category data of the multiple categories.

[0017] Multiple categories of representation information refer to information that can represent the characteristics of each category of data. It is certain attributes of the data, feature vectors, or processed abstract representations. For basic attribute data of users, the representation information is age range, occupation type, etc.; for behavior monitoring data, the representation information is click frequency, collection preference, etc. Cross-category indexing refers to associating and matching between representation information of different categories to discover potential connections and patterns.

[0018] Before cross-category indexing, the recommendation system extracts and transforms the features of multiple categories of category data into representation information suitable for indexing. The recommendation system uses natural language processing technology to process the user's text description information, extract keywords and topics, and uses machine learning algorithms to analyze the user's behavior data to obtain behavior feature vectors. These representation information will serve as the basis for cross-category indexing.

[0019] The recommendation system associates and matches multiple categories of representation information through cross-category indexing. This is achieved through a variety of technical means, such as similarity calculation, association rule mining, etc. Similarity calculation refers to calculating the similarity between the representation information of different categories of data. Optional similarity measurement methods include Euclidean distance, cosine similarity, etc. The calculation formula of Euclidean distance is: , where x and y are two information vectors, and n is the dimension of the vector. The formula for calculating cosine similarity is: ,in is the dot product of the vectors, and are the moduli of the vectors respectively. Based on the results of similarity calculation, the recommendation system finds the representation information of other categories of data that is most similar to the representation information of the current category data, thereby realizing cross-category indexing. Association rule mining is to discover potential rules and patterns by mining the associations between different categories of data. The recommendation system uses association rule mining algorithms such as the Apriori algorithm to find frequent item sets and association rules between different categories of data, thereby realizing cross-category indexing.

[0020] Through cross-category indexing, the recommendation system obtains multi-category index portrait elements for each category data of multiple categories. Multi-category index portrait elements refer to elements obtained through cross-category indexing that can reflect the characteristics of users at multiple category data levels. These elements are the user's interest preferences, consumption habits, behavior patterns, etc. A young working woman, whose basic attribute data characterizes her age between 25-30 years old and her occupation is a company white-collar worker, and whose behavior monitoring data characterizes her frequent clicks on fashion and beauty products and the purchase of high-end skin care products. Through cross-category indexing, the recommendation system obtains multi-category index portrait elements of the user, such as "a young working woman with a high interest in fashion and beauty and a certain consumption capacity."

[0021] Step 300: Determine a plurality of pending portrait elements for the user data based on the respective multi-category index portrait elements of the plurality of categories of category data.

[0022] Multi-category indexing portrait elements are obtained by cross-category indexing the representation information of multiple categories of category data. These elements reflect the characteristics of users at different levels of category data. However, there are a large number of these elements and there is a certain redundancy and uncertainty. Therefore, the recommendation system needs to screen and determine them to obtain more targeted and representative pending portrait elements.

[0023] When determining multiple pending portrait elements, the recommendation system conducts a comprehensive analysis of the multi-category index portrait elements of each of the category data of multiple categories. This involves the evaluation of the importance, relevance and reliability of the elements. Importance evaluation refers to the judgment of the importance of a certain portrait element in describing the user's characteristics and preferences. For a user of an e-commerce platform, "preference for buying high-end electronic products" is more important than the portrait element of "occasionally browsing household goods" because the former can better reflect the user's main consumption tendency. Relevance evaluation refers to the judgment of the degree of association between the portrait element and the user's data. If the user's behavior monitoring data is mainly concentrated on clothing products, then the portrait elements related to clothing have a high relevance. Reliability evaluation refers to the judgment of the credibility and stability of the portrait element. Some portrait elements obtained based on a small amount of data or accidental behavior have low reliability and need to be screened and excluded. One method is based on statistical analysis, which evaluates the importance and reliability of the portrait element by calculating statistical indicators such as the frequency and distribution of the portrait element. The number of occurrences of each portrait element in the multi-category index portrait element is counted, and the elements with more occurrences are more important and reliable. Another method is based on machine learning methods, which use classification algorithms, clustering algorithms, etc. to classify and cluster portrait elements, so as to discover the relationship and pattern between different portrait elements. The recommendation system uses the K-means clustering algorithm to divide the profile elements into different clusters. The elements in each cluster have similar characteristics, and then evaluates the importance and relevance of the elements in each cluster based on the size and characteristics of the cluster.

[0024] The recommendation system also considers the correlation and complementarity between portrait elements. Some portrait elements are not important individually, but combined with other elements, they provide more comprehensive and accurate user information. The two portrait elements "like outdoor sports" and "buy sports equipment" only reflect some of the user's characteristics individually, but combined together, they more clearly depict the user's sports preferences and consumption behavior. The recommendation system mines the association rules between portrait elements and discovers these complementary element combinations, thereby better determining the pending portrait elements.

[0025] Step 400, performing portrait element inference based on the multiple pending portrait elements and the multiple categories of category data to obtain a user portrait element for the user data, and generating a target user portrait of the target user based on the user portrait element; The multiple pending portrait elements are determined in step 300 based on the respective multi-category index portrait elements of the multiple categories of category data. They represent the characteristic elements related to the user that are initially extracted from different perspectives. The multiple categories of category data contain original information such as the user's basic attribute data and behavior monitoring data. The recommendation system combines the two and deeply explores the user's potential characteristics and needs through portrait element reasoning.

[0026] Image element reasoning uses algorithms and models to discover hidden patterns and relationships from pending image elements and category data. Recommendation systems use machine learning algorithms, such as decision trees and neural networks, to analyze and model these data. The decision tree algorithm divides and classifies the data to construct a decision tree, and infers the user's image elements based on different feature conditions. Neural networks simulate the neuron structure of the human brain to perform nonlinear mapping and learning on the data, thereby mining complex patterns and relationships in the data.

[0027] When inferring portrait elements, the recommendation system first needs to perform feature fusion on multiple pending portrait elements and multiple categories of category data. Feature fusion refers to the integration of data features from different sources and types to form a more comprehensive and representative feature set. For basic attribute data and pending portrait elements, the recommendation system concatenates or weightedly combines their feature vectors to obtain a new fused feature vector. Assuming that the pending portrait elements include "like outdoor sports" and the basic attribute data include "age 30" and "male", the recommendation system converts this information into feature vectors and then fuses them to obtain a new vector containing more information.

[0028] Next, the recommendation system uses the fused features to infer the profile elements. Taking the decision tree algorithm as an example, the recommendation system makes judgments and selections on the nodes of the decision tree based on the feature values ​​in the fused features, and finally reaches the leaf node. The label corresponding to the leaf node is the user profile element obtained by inference. Assume that a branch of the decision tree is divided according to "age" and "whether to like outdoor sports". When the age in the fused feature is 30 years old and likes outdoor sports, it is inferred along the corresponding branch that the user is an "active outdoor enthusiast" profile element.

[0029] For the neural network algorithm, the recommendation system takes the fusion features as input, passes through the forward propagation process of the neural network, and undergoes nonlinear transformations of multiple hidden layers, and finally outputs the inferred user profile elements. The training process of the neural network is to continuously adjust the weights and biases of the network through a large amount of historical data, so that the output of the network is as close to the real user profile elements as possible.

[0030] Through the reasoning of portrait elements, the recommendation system obtains user portrait elements for user data. These user portrait elements are more accurate and detailed descriptions of user characteristics and needs. They comprehensively consider the pending portrait elements and multiple categories of category data. User portrait elements include "fashion lovers with high spending power" and "health-conscious housewives", which can more comprehensively reflect users' consumption preferences, lifestyles, and values.

[0031] After obtaining the user profile elements, the recommendation system generates a target user profile of the target user based on these elements. The target user profile is an overall description of the target user. It integrates and organizes multiple user profile elements to form a structured and intuitive user feature model. The recommendation system uses a graphical method to display the target user profile, such as making a user profile card that contains the user's basic information, interests, hobbies, consumption habits and other key features.

[0032] Step 500: recommend products to the target user based on the target user portrait.

[0033] The target user portrait is a comprehensive description of the target user's characteristics, needs, interests, and consumption habits. Based on this portrait, the recommendation system selects the products that are most attractive to the target user from the massive products on the e-commerce platform. To achieve this goal, the recommendation system uses recommendation algorithms and strategies to match and associate the characteristics in the target user portrait with the attributes of the product.

[0034] The recommendation system uses a content-based recommendation algorithm. This algorithm mainly matches the content characteristics of the product with the interest preferences in the user portrait. If the target user portrait shows that the user is a "young technology enthusiast with a high interest and willingness to consume smartphones and digital accessories", the recommendation system will select products that meet this feature from the product library of the e-commerce platform, such as new smartphones, high-performance tablets, fashionable digital headphones, etc. The recommendation system will analyze the attributes of these products, such as brand, model, performance parameters, price, etc., and then compare them with the needs in the user portrait, and select the most matching products for recommendation. The content-based recommendation algorithm uses similarity calculation to measure the degree of match between the product and the user portrait. The optional similarity measurement methods include cosine similarity. The recommendation system will sort the products according to the cosine similarity value. The higher the similarity value, the more likely the product will be recommended to the user.

[0035] The recommendation system also uses collaborative filtering recommendation algorithms. Collaborative filtering algorithms are divided into user-based collaborative filtering and item-based collaborative filtering. User-based collaborative filtering algorithms find other users with similar interests and behaviors as the target user, and then recommend products that these similar users like to the target user. Assuming that the target user's profile shows that the user likes outdoor sports equipment, the recommendation system will look for other users who also like outdoor sports equipment in the user database, analyze the products recently purchased or collected by these similar users, such as professional hiking shoes, outdoor tents, etc., and then recommend these products to the target user. Item-based collaborative filtering algorithms make recommendations based on the similarity between products. If the target user buys a sports watch, the recommendation system will analyze other products similar to this sports watch, such as other watches with the same functions or brands, and then recommend these similar products to the user.

[0036] In addition to the two optional recommendation algorithms mentioned above, the recommendation system also combines multiple algorithms for hybrid recommendation to improve the accuracy and diversity of recommendations. The recommendation system first uses a content-based recommendation algorithm to filter out a set of products that match the target user profile, and then uses a collaborative filtering algorithm to further filter and sort this set to obtain more accurate recommendation results.

[0037] When making product recommendations, the recommendation system also considers the real-time information and dynamic changes of products. Information such as product prices, inventory, and promotions on e-commerce platforms are constantly updated. The recommendation system obtains this information in a timely manner and takes it into consideration when making recommendation decisions. If a product is on a limited-time discount, the recommendation system will prioritize recommending it to target users to increase their willingness to buy.

[0038] The recommendation system also needs to display personalized recommendations based on the different scenarios and usage habits of the target users. For users browsing the e-commerce platform on a computer, the recommendation system displays recommended products in a list format, providing detailed product information and pictures; for users browsing on a mobile device, the recommendation system displays recommended products in a card-like or waterfall-like format to adapt to the screen characteristics of the mobile device and user operation habits.

[0039] As an implementation manner, the category data based on the multiple categories are each cross-category indexed at the level of the representation information of the multiple categories to obtain the multi-category index portrait elements of the category data of the multiple categories, including: Step 210, determining a portrait element mapping relationship set; the portrait element mapping relationship set includes comparison category representation information of the comparison user data under the multiple categories and comparison portrait elements corresponding to the comparison category representation information; Step 220, based on the commonality measurement results between the category data of the multiple categories and the control category representation information of the control user data under the multiple categories, index in the portrait element mapping relationship set to obtain the multi-category index portrait elements of each of the category data of the multiple categories.

[0040] In step 210, the recommendation system first needs to determine the portrait element mapping relationship set, which is the basis for subsequent cross-category indexing. The portrait element mapping relationship set is a set that stores the corresponding relationship between the control user data and the control portrait elements, where the control category representation information of the control user data under multiple categories is regarded as a known feature pattern, and the control portrait element is the user feature description corresponding to these feature patterns. In order to construct this mapping relationship set, the recommendation system collects a large amount of historical user data and classifies and annotates these data. The recommendation system extracts the basic attribute data and behavior monitoring data of many users from the database of the e-commerce platform, divides these data into different categories, such as age, occupation, consumption habits, etc., and then analyzes the data features under each category, groups data with similar features, and annotates the corresponding control portrait elements for each group of data. For example, for a user group between the ages of 20 and 30, whose occupation is a student or a beginner, and who often buys fashion trend products, the recommendation system annotates the corresponding control portrait element as "young fashion consumer group".

[0041] The recommendation system needs to use data processing and analysis technology when constructing a portrait element mapping relationship set. The recommendation system uses a data mining algorithm to perform cluster analysis on historical user data, clustering similar user data into different clusters, each cluster representing a typical user feature pattern. The recommendation system uses the K-means clustering algorithm, which iteratively distributes data points to different clusters, so that the similarity of data points within a cluster is the highest and the similarity of data points between clusters is the lowest. After clustering is completed, the recommendation system labels each cluster with the corresponding comparison portrait element, thereby forming a portrait element mapping relationship set.

[0042] In step 220, commonality measurement refers to the process of measuring the similarity between two types of data. The recommendation system calculates the similarity between category data of multiple categories and the control category representation information, finds the control category representation information that is most similar to the current category data, and then determines the corresponding control portrait element as the multi-category index portrait element. Optional methods include Euclidean distance, cosine similarity, etc.

[0043] After measuring the commonality, the recommendation system obtains the similarity values ​​between the category data of multiple categories and the control category representation information. Based on these similarity values, the recommendation system indexes in the portrait element mapping relationship set, finds the control category representation information with the highest similarity to the current category data, and then uses the control portrait element corresponding to the control category representation information as the multi-category index portrait element of the current category data. For example, the recommendation system analyzes the basic attribute data and behavior monitoring data of a user, and through calculating the similarity, finds that the user's data has the highest similarity with the control category representation information corresponding to the "young fashion consumer group" in the portrait element mapping relationship set. In this case, the recommendation system will use "young fashion consumer group" as the multi-category index portrait element of the user's data.

[0044] As an implementation manner, the commonality measurement results between the category data based on the multiple categories and the comparison category representation information of the comparison user data under the multiple categories are indexed in the portrait element mapping relationship set to obtain the multi-category index portrait elements of the category data of the multiple categories, including: Step 221, performing commonality measurement on each category of the multiple categories of category data and the comparison category representation information of the comparison user data under the multiple categories, to obtain a commonality measurement result for each category of data; Step 222, from the common measurement results of each category of data, determine that each category of data represents a target measurement result that meets the common requirements; Step 223, based on the control category representation information corresponding to the target measurement result of each category of data, index respectively in the image element mapping relationship set to obtain the multi-category index image element of each category of data.

[0045] In step 221, multiple categories of category data have been determined in the previous steps, such as the user's basic attribute data and behavior monitoring data, and the control category representation information is stored in the portrait element mapping relationship set, which is the feature representation of the control user data under multiple categories. Optional methods include Manhattan distance. Manhattan distance refers to the sum of the distances between two points along the coordinate axis in a multidimensional space, and its calculation formula is: By measuring the commonality of each category of data, the commonality measurement results reflect the similarity between each category of data and the representation information of each control category.

[0046] In step 222, since in step 221, each category of data is measured for commonality with multiple reference category representation information, multiple similarity values ​​are obtained, and the recommendation system selects the similarity value that meets the commonality requirements as the target measurement result. The commonality requirements are set according to specific business needs and data characteristics, and a similarity threshold is set. Only measurement results with similarity values ​​greater than the threshold are considered to meet the commonality requirements.

[0047] In this way, the recommendation system selects the measurement results corresponding to the comparison category representation information that is most similar to the current category data from a large number of common measurement results, providing an accurate basis for subsequent indexing operations.

[0048] In step 223, the correspondence between the control category representation information and the control portrait elements is stored in the portrait element mapping relationship set. The recommendation system searches for the corresponding control portrait elements in this relationship set according to the control category representation information corresponding to the target measurement result. These control portrait elements are the multi-category index portrait elements of each category data.

[0049] As an implementation mode, the multiple categories include monitoring data categories and attribute data categories; the multiple categories of category data include behavior monitoring data and basic attribute data of the user data; the portrait element mapping relationship set includes behavior comparison category representation information of the comparison user data under the monitoring data category and behavior comparison portrait elements corresponding to the behavior comparison category representation information, and attribute comparison category representation information of the comparison user data under the attribute data category and attribute comparison portrait elements corresponding to the attribute comparison category representation information; In step 220, the commonality measurement results between the category data based on the multiple categories and the comparison category representation information of the comparison user data under the multiple categories are indexed in the portrait element mapping relationship set to obtain the multi-category index portrait elements of the category data of the multiple categories, including: Step 10, extracting representation information based on the behavior monitoring data and the basic attribute data to obtain behavior data representation and attribute data representation; Step 20, performing behavior representation commonality measurement on the behavior data representation, the behavior comparison category representation information and the attribute comparison category representation information respectively, and obtaining a multi-category index portrait element of the behavior monitoring data in the behavior comparison portrait element and the attribute comparison portrait element based on the behavior representation commonality measurement result; Step 30, perform attribute representation commonality measurement on the attribute data representation, the behavior comparison category representation information and the attribute comparison category representation information respectively, and obtain a multi-category index portrait element of the basic attribute data in the behavior comparison portrait element and the attribute comparison portrait element based on the attribute representation commonality measurement result.

[0050] In step 10, behavior monitoring data refers to the records of various user operations on the e-commerce platform, such as clicking, collecting, and purchasing products; basic attribute data refers to some inherent characteristic information of users, such as age, gender, occupation, etc. Representation information extraction is the process of converting raw data into more representative and computational feature representations. These feature representations can reflect the core characteristics and internal laws of the data, and provide a basis for subsequent commonality measurement and portrait element indexing.

[0051] For extracting information representing behavior monitoring data, an optional method used by the recommendation system is to convert the behavior data into a vector representation. The recommendation system counts the number of clicks, favorites, and purchases of different categories of products by users within a certain period of time, and combines these statistical values ​​into a vector. Assuming that a user clicks on mobile phone products 5 times, computer products 3 times, and home appliance products 2 times in a week, the behavior data representation is represented as a vector (5, 3, 2). The recommendation system also considers factors such as the time sequence and frequency distribution of the behavior to perform more complex feature extraction on the behavior data. For example, using time series analysis methods, the user's behavior is arranged in chronological order to extract characteristics such as the periodicity and trend of the behavior.

[0052] For the extraction of representation information of basic attribute data, the recommendation system encodes and quantifies the attribute data. For age attributes, it is divided into different age groups, such as 18-25 years old, 26-35 years old, etc., and then each age group is represented by the corresponding code; for occupation attributes, an occupation classification system is established to encode different occupations. Assuming that the user is 28 years old and the occupation is a programmer, after encoding, the attribute data representation is represented as a vector (2, 5), where 2 represents the code of the 26-35 age group and 5 represents the code of the programmer occupation.

[0053] In step 20, the behavior comparison category representation information is the comparison feature representation related to the behavior data in the portrait element mapping relationship set, and the attribute comparison category representation information is the comparison feature representation related to the attribute data. The behavior representation commonality measurement is the process of measuring the similarity between the behavior data representation and the comparison category representation information. By calculating the similarity, the recommendation system finds the comparison category representation information that is most similar to the behavior data representation, and then determines the corresponding comparison portrait element.

[0054] In step 30, the process of measuring the commonality of attribute representation is similar to that of measuring the commonality of behavior representation, and similarity is calculated to find the comparison category representation information that is most similar to the attribute data representation, and then determine the corresponding comparison portrait element.

[0055] As an implementation mode, the behavior data representation is respectively measured with the behavior comparison category representation information and the attribute comparison category representation information, and based on the behavior representation commonality measurement result, a multi-category index portrait element of the behavior monitoring data is obtained in the behavior comparison portrait element and the attribute comparison portrait element, including: Step 21, performing behavior representation commonality measurement on the behavior data representation and the behavior comparison category representation information to obtain a first behavior representation commonality measurement result; Step 22, based on the behavior comparison category representation information corresponding to the first behavior representation commonality measurement result representing the satisfaction of the commonality requirement, determining a first behavior category index portrait element in the behavior comparison portrait element; Step 23, performing behavior representation commonality measurement on the behavior data representation and the attribute comparison category representation information to obtain a second behavior representation commonality measurement result; Step 24, based on the attribute comparison category representation information corresponding to the second behavior representation commonality measurement result representing the satisfaction of the commonality requirement, determining the first attribute category index portrait element in the attribute comparison portrait element; Step 25, based on the first behavior category index portrait element and the first attribute category index portrait element, obtain a multi-category index portrait element of the behavior monitoring data.

[0056] In step 21, the behavior data representation is the result obtained after the representation information of the user behavior monitoring data is extracted, which reflects the user's behavior characteristics in a form that is easy for computer processing and analysis. The behavior comparison category representation information is the behavior comparison category representation information of the user data under the monitoring data category in the portrait element mapping relationship set, which represents a set of representative behavior patterns. The behavior representation commonality measurement is an operation to measure the similarity between the behavior data representation and the behavior comparison category representation information, and through this measurement, the comparison behavior pattern that is most similar to the user's current behavior is found.

[0057] In step 22, satisfying the commonality requirement means that the similarity score reaches a certain threshold, which is set according to business requirements and data characteristics. The behavior comparison portrait elements are portrait elements corresponding to the behavior comparison category representation information in the portrait element mapping relationship set, which describe the characteristics and preferences of users with corresponding behavior patterns.

[0058] In step 23, the attribute comparison category representation information is the attribute comparison category representation information of the user data under the attribute data category in the portrait element mapping relationship set, which represents a set of representative user attribute features. Although the behavioral data representation mainly reflects the user's behavioral information, the potential association between user behavior and attributes is discovered by measuring the commonality with the attribute comparison category representation information. If the data distribution of the attribute comparison category representation information is relatively discrete, a measurement method that is more suitable for processing discrete data is selected, such as the Hamming distance. The Hamming distance is used to measure the number of different characters in corresponding positions between two equal-length strings, and is more effective when processing attribute comparison category representation information in the form of binary vectors.

[0059] In step 24, similar to step 22, satisfying the commonality requirement means that the similarity score reaches a preset threshold. The attribute comparison portrait elements are portrait elements corresponding to the attribute comparison category representation information in the portrait element mapping relationship set, and they describe the characteristics of users with corresponding attribute features.

[0060] In step 25, the index portrait elements obtained from the two perspectives of behavior and attribute are integrated to form a more comprehensive and accurate multi-category index portrait element about the behavior monitoring data.

[0061] When implementing step 25, the recommendation system uses a splicing method to integrate the first behavior category index portrait element and the first attribute category index portrait element. However, in actual applications, the integration method needs to be optimized according to business needs. For example, different weights are assigned according to the importance of different portrait elements, and then weighted combinations are performed.

[0062] Steps 21-25 measure the commonality of the behavior data representation with the behavior control category representation information and the attribute control category representation information, determine the corresponding index portrait elements, and finally integrate the multi-category index portrait elements of the behavior monitoring data. The recommendation system can deeply analyze the user's behavior characteristics from multiple angles and explore the potential correlation between user behavior and attributes.

[0063] As an implementation manner, the attribute data representation is respectively measured with the behavior comparison category representation information and the attribute comparison category representation information, and based on the attribute representation commonality measurement result, a multi-category index portrait element of the basic attribute data is obtained in the behavior comparison portrait element and the attribute comparison portrait element, including: Step 31, performing attribute representation commonality measurement on the attribute data representation and the behavior comparison category representation information to obtain a first attribute representation commonality measurement result; Step 32, based on the behavior comparison category representation information corresponding to the first attribute representation commonality measurement result representing the satisfaction of the commonality requirement, determining a second behavior category index portrait element in the behavior comparison portrait element; Step 33, performing attribute representation commonality measurement on the attribute data representation and the attribute comparison category representation information to obtain a second attribute representation commonality measurement result; Step 34, based on the attribute comparison category representation information corresponding to the second attribute representation commonality measurement result representing the satisfaction of the commonality requirement, determine the second attribute category index portrait element in the attribute comparison portrait element; Step 35, based on the second behavior category index portrait element and the second attribute category index portrait element, obtain the multi-category index portrait element of the basic attribute data.

[0064] In step 31, the attribute data representation is a representation obtained after feature extraction and conversion of the user's basic attribute data, which can reflect the core characteristics of the user's basic attributes. The behavior comparison category representation information is a comparison feature representation related to the user's behavior in the portrait element mapping relationship set, which contains typical characteristics under different behavior patterns. The purpose of the attribute representation commonality measurement is to measure the similarity between the attribute data representation and the behavior comparison category representation information, and through this measurement, the potential connection between the user's basic attributes and certain behavior patterns can be discovered.

[0065] In step 32, the commonality requirement is a pre-set similarity threshold. Only when the similarity value in the first attribute representation commonality measurement result is greater than the threshold, the corresponding behavior comparison category representation information and the attribute data representation are considered to have sufficient similarity and meet the commonality requirement.

[0066] In step 33, the attribute comparison category representation information is a comparison feature representation related to the user's basic attributes in the portrait element mapping relationship set, which covers typical features under different attribute combinations. This step is similar to step 31, and also uses a similarity measurement method to measure the similarity between the attribute data representation and the attribute comparison category representation information, except that the comparison information here is related to the attribute.

[0067] In step 34, similar to step 32, a similarity threshold needs to be set as a criterion for determining the commonality requirement.

[0068] In step 35, this step is to integrate the results of the previous steps, and synthesize the features related to the basic attribute data mined from the two perspectives of behavior and attribute, so as to obtain a more comprehensive and accurate multi-category index portrait element of the basic attribute data.

[0069] As another implementation, in step 10, the process of extracting characterization information based on the behavior monitoring data and the basic attribute data to obtain the behavior data characterization and the attribute data characterization is completed by using a multi-category characterization analysis network, and the multi-category characterization analysis network is trained by the following process: Step a, obtaining training binary data, wherein the training binary data includes behavior monitoring training data and basic attribute training data; Step b, extracting representation information from the behavior monitoring training data and the basic attribute training data through the initialized initial multi-category representation analysis network, to obtain the behavior data representation of the behavior monitoring training data and the attribute data representation of the basic attribute training data; Step c: based on the behavior data representation and the attribute data representation, iteratively debug the initial multi-category representation analysis network until a preset debugging stop condition is reached to obtain the multi-category representation analysis network.

[0070] In step a, behavior monitoring training data refers to relevant information collected from a large number of historical user behaviors, covering various user operations on products on e-commerce platforms, such as clicks, favorites, purchases, forwarding, etc. These behavior data can reflect user interests, preferences and consumption habits. Basic attribute training data refers to some inherent characteristic information about users, such as age, gender, occupation, region, etc. These attribute data provide a basis for understanding the basic background and characteristics of users.

[0071] In step b, this process is regarded as the process of feature extraction and transformation of data by the network. The network attempts to mine the information that best represents the data features from the original data. The initial multi-category representation analysis network is a neural network model with a predefined structure. Before training begins, its parameters are randomly initialized.

[0072] In step c, the purpose of iterative debugging is to adjust the parameters of the network so that the behavioral data representation and attribute data representation of the network output can better reflect the characteristics and internal laws of the input data.

[0073] The recommendation system uses a loss function to measure the difference between the representation information output by the network and the expected result. The optional loss function is the mean square error loss function (MSE), which is calculated as , where n is the number of samples, is the expected result, is the output of the network.

[0074] In each iteration, the recommendation system first calculates the loss value between the behavior data representation and attribute data representation of the current network output and the expected result. The expected result is the ideal representation information obtained based on prior knowledge or expert annotation. Then, the recommendation system uses an optimization algorithm to adjust the parameters of the network to reduce the loss value. Optional optimization algorithms include stochastic gradient descent, adaptive moment estimation, etc. The update formula of the stochastic gradient descent algorithm (SGD) is ,in is the network parameter at the tth iteration, is the learning rate, is the loss function with respect to the parameters gradient.

[0075] The recommendation system repeats this iterative process continuously, and each iteration updates the network parameters in the direction of reducing the loss value. When the loss value decreases below the preset threshold, or the number of iterations reaches the preset maximum number, the recommendation system considers that the debugging stop condition has been reached, and the network obtained at this time is the trained multi-category representation analysis network.

[0076] As an implementation mode, the multiple categories include monitoring data categories and attribute data categories; the multiple categories of category data include behavior monitoring data and basic attribute data; The performing image element reasoning based on the multiple undetermined image elements and the multiple categories of category data to obtain a user image element for the user data includes: Step 410, performing feature fusion on the basic attribute data and the multiple undetermined portrait elements to obtain first fused representation information; and performing feature fusion on the behavior monitoring data and the multiple undetermined portrait elements to obtain second fused representation information; Step 420: Perform portrait element inference based on the first fused representation information and the second fused representation information to obtain a user portrait element for the user data.

[0077] In step 410, feature fusion refers to integrating data features from different sources or types to form a more comprehensive and representative feature set, thereby being able to more accurately reflect the characteristics and needs of users.

[0078] The recommendation system combines the basic attribute data and multiple pending portrait elements for feature fusion, aiming to combine the basic features of the user with the initially refined feature elements, and to mine deeper and more comprehensive user feature information. For the feature fusion of basic attribute data and pending portrait elements, the recommendation system adopts multiple methods. One optional method is to splice the representation information after extracting the representation information separately. The recommendation system converts the age, gender, occupation and other information in the basic attribute data into feature vectors, and converts the pending portrait elements into corresponding feature vectors, and then splices the two vectors together to form a new vector, which is the first fused representation information. Assuming that the user's age in the basic attribute data is 30 years old, the gender is male, and the occupation is engineer, the feature vector converted is (30, 1, 2) (where 1 represents male and 2 represents the engineer occupation code); the pending portrait elements are "technology product enthusiasts" and "high consumption ability", and the feature vector converted is (1, 1) (assuming 1 represents the possession of this feature), then the first fused representation information is the spliced ​​vector (30, 1, 2, 1, 1).

[0079] The recommendation system fuses the behavior monitoring data with multiple pending portrait elements in order to combine the user's behavior characteristics with the initially refined feature elements and further explore the user's interests and needs. The same splicing method is used for the feature fusion of behavior monitoring data and pending portrait elements. The recommendation system converts the number of clicks, favorite product lists, purchase product details and other information in the behavior monitoring data into feature vectors, converts the pending portrait elements into corresponding feature vectors, and then splices the two vectors together to form the second fusion representation information.

[0080] In step 420, the portrait element reasoning requires the recommendation system to use algorithms and models to discover hidden rules and relationships from the fused feature information, so as to infer the user portrait elements that can accurately describe the user's characteristics and needs.

[0081] As an implementation manner, the performing portrait element reasoning based on the first fused representation information and the second fused representation information to obtain a user portrait element for the user data includes: Step 421, performing canonical mapping on the first fused representation information through a cross-modal canonical mapping network to obtain first canonical mapping fused representation information; Step 422, performing canonical mapping on the second fused representation information through the cross-modal canonical mapping network to obtain second canonical mapping fused representation information; Step 423: Performing portrait element inference based on the first canonical mapping fusion representation information and the second canonical mapping fusion representation information through a portrait element inference network to obtain a user portrait element for the user data.

[0082] In step 421, the first fused representation information is obtained by feature fusion of basic attribute data and multiple pending portrait elements in step 410. It integrates the basic features of the user and the initially refined feature elements. However, since data from different sources have different feature distributions and scales, directly using this information for portrait element reasoning will lead to poor model performance. The role of the cross-modal canonical mapping network is to process the first fused representation information to make its feature distribution and scale more uniform, which is convenient for subsequent network processing.

[0083] The cross-modal canonical mapping network is a trained neural network model that learns the mapping relationship between different modal data and converts the input first fusion representation information into a form that is more suitable for image element reasoning network processing. The recommendation system inputs the first fusion representation information into the cross-modal canonical mapping network, and the network transforms the input information according to its internal parameters. This transformation is regarded as a process of normalizing and aligning the data, making different features comparable.

[0084] For example, assume that the first fusion representation information contains the user's age, occupation code, and feature vectors corresponding to the pending portrait elements. The age range is between 18 and 80 years old, the occupation code is an integer from 1 to 100, and the feature vector corresponding to the pending portrait element has a value between 0 and 1. The scales of these different features vary greatly. If they are directly input into the portrait element inference network, the influence of some features will be too large, while the influence of other features will be ignored. The cross-modal canonical mapping network processes these features and maps them into a unified feature space so that each feature can play a reasonable role in the reasoning process.

[0085] The network structure is a multi-layer perceptron (MLP), which includes an input layer, a hidden layer, and an output layer. The number of neurons in the input layer is determined by the dimension of the first fusion representation information, and the number of neurons in the output layer is determined by the input requirements of the subsequent image element inference network. The recommendation system uses batch normalization and other techniques to accelerate network training and improve stability. The formula for batch normalization is ,in is the input data, is the mean of the mini-batch data, is the variance of the mini-batch data, is a small constant used to avoid division by zero errors. After batch normalization, the mean of the data is 0 and the variance is 1, which helps the network converge faster.

[0086] In step 422, the second fused representation information is obtained by feature fusion of the behavior monitoring data and multiple pending portrait elements, which reflects the user's behavior characteristics and the initially refined feature elements. Similar to the first fused representation information, the second fused representation information also has the problem of different feature scales and inconsistent distributions, so it needs to be processed through a cross-modal canonical mapping network.

[0087] The recommendation system inputs the second fused representation information into the trained cross-modal canonical mapping network, which transforms it according to the same mapping rules and converts it into a representation in a unified feature space. For example, the second fused representation information contains the number of clicks, the number of favorite items, and the feature vectors corresponding to the pending portrait elements. The value ranges of the number of clicks and the number of favorite items vary greatly. Through the processing of the cross-modal canonical mapping network, these features will be mapped to a scale, so that they can play a more fair role in the subsequent portrait element reasoning.

[0088] The training of the cross-modal canonical mapping network is based on a large amount of historical data. The recommendation system uses the first training behavior data representation of the first training user data, the first training attribute data representation, and the prior user portrait to debug and obtain the transition cross-modal canonical mapping network. In this process, the network learns the mapping relationship between different modal data and can effectively perform canonical mapping on the input information. For the canonical mapping of the second fusion representation information, the recommendation system directly uses the trained cross-modal canonical mapping network to ensure the consistency and accuracy of the processing.

[0089] In step 423, the profile element inference network is a neural network model specifically used to infer user profile elements from the feature information after canonical mapping. It receives the first canonical mapping fusion representation information and the second canonical mapping fusion representation information as input, and through a series of calculations and transformations, outputs user profile elements that can accurately describe user characteristics and needs.

[0090] When training the portrait element inference network, the recommendation system uses the second training behavior specification mapping representation information, the second training attribute specification mapping representation information, and the prior portrait elements of the second training user data. Prior portrait elements are known portrait elements that accurately describe user characteristics and are used as training labels. The recommendation system inputs the second training behavior specification mapping representation information and the second training attribute specification mapping representation information into the initialized initial portrait element inference network, and the network outputs the inferred portrait elements. Then, the recommendation system optimizes the transition cross-modal specification mapping network and the initial portrait element inference network based on the inferred portrait elements and the prior portrait elements, and then performs collaborative debugging again until the preset debugging stop condition is reached, and the converged cross-modal specification mapping network and portrait element inference network are obtained.

[0091] When the portrait element inference network training is completed, the recommendation system inputs the first canonical mapping fusion representation information and the second canonical mapping fusion representation information of the new user into the trained network, and the network outputs the user portrait element for the user data.

[0092] As another implementation, the cross-modal specification mapping network and the image element reasoning network are trained by the following steps: Step S10, based on the first training behavior data representation of the first training user data, the first training attribute data representation and the prior user portrait, debugging to obtain a transition cross-modal norm mapping network, wherein the first training behavior data representation is obtained by mining the representation information of the behavior monitoring data of the first training user data; the first training attribute data representation is obtained by mining the representation information of the basic attribute data of the first training user data; The first training user data is a part of the data selected by the recommendation system from a large amount of user data for training. It contains the user's behavior monitoring data and basic attribute data. The first training behavior data representation is obtained by mining the representation information of the behavior monitoring data in the first training user data. This process involves converting the user's behavior data into a more representative and computational feature vector.

[0093] The first training attribute data representation is obtained by mining the representation information of the basic attribute data of the first training user data. The basic attribute data includes information such as the user's age, gender, occupation, and region. The recommendation system encodes and quantifies these attribute data and converts them into feature vectors. For the age attribute, it is divided into different age groups, such as 18-25 years old, 26-35 years old, etc., and then each age group is represented by the corresponding code; for the occupation attribute, an occupation classification system is established to encode different occupations. The prior user portrait is a pre-known, accurate description of the user characteristics corresponding to the first training user data, which is based on historical data, expert experience or other reliable sources. The prior user portrait contains information about the user's interests, hobbies, consumption habits, values, etc.

[0094] The transition cross-modal canonical mapping network is a network that has not been fully trained. Its function is to perform canonical mapping on data of different modalities so that these data have better alignment and comparability in the feature space, which is convenient for subsequent portrait element reasoning. When debugging the transition cross-modal canonical mapping network, the recommendation system continuously adjusts the parameters of the network so that the network's mapping results for the first training behavior data representation and the first training attribute data representation can be closer to the characteristics reflected by the prior user portrait.

[0095] When debugging the transition cross-modal canonical mapping network, the recommendation system inputs the first training behavior data representation and the first training attribute data representation into the initialized initial cross-modal canonical mapping network. The network maps the input data to obtain the mapped representation information. Then, the recommendation system uses an initialized initial portrait element inference network to predict the user portrait of the mapped representation information and obtain the predicted user portrait for the first training user data. There is a certain difference between the predicted user portrait and the prior user portrait, and the recommendation system uses a loss function to measure this difference.

[0096] The recommendation system uses an optimization algorithm (such as stochastic gradient descent or adaptive moment estimation) to adjust the parameters of the initial cross-modal norm mapping network according to the value of the loss function, so that the value of the loss function continues to decrease.

[0097] Step S20, obtaining a second training behavior data representation, a second training attribute data representation, and a priori portrait elements of the second training user data; the second training behavior data representation is obtained by mining the representation information of the behavior monitoring data of the second training user data; the second training attribute data representation is obtained by mining the representation information of the basic attribute data of the second training user data; To obtain the second training behavior data representation, the recommendation system uses natural language processing technology to process and analyze the user's behavior monitoring data. For example, if the behavior monitoring data is a user's comment on a product, the recommendation system uses word embedding technology to convert each word in the comment into a vector, and then processes these vectors through a deep learning model (such as a long short-term memory network LSTM) to obtain a representation of the entire comment. In addition, the recommendation system also uses feature engineering methods to extract some statistical features from the behavior monitoring data, such as the number of browsed products, the frequency of clicking on products, etc., and then combines these features into a vector as the second training behavior data representation. Assume that the user's behavior monitoring data is a list containing multiple product browsing records, each record includes information such as the name of the product and the browsing time. The recommendation system calculates the number of products browsed by users in different time periods, as well as the browsing ratio of different types of products, and uses these statistical information as features to combine them into a vector. Its calculation formula is expressed as: ,in represents the value of the i-th feature, for example is the number of items viewed in the past day. The browsing ratio of electronic products, etc.

[0098] To obtain the second training attribute data representation, the recommendation system uses one-hot encoding or embedding technology to process the basic attribute data. One-hot encoding converts each attribute value into a binary vector, in which only one element is 1 and the rest are 0, representing the value of the attribute. For example, for the user's gender attribute, male is represented as [1, 0] and female is represented as [0, 1]. Embedding technology maps the attribute value to a low-dimensional vector space so that vectors with similar attribute values ​​are close in space. For example, for the user's age attribute, an embedding layer is used to map the age value to a 10-dimensional vector space. The recommendation system combines these processed attribute vectors to obtain the second training attribute data representation. Assuming that the user's basic attribute data includes age, gender, and occupation, after one-hot encoding and embedding processing, vectors A, G, and O are obtained respectively. Then the second training attribute data representation P is expressed as P = [A; G; O], where ";" represents the concatenation operation of the vector.

[0099] To obtain prior portrait elements, the recommendation system refers to historical data or expert knowledge. Historical data refers to user data that has been collected and processed by the recommendation system in the past. By analyzing and mining these data, some user behavior patterns and preference rules are discovered. For example, by analyzing the shopping data of a large number of young male programmers, it is found that they generally have a high demand for electronic products, especially high-performance computers and mobile phones. Expert knowledge comes from the experience and judgment of professionals in the field. They predict and judge the user's portrait elements based on their professional knowledge and experience. The recommendation system combines historical data and expert knowledge to determine the prior portrait elements. For example, the recommendation system uses Bayes' theorem to integrate historical data and expert knowledge. Assuming that H represents a certain portrait element of the user (such as interest in mobile phones), D represents historical data, and E represents expert knowledge, then according to Bayes' theorem, the probability of the prior portrait element is expressed as: , where P(H) is the prior probability, P(D, E|H) is the likelihood probability, and P(D,E) is the evidence probability. The recommendation system determines the prior profile elements by calculating this probability.

[0100] Step S30, performing canonical mapping on the second training behavior data representation through the transitional cross-modal canonical mapping network to obtain second training behavior canonical mapping representation information; and performing canonical mapping on the second training attribute data representation through the transitional cross-modal canonical mapping network to obtain second training attribute canonical mapping representation information; When implementing step S30, the recommendation system uses a fully connected layer to linearly transform the input representation information, and then introduces nonlinear factors through the activation function to map the input data to a new feature space. Assuming that the second training behavior data is represented as a vector, the second training attribute data is represented as a vector, and the fully connected layer in the transition cross-modal norm mapping network is represented as Y = WX + b, where W is the weight matrix, b is the bias vector, and X is or , Y is the output vector after the normalized mapping. The activation function is selected as ReLU (Rectified Linear Unit) function, and its formula is , by processing the output of the fully connected layer through the ReLU function, the network's expressive power is enhanced. Another technical approach is to perform batch normalization on the input representation information during the canonical mapping process, so that the data has a similar distribution at each layer of the input.

[0101] When performing normative mapping, the recommendation system also needs to consider the characteristics of different modal data. For the second training behavior data representation, since it reflects the user's behavior information, it has high dynamics and diversity. For example, the user's behavior will change with time, promotional activities and other factors. Therefore, in the normative mapping process, the time factor is introduced, and the behavior data representation is processed using the time series analysis method. The behavior data representation is regarded as a time series, and it is modeled using a recurrent neural network (RNN) or its variants (such as LSTM, GRU) to capture the time dependency in the behavior data. When processing the second training attribute data representation, since the attribute data is relatively stable, such as the user's age, gender and other information will not change in the short term, the focus is on the feature selection and dimensionality compression of the data. Use methods such as principal component analysis (PCA) to reduce the dimension of the attribute data representation, remove redundant information in the data, and improve the computational efficiency of the model.

[0102] In practical applications, recommendation systems may encounter problems such as excessive data volume and limited computing resources. To solve these problems, recommendation systems use distributed computing technology. The computing tasks of the transition cross-modal norm mapping network are assigned to multiple computing nodes for parallel processing, thereby increasing the computing speed. The recommendation system also uses model compression technology, such as pruning and quantization, to compress the transition cross-modal norm mapping network, reduce the number of model parameters, and reduce computational complexity.

[0103] Step S40, performing image element reasoning based on the second training behavior specification mapping representation information and the second training attribute specification mapping representation information through the initialized initial image element reasoning network to obtain an inferred image element for the second training user data; When implementing step S40, an optional technical means adopted by the recommendation system is to use a multilayer perceptron. The multilayer perceptron is a feedforward neural network, which is composed of an input layer, a hidden layer, and an output layer. The recommendation system concatenates the second training behavior specification mapping representation information and the second training attribute specification mapping representation information into an input vector, and inputs it into the input layer of the multilayer perceptron. Assume that the second training behavior specification mapping representation information is a vector , dimension is m, the second training attribute specification mapping representation information is a vector , the dimension is n, then the input vector X = [ ; ], with a dimension of m + n. The hidden layer of the multilayer perceptron processes the input vector through a series of linear transformations and nonlinear activation functions. Nonlinear activation functions such as Sigmoid function Or ReLU function, etc., are used to introduce nonlinear factors and enhance the network's expressive power. After being processed by multiple hidden layers, the output layer outputs the estimated image element vector Y, whose dimension depends on the number of image elements that need to be inferred.

[0104] Another technical means is to use a convolutional neural network. If the second training behavior specification mapping representation information and the second training attribute specification mapping representation information are represented as a two-dimensional or multi-dimensional data structure, for example, the behavior data is arranged in a matrix form according to a time series, the recommendation system uses a convolutional neural network to infer the portrait elements. The convolutional neural network processes the input data through a convolution layer, a pooling layer, and a fully connected layer. The convolution layer uses a convolution kernel to perform a sliding convolution operation on the input data to extract local features of the data. Assuming that the input data is X and the convolution kernel is K, the convolution operation is represented as Y = X *K, where * represents the convolution operation. The pooling layer is used to downsample the output of the convolution layer to reduce the dimension of the data while retaining important feature information. The fully connected layer linearly transforms the output of the pooling layer and finally outputs the inferred portrait element.

[0105] When the recommendation system is inferring the elements of a profile, it also needs to consider the interaction between different features. The attention mechanism is used to automatically assign weights to different features, so that the network pays more attention to important features. The attention mechanism calculates the attention score of each feature and then performs a weighted summation of the features based on the score. Assume that the input feature is , the attention score is calculated by an attention function, such as ,in Represents the attention score of the i-th feature. Then the attention score is normalized by the Softmax function to obtain the attention weight Finally, the features are weighted and summed according to the attention weights to obtain the weighted feature representation .

[0106] In practical applications, recommendation systems face problems such as data noise and overfitting. To solve these problems, recommendation systems use regularization methods, such as L1 and L2 regularization. L2 regularization constrains the complexity of the model by adding the sum of squares of weights to the loss function. Its loss function is expressed as , where L is the original loss function, is the regularization parameter, is the weight of the model. The recommendation system also uses Dropout technology to randomly discard some neurons during the training process to prevent excessive dependence between neurons, thereby improving the generalization ability of the model. The specific method is to randomly set the output of some neurons to 0 with a certain probability p in each training batch, thus forcing the network to learn more robust features.

[0107] Step S50, after optimizing the transition cross-modal specification mapping network and the initial portrait element reasoning network based on the inferred portrait elements and the prior portrait elements, collaborative debugging is performed again until the preset debugging stop condition is reached to obtain the converged cross-modal specification mapping network and portrait element reasoning network.

[0108] When implementing step S50, the recommendation system uses the following technical means to complete the optimization and collaborative debugging of the network. First, the recommendation system defines a loss function to measure the difference between the estimated portrait element and the prior portrait element. For example, the mean square error loss function is used to measure the average square error between two vectors. The smaller the error, the closer the estimated portrait element is to the prior portrait element.

[0109] The recommendation system uses the gradient descent algorithm to minimize the loss function. The gradient descent algorithm is an iterative optimization algorithm that calculates the gradient of the loss function with respect to the network parameters and then updates the network parameters in the opposite direction of the gradient so that the loss function gradually decreases.

[0110] During the optimization process, the recommendation system collaboratively debugs the transition cross-modal norm mapping network and the initial image element inference network. This means that when updating the network parameters, the parameters of both networks need to be considered at the same time. The two networks are considered as a whole, the gradient of the loss function with respect to the entire network parameters is calculated, and then updated together.

[0111] The recommendation system also uses batch training to improve training efficiency. Batch training means selecting a small batch of data from the training data for training each time, rather than using the entire training data set. Assuming that there are N samples in the training data set, and m samples are selected as a small batch each time, the training data set is divided into In each iteration, the recommendation system uses a small batch of data to calculate the loss function and gradient, and then updates the network parameters. This method reduces the amount of calculation and speeds up training.

[0112] As an implementation manner, the debugging and obtaining of a transition cross-modal norm mapping network based on the first training behavior data representation, the first training attribute data representation and the prior user portrait of the first training user data includes: Step S11, obtaining a first training behavior data representation, a first training attribute data representation, and a priori user portrait of first training user data; Step S12, performing canonical mapping on the first training behavior data representation through the initialized initial cross-modal canonical mapping network to obtain first training behavior canonical mapping representation information; performing canonical mapping on the first training attribute data representation through the initialized initial cross-modal canonical mapping network to obtain first training attribute canonical mapping representation information; Step S13, performing user portrait prediction on the first training behavior specification mapping representation information and the first training attribute specification mapping representation information through the initialized initial portrait element inference network to obtain a predicted user portrait for the first training user data; Step S14, iteratively debugging the initial cross-modal norm mapping network based on the predicted user portrait and the prior user portrait until a preset debugging stop condition is reached to obtain a transitional cross-modal norm mapping network.

[0113] In step S11, the first training user data is a data set used for the initial training model, which contains information about a large number of users. The first training behavior data representation is the result of characterization information mining of the behavior monitoring data in the first training user data, which converts the user's behavior information into a form that is convenient for computer processing and analysis. The first training attribute data representation is obtained by characterization information mining of the basic attribute data of the first training user data, reflecting the basic attribute characteristics of the user. The prior user portrait is a user portrait known in advance for the first training user data, which is obtained based on historical data, expert knowledge or other reliable sources and serves as a reference standard for subsequent model training.

[0114] When implementing step S11, the recommendation system uses a variety of technical means to obtain these data. For the first training behavior data representation, the recommendation system uses deep learning models such as recurrent neural networks (RNN) or their variants, long short-term memory networks (LSTM), and gated recurrent units (GRU) to process behavior sequence data. Taking LSTM as an example, it can effectively process behavior data with time sequence and capture long-term dependencies between behaviors. Assuming that the behavior monitoring data is a time series of commodity browsing records, LSTM takes the browsing information of each time step as input, and outputs a vector that can represent the characteristics of the entire behavior sequence through its internal memory unit and gating mechanism, that is, the first training behavior data representation. For the first training attribute data representation, the recommendation system uses unique hot encoding, embedding technology, etc. to process the basic attribute data. For example, for the user's gender attribute, unique hot encoding is used to represent males as [1, 0] and females as [0, 1]; for the user's age attribute, an embedding layer is used to map the age value to a low-dimensional vector space. The prior user portrait is obtained by collecting and analyzing historical sales data, user feedback information, etc., and is also determined with the help of expert experience and judgment.

[0115] In step S12, the initial cross-modal canonical mapping network is a network in an initial state, and its parameters are randomly initialized. The purpose of canonical mapping is to align the features of data (behavior data and attribute data) of different modalities so that they have a unified format and scale for subsequent network processing.

[0116] In step S13, the initial portrait element inference network is also a network in an initialized state. It receives the behavior and attribute representation information after canonical mapping, and outputs the predicted user portrait through calculation and processing within the network.

[0117] In step S14, the core of this step is to adjust the parameters of the initial cross-modal norm mapping network by comparing the difference between the predicted user portrait and the prior user portrait, so that the prediction result is closer to the prior user portrait.

[0118] In order to measure the difference between the predicted user profile and the prior user profile, the recommendation system defines a loss function. For example, the mean square error loss function is used to calculate the average square error between two vectors. The smaller the error, the closer the prediction result is to the prior user profile.

[0119] As an implementation manner, the determining of a plurality of pending portrait elements for the user data based on the respective multi-category index portrait elements of the plurality of categories of category data includes: Step 310, performing element frequency analysis on the multi-category index image elements of the plurality of categories of category data to obtain element frequency analysis results; Step 320: Use the portrait element with the largest number of occurrences included in the element frequency analysis result as the pending portrait element for the user data.

[0120] In step 310, the multiple categories of category data contain information about different aspects of the user, such as the user's basic attribute data and behavior monitoring data, and the multi-category indexed portrait elements are portrait elements related to these category data obtained by indexing in the portrait element mapping relationship set. Element frequency analysis refers to counting the number of times each portrait element appears in all multi-category indexed portrait elements to understand the frequency of occurrence of each portrait element.

[0121] When implementing step 310, the recommendation system uses a hash table data structure to implement element frequency analysis. The hash table efficiently stores and searches for elements, using each portrait element as a key and the number of occurrences as a value stored in the hash table. When traversing multi-category index portrait elements, if the element is already in the hash table, its corresponding value is increased by 1; if the element is not in the hash table, it is inserted into the hash table and the value is initialized to 1. Its time complexity is O(n), where n is the number of multi-category index portrait elements, and element frequency analysis can be completed quickly.

[0122] In step 320, the pending portrait elements are intermediate results before the final user portrait elements are determined. Based on the results of element frequency analysis, the portrait elements with the highest frequency of occurrence are selected because these portrait elements can better represent the user's characteristics and preferences.

[0123] When implementing step 320, the recommendation system traverses the hash table to find the key with the largest value, that is, the portrait element with the most occurrences. If there are multiple portrait elements with the same number of occurrences and all of them are maximum values, the recommendation system selects according to a preset rule, such as selecting the first element in alphabetical order of the portrait elements, or setting priorities for different portrait elements according to business needs and selecting the element with the highest priority.

[0124] From the overall process point of view, steps 310-320 play a connecting role in the entire e-commerce product recommendation method. In the previous steps, the recommendation system obtains multi-category index portrait elements of various categories of category data through cross-category indexing. These portrait elements reflect the characteristics of users from different angles. Steps 310-320 screen and refine these portrait elements, and find the most representative portrait elements as pending portrait elements through element frequency analysis, providing more focused information for subsequent portrait element reasoning and user portrait generation.

[0125] Figure 2 A hardware entity diagram of an e-commerce product recommendation system provided by an embodiment of the present invention is shown in FIG. Figure 2As shown, the hardware entity of the e-commerce product recommendation system P includes: a processor P1 and a memory P2, wherein the memory P2 stores a computer program that can be run on the processor P1, and when the processor P1 executes the program, the steps in the method of any of the above embodiments are implemented.

[0126] The above description is only an implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An e-commerce product recommendation method based on big data analysis, characterized in that: The method comprises: Determine user data to be analyzed of a target user, and determine multiple categories of category data by mining the user data; Based on the multiple categories of category data, cross-category indexing is performed on each of the multiple categories of representation information to obtain a multi-category index portrait element of each of the multiple categories of category data; Determining a plurality of pending portrait elements for the user data based on the respective multi-category index portrait elements of the plurality of categories of category data; The multiple categories include monitoring data categories and attribute data categories; the multiple categories of category data include behavior monitoring data and basic attribute data, the basic attribute data and the multiple pending portrait elements are feature fused to obtain first fused representation information; and the behavior monitoring data and the multiple pending portrait elements are feature fused to obtain second fused representation information; Performing canonical mapping on the first fused representation information through a cross-modal canonical mapping network to obtain first canonical mapping fused representation information; Performing canonical mapping on the second fused representation information through the cross-modal canonical mapping network to obtain second canonical mapping fused representation information; Performing portrait element reasoning based on the first canonical mapping fusion representation information and the second canonical mapping fusion representation information through a portrait element reasoning network to obtain a user portrait element for the user data, and generating a target user portrait of the target user based on the user portrait element; Recommend products to the target user based on the target user portrait.

2. The e-commerce product recommendation method based on big data analysis according to claim 1, characterized in that: The category data based on the multiple categories are each cross-category indexed at the level of the representation information of the multiple categories to obtain the multi-category index portrait elements of the category data of the multiple categories, including: Determine a portrait element mapping relationship set; the portrait element mapping relationship set includes comparison category representation information of comparison user data under the multiple categories and comparison portrait elements corresponding to the comparison category representation information; Based on the commonality measurement results between the category data of the multiple categories and the control category representation information of the control user data under the multiple categories, the image element mapping relationship set is indexed to obtain the multi-category indexed image elements of each of the category data of the multiple categories.

3. The e-commerce product recommendation method based on big data analysis according to claim 2, characterized in that: The commonality measurement results between the category data based on the multiple categories and the comparison category representation information of the comparison user data under the multiple categories are indexed in the image element mapping relationship set to obtain the multi-category index image elements of the category data of the multiple categories, including: Performing commonality measurement on each category of the multiple categories of category data and the comparison category representation information of the comparison user data under the multiple categories, respectively, to obtain a commonality measurement result for each category of data; From the common measurement results of each category of data, determine that each category of data represents a target measurement result that meets the common requirements; Based on the control category representation information corresponding to the target measurement result of each category of data, the image element mapping relationship set is indexed respectively to obtain the multi-category index image element of each category of data.

4. The e-commerce product recommendation method based on big data analysis according to claim 2, characterized in that: The multiple categories include monitoring data categories and attribute data categories; the multiple categories of category data include behavior monitoring data and basic attribute data of the user data; the portrait element mapping relationship set includes behavior comparison category representation information of the comparison user data under the monitoring data category and behavior comparison portrait elements corresponding to the behavior comparison category representation information, and attribute comparison category representation information of the comparison user data under the attribute data category and attribute comparison portrait elements corresponding to the attribute comparison category representation information; The commonality measurement results between the category data based on the multiple categories and the comparison category representation information of the comparison user data under the multiple categories are indexed in the image element mapping relationship set to obtain the multi-category index image elements of the category data of the multiple categories, including: Based on the behavior monitoring data and the basic attribute data, respectively, representation information is extracted to obtain behavior data representation and attribute data representation; Performing behavior representation commonality measurement on the behavior data representation, the behavior comparison category representation information and the attribute comparison category representation information respectively, and obtaining a multi-category index portrait element of the behavior monitoring data in the behavior comparison portrait element and the attribute comparison portrait element based on the behavior representation commonality measurement result; The attribute data representation is respectively subjected to attribute representation commonality measurement with the behavior comparison category representation information and the attribute comparison category representation information, and based on the attribute representation commonality measurement result, a multi-category index portrait element of the basic attribute data is obtained in the behavior comparison portrait element and the attribute comparison portrait element.

5. The e-commerce product recommendation method based on big data analysis according to claim 4, characterized in that: The method of measuring the commonality of the behavior data representation with the behavior comparison category representation information and the attribute comparison category representation information, and obtaining a multi-category index portrait element of the behavior monitoring data in the behavior comparison portrait element and the attribute comparison portrait element based on the behavior comparison category representation result, includes: Performing behavior representation commonality measurement on the behavior data representation and the behavior control category representation information to obtain a first behavior representation commonality measurement result; Determining a first behavior category index portrait element in the behavior comparison portrait element based on the behavior comparison category representation information corresponding to the first behavior representation commonality measurement result representing the satisfaction of the commonality requirement; Performing behavior representation commonality measurement on the behavior data representation and the attribute comparison category representation information to obtain a second behavior representation commonality measurement result; Based on the attribute comparison category representation information corresponding to the second behavior representation commonality measurement result representing the satisfaction of the commonality requirement, determining the first attribute category index portrait element in the attribute comparison portrait element; Based on the first behavior category index portrait element and the first attribute category index portrait element, a multi-category index portrait element of the behavior monitoring data is obtained.

6. The e-commerce product recommendation method based on big data analysis according to claim 4, characterized in that: The method of performing attribute representation commonality measurement on the attribute data representation, the behavior comparison category representation information and the attribute comparison category representation information, and obtaining a multi-category index portrait element of the basic attribute data in the behavior comparison portrait element and the attribute comparison portrait element based on the attribute representation commonality measurement result, includes: Performing attribute representation commonality measurement on the attribute data representation and the behavior comparison category representation information to obtain a first attribute representation commonality measurement result; Determine a second behavior category index portrait element in the behavior comparison portrait element based on the behavior comparison category representation information corresponding to the first attribute representation commonality measurement result representing the satisfaction of the commonality requirement; Performing attribute representation commonality measurement on the attribute data representation and the attribute comparison category representation information to obtain a second attribute representation commonality measurement result; Determine a second attribute category index portrait element in the attribute comparison portrait element based on the attribute comparison category representation information corresponding to the second attribute representation commonality measurement result representing the satisfaction of the commonality requirement; Based on the second behavior category index portrait element and the second attribute category index portrait element, a multi-category index portrait element of the basic attribute data is obtained.

7. The e-commerce product recommendation method based on big data analysis according to claim 1, characterized in that: The cross-modal specification mapping network and the image element reasoning network are trained by the following steps: Based on a first training behavior data representation of first training user data, a first training attribute data representation, and a priori user portrait, debugging to obtain a transition cross-modal norm mapping network, wherein the first training behavior data representation is obtained by mining representation information of behavior monitoring data of the first training user data; The first training attribute data representation is obtained by mining the representation information of the basic attribute data of the first training user data; Acquire a second training behavior data representation, a second training attribute data representation, and a priori portrait elements of the second training user data; The second training behavior data representation is obtained by mining the representation information of the behavior monitoring data of the second training user data; The second training attribute data representation is obtained by mining the representation information of the basic attribute data of the second training user data; Performing canonical mapping on the second training behavior data representation through the transition cross-modal canonical mapping network to obtain second training behavior canonical mapping representation information; and performing canonical mapping on the second training attribute data representation through the transition cross-modal canonical mapping network to obtain second training attribute canonical mapping representation information; Performing image element reasoning based on the second training behavior specification mapping representation information and the second training attribute specification mapping representation information through the initialized initial image element reasoning network to obtain an inferred image element for the second training user data; After optimizing the transition cross-modal specification mapping network and the initial portrait element reasoning network based on the inferred portrait elements and the prior portrait elements, collaborative debugging is performed again until a preset debugging stop condition is reached to obtain a converged cross-modal specification mapping network and portrait element reasoning network.

8. The e-commerce product recommendation method based on big data analysis according to claim 7, characterized in that: The debugging and obtaining of a transition cross-modal norm mapping network based on the first training behavior data representation, the first training attribute data representation and the prior user portrait of the first training user data includes: Obtaining a first training behavior data representation, a first training attribute data representation, and a priori user portrait of first training user data; Performing canonical mapping on the first training behavior data representation through the initialized initial cross-modal canonical mapping network to obtain first training behavior canonical mapping representation information; performing canonical mapping on the first training attribute data representation through the initialized initial cross-modal canonical mapping network to obtain first training attribute canonical mapping representation information; Performing user profile prediction on the first training behavior specification mapping representation information and the first training attribute specification mapping representation information through the initialized initial profile element inference network to obtain a predicted user profile for the first training user data; The initial cross-modal canonical mapping network is iteratively debugged based on the predicted user portrait and the prior user portrait until a preset debugging stop condition is reached to obtain a transitional cross-modal canonical mapping network.

9. The e-commerce product recommendation method based on big data analysis according to claim 1, characterized in that: The determining of a plurality of pending portrait elements for the user data based on the respective multi-category index portrait elements of the plurality of categories of category data comprises: Performing element frequency analysis on the multi-category index image elements of each of the multiple categories of category data to obtain an element frequency analysis result; The portrait element with the largest number of occurrences included in the element frequency analysis result is used as a pending portrait element for the user data.

10. An e-commerce product recommendation system based on big data analysis, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the program, the steps in the method according to any one of claims 1 to 9 are implemented.

Images