CN119295190A

CN119295190A - An intelligent product selection management method and system based on machine learning

Info

Publication number: CN119295190A
Application number: CN202411832888.5A
Authority: CN
Inventors: 罗志青; 苏维勇; 韩国峰; 胡冬根; 谢伟
Original assignee: Guangzhou Ruoyuchen Tech Co ltd
Current assignee: Guangzhou Ruoyuchen Tech Co ltd
Priority date: 2024-12-13
Filing date: 2024-12-13
Publication date: 2025-01-10

Abstract

The embodiment of the present application belongs to the field of artificial intelligence, and relates to an intelligent product selection management method based on machine learning, including: generating customer preference portraits of each customer based on historical purchase data and browsing data; embedding attributes and knowledge graphs of product information of each product to obtain product portraits of each product; constructing a customer product interaction graph based on historical purchase data and browsing data; performing representation learning on the customer product interaction graph to obtain product node features of each product node; generating product comprehensive features of each product based on the product portraits of each product and the product node features of its corresponding product nodes; combining the customer preference portraits of target customers and the product comprehensive features of each product to obtain combined features and inputting them into a matching model to obtain the interest between the target customer and each product; determining the target product among each product based on the obtained interest to make product recommendations. The present application improves the accuracy of product selection and recommendation.

Description

Intelligent option management method and system based on machine learning

Technical Field

The present application relates to the field of artificial intelligence, and in particular, to a machine learning-based intelligent option management method, apparatus, computer device, and storage medium.

Background

In the e-commerce platform, the number of products is huge and the variety is great, and consumers often face the problems of information overload and difficult decision making when selecting the products. Meanwhile, the purchase preference and demand of different consumers are greatly different, and the traditional product recommendation technology faces a plurality of technical challenges, so that how to accurately capture and understand the demands of consumers is a big problem. Secondly, the attribute feature dimensions of the product are many, complex association relations may exist between different attributes, and how to build a comprehensive and accurate product portrait is another challenge. Moreover, in the context of massive products and user data, high requirements are also put forward on how to efficiently construct the matching relationship between users and products. Therefore, there is a need for an intelligent product selection and recommendation method to improve the accuracy of product selection and recommendation.

Disclosure of Invention

The embodiment of the application aims to provide an intelligent option management method, device, computer equipment and storage medium based on machine learning, so as to solve the problem of accuracy of product option and recommendation.

In order to solve the technical problems, the embodiment of the application provides an intelligent option management method based on machine learning, which adopts the following technical scheme:

Generating a customer preference portrait of each customer through a preset portrait generation algorithm based on historical purchase data and browsing data of each customer;

Obtaining product information of each product, and carrying out attribute embedding and knowledge graph embedding on the product information of each product to obtain product images of each product;

Constructing a customer product interaction diagram according to the obtained historical purchase data and browsing data, wherein the customer product interaction diagram comprises customer nodes and product nodes;

performing representation learning on the customer product interaction graph to obtain product node characteristics of each product node;

generating product comprehensive characteristics of each product according to the product portraits of each product and the product node characteristics of the corresponding product nodes of each product;

Combining the customer preference portrait of the target customer and the product comprehensive characteristics of each product to obtain combined characteristics, and inputting each combined characteristic into a matching model to obtain the interestingness between the target customer and each product;

And determining target products in the products according to the obtained interestingness, and recommending the products to the target clients according to the determined target products.

Further, the step of generating the customer preference portraits of the customers through a preset portrayal generation algorithm based on the historical purchase data and the browsing data of the customers comprises the following steps:

Calculating the client similarity among the clients through a collaborative filtering algorithm based on the historical purchase data of the clients, and determining the similar clients of target clients from the clients according to the obtained client similarity;

determining purchase preference products of each similar client according to the historical purchase data of each similar client, and constructing a first product set of the target client according to each purchase preference product;

calculating the product similarity among products through a content-based recommendation algorithm based on the browsing data of each client;

determining browsing preference products similar to the current browsing products of the target clients according to the obtained product similarity, and adding each browsing preference product into the first product set to obtain a second product set of the target clients;

and generating a client preference portrait of the target client according to the second product set, the historical purchase data and the browse data of the target client, and obtaining the client preference portraits of the clients.

Further, after the step of obtaining the second product set of the target customer, the method further includes:

processing the historical purchase data of each client through an association rule mining algorithm to obtain product purchase association information;

And adding purchase related products in the second product set according to the product purchase related information.

Further, the step of obtaining the product image of each product by performing attribute embedding and knowledge graph embedding on the product information of each product includes:

preprocessing unstructured data in product information of each product to obtain structured data of each product;

extracting attribute information from the structured data of each product to obtain structured attribute data of each product;

mapping the structured attribute data of each product into attribute embedded representations respectively;

Based on the attribute embedded representation of each product, determining the association relation between the products through similarity calculation and a clustering algorithm, and constructing a product knowledge graph according to the obtained association relation;

And fusing the attribute embedded representation and the knowledge graph embedded representation of each product to obtain the product image of each product.

Further, the step of performing representation learning on the customer product interaction graph to obtain product node characteristics of each product node includes:

generating a node sequence on the customer product interaction diagram through a random walk algorithm;

and inputting the node sequence into a skip-gram model to generate product node characteristics of each product node in the customer product interaction graph by maximizing node co-occurrence probability.

Further, the method further comprises:

For each product node in the customer product interaction graph, determining neighbor nodes of the product node;

calculating node similarity between the product node and each neighboring node;

Determining the attention weight of each neighbor node according to the obtained node similarity;

Aggregating the product node characteristics of each neighbor node according to the obtained attention weight to obtain aggregated neighbor information;

And updating the product node characteristics of the product node according to the aggregation neighbor information.

Further, the step of generating the product comprehensive characteristics of each product according to the product portrait of each product and the product node characteristics of the corresponding product node comprises the following steps:

For each product, extracting a product embedding vector of the product from a product representation of the product;

and combining the product embedding vector with the product node characteristics of the product nodes corresponding to the product to obtain the product comprehensive characteristics of the product.

In order to solve the technical problems, the embodiment of the application also provides an intelligent option management device based on machine learning, which adopts the following technical scheme:

The preference generation module is used for generating client preference portraits of all clients through a preset portrayal generation algorithm based on historical purchase data and browsing data of all clients;

The product generation module is used for acquiring product information of each product, and carrying out attribute embedding and knowledge graph embedding on the product information of each product to obtain a product image of each product;

The interactive construction module is used for constructing a customer product interactive graph according to the obtained historical purchase data and browsing data, wherein the customer product interactive graph comprises customer nodes and product nodes;

The feature generation module is used for carrying out representation learning on the customer product interaction graph to obtain product node features of each product node;

the comprehensive generation module is used for generating product comprehensive characteristics of each product according to the product portrait of each product and the product node characteristics of the corresponding product nodes;

The combination calculation module is used for respectively combining the customer preference portrait of the target customer and the product comprehensive characteristics of each product to obtain combination characteristics, and inputting each combination characteristic into a matching model to obtain the interestingness between the target customer and each product;

And the product recommendation module is used for determining target products in the products according to the obtained interestingness and recommending the products to the target clients according to the determined target products.

In order to solve the above technical problem, an embodiment of the present application further provides a computer device, where the computer device includes a memory and a processor, where the memory stores computer readable instructions, and the processor implements the steps of the intelligent option management method based on machine learning as described above when executing the computer readable instructions.

In order to solve the above technical problem, an embodiment of the present application further provides a computer readable storage medium, where computer readable instructions are stored on the computer readable storage medium, and when the computer readable instructions are executed by a processor, the steps of an intelligent option management method based on machine learning are implemented as described above.

Compared with the prior art, the method and the device have the advantages that the client preference image of each client is generated through historical purchase data and browsing data, attribute embedding and knowledge graph embedding are carried out on product information of each product to generate a high-quality product image, a client product interaction graph is constructed and representation learning is carried out, implicit correlation representation among product nodes is further enhanced, product node characteristics of the product image of each product and corresponding product nodes are combined, complex relations among the clients are converted into product comprehensive characteristics of multi-dimensional representation, the client preference image and the product comprehensive characteristics are input into a matching model to accurately calculate interestingness, personalized recommendation is achieved, accuracy of product selection and recommendation is improved, and conversion rate and client satisfaction of product recommendation are improved.

Drawings

In order to more clearly illustrate the solution of the present application, a brief description will be given below of the drawings required for the description of the embodiments of the present application, it being apparent that the drawings in the following description are some embodiments of the present application, and that other drawings may be obtained from these drawings without the exercise of inventive effort for a person of ordinary skill in the art.

FIG. 1 is an exemplary system architecture diagram in which the present application may be applied;

FIG. 2 is a flow chart of one embodiment of a machine learning based intelligent choice management method in accordance with the present application;

FIG. 3 is a schematic diagram of one embodiment of a machine learning based intelligent option management device in accordance with the present application;

FIG. 4 is a schematic structural diagram of one embodiment of a computer device in accordance with the present application.

Detailed Description

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs, the terms used in the description herein are used for the purpose of describing particular embodiments only and are not intended to limit the application, and the terms "comprising" and "having" and any variations thereof in the description of the application and the claims and the above description of the drawings are intended to cover non-exclusive inclusions. The terms first, second and the like in the description and in the claims or in the above-described figures, are used for distinguishing between different objects and not necessarily for describing a sequential or chronological order.

Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearances of such phrases in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those of skill in the art will explicitly and implicitly appreciate that the embodiments described herein may be combined with other embodiments.

In order to make the person skilled in the art better understand the solution of the present application, the technical solution of the embodiment of the present application will be clearly and completely described below with reference to the accompanying drawings.

As shown in fig. 1, the system architecture 100 may include a terminal device 101, a network 102, and a server 103, where the terminal device 101 may be a notebook 1011, a tablet 1012, or a cell phone 1013. Network 102 is the medium used to provide communication links between terminal device 101 and server 103. Network 102 may include various connection types such as wired, wireless communication links, or fiber optic cables.

A user may interact with the server 103 via the network 102 using the terminal device 101 to receive or send messages or the like. Various communication client applications, such as a web browser application, a shopping class application, a search class application, an instant messaging tool, a mailbox client, social platform software, and the like, may be installed on the terminal device 101.

The terminal device 101 may be various electronic devices having a display screen and supporting web browsing, and the terminal device 101 may be an electronic book reader, an MP3 player (Moving Picture Experts Group Audio Layer III, moving picture experts compression standard audio layer III), an MP4 (Moving Picture Experts Group Audio Layer IV, moving picture experts compression standard audio layer IV) player, a laptop portable computer, a desktop computer, or the like, in addition to the notebook 1011, the tablet 1012, or the mobile phone 1013.

The server 103 may be a server providing various services, such as a background server providing support for pages displayed on the terminal device 101.

It should be noted that, the intelligent option management method based on machine learning provided by the embodiment of the application is generally executed by a server, and correspondingly, the intelligent option management device based on machine learning is generally arranged in the server.

It should be understood that the number of terminal devices, networks and servers in fig. 1 is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation.

With continued reference to FIG. 2, a flow chart of one embodiment of a machine learning based intelligent option management method in accordance with the present application is shown. The intelligent option management method based on machine learning comprises the following steps:

Step S201, generating the customer preference portrait of each customer through a preset portrait generation algorithm based on the historical purchase data and browsing data of each customer.

In this embodiment, an electronic device (for example, a server shown in fig. 1) on which the intelligent option management method based on machine learning operates may communicate with a terminal device through a wired connection manner or a wireless connection manner. It should be noted that the wireless connection may include, but is not limited to, 3G/4G/5G connection, wiFi connection, bluetooth connection, wiMAX connection, zigbee connection, UWB (ultra wideband) connection, and other now known or later developed wireless connection.

Specifically, historical purchase data of each customer and browsing data are obtained, wherein the historical purchase data can be a past purchase record of the customer on the platform, and the browsing data can be behavior data of the customer for browsing products on the platform. These data can reflect customer preferences for different product categories, price intervals, brands, etc.

The historical purchase data and the browsing data can be processed step by step according to a preset portrait generation algorithm to generate a customer preference portrait of each customer. Customer preference portraits are generated by analyzing the historical purchasing behavior and browsing behavior of customers, aiming at capturing the customers' consumption habits, interests and demands. The portrayal generation algorithm may include steps of data cleansing, feature extraction, behavioral analysis, etc., converting the historical data into a digitized representation of the customer's preferences.

Step S202, obtaining product information of each product, and performing attribute embedding and knowledge graph embedding on the product information of each product to obtain product images of each product.

Specifically, a product representation is characteristic information describing a product, and may include structured data such as attributes, categories, brands, prices, etc. of the product, and unstructured data such as user-generated content (e.g., reviews, labels, etc.).

The generation of the product portrait is realized by attribute embedding and knowledge graph embedding. Attribute embedding is to map product attribute data (such as price, brand, etc.) into a low-dimensional space to obtain a characteristic representation thereof, and knowledge graph embedding is to construct a knowledge graph by utilizing association relations (such as similar products, complementary products, etc.) among products to further enhance the representation of product characteristics. In the knowledge graph, nodes can be products, and edges can be relationships among the products. The association between the product attribute data and the product comes from the product information.

Step S203, constructing a customer product interaction diagram according to the obtained historical purchase data and browsing data, wherein the customer product interaction diagram comprises customer nodes and product nodes.

Specifically, a customer product interaction graph may be constructed from historical purchase data and browsing data for each customer. Customer product interaction graphs describe the relationship between a customer and a product in which nodes represent the customer or the product and edges represent interactions (e.g., purchase, browse, etc.) between the customer and the product. The graph structure helps discover underlying relationships between customers and products.

The customer product interaction graph is an interaction network between the user and the product. For example, if a customer purchases a product, the customer node is connected with the product node in the customer product interaction diagram to form an edge.

And step S204, performing representation learning on the customer product interaction graph to obtain the product node characteristics of each product node.

In particular, representation learning is the training of each node (here, product nodes) in a customer product interaction graph with a low-dimensional vector representation that fully expresses the characteristics of the nodes and the relationships between the nodes. The graph shows that the learning method may include DeepWalk, node2vec, etc.

The vectors representing the product nodes are called product node features, and the product node features of each product node are embedded vectors obtained through a graph learning method, and the feature vectors can capture the relation and the similarity with other product nodes.

Step S205, product comprehensive characteristics of each product are generated according to the product portraits of each product and the product node characteristics of the corresponding product nodes.

Specifically, the product comprehensive feature is to combine the attribute of the product (such as brand, price, type, etc. and product portrait of the product) with the node feature of the product obtained by the graph representation learning to form the multidimensional feature representation of the product. This feature representation can reflect both the basic attributes of the product and incorporate similarities or relationships with other products.

And S206, respectively combining the customer preference portrait of the target customer and the product comprehensive characteristics of each product to obtain combined characteristics, and inputting each combined characteristic into a matching model to obtain the interestingness between the target customer and each product.

Specifically, the customer preference portraits of the target customers and the product synthesis features of the respective products are combined (e.g., stitched, weighted, etc.) to form a complete representation to fully reflect the relationship between the target customers and each product. The combined characteristics obtained after combination are input into a trained matching model (such as a neural network model, a support vector machine and the like), and the interestingness between a target customer and each product is calculated through the matching model, wherein the higher the interestingness is, the more likely the customer is interested in the product.

And S207, determining target products in the products according to the obtained interestingness, and recommending the products to the target clients according to the determined target products.

Specifically, according to the interest degree of the target clients on each product, selecting the product with higher interest degree from the products as the target product. The target products can be meeting points of potential demands of the clients, and the target clients are recommended according to the target products, so that the purchasing conversion rate of the clients is improved, and the accuracy of product selection and product recommendation is improved.

In the embodiment, a customer preference image of each customer is generated through historical purchase data and browsing data, attribute embedding and knowledge graph embedding are carried out on product information of each product to generate a high-quality product image, a customer product interaction image is constructed and represented and learned, implicit correlation representation among product nodes is further enhanced, product image of each product and product node characteristics of corresponding product nodes are combined, complex relations among the customers and the products are converted into product comprehensive characteristics of multi-dimensional representation, and the customer preference image and the product comprehensive characteristics are input into a matching model to accurately calculate interestingness, so that personalized recommendation is realized, accuracy of product selection and recommendation is improved, and conversion rate of product recommendation and customer satisfaction are improved.

Further, the step S201 may include calculating a customer similarity between customers through a collaborative filtering algorithm based on historical purchase data of the customers, determining similar customers of the target customers among the customers according to the obtained customer similarity, the target customers from the customers, determining purchase preference products of the similar customers according to the historical purchase data of the similar customers, constructing a first product set of the target customers according to the purchase preference products, calculating a product similarity between the products through a content-based recommendation algorithm based on the browsing data of the customers, determining browsing preference products similar to the current browsing products of the target customers according to the obtained product similarity, and adding the browsing preference products to the first product set to obtain a second product set of the target customers, generating a customer preference portrait of the target customers according to the second product set of the target customers, the historical purchase data and the browsing data, and obtaining the customer preference portrait of the target customers.

Specifically, through a collaborative filtering algorithm, the client similarity between the target client and other clients is calculated according to historical purchase data of the target client and other clients. The calculation of customer similarity may employ user-based collaborative filtering to find similar customers by comparing purchasing behavior (e.g., products purchased, frequency of purchase, etc.).

Collaborative filtering is a recommendation algorithm that makes recommendations mainly by analyzing the similarity between users or the similarity between items. Based on the collaborative filtering method of the user, products which are possibly interested by the user are predicted according to the similarity between the user and other users.

And finding other clients which are most similar to the purchasing behavior of the target client according to the calculated client similarity, wherein the purchasing preferences of the similar clients can recommend proper products for the target client. The products they prefer to purchase, i.e., purchase preferred products, are extracted from the purchase history data of similar customers, and the extracted purchase preferred products are determined as the first product set of the target customer. Purchase preference products are extrapolated from the purchase preferences of similar customers and reflect the potential interests of the target customer.

And calculating the similarity between the current browsing product of the client and other products by adopting a content-based recommendation algorithm based on the browsing data of the target client. This step focuses on the properties of the product itself, such as category, brand, price, etc. And selecting products with higher similarity from other products similar to the currently browsed products of the target clients based on the product similarity, and adding the products with higher similarity to the first product set to obtain a second product set. These products reflect the current interests and needs of the target customer.

And integrating the historical purchase data, the browsing data and the second product set of the target client to generate a client preference portrait of the target client. It describes the interest level and purchasing tendency of the target customer for different categories of products. In one embodiment, the customer preference portrait may also include customer information for the target customer.

In the embodiment, through combining the historical purchase data and the browsing data, the recommendation can be generated based on the past behavior of the client, the content to be recommended can be adjusted according to the current browsing preference, the diversity of the recommendation is expanded, more personalized and accurate product recommendation is facilitated, and the real-time browsing data can be timely adapted to the change of the interest of the client, so that more accurate recommendation is provided.

Further, after the step of obtaining the second product set of the target customer, the method further comprises the steps of processing historical purchase data of each customer through an association rule mining algorithm to obtain product purchase association information, and adding purchase association products into the second product set according to the product purchase association information.

Specifically, association rule mining is a technique in data mining for finding relationships between different things. It is created by finding relationships or "rules" between "item sets" from a large amount of data. The most common algorithms are the Apriori algorithm and the FP-growth algorithm, which can mine rules such as "buy a person also typically buy B". The algorithm can be applied to retail, recommendation systems and other scenes for finding patterns and relationships hidden in data.

And (3) through analyzing historical purchase data of all clients, utilizing an association rule mining algorithm to find potential purchase relations among products, and obtaining product purchase association information. For example, if a large number of customers purchase "cell phones" while also purchasing "headphones", then there is some purchasing relationship between "headphones" and "cell phones".

The product purchase association information may be a rule such as "purchase a, B is also frequently purchased". Each rule may have a degree of support (representing the frequency of occurrence of the rule in the purchase data) and a degree of confidence (representing the probability of the rule being established) that help evaluate the strength and reliability of the association rule.

And adding the associated product to the second product set of the target customer according to the obtained product purchase association information. For example, if a target customer browses or has purchased a product, the system may recommend some of the associated products based on the association rules for that product in the historical purchase data. For example, if the customer has browsed the smartwatch and the purchase history indicates that there is a strong association of "smartwatch" with "sports shoe", then sports shoes are added to the second product set, thereby updating the second product set to form a richer product set. This may include not only the current products of interest to the customer, but also related products of potential interest to the customer, further enhancing the accuracy and diversity of the recommendation.

In the embodiment, the historical purchase data of each customer is processed through the association rule mining algorithm, and the product purchase association information is introduced, so that the product recommendation can be expanded, some potentially relevant products which are not browsed or purchased are recommended to the target customer, the diversity and coverage of the recommendation result are improved, and the recommendation accuracy is improved.

Further, the step of obtaining the product image of each product by performing attribute embedding and knowledge graph embedding on the product information of each product may include preprocessing unstructured data in the product information of each product to obtain structured data of each product, extracting attribute information from the structured data of each product to obtain structured attribute data of each product, mapping the structured attribute data of each product to attribute embedding representations respectively, determining association relations between each product through similarity calculation and clustering algorithm based on the attribute embedding representations of each product, and constructing a product knowledge graph according to the obtained association relations, and fusing the attribute embedding representations and the knowledge graph embedding representations of each product to obtain the product image of each product.

Specifically, the product information has unstructured data (e.g., product description, user evaluation, pictures, etc.), and first, preprocessing operations such as text processing, image processing, etc. are performed to convert the information into structured data. For example, text in a product description may be extracted by natural language processing techniques, emotion analysis, and entity recognition, and pictures may be converted to structured data by extracting features by computer vision algorithms.

And extracting the attribute information of the products from the structured data of the products to obtain the structured attribute data of the products, such as brands, categories, prices, sizes and the like.

The structured attribute data for each product is mapped to an attribute embedded representation. For example, for information of brands, types, sizes, etc. of products, they are converted into vector representations by the embedding layer. The embedded purpose is to convert the discrete attributes of the product into a low-dimensional representation with semantics, capture the semantic relationships between the attributes, and facilitate subsequent computation and analysis. This process is typically learned through a neural network to obtain better representation capabilities.

And calculating the similarity between different products based on the attribute embedded representation of the products by using a similarity calculation method. For example, the similarity between the vectors of each product can be calculated by using the measurement modes such as cosine similarity, manhattan distance and the like, so as to judge which products have similarity and association relation on the attributes.

The products may also be grouped using a clustering algorithm. For example, by K-means clustering, products with similar characteristics can be classified into the same class, and products in each class have an association relationship.

Based on the obtained association relationship, a product knowledge graph can be constructed, which contains information such as similarity relationship and association relationship among products. A knowledge graph is a knowledge base represented by a graph structure in which nodes represent entities (e.g., products) and edges represent relationships between the entities. In the product recommendation system, the knowledge graph establishes a correlation network among products by integrating the relations (such as similarity, purchasing correlation and the like) of various products, so that the recommendation system is helped to better understand the relations among the products.

After the knowledge graph is constructed, the attribute embedded representation of the product is fused with the knowledge graph embedded representation of the product in the knowledge graph. The manner of fusion may be stitching, weighted averaging, etc. And obtaining a product image of each product through fusion, wherein the image contains basic attribute information of the product and also combines relationship and similarity information between the product and other products.

In the embodiment, attribute information is extracted from product information, association relations among products are determined based on attribute embedded representations of the products, a product knowledge graph is constructed according to the association relations, potential semantic relations and relations among the products are revealed, the attribute embedded representations and the knowledge graph embedded representations of the products are comprehensively embedded, a comprehensive and accurate product portrait is generated, the portrait not only comprises basic attributes of the products, but also combines similarity and association information among the products, complex relations among the products can be better understood, and follow-up improvement of correlation and accuracy of product recommendation is facilitated.

Further, the step S204 may include generating a node sequence on the customer product interaction graph by a random walk algorithm, and inputting the node sequence into a skip-gram model to generate product node characteristics of each product node in the customer product interaction graph by maximizing node co-occurrence probability.

Specifically, a product interaction diagram is constructed according to the purchase history data, browsing records and other interaction data of each customer. In this graph, the nodes include customer nodes and product nodes,

Edges represent customer interactions with a product (e.g., purchasing or browsing). Edges between products reflect their degree of association in customer interactions.

And performing wandering in the customer product interaction diagram through a random walk algorithm to generate a node sequence. The random walk algorithm (Random Walk Algorithm) is a graph algorithm, and each node selects a neighboring node as the next access node according to a certain probability. The method simulates the roaming process in the graph and is widely applied to the fields of graph embedding, social network analysis and the like. In the recommendation system, random walks may be used to generate a sequence of nodes of a product interaction graph, revealing potential associations between products. In the application, the algorithm starts from a certain product node, and randomly selects an adjacent node (namely, related product) as the next access node. By generating multiple node sequences through multiple iterations, the model can capture the interrelationship and similarity between products.

A node sequence refers to a series of nodes (product nodes in the present application) generated by a random walk algorithm, whose order reflects their relationship and interaction frequency in the graph. Through the sequence of nodes, potential associations between different products can be revealed.

The node sequence is input into the Skip-gram model. The Skip-gram model is a Word-embedding (Word 2 Vec) based model used in natural language processing to learn a vector representation of vocabulary. In the recommendation system, skip-gram models are used to learn node (product) features. Skip-grams generate embedded representations for each node by maximizing the probability of co-occurrence of neighboring nodes. In the graph embedding task, it is able to capture semantic relationships between nodes.

In the application, the Skip-gram model aims at learning an embedded vector (namely product node characteristics) of each product node by maximizing the probability of node co-occurrence, and the vector can be used for representing the semantic relation of the product in the interaction graph. The model reflects their similarity in the interaction graph by constantly adjusting the embedded vectors so that nodes that frequently occur together in the sequence have similar vector representations.

The generated node characteristics of the product not only reflect the static information (such as category, price and the like) of the product, but also contain the dynamic association information learned through the client behavior and interaction relationship.

In this embodiment, by using a random walk algorithm and Skip-gram model, the feature representation of the product nodes can capture the relevance between products more accurately, and the features learned through the client behavior and interaction data can reflect the potential similarity of the products more accurately, so that the accuracy of understanding and predicting the preference of the clients by the recommendation system is enhanced.

Further, the intelligent option management method based on machine learning can further comprise the steps of determining neighbor nodes of the product nodes for each product node in the customer product interaction diagram, calculating node similarity between the product nodes and the neighbor nodes, determining attention weights of the neighbor nodes according to the obtained node similarity, aggregating product node characteristics of the neighbor nodes according to the obtained attention weights to obtain aggregated neighbor information, and updating the product node characteristics of the product nodes according to the aggregated neighbor information.

Specifically, for each product node in the customer product interaction graph, searching nodes directly connected with the product node in the customer product interaction graph, wherein the nodes are neighbor nodes of the target node. Neighbor nodes generally refer to products that have a strong contact with the target product during a customer's purchase, browsing, or other interaction. For example, a certain product A may be a neighbor with products B, C, D, etc., because they are often purchased or viewed by the same group of customers.

Similarity is calculated for each pair of product nodes and its neighbors, for example by cosine similarity or Jaccard similarity based on customer interactions. The higher the similarity, the closer the association of the two products in the customer interaction.

And distributing an attention weight to each neighbor node according to the calculated node similarity. The attention weight reflects the importance of the neighbor node in the updating of the node characteristics of the target product. The neighbor nodes with high similarity have higher weights, and the neighbor nodes with low similarity have lower weights.

The attention weights are used to weight and sum the features of the neighbor nodes to obtain a new feature representation, namely the aggregated neighbor information. The weighted feature can better capture the influence of the neighbor nodes in the recommendation, thereby providing more accurate product recommendation.

The product node characteristics of the product nodes are updated according to the aggregated neighbor information, and after the product node characteristics are updated, the product node characteristics not only reflect the information of the product nodes, but also integrate the information of the neighbor nodes related to the product node characteristics, so that the system can be helped to more accurately understand and recommend the products possibly interested by the user.

In the embodiment, the neighbor nodes of the product nodes are determined, the node similarity between the product nodes and each neighbor node is calculated, the attention weight of each neighbor node is determined according to the node similarity, the characteristics of the neighbor nodes are introduced, the characteristics are weighted and aggregated according to the node similarity and the attention mechanism, the potential relation among the products is well excavated, the product node characteristics of the products are more comprehensively represented, the potential interests of the users can be more accurately reflected, the deep relation among the products is captured, and more personalized and accurate product recommendation is well provided.

Further, the step S205 may include extracting, for each product, a product embedding vector of the product from a product image of the product, and merging the product embedding vector with product node features of product nodes corresponding to the product to obtain product comprehensive features of the product.

Specifically, for each product, a product embedding vector of the product is extracted from the product portrait generated before, which is a low-dimensional vector representation generated by embedding various information of the product, captures various characteristics of the product, such as static characteristics (such as brands, categories and prices) of the product, and possibly fuses behavior data (such as purchase history and browsing records) of the user.

And obtaining product node characteristics of product nodes corresponding to the products, and combining the product embedded vectors with the product node characteristics, wherein the combination is used for realizing comprehensive fusion of information. The product node characteristics can comprise basic information and historical behavior data of the product, and the product embedding vector captures potential relations between the product and other products and users in a more complex mode. When the two information are combined, the two information can be combined in a splicing and weighting mode to form a more multidimensional comprehensive characteristic representation, namely the product comprehensive characteristic of the product.

The comprehensive characteristics of the products are final representations of the products used in the recommendation system, each product can be described more accurately, the static attribute and the dynamically generated behavior characteristics are combined, richer information can be provided for a subsequent recommendation model, and the system is helped to recommend users more accurately.

In the embodiment, the product embedded vector of the product is combined with the product node characteristic, the generated product comprehensive characteristic can more comprehensively reflect the multidimensional information of the product, the fused characteristic enables a recommendation system to more accurately understand the potential relationship between the product and a user, the dominant attribute and the recessive relationship of the product can be captured at the same time, the understanding depth of the product is improved, more accurate personalized recommendation is provided, and the accuracy and the user satisfaction of recommendation are enhanced.

The embodiment of the application can acquire and process the related data based on the artificial intelligence technology. Wherein artificial intelligence (ARTIFICIAL INTELLIGENCE, AI) is the theory, method, technique, and application system that uses a digital computer or a digital computer-controlled machine to simulate, extend, and expand human intelligence, sense the environment, acquire knowledge, and use knowledge to obtain optimal results.

Artificial intelligence infrastructure technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a robot technology, a biological recognition technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and other directions.

Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by computer readable instructions stored in a computer readable storage medium that, when executed, may comprise the steps of the embodiments of the methods described above. The storage medium may be a nonvolatile storage medium such as a magnetic disk, an optical disk, a Read-Only Memory (ROM), or a random access Memory (Random Access Memory, RAM).

It should be understood that, although the steps in the flowcharts of the figures are shown in order as indicated by the arrows, these steps are not necessarily performed in order as indicated by the arrows. The steps are not strictly limited in order and may be performed in other orders, unless explicitly stated herein. Moreover, at least some of the steps in the flowcharts of the figures may include a plurality of sub-steps or stages that are not necessarily performed at the same time, but may be performed at different times, the order of their execution not necessarily being sequential, but may be performed in turn or alternately with other steps or at least a portion of the other steps or stages.

With further reference to fig. 3, as an implementation of the method shown in fig. 2, the present application provides an embodiment of an intelligent option management apparatus based on machine learning, where the embodiment of the apparatus corresponds to the embodiment of the method shown in fig. 2, and the apparatus may be specifically applied to various electronic devices.

As shown in fig. 3, the intelligent option management device 300 based on machine learning according to the present embodiment includes a preference generation module 301, a product generation module 302, an interaction construction module 303, a feature generation module 304, a comprehensive generation module 305, a combination calculation module 306, and a product recommendation module 307, wherein:

The preference generation module 301 is configured to generate a customer preference portrait of each customer through a preset portrait generation algorithm based on historical purchase data and browsing data of each customer.

The product generating module 302 is configured to obtain product information of each product, and perform attribute embedding and knowledge graph embedding on the product information of each product to obtain a product image of each product.

The interaction construction module 303 is configured to construct a customer product interaction diagram according to the obtained historical purchase data and browsing data, where the customer product interaction diagram includes customer nodes and product nodes.

And the feature generation module 304 is configured to perform representation learning on the customer product interaction graph to obtain product node features of each product node.

The comprehensive generation module 305 is configured to generate product comprehensive features of each product according to the product representation of each product and the product node features of the corresponding product nodes.

And the combination calculation module 306 is used for respectively combining the customer preference portrait of the target customer and the product comprehensive characteristics of each product to obtain combination characteristics, and inputting each combination characteristic into the matching model to obtain the interestingness between the target customer and each product.

The product recommending module 307 is configured to determine a target product among the products according to the obtained interestingness, and recommend a product to the target client according to the determined target product.

In some alternative implementations of the present embodiment, the preference generation module 301 may include a customer determination sub-module, a purchase determination sub-module, a similarity calculation sub-module, a browse determination sub-module, and a preference generation sub-module, wherein:

And the client determination sub-module is used for calculating the client similarity among the clients through a collaborative filtering algorithm based on the historical purchase data of the clients, determining the similar clients of the target clients from the clients according to the obtained client similarity.

And the purchase determination submodule is used for determining purchase preference products of the similar clients according to the historical purchase data of the similar clients and constructing a first product set of the target clients according to the purchase preference products.

And the similarity calculation sub-module is used for calculating the product similarity between the products through a content-based recommendation algorithm based on the browsing data of each client.

And the browsing determination submodule is used for determining browsing preference products similar to the current browsing products of the target clients according to the obtained product similarity, and adding the browsing preference products into the first product set to obtain a second product set of the target clients.

And the preference generation sub-module is used for generating a client preference portrait of the target client according to the second product set, the historical purchase data and the browse data of the target client and obtaining the client preference portrait of each client.

In some alternative implementations of the present embodiment, the preference generation module 301 may further include an association determination sub-module and an association addition sub-module, wherein:

and the association determination submodule is used for processing the historical purchase data of each customer through an association rule mining algorithm to obtain product purchase association information.

And the association adding sub-module is used for adding the purchase association products in the second product set according to the product purchase association information.

In some alternative implementations of the present embodiment, the product generation module 302 may include a preprocessing sub-module, an attribute extraction sub-module, an attribute mapping sub-module, a relationship determination sub-module, and an embedded fusion sub-module, where:

And the preprocessing sub-module is used for preprocessing unstructured data in the product information of each product to obtain structured data of each product.

And the attribute extraction sub-module is used for extracting attribute information from the structured data of each product to obtain the structured attribute data of each product.

And the attribute mapping sub-module is used for mapping the structured attribute data of each product into attribute embedded representations respectively.

And the relation determination submodule is used for determining the association relation among the products through similarity calculation and a clustering algorithm based on the attribute embedded representation of each product, and constructing a product knowledge graph according to the obtained association relation.

And the embedding fusion sub-module is used for fusing the attribute embedding representation and the knowledge map embedding representation of each product to obtain the product image of each product.

In some alternative implementations of the present embodiment, the feature generation module 304 may include a sequence generation sub-module and a feature determination sub-module, wherein:

And the sequence generation sub-module is used for generating a node sequence on the customer product interaction graph through a random walk algorithm.

And the characteristic determination submodule is used for inputting the node sequence into the skip-gram model so as to generate the product node characteristics of each product node in the customer product interaction diagram by maximizing the node co-occurrence probability.

In some alternative implementations of the present embodiment, the feature generation module 304 may further include a neighbor determination sub-module, a similarity calculation sub-module, a weight determination sub-module, an aggregate determination sub-module, and a feature update sub-module, wherein:

And the neighbor determination submodule is used for determining neighbor nodes of the product nodes for each product node in the customer product interaction graph.

And the similarity calculation sub-module is used for calculating the node similarity between the product node and each neighbor node.

And the weight determination submodule is used for determining the attention weight of each neighbor node according to the obtained node similarity.

And the aggregation determination submodule is used for aggregating the product node characteristics of each neighbor node according to the obtained attention weight to obtain aggregation neighbor information.

And the characteristic updating sub-module is used for updating the product node characteristics of the product node according to the aggregated neighbor information.

In some alternative implementations of the present embodiment, the comprehensive generation module 305 may include a vector extraction sub-module and a merging sub-module, where:

and the vector extraction submodule is used for extracting the product embedding vector of the product from the product image of the product for each product.

And the merging sub-module is used for merging the product embedding vector with the product node characteristics of the product nodes corresponding to the product to obtain the product comprehensive characteristics of the product.

In order to solve the technical problems, the embodiment of the application also provides computer equipment. Referring specifically to fig. 4, fig. 4 is a basic structural block diagram of a computer device according to the present embodiment.

The computer device 4 comprises a memory 41, a processor 42, a network interface 43 communicatively connected to each other via a system bus. It is noted that only a computer device 4 having a memory 41, a processor 42, a network interface 43 is shown in the figures, but it is understood that not all illustrated components are required to be implemented and that more or fewer components may be implemented instead. It will be appreciated by those skilled in the art that the computer device herein is a device capable of automatically performing numerical calculation and/or information processing according to a preset or stored instruction, and its hardware includes, but is not limited to, a microprocessor, an Application SPECIFIC INTEGRATED Circuit (ASIC), a Programmable gate array (Field-Programmable GATE ARRAY, FPGA), a digital Processor (DIGITAL SIGNAL Processor, DSP), an embedded device, and the like.

The computer equipment can be a desktop computer, a notebook computer, a palm computer, a cloud server and other computing equipment. The computer equipment can perform man-machine interaction with a user through a keyboard, a mouse, a remote controller, a touch pad or voice control equipment and the like.

The memory 41 includes at least one type of readable storage medium including flash memory, hard disk, multimedia card, card memory (e.g., SD or DX memory, etc.), random Access Memory (RAM), static Random Access Memory (SRAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), programmable Read Only Memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the storage 41 may be an internal storage unit of the computer device 4, such as a hard disk or a memory of the computer device 4. In other embodiments, the memory 41 may also be an external storage device of the computer device 4, such as a plug-in hard disk, a smart memory card (SMART MEDIA CARD, SMC), a Secure Digital (SD) card, a flash memory card (FLASH CARD) or the like, which are provided on the computer device 4. Of course, the memory 41 may also comprise both an internal memory unit of the computer device 4 and an external memory device. In this embodiment, the memory 41 is generally used for storing an operating system and various application software installed on the computer device 4, such as a computer readable instruction of an intelligent option management method based on machine learning. Further, the memory 41 may be used to temporarily store various types of data that have been output or are to be output.

The processor 42 may be a central processing unit (Central Processing Unit, CPU), controller, microcontroller, microprocessor, or other data processing chip in some embodiments. The processor 42 is typically used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is configured to execute computer readable instructions stored in the memory 41 or process data, for example, execute computer readable instructions of the intelligent option management method based on machine learning.

The network interface 43 may comprise a wireless network interface or a wired network interface, which network interface 43 is typically used for establishing a communication connection between the computer device 4 and other electronic devices.

The computer device provided in the embodiment may execute the above-mentioned intelligent option management method based on machine learning. The intelligent option management method based on machine learning may be an intelligent option management method based on machine learning according to the above embodiments.

The present application also provides another embodiment, namely, a computer readable storage medium storing computer readable instructions executable by at least one processor to cause the at least one processor to perform the steps of a machine learning-based intelligent option management method as described above.

The application provides an intelligent option management system based on machine learning, which is also a recommendation system and comprises computer equipment and a computer readable storage medium, wherein the computer equipment is provided with an intelligent option management device based on machine learning.

From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) comprising instructions for causing a terminal device (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to perform the method according to the embodiments of the present application.

It is apparent that the above-described embodiments are only some embodiments of the present application, but not all embodiments, and the preferred embodiments of the present application are shown in the drawings, which do not limit the scope of the patent claims. This application may be embodied in many different forms, but rather, embodiments are provided in order to provide a thorough and complete understanding of the present disclosure. Although the application has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that modifications may be made to the embodiments described in the foregoing description, or equivalents may be substituted for elements thereof. All equivalent structures made by the content of the specification and the drawings of the application are directly or indirectly applied to other related technical fields, and are also within the scope of the application.

Claims

1. The intelligent option management method based on machine learning is characterized by comprising the following steps:

2. The intelligent option management method based on machine learning according to claim 1, wherein the step of generating the customer preference portraits of the customers through a preset portrayal generation algorithm based on the historical purchase data and browsing data of the customers comprises:

3. The machine learning based intelligent choice management method of claim 2, further comprising, after said step of obtaining said second set of products for said target customer:

4. The intelligent option management method based on machine learning according to claim 1, wherein the step of performing attribute embedding and knowledge graph embedding on the product information of each product to obtain a product representation of each product comprises:

5. The intelligent option management method based on machine learning of claim 1, wherein the step of performing representation learning on the customer product interaction graph to obtain product node characteristics of each product node comprises:

6. The machine learning based intelligent choice management method of claim 5, further comprising:

calculating node similarity between the product node and each neighboring node;

7. The intelligent option management method based on machine learning according to claim 1, wherein the step of generating the product comprehensive characteristics of each product according to the product representation of each product and the product node characteristics of the corresponding product node comprises:

8. An intelligent option management device based on machine learning, which is characterized by comprising:

9. A computer device comprising a memory and a processor, the memory having stored therein computer readable instructions which when executed by the processor implement the steps of a machine learning based intelligent option management method as claimed in any one of claims 1 to 7.

10. A computer readable storage medium having stored thereon computer readable instructions which when executed by a processor implement the steps of a machine learning based intelligent option management method according to any of claims 1 to 7.