Product recommendation method, device, equipment, storage medium and program product

Abstract

The application discloses a product recommendation method and device, equipment, a storage medium and a program product, wherein the method comprises the following steps: performing entity extraction and attribute extraction based on the structure data and non-structure data of a plurality of products to obtain the entities and entity metadata corresponding to the plurality of products; determining a product knowledge base based on the entities and entity metadata corresponding to the plurality of products; determining a user product preference matrix based on the user behavior data of a plurality of users and the product knowledge base; performing correlation analysis on the plurality of users and the plurality of products based on the user product preference matrix to obtain the correlation coefficients between the plurality of users and the plurality of products; and performing product recommendation on the plurality of users based on the correlation coefficients between the plurality of users and the plurality of products and a product penalty coefficient.

Description

Technical Field

[0001] This application relates to the field of cloud services, and more particularly to a product recommendation method, apparatus, device, storage medium, and program product. Background Technology

[0002] With the increasing number and variety of public cloud products covering computing, storage, networking, security, big data, and other fields, a key challenge facing public clouds is how to recommend suitable products to both new and existing users to improve user satisfaction. However, current product recommendation methods are limited to common algorithms or basic improvements on existing algorithms, making them less adaptable to continuous product updates and iterations. This results in incomplete data mining of product and user information, hindering the acquisition of more effective data. Summary of the Invention

[0003] To address the aforementioned technical problems, embodiments of this application provide a product recommendation method, apparatus, device, storage medium, and program product.

[0004] The product recommendation method provided in this application includes:

[0005] Entity extraction and attribute extraction are performed based on the structured and unstructured data of multiple products to obtain the entities and entity metadata corresponding to the multiple products.

[0006] Based on the entities and entity metadata corresponding to the multiple products, a product knowledge base is determined;

[0007] Based on user behavior data from multiple users and the product knowledge base, a user product preference matrix is ​​determined.

[0008] Based on the user product preference matrix, an association analysis is performed on the multiple users and the multiple products to obtain the correlation coefficients between the multiple users and the multiple products;

[0009] Product recommendations are made to the multiple users based on the correlation coefficients between the multiple users and the multiple products, and the product penalty coefficients.

[0010] The product recommendation device provided in this application embodiment includes:

[0011] The processing unit is used to extract entities and attributes based on the structured and unstructured data of multiple products, and obtain the entities and entity metadata corresponding to the multiple products.

[0012] The determining unit is used to determine a product knowledge base based on the entities and entity metadata corresponding to the multiple products; and to determine a user product preference matrix based on user behavior data of multiple users and the product knowledge base.

[0013] The processing unit is further configured to perform correlation analysis on the multiple users and the multiple products based on the user product preference matrix to obtain the correlation coefficient between the multiple users and the multiple products; and to make product recommendations to the multiple users based on the correlation coefficient and product penalty coefficient between the multiple users and the multiple products.

[0014] The processing device provided in this application includes a processor and a memory, the memory being used to store a computer program, and the processor being used to call and run the computer program stored in the memory to execute any of the above-described product recommendation methods.

[0015] The computer-readable storage medium provided in this application embodiment is used to store a computer program that causes a computer to execute any of the above-described product recommendation methods.

[0016] The computer program product provided in this application includes computer program instructions that cause a computer to execute any of the above-described product recommendation methods.

[0017] In the technical solution of this application embodiment, entity extraction and attribute extraction are performed based on the structured and unstructured data of multiple products to obtain entities and entity metadata corresponding to multiple products; a product knowledge base is determined based on the entities and entity metadata corresponding to multiple products; a user product preference matrix is ​​determined based on user behavior data of multiple users and the product knowledge base; correlation analysis is performed on multiple users and multiple products based on the user product preference matrix to obtain the correlation coefficient between multiple users and multiple products; and product recommendations are made for multiple users based on the correlation coefficient and product penalty coefficient between multiple users and multiple products. In this way, by leveraging structured and unstructured data from public cloud products, we can deeply explore the relationships between products, extract entities and attributes from product data to construct product vector models for the cloud product domain, obtain semantic relationships between products and between products and attributes, and fully mine user behavior data to obtain relationships between users and between users and products, allowing users of the same type to be clustered together. By mixing product vector models with user behavior data and combining product penalty coefficients to obtain improved Pearson correlation coefficients, we can build entity vectorization and user similarity models between products and between products and attributes for hybrid recommendation. This solves the problems of incomplete user and product information mining and cold start in recommendation systems, enabling accurate and rapid recommendation of suitable products to users. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the product recommendation method provided in the embodiments of this application;

[0019] Figure 2This is a flowchart illustrating the hybrid recommendation method based on cloud products provided in an embodiment of this application;

[0020] Figure 3 This is a schematic diagram of the product recommendation device provided in the embodiments of this application;

[0021] Figure 4 This is a schematic diagram of the processing device provided in the embodiments of this application. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0023] It should be noted that, in the embodiments of this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, in the embodiments of this application, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0024] In the description of the embodiments of this application, the term "correspondence" may indicate that there is a direct or indirect correspondence between two things, or that there is an association between two things, or that there is a relationship of instruction and being instructed, configuration and being configured, etc.

[0025] To facilitate understanding of the technical solutions of the embodiments of this application, the relevant technologies of the embodiments of this application are described below. The following relevant technologies are optional solutions and can be combined with the technical solutions of the embodiments of this application in any way, and they all fall within the protection scope of the embodiments of this application.

[0026] The number of public cloud products is increasing daily, and their business offerings are becoming increasingly diverse, covering various fields such as computing, storage, networking, security, and big data. Currently, some public cloud vendors offer recommendations to users. However, how to recommend suitable products to both new and existing users and improve user satisfaction is a significant challenge facing the public cloud market. The recommendation algorithms used in relevant methods are mostly common algorithms and some basic improvements. Essentially, they can be divided into three types: content-based recommendation algorithms, collaborative filtering-based recommendation algorithms, and model-based recommendation algorithms.

[0027] The basic principle of content-based recommendation is to discover correlations between items or content based on the content's metadata, and then recommend similar items to the user based on their preferences. For example, if a user purchases an Elastic Public IP cloud product, the content-based recommendation algorithm might discover a strong correlation between the Elastic Public IP and IPv6 bandwidth in terms of content, and thus provide recommendations. Since this recommendation method doesn't require user data, it avoids cold start and sparsity issues. However, due to the priority given to the content's attributes, it's difficult to effectively obtain more data.

[0028] Collaborative filtering-based recommendation algorithms can be divided into user-based and item-based algorithms. User-based collaborative filtering algorithms discover similar users to recommend to, by calculating user similarity to recommend users with similar tastes. Item-based collaborative filtering algorithms calculate the similarity between items and then recommend similar items to users based on their historical preference information, such as ratings. This algorithm can better support users' historical preferences and is widely used; however, because it relies on historical data, it suffers from the cold start problem, and the recommendation effect also depends on users' historical preferences.

[0029] Model-based recommendation algorithms typically use machine learning algorithms to build models of target users, then predict and score user preferences for recommendations. Commonly used machine learning models include linear regression, support vector machines, decision trees, random forests, and convolutional neural networks. This training method requires large datasets, has a relatively long training process, and necessitates continuous updates with the latest data to fine-tune the model.

[0030] The accuracy of the aforementioned machine learning models largely depends on the training set, making the selection of the training set crucial. Currently, structured data is commonly used, but it is often small in volume and has a relatively fixed structure. Therefore, it is particularly important to find ways to extract effective information from large amounts of unstructured data to improve the accuracy of recommendation algorithms.

[0031] With numerous public cloud products available, many of which are interdependent and interconnected, and even the same product existing in different specifications, coupled with continuous product updates and iterations, the demand for public cloud products is growing as the user base expands. The increasing volume of product and user data makes it crucial to extract valuable information from this vast amount of data to create datasets with key characteristics for user recommendations. Therefore, more personalized product recommendations require a highly accurate recommendation method that aligns with the needs and data characteristics of both new and existing users, allowing users to find suitable cloud services more quickly and ultimately improving user satisfaction.

[0032] To address the aforementioned technical issues, this application proposes a product recommendation method. It combines the data characteristics of public cloud products and massive amounts of user data to extract a dataset with product relationships. This dataset serves as input for product recommendations. Compared to the original, discrete, massive data, the extracted data can initially reflect the product preferences of users with similar interests. Based on user data from public cloud products, an improved Pearson correlation coefficient is obtained by incorporating a popular product penalty. This method combines entity vectorization between products and between products and attributes with user similarity models, enabling faster and more accurate product recommendations. This solves the problems of incomplete user and product information mining in recommendation systems and the cold start problem of single collaborative filtering recommendation models.

[0033] To facilitate understanding of the technical solutions of the embodiments of this application, the technical solutions of this application are described in detail below through specific embodiments. The above-mentioned related technologies are optional solutions and can be arbitrarily combined with the technical solutions of the embodiments of this application, all of which fall within the protection scope of the embodiments of this application. The embodiments of this application include at least some of the following contents.

[0034] This application provides a product recommendation method. Figure 1 This is a flowchart illustrating the product recommendation method provided in the embodiments of this application, such as... Figure 1 As shown, the method includes the following steps:

[0035] Step 101: Based on the structured and unstructured data of multiple products, perform entity extraction and attribute extraction to obtain the entities and entity metadata corresponding to multiple products.

[0036] In this embodiment, product data in relevant fields (such as network security, big data, etc.) is first obtained, including structured and unstructured data of multiple products. To extract an abstract representation of the product data, feature extraction is required on the structured and unstructured data of multiple products to obtain feature information of multiple products. The feature information of each product includes one or more features, and there is a corresponding relationship between each product and its one or more features. In order to convert each product and its one or more features into fields for data models, entity extraction and attribute extraction are required on multiple products and their feature information to obtain the entity and entity metadata corresponding to each product in the multiple products.

[0037] In this context, an entity refers to a specific transaction or concept with a unique identifier, used to identify each different product. Entity metadata describes the structured information of an entity; it is an abstract definition of the entity, used to define the entity's common characteristics, including basic information such as entity type, entity attributes, and entity attribute values. For example, in the concept of cloud products, an entity is a specific instance or concept node, and entity metadata is all the key-value pairs describing that node. Furthermore, in a knowledge graph, an entity is a node in the graph, responsible for "connecting edges," and entity metadata is the complete attribute combination of the node, responsible for "filtering" or "displaying."

[0038] In some implementations, step 101 specifically includes:

[0039] Feature extraction is performed on the structural and unstructured data of multiple products to obtain the feature information of multiple products;

[0040] A product matrix is ​​determined based on multiple products and their characteristic information.

[0041] Entity and attribute extraction are performed based on the product ontology model and product matrix to obtain entities and entity metadata corresponding to multiple products.

[0042] Specifically, the process begins by extracting features from both structured and unstructured data of multiple products to obtain their feature information. Based on the correspondence between each product and one or more of its features, a product matrix is ​​constructed. Next, based on product characteristics, a product ontology model is defined in the form of a quintuple, including a set of product concepts, a hierarchical structure of product concepts, a set of product attributes, rules governing relationships between products, and product instances. Then, based on the core elements of the defined product ontology model, entity extraction and attribute extraction are performed on each product and its features in the product matrix to obtain the entity and entity metadata corresponding to each product in the multiple products.

[0043] The product ontology model consists of several key elements. The product concept set refers to the set of product categories, with each category containing products of the same type. The hierarchical structure of product concepts refers to the parent-child relationships between product categories. The set of product attributes refers to the set of attribute information included in each product. The rules governing relationships between products refer to the connections or degrees of connection between one product and another. The product ontology model is a semantic data model that uses structured attribute information (the aforementioned product concept set, hierarchical structure of product concepts, set of product attributes, rules governing relationships between products, and product instances) to represent the inherent characteristics of a product. In other words, it answers the question "What kind of product is this?" based on the product's own physical attributes. This structured attribute information constitutes the multiple core elements of the product ontology model.

[0044] The product matrix includes the correspondence between each product and its feature information, meaning there is a one-to-one correspondence between each product and each of its features.

[0045] In some implementations, entity extraction and attribute extraction are performed based on the product ontology model and product matrix to obtain entities and entity metadata corresponding to multiple products, including:

[0046] Identify the core elements corresponding to multiple products in the product ontology model;

[0047] Based on the correspondence between multiple core elements and the feature information of each product, entity extraction and attribute extraction are performed on the product matrix to obtain the entities and entity metadata corresponding to multiple products.

[0048] Here, since the product ontology model is defined in the form of quintuples based on product characteristics, we can first determine the multiple core elements included in the product ontology model, that is, the multiple core elements corresponding to multiple products. Each core element specifies which attributes each product can include, what relationships can exist between products, and what hierarchical relationships can exist between the attributes of products. For example, a cloud product can include attributes such as functional specifications, computing specifications, network specifications, pricing calculation, and lifecycle. The relationships between one cloud product and another cloud product can include resource dependencies, mounting relationships, compatibility, subclassing, etc., represented in the form of triples such as (FunctionCompute, dependsOn, LogService), (Disk, attachTo, ECS), (c6.large, compatibleWith, CentOS7), and (ECS, subProductOf, Compute).

[0049] Based on the above provisions and the one-to-one correspondence between each product in the product matrix and each of its features, entity extraction and attribute extraction can be performed on the product and feature information in the product matrix to obtain the entity and entity metadata corresponding to each product in multiple products, i.e., the attribute information corresponding to the entity.

[0050] Step 102: Determine the product knowledge base based on the entities and entity metadata corresponding to multiple products.

[0051] In this embodiment of the application, after obtaining the entity and entity metadata corresponding to each product, the entity and entity metadata corresponding to each product can be represented in the form of triples to represent the product entity and its own attributes as well as the association between the product entity and the product entity. Each triple is then combined to form a product knowledge graph, i.e., a product knowledge base, in the relevant field.

[0052] In some implementations, step 102 specifically includes:

[0053] Based on the entity metadata corresponding to multiple products, determine the entity attributes and entity attribute values ​​for each product among the multiple products;

[0054] Based on the entity, entity attributes, and entity attribute values ​​corresponding to each product, determine the association between each product and the entities corresponding to other products, the association between the entity and entity attributes corresponding to each product, and the association between the attributes corresponding to each product.

[0055] The product knowledge base is determined based on the relationships between each product and the entities corresponding to other products, the relationships between each product's entities and entity attributes, and the relationships between each product's attributes.

[0056] Specifically, firstly, based on the entity metadata corresponding to each product, the entity attributes and entity attribute values ​​corresponding to each product are determined. Secondly, based on the entity, entity attributes, and entity attribute values ​​corresponding to each product, the relationships between each product entity and its own attributes, the relationships between each product attribute, and the relationships between each product entity are determined, and represented in the form of triples, namely (entity, relation, entity), (entity, relation, attribute value), (attribute, relation, attribute), (entity, attribute, attribute value). After obtaining the above triple-form relationships, each triple can be merged sequentially according to the relational order to form a product knowledge graph, in which there are relationships between data.

[0057] Step 103: Determine the user product preference matrix based on user behavior data from multiple users and the product knowledge base.

[0058] In this embodiment of the application, after obtaining the product knowledge base of multiple products, user behavior data of multiple users for multiple products can be obtained, such as a user's browsing history, rating, click volume, etc. for a product. Combined with the relationship between each product entity and its own attributes, each product entity and other product entities, and the relationship between each product attribute in the product knowledge base, the product preferences of multiple users are comprehensively analyzed to obtain a user product preference matrix of multiple users for multiple products.

[0059] In this matrix, each row of the user product preference matrix corresponds to a user's preference for all products, and each column corresponds to a user's preference for the same product.

[0060] In some implementations, step 103 specifically includes:

[0061] Knowledge is extracted from the product knowledge base to obtain multiple sets of entity attribute relationships;

[0062] The product characteristic matrix is ​​obtained by representing the set of multiple entity attribute relationships using vectors.

[0063] Based on user behavior data and product characteristic matrix, determine user product preference matrix.

[0064] Here, we can first extract knowledge from the product knowledge base to obtain multiple sets of relationships between product entities and product attributes that are more precise, granular, and have higher confidence, namely, multiple entity attribute relationship sets. Each entity attribute relationship set is then represented as a vector, and a product feature matrix is ​​formed. Next, we combine user behavior data of multiple users for multiple products with the product feature matrix for comprehensive analysis. That is, based on the attribute information of each product and each user's browsing, rating, clicking and other behaviors for each product, we analyze each user's preference for each product and represent each user's preference for each product in vector form to form a user product preference matrix for multiple products.

[0065] In some implementations, a user product preference matrix is ​​determined based on user behavior data and a product characteristic matrix, including:

[0066] Based on user behavior data, we analyze each user's product preferences to obtain the product preference results for each user;

[0067] Based on each user's product preference results and product characteristic matrix, determine the user product preference matrix.

[0068] Specifically, behavioral feature engineering can be constructed based on each user's behavioral data on multiple products. This involves processing each user's behavioral data on multiple products into a "behavioral feature engineering," or user-product behavior matrix. Next, preference modeling is performed on each user's product behavior matrix. A preset discrete model is used to predict based on each user's product behavior matrix. By analyzing each user's selection behavior, the preference results for each user on each product are output. Finally, the preference results for each user on each product and the product feature matrix are comprehensively calculated to obtain the user-product preference matrix.

[0069] The preset discrete model is a discrete selection model, such as the Logistic Regression (Logit) model or the Probit model, and there are no restrictions on it here.

[0070] In some implementations, a user product preference matrix is ​​determined based on each user's product preference results and product characteristic matrix, including:

[0071] Based on each user's product preference results, determine the product characteristics preferred by each user;

[0072] Based on each user's preferred product characteristics, determine the user product characteristic preference matrix;

[0073] The user product preference matrix is ​​determined based on the user product feature preference matrix and the product feature matrix.

[0074] Specifically, based on each user's preference for each product, the product characteristics (i.e., product attributes) preferred by each user for each product can be determined. Then, these product characteristics are represented as vectors, forming a user product characteristic preference matrix. Next, the user product characteristic preference matrix and the product characteristic matrix are aligned in dimensions, and the product characteristics preferred by each user for each product and the corresponding product characteristics of each product are numerically transformed and normalized using the same set of semantic dimensions, resulting in a unified representation of the user product characteristic preference matrix and the product characteristic matrix. Finally, the unified representation of the user product characteristic preference matrix and the product characteristic matrix is ​​multiplied by the inner product ...

[0075] Step 104: Based on the user product preference matrix, perform correlation analysis on multiple users and multiple products to obtain the correlation coefficients between multiple users and multiple products.

[0076] In some implementations, step 104 specifically includes:

[0077] Determine the average degree of preference for each user among multiple products based on the user product preference matrix;

[0078] For each of the multiple products, determine each user's preference for that product based on the user product preference matrix;

[0079] Based on each user's average preference for multiple products and each user's preference for a particular product, a correlation analysis is performed between each user and the product to obtain the correlation coefficient between each user and the product.

[0080] Here, for each user, the user's preference for each product is first determined based on the user product preference matrix. Then, the user's preference for each product is summed and averaged to obtain the user's average preference for multiple products. Then, for each product, a comprehensive correlation analysis is performed based on the user's average preference for multiple products and the user's preference for the product to obtain the correlation coefficient between each user and the product.

[0081] The correlation coefficient between a user and a product ranges from 0 to 1. The larger the correlation coefficient, the stronger the correlation between the user and the product, and the greater the user's preference for the product.

[0082] Step 105: Recommend products to multiple users based on the correlation coefficients and product penalty coefficients between multiple users and multiple products.

[0083] In this embodiment of the application, after obtaining the correlation coefficients between multiple users and multiple products, a popular product penalty, i.e., a product penalty coefficient, can be introduced. Based on the correlation coefficient between each user and each product and the popular product penalty, one or more target users corresponding to each product can be determined, and one or more target users can be recommended to each corresponding product.

[0084] In some implementations, step 105 specifically includes:

[0085] For each product, the similarity between each user's preference for that product is determined based on the correlation coefficient between each user and the product and the product penalty coefficient.

[0086] Based on the similarity between each user's level of preference for the product, one or more target users are identified, and the product is recommended to one or more target users.

[0087] Specifically, for each product, the similarity between each user's preference for the product can be calculated based on the correlation coefficient between each user and the product, the penalty for popular products, and the Pearson coefficient. The higher the similarity (0-1), the more similar the users are in their preferred product types. If two users have a similarity exceeding the similarity threshold (e.g., 0.8) and both have a correlation coefficient exceeding the correlation threshold (e.g., 0.9), it means that these two users not only have a high degree of preference for the product but also have similar preferences in product types. Therefore, these two users can be clustered into the same type of users, identified as target users for the product, and the product can be recommended to these two target users.

[0088] In the technical solution of this application embodiment, entity extraction and attribute extraction are performed based on the structured and unstructured data of multiple products to obtain entities and entity metadata corresponding to multiple products; a product knowledge base is determined based on the entities and entity metadata corresponding to multiple products; a user product preference matrix is ​​determined based on user behavior data of multiple users and the product knowledge base; correlation analysis is performed on multiple users and multiple products based on the user product preference matrix to obtain the correlation coefficient between multiple users and multiple products; and product recommendations are made for multiple users based on the correlation coefficient and product penalty coefficient between multiple users and multiple products. In this way, by leveraging structured and unstructured data from public cloud products, we can deeply explore the relationships between products, extract entities and attributes from product data to construct product vector models for the cloud product domain, obtain semantic relationships between products and between products and attributes, and fully mine user behavior data to obtain relationships between users and between users and products, allowing users of the same type to be clustered together. By mixing product vector models with user behavior data and combining product penalty coefficients to obtain improved Pearson correlation coefficients, we can build entity vectorization and user similarity models between products and between products and attributes for hybrid recommendation. This solves the problems of incomplete user and product information mining and cold start in recommendation systems, enabling accurate and rapid recommendation of suitable products to users.

[0089] This application also provides a hybrid recommendation method based on cloud products. Figure 2 This is a flowchart illustrating the hybrid recommendation method based on cloud products provided in an embodiment of this application, such as... Figure 2 As shown, the method includes the following steps:

[0090] Step 201: Initialize the product matrix from the structured and unstructured data of cloud products.

[0091] Data on cloud products in relevant fields is acquired, and data mining is performed from structured and unstructured data to obtain one or more feature information for each cloud product. Based on each cloud product and its one or more feature information, a product matrix vector is constructed. Table 1 below shows an example of a product matrix provided in this application embodiment.

[0092] Table 1

[0093]

[0094] Step 202: Define the ontology model of the cloud product based on its characteristics.

[0095] Define a quintuple; use the quintuple form to construct the cloud product ontology model, defined as follows: ,in, This represents a collection of product concepts. This represents the hierarchical structure of all product concepts. Represents a set of product attributes. Rules for representing the relationships between products, This represents a single instance. First, a hierarchical definition of cloud product domain concepts is given, using the basic expression C1::C2 to declare the subclass relationships between product concepts. For example, "ECS::Cloud Host" indicates that the ECS class is a subclass of the Cloud Host class, and "Mixed Bandwidth::Shared Bandwidth" indicates that the Mixed Bandwidth class is a subclass of the Shared Bandwidth class.

[0096] Step 203: Extract product entities and entity attributes from the product matrix to form a product knowledge base.

[0097] Based on several core elements in the cloud product ontology model, entity and attribute extraction is performed on the product matrix. Product entities and their relationships are represented using triples, such as (entity, relation, entity), (entity, relation, attribute value), (attribute, relation, attribute), and (entity, attribute, attribute value). By extracting cloud product metadata into triples, a mapping relationship between entities and between entities and attributes is formed in the cloud domain knowledge graph. The extracted entities and attributes are then organized to form a product knowledge base, with interconnected data.

[0098] Step 204: Extract product attribute relationships from the product knowledge base to obtain the product characteristic matrix.

[0099] Based on the triplet form in the product knowledge base, knowledge extraction is performed on the product knowledge base to obtain multiple entity attribute relationship sets, and these multiple entity attribute relationship sets are represented as vectors to obtain the product characteristic matrix.

[0100] Step 205: Determine the user product preference matrix based on user behavior data and product characteristic matrix.

[0101] This involves acquiring user behavior data from multiple users regarding cloud products, analyzing each user's product preferences based on this data, determining the product characteristics preferred by each user, and then analyzing the correlation between users based on these product characteristics. Assume... For users Preferred products and users Preferred products The Tanimoto coefficient is used to define user correlation based on shared features. This coefficient is used to calculate the correlation of attributes with sparse features and non-binary values. The more shared attributes two users have, the larger the Tanimoto coefficient. The Tanimoto coefficient reflects the degree of correlation between two users. The larger the value, the more similar the two users are. The formula for its definition is as follows:

[0102] (1)

[0103] User behavior data can effectively reflect a user's dependence on and preference for a particular product, such as browsing history, ratings, and click-through rates. Therefore, a user-product preference matrix can be constructed based on user behavior data and product feature matrices. This matrix represents the degree of user preference for a product. Assume there are m users. n products The input dataset is an m×n user product preference matrix. Its definition formula is as follows:

[0104] (2)

[0105] Step 206: Determine the similarity between products preferred by each user based on the user product preference matrix and the popular product penalty.

[0106] This application improves upon the Pearson correlation coefficient by incorporating a penalty for popular products. Since almost all users will order a popular product, its contribution to the product similarity calculation will be relatively small. The specific calculation formula is as follows:

[0107] (3)

[0108] in, Indicates user For the product The degree of preference, Indicates user For the product The degree of preference is indicated by a value closer to 1, representing a stronger preference. Indicates user Average preference for all products Indicates user Average preference for all products Indicates user For the product The degree of preference and users For the product The similarity between user preferences is considered; the closer the similarity is to 1, the stronger the user's preference. and For the product The more similar their preferences are.

[0109] Step 207: Recommend products to each user based on the similarity between the products they prefer.

[0110] According to the above formula (3), for each product, the similarity between each user's preference for the product is determined based on the correlation coefficient between each user and the product. If the similarity between two users' preferences for the product exceeds the similarity threshold (e.g., 0.8) and the correlation coefficient between these two users and the product exceeds the correlation threshold (e.g., 0.9), then these two users can be clustered into users of the same type, and the product can be uniformly recommended to these two users.

[0111] The technical solution provided in this application embodiment extracts entities and attributes from the structured and unstructured data of the original cloud products, constructs a product vector model in the cloud product domain, and obtains semantic relationships between products and between products and attributes. This solves the problem of incomplete user and product information mining caused by directly using the original data as input data. It also fully mines user behavior information related to cloud products from the information of all registered users in the existing public cloud system, so that users of the same type are clustered together. Furthermore, it mixes the entity vectorization and user similarity models between products and between products and attributes, and adds an improved Pearson coefficient with a popular product penalty for product recommendation, making the calculated recommendation results more accurate and improving the accuracy of product recommendation. This solves the cold start problem of related single collaborative filtering algorithms.

[0112] This application also proposes a product recommendation device. Figure 3 This is a schematic diagram of the product recommendation device provided in the embodiments of this application, such as... Figure 3 As shown, the device includes:

[0113] Processing unit 301 is used to extract entities and attributes based on the structured and unstructured data of multiple products, and obtain entities and entity metadata corresponding to multiple products.

[0114] Unit 302 is used to determine the product knowledge base based on the entities and entity metadata corresponding to multiple products; and to determine the user product preference matrix based on the user behavior data and product knowledge base of multiple users.

[0115] The processing unit 301 is also used to perform correlation analysis on multiple users and multiple products based on the user product preference matrix to obtain the correlation coefficient between multiple users and multiple products; and to make product recommendations for multiple users based on the correlation coefficient between multiple users and multiple products and the product penalty coefficient.

[0116] In some embodiments, the processing unit 301 is specifically used for:

[0117] Feature extraction is performed on the structural and unstructured data of multiple products to obtain the feature information of multiple products;

[0118] A product matrix is ​​determined based on multiple products and their characteristic information.

[0119] Entity and attribute extraction are performed based on the product ontology model and product matrix to obtain entities and entity metadata corresponding to multiple products. The product ontology model refers to a semantic data model that uses structured attribute information to represent the inherent characteristics of a product.

[0120] The product matrix includes the correspondence between each product and its feature information among multiple products.

[0121] In some embodiments, the processing unit 301 is further specifically used for:

[0122] Identify the core elements corresponding to multiple products in the product ontology model;

[0123] Based on the correspondence between multiple core elements and the feature information of each product, entity extraction and attribute extraction are performed on the product matrix to obtain the entities and entity metadata corresponding to multiple products.

[0124] In some implementations, the determining unit 302 is specifically used for:

[0125] Based on the entity metadata corresponding to multiple products, determine the entity attributes and entity attribute values ​​for each product among the multiple products;

[0126] Based on the entity, entity attributes, and entity attribute values ​​corresponding to each product, determine the association between each product and the entities corresponding to other products, the association between the entity and entity attributes corresponding to each product, and the association between the attributes corresponding to each product.

[0127] The product knowledge base is determined based on the relationships between each product and the entities corresponding to other products, the relationships between each product's entities and entity attributes, and the relationships between each product's attributes.

[0128] In some embodiments, the determining unit 302 is further specifically used for:

[0129] Knowledge is extracted from the product knowledge base to obtain multiple sets of entity attribute relationships;

[0130] The product characteristic matrix is ​​obtained by representing the set of multiple entity attribute relationships using vectors.

[0131] Based on user behavior data and product characteristic matrix, determine user product preference matrix.

[0132] In some embodiments, the determining unit 302 is further specifically used for:

[0133] Based on user behavior data, we analyze each user's product preferences to obtain the product preference results for each user;

[0134] Based on each user's product preference results and product characteristic matrix, determine the user product preference matrix.

[0135] In some embodiments, the determining unit 302 is further specifically used for:

[0136] Based on each user's product preference results, determine the product characteristics preferred by each user;

[0137] Based on each user's preferred product characteristics, determine the user product characteristic preference matrix;

[0138] The user product preference matrix is ​​determined based on the user product feature preference matrix and the product feature matrix.

[0139] In some embodiments, the processing unit 301 is further specifically used for:

[0140] Determine the average degree of preference for each user among multiple products based on the user product preference matrix;

[0141] For each of the multiple products, determine each user's preference for that product based on the user product preference matrix;

[0142] Based on each user's average preference for multiple products and each user's preference for a particular product, a correlation analysis is performed between each user and the product to obtain the correlation coefficient between each user and the product.

[0143] In some embodiments, the processing unit 301 is further specifically used for:

[0144] For each product, the similarity between each user's preference for that product is determined based on the correlation coefficient between each user and the product and the product penalty coefficient.

[0145] Based on the similarity between each user's level of preference for the product, one or more target users are identified, and the product is recommended to one or more target users.

[0146] Those skilled in the art should understand that Figure 3 The functions of each unit in the product recommendation device shown can be understood by referring to the relevant descriptions of the aforementioned method. Figure 3 The functions of each unit in the product recommendation device shown can be implemented by a program running on a processor or by specific logic circuits.

[0147] Figure 4 This is a schematic diagram of the processing device provided in an embodiment of this application. The processing device may be a terminal device or a network device. Figure 4 The processing device shown includes a processor 401, which can call and run computer programs from memory to implement the methods in the embodiments of this application.

[0148] Optionally, such as Figure 4 As shown, the processing device may further include a memory 402. The processor 401 can retrieve and run computer programs from the memory 402 to implement the methods described in this embodiment.

[0149] The memory 402 can be a separate device independent of the processor 401, or it can be integrated into the processor 401.

[0150] Optionally, such as Figure 4 As shown, the processing device may also include a transceiver 403, which the processor 401 can control to communicate with other devices. Specifically, it can send information or data to other devices or receive information or data sent by other devices.

[0151] The transceiver 403 may include a transmitter and a receiver. The transceiver 403 may further include an antenna, and the number of antennas may be one or more.

[0152] The processing device may specifically be the product recommendation device in the embodiments of this application, and the processing device can implement the corresponding processes of the various methods in the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0153] It should be understood that the processor in the embodiments of this application may be an integrated circuit chip with signal processing capabilities. In implementation, the steps of the above method embodiments can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor described above can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0154] It is understood that the memory in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DR RAM). It should be noted that the memory used in the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0155] This application also provides a computer-readable storage medium for storing a computer program. This computer-readable storage medium can be applied to the processing device in this application embodiment, and the computer program causes the computer to execute the corresponding processes implemented by the various methods in this application embodiment; for brevity, further details are omitted here.

[0156] This application also provides a computer program product, including computer program instructions. This computer program product can be applied to the processing device in this application embodiment, and the computer program instructions cause the computer to execute the corresponding processes implemented by the various methods in this application embodiment; for brevity, further details are omitted here.

[0157] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0158] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0159] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0160] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0161] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0162] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0163] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A product recommendation method, characterized in that, The method includes: Entity extraction and attribute extraction are performed based on the structured and unstructured data of multiple products to obtain the entities and entity metadata corresponding to the multiple products. Based on the entities and entity metadata corresponding to the multiple products, a product knowledge base is determined; Based on user behavior data from multiple users and the product knowledge base, a user product preference matrix is ​​determined. Based on the user product preference matrix, an association analysis is performed on the multiple users and the multiple products to obtain the correlation coefficients between the multiple users and the multiple products; Product recommendations are made to the multiple users based on the correlation coefficient between the multiple users and the multiple products and the product penalty coefficient. The step of extracting entities and attributes from the structured and unstructured data of multiple products to obtain the entities and entity metadata corresponding to the multiple products includes: Feature extraction is performed on the structural and unstructured data of the multiple products to obtain the feature information of the multiple products; Based on the plurality of products and their feature information, a product matrix is ​​determined; wherein, the product matrix includes the correspondence between each of the plurality of products and its feature information; Based on the product ontology model and the product matrix, entity extraction and attribute extraction are performed to obtain the entities and entity metadata corresponding to the multiple products; wherein, the product ontology model refers to a semantic data model that uses structured attribute information to represent the inherent characteristics of a product, and the product ontology model includes a set of product concepts, a hierarchical structure of product concepts, a set of product attributes, rules for the relationship between products, and product instances. The determination of the user product preference matrix based on user behavior data from multiple users and the product knowledge base includes: Knowledge is extracted from the product knowledge base to obtain multiple sets of entity attribute relationships; The product characteristic matrix is ​​obtained by representing the set of multiple entity attribute relationships using vectors. Based on user behavior data, behavioral feature engineering is used to construct a product behavior matrix for each user. The product behavior matrix of each user is used to model preferences. A preset discrete model is used to predict based on the product behavior matrix of each user. By analyzing the selection behavior of each user, the product preference results of each user for each product are output. Based on each user's product preference results for each product, determine the product characteristics that each user prefers for each product; based on the product characteristics that each user prefers for each product, determine the user product characteristic preference matrix; Align the user product feature preference matrix and the product feature matrix with dimensions, and use the same set of semantic dimensions to perform numerical conversion and normalization on the product features that each user prefers for each product and the product features corresponding to each product, to obtain a unified representation of the user product feature preference matrix and the product feature matrix. The user product characteristic preference matrix and the product characteristic matrix of the unified representation are multiplied by the matrix inner product. The product characteristic component of each user's preference for each product is multiplied by all product characteristic components corresponding to that product in each dimension and summed to obtain the degree of preference of each user for each product. The user product preference matrix is ​​determined based on each user's preference for each product.

2. The method according to claim 1, characterized in that, The product matrix includes the correspondence between each product and its feature information among the plurality of products; The entity extraction and attribute extraction based on the product ontology model and the product matrix yields the entities and entity metadata corresponding to the multiple products, including: Identify the core elements corresponding to the multiple products in the product ontology model; Based on the correspondence between the multiple core elements and the feature information of each product, entity extraction and attribute extraction are performed on the product matrix to obtain the entities and entity metadata corresponding to the multiple products.

3. The method according to claim 1, characterized in that, The process of determining a product knowledge base based on the entities and entity metadata corresponding to the multiple products includes: Based on the entity metadata corresponding to the multiple products, determine the entity attributes and entity attribute values ​​corresponding to each of the multiple products; Based on the entity, entity attributes, and entity attribute values ​​corresponding to each product, determine the association relationships between the entities corresponding to each product and other products, the association relationships between the entities and entity attributes corresponding to each product, and the association relationships between the attributes corresponding to each product. The product knowledge base is determined based on the relationships between the entities corresponding to each product and other products, the relationships between the entities corresponding to each product and their attributes, and the relationships between the attributes corresponding to each product.

4. The method according to any one of claims 1 to 3, characterized in that, The step of performing correlation analysis on the multiple users and multiple products based on the user product preference matrix to obtain the correlation coefficients between the multiple users and multiple products includes: The average degree of preference of each user for the multiple products is determined based on the user product preference matrix. For each of the plurality of products, the degree of preference of each user for that product is determined based on the user product preference matrix; Based on each user's average preference for the multiple products and each user's preference for the product, a correlation analysis is performed between each user and the product to obtain the correlation coefficient between each user and the product.

5. The method according to claim 4, characterized in that, The step of recommending products to the multiple users based on the correlation coefficient and product penalty coefficient between the multiple users and the multiple products includes: For each product, the similarity between each user's preference for that product is determined based on the correlation coefficient between each user and the product and the product's penalty coefficient. Based on the similarity between the degree of preference of each user for the product, one or more target users are identified, and the product is recommended to the one or more target users.

6. A product recommendation device, characterized in that, The device includes: The processing unit is used to extract entities and attributes based on the structured and unstructured data of multiple products, and obtain the entities and entity metadata corresponding to the multiple products. The determining unit is used to determine a product knowledge base based on the entities and entity metadata corresponding to the multiple products; and to determine a user product preference matrix based on user behavior data of multiple users and the product knowledge base. The processing unit is further configured to perform correlation analysis on the multiple users and the multiple products based on the user product preference matrix to obtain the correlation coefficient between the multiple users and the multiple products; and to recommend products to the multiple users based on the correlation coefficient between the multiple users and the multiple products and the product penalty coefficient. The processing unit is specifically used for: Feature extraction is performed on the structural and unstructured data of the multiple products to obtain the feature information of the multiple products; Based on the plurality of products and their feature information, a product matrix is ​​determined; wherein, the product matrix includes the correspondence between each of the plurality of products and its feature information; Based on the product ontology model and the product matrix, entity extraction and attribute extraction are performed to obtain the entities and entity metadata corresponding to the multiple products; wherein, the product ontology model refers to a semantic data model that uses structured attribute information to represent the inherent characteristics of a product, and the product ontology model includes a set of product concepts, a hierarchical structure of product concepts, a set of product attributes, rules for the relationship between products, and product instances. Determine the unit, specifically for: Knowledge is extracted from the product knowledge base to obtain multiple sets of entity attribute relationships; The product characteristic matrix is ​​obtained by representing the set of multiple entity attribute relationships using vectors. Based on user behavior data, behavioral feature engineering is used to construct a product behavior matrix for each user. The product behavior matrix of each user is used to model preferences. A preset discrete model is used to predict based on the product behavior matrix of each user. By analyzing the selection behavior of each user, the product preference results of each user for each product are output. Based on each user's product preference results for each product, determine the product characteristics that each user prefers for each product; based on the product characteristics that each user prefers for each product, determine the user product characteristic preference matrix; Align the user product feature preference matrix and the product feature matrix with dimensions, and use the same set of semantic dimensions to perform numerical conversion and normalization on the product features that each user prefers for each product and the product features corresponding to each product, to obtain a unified representation of the user product feature preference matrix and the product feature matrix. The user product characteristic preference matrix and the product characteristic matrix of the unified representation are multiplied by the matrix inner product. The product characteristic component of each user's preference for each product is multiplied by all product characteristic components corresponding to that product in each dimension and summed to obtain the degree of preference of each user for each product. The user product preference matrix is ​​determined based on each user's preference for each product.

7. A processing device, characterized in that, include: A processor and a memory for storing a computer program, the processor for calling and running the computer program stored in the memory to perform the method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, Used to store a computer program that causes a computer to perform the method as described in any one of claims 1 to 5.

9. A computer program product, characterized in that, It includes computer program instructions that cause a computer to perform the method as described in any one of claims 1 to 5.

Images