WO2022126901A1

WO2022126901A1 - Commodity recommendation method and related device thereof

Info

Publication number: WO2022126901A1
Application number: PCT/CN2021/082934
Authority: WO
Inventors: 陈浩; 谯轶轩; 高鹏
Original assignee: 平安科技（深圳）有限公司
Priority date: 2020-12-18
Filing date: 2021-03-25
Publication date: 2022-06-23
Also published as: CN112633973B; CN112633973A

Abstract

Embodiments of the present application relate to the field of artificial intelligence, and relates to a commodity recommendation method. The method mainly comprises: after obtaining a historical commodity browsing record, separately determining a commodity node embedding vector and a commodity attribute node embedding node, then merging or splicing the two embedding vectors to obtain a comprehensive embedding vector of a commodity node, and performing commodity recommendation by using the comprehensive embedding vector. Since the comprehensive embedding vector represents the behavior information of a commodity and the attribute information of the commodity, compared with the prior art, the represented information is richer, and therefore, a more accurate commodity recommendation effect can be achieved. The present application further provides a commodity recommendation apparatus, a computer device, and a storage medium. In addition, the present application further relates to a blockchain technology, and the historical commodity browsing record of a user can be stored in a blockchain, thereby improving the security and stability of the storage of the historical commodity browsing record.

Description

A product recommendation method and related equipment

This application is based on the Chinese invention patent application with the application number 202011504942.5 filed on December 18, 2020, titled "A Commodity Recommendation Method and Related Equipment", and claims its priority.

technical field

The present application relates to the technical field of artificial intelligence, and in particular, to a product recommendation method, device, computer equipment and storage medium.

Background technique

Graph embedding is one of the hot research fields in recommender systems and graph social networks in recent years. By learning the information contained in the nodes in the graph, the nodes are mapped to a quantifiable space, that is, the node embedding in the graph is generated. In downstream tasks, the complex relationship between nodes in the network can be more deeply understood by performing tasks such as similarity calculation, classification, and clustering on node embedding.

When the existing graph embedding technology is applied in the product recommendation scenario, the historical browsing order records of the products can be obtained first, and a graph network of commodity nodes can be constructed based on the historical browsing order records. Then, based on the random walk method, the topology structure between the nodes in each graph in the constructed commodity node graph network is transformed into a text-like sequence structure. Further, the word2vec model combined with the method of negative sampling is used to embed the generated sequence structure into a new space, and the structural relationship of the browsing order between each node in the graph network is mined, so as to determine the embedding vector of each commodity node in the commodity node graph network, That is, the embedding vector. Finally, use the determined embedding vector of each commodity node to recommend similar commodities.

The above-determined embedding vector of commodity nodes can realize commodity recommendation to a certain extent, but the inventor found that the commodity node embedding vector determined only by the above-mentioned means only contains the order information between commodity nodes, and contains a single information. When recommended, the effect of product recommendation is not significant.

SUMMARY OF THE INVENTION

The purpose of the embodiments of the present application is to provide a method, apparatus, computer equipment and storage medium for recommending products, which are mainly used to solve the technical problem of poor product recommendation effect due to the single information contained in the embedded vector of existing product nodes.

In order to solve the above-mentioned technical problems, the embodiment of the present application provides a method for recommending commodities, which adopts the following technical solutions:

Get the historical commodity browsing records of multiple users;

Generate a sequence of commodity nodes and a sequence of commodity attribute nodes according to the historical commodity browsing records;

Combined with the negative sampling strategy, the commodity node sequence and the commodity attribute node sequence are used as training corpus, respectively input the word2vec model, and the commodity node embedding vector and the commodity attribute node embedding vector are obtained by calculation;

Perform vector splicing or vector summation operations on the commodity node embedding vector and the commodity attribute node embedding vector to obtain a comprehensive embedding vector corresponding to each commodity;

Determine the target product that the user visits;

It is determined that the product corresponding to the comprehensive embedding vector close to the comprehensive embedding vector of the target product is the recommended product determined for the target product.

In order to solve the above technical problems, the embodiment of the present application also provides a product recommendation device, which adopts the following technical solutions:

an acquisition unit, used to acquire historical commodity browsing records of multiple users;

a generating unit, generating a sequence of commodity nodes and a sequence of commodity attribute nodes according to the historical commodity browsing records;

The training unit is used for combining the negative sampling strategy, using the commodity node sequence and the commodity attribute node sequence as training corpus, respectively inputting the word2vec model, and calculating the commodity node embedding vector and the commodity attribute node embedding vector;

a computing unit, configured to perform vector splicing or vector summation operations on the commodity node embedding vector and the commodity attribute node embedding vector to obtain a comprehensive embedding vector corresponding to each commodity;

A response unit, used to determine the target commodity accessed by the user;

The recommendation unit is configured to determine the product corresponding to the comprehensive embedding vector similar to the comprehensive embedding vector of the target product, which is the recommended product determined for the target product.

In order to solve the above-mentioned technical problems, the embodiment of the present application also provides a computer device, which adopts the following technical solutions:

A computer device, comprising a memory and a processor, wherein computer-readable instructions are stored in the memory, and when the processor executes the computer-readable instructions, the processor implements the following steps of a commodity recommendation method:

Get the historical commodity browsing records of multiple users;

Determine the target product that the user visits;

It is determined that the product corresponding to the comprehensive embedding vector that is similar to the comprehensive embedding vector of the target product is the recommended product determined for the target product.

In order to solve the above technical problems, the embodiments of the present application also provide a computer-readable storage medium, which adopts the following technical solutions:

A computer-readable storage medium, wherein 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 the following commodity recommendation method are implemented:

Get the historical commodity browsing records of multiple users;

Determine the target product that the user visits;

Compared with the prior art, the embodiments of the present application mainly have the following beneficial effects:

In the embodiment of the present application, after obtaining the historical commodity browsing records of multiple users, a commodity node sequence and a commodity attribute node sequence are further generated according to the historical commodity browsing records. Then, combined with the negative sampling strategy, the commodity node sequence and commodity attribute node sequence are used as training corpus, and input into the word2vec model respectively, and the commodity node embedding vector and commodity attribute node embedding vector are calculated. Finally, perform vector splicing or vector summation operations on the commodity node embedding vector and commodity attribute node embedding vector to obtain the comprehensive embedding vector corresponding to each commodity.

To sum up, compared with the traditional graph embedding method to calculate the embedding vector of the commodity, the commodity recommendation method that integrates commodity attribute information proposed in this application can better combine the structural information and attribute information of the nodes in the graph network. In the similar commodity scenario of the recommendation system, the commodity comprehensive embedding vector calculated in this application not only contains the user's behavior information, but also contains the content information of the commodity itself, which can reflect the property information of the commodity itself. The embedding vector in the prior art can achieve a more accurate product recommendation effect.

Description of drawings

In order to illustrate the solutions in the present application more clearly, the following will briefly introduce the accompanying drawings used in the description of the embodiments of the present application. For those of ordinary skill, other drawings can also be obtained from these drawings without any creative effort.

FIG. 1 is a flowchart of an embodiment of a product recommendation method of the present application;

Fig. 2 is a flow chart of a specific implementation manner of step S120 in Fig. 1;

FIG. 3 is a flowchart of another embodiment of a product recommendation method of the present application;

FIG. 4 is a schematic diagram of an embodiment of a product recommendation device of the present application;

FIG. 5 is a schematic diagram of an embodiment of the generating unit 420;

FIG. 6 is a schematic diagram of an embodiment of a computer device of the present application.

Detailed ways

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field of this application; the terms used herein in the specification of the application are for the purpose of describing specific embodiments only It is not intended to limit the application; the terms "comprising" and "having" and any variations thereof in the description and claims of this application and the above description of the drawings are intended to cover non-exclusive inclusion. The terms "first", "second" and the like in the description and claims of the present application or the above drawings are used to distinguish different objects, rather than to describe a specific order.

Reference herein to an "embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor a separate or alternative embodiment that is mutually exclusive of other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

In order to make those skilled in the art better understand the solutions of the present application, some terms in the embodiments of the present application are briefly described below.

Embedding refers to using a low-dimensional vector to represent an object, which can be a word, a commodity, or a movie, etc. The nature of the embedding vector is to make the objects corresponding to the vectors with similar distances have similar meanings. For example, the distance between embedding (Avengers) and embedding (Iron Man) will be very close, but embedding (Avengers) and embedding (Gone with the Wind) distance will be farther.

Based on the concept of embedding, the Google team published a tool for converting words into vector form, the word2vec (word to vector) algorithm. After the large-scale text is input into the word2vec model tool as the training corpus, a unique multi-dimensional word vector can be mapped for each word included in the training corpus after the training, thereby simplifying the processing of the text content. is a vector operation in a vector space. Common, for example, word2vec can be used in semantic analysis scenarios. After two words are input into the pre-trained word2vec model, the word vectors corresponding to the two words can be determined. Furthermore, by calculating the similarity of the two words in the vector space, that is, calculating the size of the cosine similarity between the two word vectors, the semantic similarity of the two words in the text is determined. The greater the cosine similarity, the closer the semantics between the two words. word2vec provides an efficient bag-of-words and skip-gram implementation for computing vector words.

According to the above description, based on the word2vec model, an embodiment of the present application proposes a method for recommending products. The final network embedding vector not only contains the user's behavior information, but also contains the content information of the product itself, which can be used to a greater extent. It characterizes the nature of the commodity itself.

For further details, referring to FIG. 1 below, a flowchart of an embodiment of a product recommendation method of the present application is shown. The described method for recommending products includes the following steps:

Step S110, obtaining historical commodity browsing records of multiple users.

In this embodiment, a commodity recommendation method runs on an electronic device (for example, a server/terminal device) on which it can receive transmissions from an external device through a wired connection or a wireless connection, or actively collect the user's history. Product browsing history. It should be pointed out that the above wireless connection methods may include but are not limited to 3G/4G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, ultra wide band (UWB) connection, and other currently known or developed in the future. wireless connection method.

The above-mentioned commodities may cover various concepts, for example, may be actual commodities in a shopping website, or may be virtual commodities such as movies, music, videos, or games, which are not specifically limited in this embodiment. The obtained historical commodity browsing records may specifically include commodity name or commodity ID information of the commodity, and attribute information of the commodity. Among them, the attribute information of the product, or the label information of the product, can be classified in various ways according to different classification standards and classification levels. For example, commonly, commodities can be classified according to attributes and characteristics such as usage, raw materials, production methods, chemical composition, and usage status, and there are no specific restrictions on the classification standards or classification levels corresponding to the attributes of specific commodities.

In some possible implementations, the acquisition of historical commodity browsing records may be collected by means of data buried points and browsing time thresholds. Trigger the operation of collecting the browsing records of the user this time.

In some possible implementations, the acquisition of historical commodity browsing records may also be for commodities under a certain commodity attribute. product.

In some possible implementations, it should be emphasized that, in order to further ensure the privacy and security of the above-mentioned historical commodity browsing records, the above-mentioned historical commodity browsing records may also be stored in a node of a blockchain.

The blockchain referred to in this application is a new application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm. Blockchain, essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information to verify its Validity of information (anti-counterfeiting) and generation of the next block. The blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.

Step S120: Generate a commodity node sequence and a commodity attribute node sequence according to the historical commodity browsing records.

In this embodiment, after the historical commodity browsing records of multiple users recorded in the system are acquired, the historical commodity browsing records can be counted to determine the commodity node set included in the historical commodity browsing records. After that, the commodity node set is processed by the commodity recommendation method of the graph network, and the commodity node sequence is output. Finally, according to a preset classification standard of commodity attributes, the classification standard corresponds to a set of commodity attributes, and the commodity node sequence is converted into a commodity attribute node sequence. Commodity node sequence, for example: commodity 1-commodity 6-commodity 3...commodity N, commodity attribute node sequence, such as category 1-category 4-category 3...category N, where the length of each sequence, that is, the included nodes The number can be set by the user in advance.

In step S130, combining the negative sampling strategy, the commodity node sequence and the commodity attribute node sequence are used as training corpus, respectively input into the word2vec model, and the commodity node embedding vector and commodity attribute node embedding vector are calculated.

In this embodiment, after obtaining the sequence of commodity nodes and the sequence of commodity attribute nodes, referring to the training method of the word2vec model commonly used in the prior art and combining with the negative sampling strategy, the sequence of commodity nodes and the sequence of commodity attribute nodes are used as training corpus, respectively input In the word2vec model, the product node embedding vector corresponding to each product and the corresponding at least one product attribute category embedding vector are finally calculated respectively.

In some possible implementations, in order to fuse the structural information of the product in the graph network with its corresponding attribute information, so that the product and category nodes can be mapped to a unified embedding space, the negative sampling strategy formulated in this application is as follows Table 1 shown:

Table 1

In a common scenario, the structure of the graph network represents the behavior information of the user clicking on the product, and the side information represents the attribute information of the product. Especially in the recommendation scenario, the number of commodity nodes is generally much more than that of category nodes. During the training process, it is hoped that commodity nodes and category nodes are embedded in the same space, and can influence, interact, and fully learn from each other. Therefore, in the process of constructing positive and negative samples of commodity nodes, θ _r % commodity nodes and (1-θ _r )% category nodes are selected to participate in its training, and because there are fewer category nodes, in order to fully train them without losing their value For its own attribute information, this application selects θ _c % category nodes and (1-θ _c )% commodity nodes to participate in training, where θ _r >θ _c .

Step S140: Perform vector splicing or vector summation operations on the commodity node embedding vector and the commodity attribute node embedding vector to obtain a comprehensive embedding vector corresponding to each commodity.

In this embodiment, after the commodity node embedding vector E _r and the category node embedding vector E _c are obtained through training, the comprehensive embedding vector corresponding to any commodity node _ri can be calculated as:

E(r _i )=E _r (r _i )||E _c (r _i )

where || represents vector concatenation or vector sum operation.

S150: Determine the target commodity accessed by the user.

In this embodiment, the target commodity may be the commodity currently being browsed by the user, or may be the commodity in the user's recent browsing record.

S160: Determine a product corresponding to a comprehensive embedding vector similar to the comprehensive embedding vector of the target product as a recommended product determined for the target product.

In this embodiment, the comprehensive embedding vector of the commodity calculated above not only contains the behavior information of the user, but also contains the content information of the commodity itself, which can reflect the property information of the commodity itself. Therefore, the relationship between the commodity and the target commodity can be determined by calculation. Commodities corresponding to the integrated embedding vector close to the integrated embedding vector are the recommended commodities determined for the target commodity.

In some possible implementations, using the commodity node embedding vector and commodity attribute node embedding vector to perform commodity recommendation at the same time may include: calculating the comprehensive embedding vector of the target commodity and the combination of all other commodities except the target commodity. The similarity between the integrated embedding vectors; it is determined that the similarity is greater than a preset threshold, or the commodities corresponding to the top N integrated embedding vectors in the order of the similarity are determined for the target commodity. recommended products. In this embodiment, the similarity is the cosine similarity. The larger the value of the cosine similarity, the closer the corresponding recommended product is to the target product. Therefore, a preset threshold or parameter N can be set to filter and determine the target product. Recommended product.

In some possible implementations, using the commodity node embedding vector and commodity attribute node embedding vector to perform commodity recommendation at the same time may include: determining the target commodity attribute corresponding to the target commodity; calculating the comprehensive embedding vector of the target commodity and In the attributes of the target product, the similarity between the comprehensive embedding vectors of all other products except the target product; the similarity is greater than the preset threshold, or the top N are ranked in the order of the similarity. The commodity corresponding to the comprehensive embedding vector is the recommended commodity determined for the target commodity. In this embodiment, the attribute of the target product corresponding to the target product is determined by setting, and only the comprehensive embedding vector corresponding to the target product and the attribute of the target product are calculated, and the sum of the comprehensive embedding vector of each product in all other products except the target product is calculated. The similarity between the two, and then determine the recommended products of the target product, reducing the calculation pressure.

To sum up, compared with the traditional graph embedding method to calculate the embedding vector of the commodity, the commodity recommendation method that integrates commodity attribute information proposed in this application can better combine the structural information and attribute information of the nodes in the graph network. In the similar commodity scenario of the recommendation system, the commodity comprehensive embedding vector calculated by this application not only contains the user's behavior information, but also contains the content information of the commodity itself, which can reflect more information on the nature of the commodity itself. Therefore, through this application The comprehensive embedding vector of the commodity calculated in the embodiment is used for commodity recommendation, which can improve the accuracy of commodity recommendation.

In some possible implementations, referring specifically to FIG. 2 , FIG. 2 is a schematic diagram of an embodiment of step S120 shown in FIG. 1 , which may include:

In step S121, the historical commodity browsing records are counted, and a graph network of commodity nodes is constructed.

In this embodiment, after obtaining the historical commodity browsing records of multiple users recorded in the system, statistics can be performed on the historical commodity browsing records, and a graph network of commodity nodes can be constructed. Among them, the commodity set r in the historical commodity browsing record is selected as a graph node of the graph network, and the set c of attribute information in the historical commodity browsing record is selected as the edge information of the graph network to be constructed.

In the recommendation system, a user’s clicked item record on the day can be expressed as R={r ₁ ,r ₂ ,…,r _i ,…,r _n }, where r _i represents the i-th item in the item set r,n Indicates the number of items clicked by the user, then the edge weight of any two nodes in the set R in the graph increases by 1. The final graph network is generated by counting the historical commodity browsing records of all users and selecting nodes whose edge weights are greater than the preset value w, where V represents the set of commodity nodes in the graph,

E represents the set of edges between items in the graph, e _ij ∈ E, e _ij >w, and w is a positive integer.

Step S122 , constructing a commodity attribute dictionary according to the graph network, where the commodity attribute dictionary includes corresponding relationships between different commodity nodes and different commodity attributes.

In this embodiment, according to the preset classification standard or classification level, the attribute information of the commodities in the graph network is counted, and the mapping dictionary D from the commodity set r to the category set c is generated according to the node information in the graph, namely:

D(r _i )=c _j

Indicates that the i-th commodity in the commodity set r belongs to the j-th category of the category set c. It should be noted that, in the mapping dictionary, each commodity may correspond to one or more categories.

Step S123, converting the graph network into a sequence of commodity nodes by random walk.

In this embodiment, the deepwalk algorithm is used for reference, and the commodity node sequence is obtained according to the commodity node set included in the graph network in a random walk manner. Among them, the Deepwalk algorithm is a graph-structured data mining algorithm that combines random walk and word2vec algorithms. Random walk refers to randomly taking a node in the graph network as the starting point to generate a sequence of commodity nodes with a preset random walk sequence length.

In some possible implementations, specifically converting the graph network into a sequence of commodity nodes through a random walk method may include: sequentially normalizing the out-degrees of commodity nodes in the graph network, and determining each commodity node The out-degree probability of ; according to the out-degree probability, a random walk method is used to generate the commodity node sequence.

Specifically, first normalize the out-degree of each commodity node in the graph network to obtain the out-degree probability between each commodity node and other adjacent commodity nodes. The closer the two commodity nodes are. Then, the commodity nodes in the graph network are converted into text-like commodity sequences L _r =[r _a ,...,r _k ,...,r _l ], where r _a ,r _k , r _l ∈ r are the commodity nodes of the random walk sequence L _r respectively, and Length(L _r ) is the given length of each random walk sequence. Select different initial walking commodity nodes from the graph network, and repeat the walking k times to generate a commodity node sequence set S _r , where K is a positive integer. Among them, in general, graphs can be divided into directed graphs and undirected graphs. All edges of a directed graph have a direction, that is, a direction from vertex to vertex is determined; while all edges of an undirected graph are bidirectional, that is, two vertices connected by an undirected edge can reach each other. In some problems, an undirected graph can be thought of as consisting of two directed edges where all edges are positive and negative. The degree of a vertex refers to the number of edges connected to the vertex. Especially for a directed graph, the number of out-edges of a vertex is called the out-degree of the vertex, and the number of in-edges of a vertex is called the in-degree of the vertex. In this embodiment, random walk is performed according to the out-degree probability obtained after out-degree normalization, and a graph network that better reflects the closeness of commodity nodes can be obtained.

It should be noted that, steps S122 and S123 do not necessarily require an execution order.

Step S124, based on the commodity attribute dictionary, convert the commodity node sequence into a commodity attribute node sequence.

In this embodiment, based on the mapping dictionary D constructed in the above step S122, the commodity node sequence set S _r in the above step S123 is converted into the corresponding category node sequence set S _c , that is, for any commodity sequence L _r in S _r , Its corresponding category sequence L _c is:

in,

are the category nodes in the sequence _Lc , respectively.

In the training process of this application, the traditional word2vec combined with negative sampling model architecture is adopted, and the structure information and attribute information of the commodity nodes of the network nodes are integrated on the basis of it, which has the advantages of simple model structure, less parameter quantity and short training time. , which can be better and widely used in real-time graph network scenarios.

In a possible implementation, referring specifically to FIG. 3 , FIG. 3 is a schematic diagram of another embodiment of a product recommendation method in the embodiment of the present application, which may include:

Step S310, obtaining historical commodity browsing records of multiple users.

It should be noted that step S310 is similar to step S110 shown in FIG. 1 , and details are not described here.

In step S320, the historical commodity browsing records are counted, and a graph network of commodity nodes is constructed.

It should be noted that step S320 is similar to step S121 shown in FIG. 2 , and details are not described here.

Step S330 , constructing corresponding commodity attribute dictionaries for different classification standards preset by the graph network respectively, wherein each classification standard is preset corresponding to a commodity attribute set.

In this embodiment, multiple classification standards can be set. For example, for books, classification standard 1 and classification standard 2 can be set, and classification standard 1 can include a set of book categories, such as {city, romance, martial arts, fantasy, Suspense, game, reasoning}, the classification standard 2 is a collection of book author names, such as {Lu Yao, Guo Jingming, Higashino Keigo, Natsume Soseki, Salinger, Shakespeare, ...} and so on. Thus, corresponding commodity attribute dictionaries under various classification systems can be constructed. For the process of constructing the commodity attribute dictionary under each classification system, reference may be made to step S122 in FIG. 2 , and details are not repeated here.

Step S340, converting the graph network into a sequence of commodity nodes by random walk.

It should be noted that step S340 is similar to step S123 shown in FIG. 2 , and details are not described here.

Step S350, based on the commodity attribute dictionary, convert the commodity node sequence into commodity attribute node sequences corresponding to different classification standards.

In this embodiment, referring to step S330, when a commodity attribute dictionary with multiple classification systems is generated, when the commodity node sequence is converted into a commodity attribute node sequence, commodity attribute nodes corresponding to different classification standards are also converted together. sequence. For the process of converting the commodity attribute node sequence to obtain the commodity attribute node sequence, reference may be made to step S124 in FIG. 2 , and details are not described here.

In step S360, combining the negative sampling strategy, the commodity node sequence and the commodity attribute node sequence are used as training corpus, respectively input the word2vec model, and the commodity node embedding vector and commodity attribute node embedding vector under different classification standards are calculated.

In this embodiment, referring to step S350, when the commodity attribute sequences of multiple classification systems are obtained after conversion, the commodity attribute sequence is used as the training corpus to calculate the commodity attribute node embedding vector, and the corresponding commodity attribute sequences under different classification systems are also obtained. Product attribute node embedding vector. For the specific calculation process of obtaining the commodity node embedding vector and commodity attribute node embedding vector under different classification standards, reference may be made to step S130 in FIG. 1 , which will not be repeated here.

Step S370: Perform a vector weighted sum operation on the commodity node embedding vector and the commodity attribute node embedding vector according to the preset weights corresponding to the commodity node embedding vectors and the corresponding weights of different classification standards, to obtain each The comprehensive embedding vector corresponding to the product.

In this embodiment, considering that the embedded vector is in the actual application scenario, the impact of the embedded vector of commodity attribute nodes corresponding to different classification standards on the comprehensive embedded vector of the product is different, so it can be the commodity node that needs to be merged to calculate the comprehensive embedded vector. The embedding vector and the corresponding commodity node embedding vector under different classification standards, set the corresponding weight value. Therefore, the vector weighted summation operation is performed on the commodity node embedding vector corresponding to each commodity and the corresponding commodity attribute node embedding vector under at least one classification standard to obtain the comprehensive embedding vector corresponding to each commodity.

For example, referring to the example in step S330, the comprehensive embedding vector may be obtained by combining commodity node embedding vector A, classification standard 1 commodity attribute node embedding vector B, and classification standard 2 commodity attribute node embedding vector C. In the actual book recommendation scenario, the priority order of importance is B-A-C. In this case, weight parameters of different sizes can be set for A, B, and C respectively, such as 0.8, 0.9., and 0.6. Therefore, when the comprehensive embedding vector corresponding to a certain book is combined and calculated, the A, B and C corresponding to the book are first determined respectively, and then the respective embedding vectors are multiplied with the corresponding weights, and the final comprehensive calculation of the book is obtained by weighted combination calculation. Embedding vector.

S380, determine the target commodity accessed by the user.

S390: Determine a product corresponding to a comprehensive embedding vector similar to the comprehensive embedding vector of the target product as a recommended product determined for the target product.

It should be noted that steps S380-S390 are similar to the foregoing steps S150-S160, and are not repeated here.

In the embodiment of the present application, considering that the embedded vector is in actual application scenarios, the impact of the embedded vector of commodity attribute nodes corresponding to different classification standards on the comprehensive embedded vector of the product is different. The embedding vector and the corresponding commodity node embedding vector under different classification standards, set the corresponding weight value. Therefore, the comprehensive embedding vector corresponding to the product obtained by the final calculation can better represent the essential properties of the product in different application scenarios. Therefore, product recommendation can be performed by using the comprehensive embedding vector of the product calculated in the embodiment of the present application, which can improve the accuracy of product recommendation. .

The present application may be used in numerous general purpose or special purpose computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, including A distributed computing environment for any of the above systems or devices, and the like. The application may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The application may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including storage devices.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. , when the program is executed, it may include the processes of the foregoing method embodiments. Wherein, the aforementioned storage medium may be a non-volatile storage medium such as a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM) or the like.

It should be understood that although the various steps in the flowchart of the accompanying drawings are sequentially shown in the order indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless explicitly stated herein, the execution of these steps is not strictly limited to the order and may be performed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but may be executed at different times, and the execution sequence is also It does not have to be performed sequentially, but may be performed alternately or alternately with other steps or at least a portion of sub-steps or stages of other steps.

Further referring to FIG. 4 , as an implementation of the method shown in FIG. 1 , the present application provides an embodiment of a product recommendation device, the device embodiment corresponds to the product recommendation method embodiment shown in FIG. 1 , the device Specifically, it can be applied to various electronic devices.

As shown in FIG. 4 , the apparatus 400 for determining a network embedding vector according to this embodiment includes:

an obtaining unit 410, configured to obtain historical commodity browsing records of multiple users;

generating unit 420, generating a sequence of commodity nodes and a sequence of commodity attribute nodes according to the historical commodity browsing records;

The training unit 430 is configured to combine the negative sampling strategy, use the commodity node sequence and the commodity attribute node sequence as training corpus, respectively input the word2vec model, and calculate and obtain the commodity node embedding vector and the commodity attribute node embedding vector;

A computing unit 440, configured to perform a vector splicing or vector sum operation on the commodity node embedding vector and the commodity attribute node embedding vector to obtain a comprehensive embedding vector corresponding to each commodity;

a response unit 450, configured to determine the target commodity accessed by the user;

The recommending unit 460 is configured to determine a product corresponding to a comprehensive embedding vector similar to the comprehensive embedding vector of the target product, which is a recommended product determined for the target product.

In some possible implementations, referring specifically to FIG. 4 , FIG. 4 is a schematic diagram of an embodiment of the generating unit 420, which may include:

The first construction subunit 421 is used to count the historical commodity browsing records and construct a graph network of commodity nodes;

The second construction subunit 422 is configured to construct a commodity attribute dictionary according to the graph network, and the commodity attribute dictionary includes the corresponding relationship between different commodity nodes and different commodity attributes;

The first conversion subunit 423 is used to convert the graph network into a sequence of commodity nodes by random walk;

The second conversion subunit 424 is configured to convert the sequence of commodity nodes into a sequence of commodity attribute nodes based on the commodity attribute dictionary.

In some possible implementations, the first conversion subunit is specifically configured to perform normalization processing on the out-degrees of commodity nodes in the graph network in turn, and determine the out-degree probability of each commodity node; The degree probability adopts a random walk method to generate the commodity node sequence.

In some possible implementations,

The second constructing subunit is specifically configured to construct corresponding commodity attribute dictionaries according to different classification standards preset by the graph network, wherein each classification standard is preset to correspond to a commodity attribute set;

a second conversion subunit, specifically configured to convert the commodity node sequence into commodity attribute node sequences corresponding to different classification standards based on the commodity attribute dictionary;

The training unit 430 is specifically configured to combine the negative sampling strategy, use the commodity node sequence and the commodity attribute node sequence as training corpus, respectively input the word2vec model, and calculate the commodity node embedding vector and commodity attribute node embedding under different classification standards. vector;

The calculation unit 440 is specifically configured to perform a vector weighted sum operation on the commodity node embedding vector and the commodity attribute node embedding vector according to the preset weights corresponding to the commodity node embedding vectors and the corresponding weights of different classification standards , to obtain the comprehensive embedding vector corresponding to each commodity.

In some possible implementations, the first construction subunit 421 is specifically configured to count the historical commodity browsing records, and select the commodity collection in the historical commodity browsing records as the graph nodes of the graph network to be constructed; determine according to the statistical results Edge weights between each commodity node; select commodity nodes whose edge weights are greater than a preset threshold to construct a graph network of commodity nodes.

In some possible implementations, the apparatus 400 for determining the network embedding vector may further include:

The recommended commodity determination unit is used for determining the target commodity accessed by the user; determining the commodity corresponding to the integrated embedding vector close to the integrated embedding vector of the target commodity is the recommended commodity determined for the target commodity.

Recommended product determination unit, including:

a first similarity calculation subunit, configured to calculate the similarity between the comprehensive embedding vector of the target commodity and the comprehensive embedding vectors of all other commodities except the target commodity;

The first recommended commodity determination subunit is used to determine that the similarity is greater than a preset threshold, or the commodities corresponding to the top N comprehensive embedding vectors in the order of the similarity are determined for the target commodity. For the recommended product, N is a positive integer.

In some possible implementations, the recommended commodity determination unit includes:

a first commodity attribute determination subunit, configured to determine the target commodity attribute corresponding to the target commodity;

a second similarity calculation subunit, configured to calculate the similarity between the comprehensive embedding vector of the target commodity and the comprehensive embedding vectors of all other commodities except the target commodity in the attributes of the target commodity;

The second recommended commodity determination subunit is used to take the similarity greater than the preset threshold, or the commodity corresponding to the top N comprehensive embedding vectors ranked according to the similarity degree, is determined for the target commodity For the recommended product, N is a positive integer.

In the embodiment of the present application, after acquiring the historical commodity browsing records of multiple users, the device 400 for determining the network embedding vector further generates commodity node sequences and commodity attribute node sequences according to the historical commodity browsing records. Then, combined with the negative sampling strategy, the commodity node sequence and commodity attribute node sequence are used as training corpus, and input into the word2vec model respectively, and the commodity node embedding vector and commodity attribute node embedding vector are calculated. Finally, perform vector splicing or vector summation operations on the commodity node embedding vector and commodity attribute node embedding vector to obtain the comprehensive embedding vector corresponding to each commodity.

To sum up, compared with the traditional graph embedding method to calculate the embedding vector of the commodity, the commodity recommendation method that integrates commodity attribute information proposed in this application can better combine the structural information and attribute information of the nodes in the graph network. In the recommendation system similarity commodity scenario, the commodity comprehensive embedding vector calculated by this application not only contains the user's behavior information, but also contains the content information of the commodity itself, which can reflect the property information of the commodity itself.

To solve the above technical problems, the embodiments of the present application also provide computer equipment. Please refer to FIG. 6 for details. FIG. 6 is a block diagram of the basic structure of a computer device according to this embodiment.

The computer device 6 includes a memory 601 , a processor 602 , and a network interface 603 that communicate with each other through a system bus. It should be pointed out that only the computer device 6 with components 601-603 is shown in the figure, but it should be understood that it is not required to implement all of the shown components, and more or less components may be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculation and/or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, special-purpose Integrated circuit (Application Specific Integrated Circuit, ASIC), programmable gate array (Field-Programmable Gate Array, FPGA), digital processor (Digital Signal Processor, DSP), embedded equipment, etc.

The computer equipment may be a desktop computer, a notebook computer, a palmtop computer, a cloud server and other computing equipment. The computer device can perform human-computer interaction with the user through a keyboard, a mouse, a remote control, a touch pad or a voice control device.

The memory 601 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, 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 memory 601 may be an internal storage unit of the computer device 6 , such as a hard disk or a memory of the computer device 6 . In other embodiments, the memory 601 may also be an external storage device of the computer device 6, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, flash memory card (Flash Card), etc. Of course, the memory 601 may also include both the internal storage unit of the computer device 6 and its external storage device. In this embodiment, the memory 601 is generally used to store the operating system and various application software installed on the computer device 6 , such as computer-readable instructions of the above-mentioned method for recommending products. In addition, the memory 601 can also be used to temporarily store various types of data that have been output or will be output.

The processor 602 may be a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 602 is typically used to control the overall operation of the computer device 6 . In this embodiment, the processor 602 is configured to execute computer-readable instructions stored in the memory 601 or process data, for example, computer-readable instructions for executing the method for recommending a commodity.

The network interface 603 may include a wireless network interface or a wired network interface, and the network interface 603 is generally used to establish a communication connection between the computer device 6 and other electronic devices.

The present application also provides another implementation manner, that is, to provide a computer-readable storage medium, the computer-readable storage medium may be non-volatile or volatile, and the computer-readable storage medium stores Computer-readable instructions, executable by at least one processor, to cause the at least one processor to perform the steps of a method for recommending an item as described above.

From the description of the above embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus a necessary general hardware platform, and of course hardware can also be used, but in many cases the former is better implementation. Based on this understanding, the technical solution of the present application can be embodied in the form of a software product in essence or in a part that contributes to the prior art, and the computer software product is stored in a storage medium (such as ROM/RAM, magnetic disk, CD-ROM), including several instructions to make a terminal device (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) execute the commodity recommendation method described in the various embodiments of this application.

Obviously, the above-described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The accompanying drawings show the preferred embodiments of the present application, but do not limit the scope of the patent of the present application. This application may be embodied in many different forms, rather these embodiments are provided so that a thorough and complete understanding of the disclosure of this application is provided. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or perform equivalent replacements for some of the technical features. . Any equivalent structure made by using the contents of the description and drawings of the present application, which is directly or indirectly used in other related technical fields, is also within the scope of protection of the patent of the present application.

Claims

A product recommendation method, comprising the following steps:

Get the historical commodity browsing records of multiple users;

Generate a sequence of commodity nodes and a sequence of commodity attribute nodes according to the historical commodity browsing records;

Combined with the negative sampling strategy, the commodity node sequence and the commodity attribute node sequence are used as training corpus, respectively input the word2vec model, and the commodity node embedding vector and the commodity attribute node embedding vector are obtained by calculation;

Perform vector splicing or vector summation operations on the commodity node embedding vector and the commodity attribute node embedding vector to obtain a comprehensive embedding vector corresponding to each commodity;

Determine the target product that the user visits;

It is determined that the product corresponding to the comprehensive embedding vector that is similar to the comprehensive embedding vector of the target product is the recommended product determined for the target product.
The product recommendation method according to claim 1, wherein the generating a product node sequence and a product attribute node sequence according to the historical product browsing records comprises:

Counting the historical commodity browsing records, and constructing a graph network of commodity nodes;

constructing a commodity attribute dictionary according to the graph network, the commodity attribute dictionary including the corresponding relationship between different commodity nodes and different commodity attributes;

Convert the graph network into a sequence of commodity nodes through random walks;

Based on the commodity attribute dictionary, the commodity node sequence is converted into a commodity attribute node sequence.
The product recommendation method according to claim 2, wherein the converting the graph network into a sequence of product nodes by a random walk method comprises:

Normalize the out-degrees of commodity nodes in the graph network in turn to determine the out-degree probability of each commodity node;

According to the out-degree probability, the commodity node sequence is generated by adopting a random walk manner.
The product recommendation method according to claim 2, wherein,

The building a commodity attribute dictionary according to the graph network includes:

Construct corresponding commodity attribute dictionaries for different preset classification standards according to the graph network, wherein each classification standard is preset to correspond to a commodity attribute set;

The converting the commodity node sequence into a commodity attribute node sequence based on the commodity attribute dictionary includes:

Based on the commodity attribute dictionary, converting the commodity node sequence into commodity attribute node sequences corresponding to different classification standards;

In combination with the negative sampling strategy, the commodity node sequence and the commodity attribute node sequence are used as training corpus, respectively input the word2vec model, and the commodity node embedding vector and commodity attribute node embedding vector are calculated and obtained, including:

Combined with the negative sampling strategy, the commodity node sequence and the commodity attribute node sequence are used as training corpus, respectively input the word2vec model, and the commodity node embedding vector and the commodity attribute node embedding vector under different classification standards are calculated to obtain;

The vector splicing or vector sum operation is performed on the commodity node embedding vector and the commodity attribute node embedding vector to obtain a comprehensive embedding vector corresponding to each commodity, including:

According to the preset weights corresponding to the commodity node embedding vectors and the weights corresponding to different classification standards, perform a vector weighted sum operation on the commodity node embedding vectors and the commodity attribute node embedding vectors to obtain the corresponding value of each commodity. The synthetic embedding vector.
The product recommendation method according to any one of claims 2-4, wherein the counting the historical product browsing records and constructing a graph network of product nodes, comprising:

Counting the historical commodity browsing records, and selecting the commodity collection in the historical commodity browsing records as the graph nodes of the graph network to be constructed;

Determine the edge weight between each commodity node according to the statistical result;

Select commodity nodes whose edge weights are greater than a preset threshold to construct a graph network of commodity nodes.
The product recommendation method according to any one of claims 1-4, wherein the product corresponding to the comprehensive embedding vector that is similar to the comprehensive embedding vector of the target product is determined for the target product. Recommended items, including:

Calculate the similarity between the comprehensive embedding vector of the target product and the comprehensive embedding vector of all other products except the target product;

It is determined that the similarity is greater than a preset threshold, or, the products corresponding to the top N comprehensive embedding vectors in the order of the similarity are the recommended products determined for the target product, and N is a positive integer.
The product recommendation method according to any one of claims 1-4, wherein the product corresponding to the comprehensive embedding vector determined to be similar to the comprehensive embedding vector of the target product is the recommended product determined for the target product ,include:

determining the target commodity attribute corresponding to the target commodity;

Calculate the similarity between the comprehensive embedding vector of the target commodity and the comprehensive embedding vector of all other commodities except the target commodity in the attributes of the target commodity;

Taking the product whose similarity is greater than the preset threshold, or the product corresponding to the top N comprehensive embedding vectors in the order of similarity, is the recommended product determined for the target product, and N is a positive integer.
A product recommendation device, comprising:

an acquisition unit, used to acquire historical commodity browsing records of multiple users;

a generating unit, generating a sequence of commodity nodes and a sequence of commodity attribute nodes according to the historical commodity browsing records;

The training unit is used for combining the negative sampling strategy, using the commodity node sequence and the commodity attribute node sequence as training corpus, respectively inputting the word2vec model, and calculating the commodity node embedding vector and the commodity attribute node embedding vector;

a computing unit, configured to perform vector splicing or vector summation operations on the commodity node embedding vector and the commodity attribute node embedding vector to obtain a comprehensive embedding vector corresponding to each commodity;

A response unit, used to determine the target commodity accessed by the user;

A recommending unit, configured to determine a product corresponding to a comprehensive embedding vector similar to the comprehensive embedding vector of the target product, which is a recommended product determined for the target product.
A computer device, comprising a memory and a processor, wherein computer-readable instructions are stored in the memory, and when the processor executes the computer-readable instructions, the processor implements the following steps of a commodity recommendation method:

Get the historical commodity browsing records of multiple users;

Generate a sequence of commodity nodes and a sequence of commodity attribute nodes according to the historical commodity browsing records;

Combined with the negative sampling strategy, the commodity node sequence and the commodity attribute node sequence are used as training corpus, respectively input the word2vec model, and the commodity node embedding vector and the commodity attribute node embedding vector are obtained by calculation;

Perform vector splicing or vector summation operations on the commodity node embedding vector and the commodity attribute node embedding vector to obtain a comprehensive embedding vector corresponding to each commodity;

Determine the target product that the user visits;

It is determined that the product corresponding to the comprehensive embedding vector that is similar to the comprehensive embedding vector of the target product is the recommended product determined for the target product.
The computer device according to claim 9, wherein the generating a sequence of commodity nodes and a sequence of commodity attribute nodes according to the historical commodity browsing records comprises:

Counting the historical commodity browsing records, and constructing a graph network of commodity nodes;

constructing a commodity attribute dictionary according to the graph network, the commodity attribute dictionary including the corresponding relationship between different commodity nodes and different commodity attributes;

Convert the graph network into a sequence of commodity nodes through random walks;

Based on the commodity attribute dictionary, the commodity node sequence is converted into a commodity attribute node sequence.
The computer device according to claim 10, wherein the converting the graph network into a sequence of commodity nodes through random walks comprises:

Normalize the out-degrees of commodity nodes in the graph network in turn to determine the out-degree probability of each commodity node;

According to the out-degree probability, the commodity node sequence is generated by adopting a random walk manner.
The computer device of claim 10, wherein,

The building a commodity attribute dictionary according to the graph network includes:

Construct corresponding commodity attribute dictionaries for different preset classification standards according to the graph network, wherein each classification standard is preset to correspond to a commodity attribute set;

The converting the commodity node sequence into a commodity attribute node sequence based on the commodity attribute dictionary includes:

Based on the commodity attribute dictionary, converting the commodity node sequence into commodity attribute node sequences corresponding to different classification standards;

In combination with the negative sampling strategy, the commodity node sequence and the commodity attribute node sequence are used as training corpus, respectively input the word2vec model, and the commodity node embedding vector and commodity attribute node embedding vector are calculated and obtained, including:

Combined with the negative sampling strategy, the commodity node sequence and the commodity attribute node sequence are used as training corpus, respectively input the word2vec model, and the commodity node embedding vector and the commodity attribute node embedding vector under different classification standards are calculated to obtain;

The vector splicing or vector sum operation is performed on the commodity node embedding vector and the commodity attribute node embedding vector to obtain a comprehensive embedding vector corresponding to each commodity, including:

According to the preset weights corresponding to the commodity node embedding vectors and the weights corresponding to different classification standards, perform a vector weighted sum operation on the commodity node embedding vectors and the commodity attribute node embedding vectors to obtain the corresponding value of each commodity. The synthetic embedding vector.
The computer device according to any one of claims 10-12, wherein the counting the historical commodity browsing records and constructing a graph network of commodity nodes, comprising:

Counting the historical commodity browsing records, and selecting the commodity collection in the historical commodity browsing records as the graph nodes of the graph network to be constructed;

Determine the edge weight between each commodity node according to the statistical result;

Select commodity nodes whose edge weights are greater than a preset threshold to construct a graph network of commodity nodes.
The computer device according to any one of claims 9-12, wherein the determined product corresponding to a comprehensive embedding vector similar to the comprehensive embedding vector of the target product is a recommendation determined for the target product Goods, including:

Calculate the similarity between the comprehensive embedding vector of the target product and the comprehensive embedding vector of all other products except the target product;

It is determined that the similarity is greater than a preset threshold, or, the products corresponding to the top N comprehensive embedding vectors in the order of the similarity are the recommended products determined for the target product, and N is a positive integer.
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 a commodity recommendation method described below are implemented:

Get the historical commodity browsing records of multiple users;

Generate a sequence of commodity nodes and a sequence of commodity attribute nodes according to the historical commodity browsing records;

Combined with the negative sampling strategy, the commodity node sequence and the commodity attribute node sequence are used as training corpus, respectively input the word2vec model, and the commodity node embedding vector and the commodity attribute node embedding vector are obtained by calculation;

Perform vector splicing or vector summation operations on the commodity node embedding vector and the commodity attribute node embedding vector to obtain a comprehensive embedding vector corresponding to each commodity;

Determine the target product that the user visits;

It is determined that the product corresponding to the comprehensive embedding vector that is similar to the comprehensive embedding vector of the target product is the recommended product determined for the target product.
The computer-readable storage medium according to claim 15, wherein the generating a sequence of commodity nodes and a sequence of commodity attribute nodes according to the historical commodity browsing records comprises:

Counting the historical commodity browsing records, and constructing a graph network of commodity nodes;

Build a commodity attribute dictionary according to the graph network, and the commodity attribute dictionary includes the corresponding relationship between different commodity nodes and different commodity attributes;

Convert the graph network into a sequence of commodity nodes through random walks;

Based on the commodity attribute dictionary, the commodity node sequence is converted into a commodity attribute node sequence.
The computer-readable storage medium of claim 16, wherein the converting the graph network into a sequence of commodity nodes by random walks comprises:

Normalize the out-degrees of commodity nodes in the graph network in turn to determine the out-degree probability of each commodity node;

According to the out-degree probability, the commodity node sequence is generated by adopting a random walk manner.
The computer-readable storage medium of claim 16, wherein,

The building a commodity attribute dictionary according to the graph network includes:

Construct corresponding commodity attribute dictionaries for different preset classification standards according to the graph network, wherein each classification standard is preset to correspond to a commodity attribute set;

The converting the commodity node sequence into a commodity attribute node sequence based on the commodity attribute dictionary includes:

Based on the commodity attribute dictionary, converting the commodity node sequence into commodity attribute node sequences corresponding to different classification standards;

In combination with the negative sampling strategy, the commodity node sequence and the commodity attribute node sequence are used as training corpus, respectively input the word2vec model, and the commodity node embedding vector and commodity attribute node embedding vector are calculated and obtained, including:

Combined with the negative sampling strategy, the commodity node sequence and the commodity attribute node sequence are used as training corpus, respectively input the word2vec model, and the commodity node embedding vector and the commodity attribute node embedding vector under different classification standards are obtained by calculation;

The vector splicing or vector sum operation is performed on the commodity node embedding vector and the commodity attribute node embedding vector to obtain a comprehensive embedding vector corresponding to each commodity, including:

According to the preset weights corresponding to the commodity node embedding vectors and the weights corresponding to different classification standards, perform a vector weighted sum operation on the commodity node embedding vectors and the commodity attribute node embedding vectors to obtain the corresponding value of each commodity. The synthetic embedding vector.
The computer-readable storage medium according to any one of claims 16-18, wherein the counting the historical commodity browsing records to construct a graph network of commodity nodes, comprising:

Counting the historical commodity browsing records, and selecting the commodity collection in the historical commodity browsing records as the graph nodes of the graph network to be constructed;

Determine the edge weight between each commodity node according to the statistical result;

Select commodity nodes whose edge weights are greater than a preset threshold to construct a graph network of commodity nodes.
The computer-readable storage medium according to any one of claims 16-18, wherein the commodity corresponding to the integrated embedding vector that is similar to the integrated embedding vector of the target commodity is determined for the target commodity. Identified recommended products, including:

Calculate the similarity between the comprehensive embedding vector of the target product and the comprehensive embedding vector of all other products except the target product;

It is determined that the similarity is greater than a preset threshold, or, the products corresponding to the top N comprehensive embedding vectors in the order of the similarity are the recommended products determined for the target product, and N is a positive integer.