CN115936805A

CN115936805A - Commodity recommendation method, commodity recommendation device, commodity recommendation equipment and commodity recommendation medium

Info

Publication number: CN115936805A
Application number: CN202211515071.6A
Authority: CN
Inventors: 黄丕帅
Original assignee: Guangzhou Huanju Shidai Information Technology Co Ltd
Current assignee: Guangzhou Huanju Shidai Information Technology Co Ltd
Priority date: 2022-11-29
Filing date: 2022-11-29
Publication date: 2023-04-07

Abstract

The application relates to a commodity recommendation method, a commodity recommendation device, commodity recommendation equipment and commodity recommendation media in the technical field of e-commerce, wherein the method comprises the following steps: acquiring commodity pictures of a full quantity of commodities, and determining a plurality of commodity labels corresponding to each commodity by adopting a preset image-text matching model, wherein the commodity labels belong to members of a preset commodity label set; determining co-occurrence scores between every two nodes by taking each commodity label as a node, and constructing a knowledge graph; calculating a recommendation score among the commodities according to the co-occurrence scores of the nodes corresponding to the commodities in the knowledge graph and the nodes corresponding to other commodities; and screening out recommended commodities corresponding to the recommendation scores meeting the preset conditions for each commodity. The commodity recommendation method and the commodity recommendation system provide a commodity recommendation solution for the cross-border e-commerce platform based on the independent sites, and service experience can be improved.

Description

Commodity recommendation method, commodity recommendation device, commodity recommendation equipment and commodity recommendation medium

Technical Field

The present application relates to the field of e-commerce technologies, and in particular, to a method for recommending a commodity and a corresponding apparatus, computer device, and computer-readable storage medium.

Background

The cross-border e-commerce is an industry rapidly developed in recent years, the independent site is a new form of the cross-border e-commerce, and the cross-border e-commerce has high autonomy and is more flexible in operation, so that the restriction of a plurality of platform rules is avoided. Compared with the traditional e-commerce platform, the quantity of online stores is different, and the quantity of commodities of some stores is hundreds of items, sometimes thousands or even tens of thousands of items. Based on the particularity of the independent site, the requirement different from the traditional E-commerce platform is provided for the commodity recommendation of the independent site, and the general commodity recommendation method in the industry is not completely applicable in the scene.

In the conventional technology, whether other commodities are similar to a target commodity is judged according to the similarity between the target commodity and commodity titles corresponding to the other commodities, so that the similar other commodities are used as recommended commodities of the target commodity, but merchants of a cross-border e-commerce platform based on independent sites are relatively independent, the titles of the commodities edited by the merchants have no fixed format and usually contain most of personalized edited text contents, the accuracy of the similarity is seriously influenced, the recommended commodities are difficult to ensure to be similar to the target commodity, and the commodity recommendation effect is poor.

The applicant has made corresponding researches on the shortcomings of the conventional technology.

Disclosure of Invention

A primary object of the present application is to solve at least one of the above problems and provide a method for recommending merchandise, and a corresponding apparatus, computer device, and computer-readable storage medium.

In order to meet various purposes of the application, the following technical scheme is adopted in the application:

a merchandise recommendation method adapted to one of the objects of the present application is provided, including the steps of:

acquiring commodity pictures of a full quantity of commodities, and determining a plurality of commodity labels corresponding to each commodity by adopting a preset image-text matching model, wherein the commodity labels belong to members of a preset commodity label set;

determining co-occurrence scores between every two nodes by taking each commodity label as a node, and constructing a knowledge graph;

calculating a recommendation score among the commodities according to the co-occurrence scores of the nodes corresponding to the commodities in the knowledge graph and the nodes corresponding to other commodities;

and screening out recommended commodities corresponding to the recommendation scores meeting the preset conditions for each commodity.

In a further embodiment, the method for obtaining the commodity pictures of the whole commodities and determining a plurality of commodity labels corresponding to each commodity by adopting a preset image-text matching model comprises the following steps:

acquiring commodity pictures of a full quantity of commodities, extracting image characteristic information of the commodity pictures by an image encoder in a preset image-text matching model, and extracting text characteristic information corresponding to each commodity label in a preset commodity label set by a text encoder in the preset image-text matching model;

and calculating the similarity between the image characteristic information corresponding to the commodity picture and the text characteristic information corresponding to each commodity label aiming at each commodity, and screening out a plurality of commodity labels with the similarity being greater than a preset threshold value.

In a further embodiment, each commodity label is taken as a node, the co-occurrence score between every two nodes is determined, and a knowledge graph is constructed, and the method comprises the following steps:

taking each commodity label as a node, if two nodes are the same, taking the co-occurrence score as a preset value, otherwise, calculating the number of commodities as the co-occurrence times when every two nodes belong to the same commodity at the same time, normalizing the co-occurrence times, and obtaining the co-occurrence score between every two nodes;

and respectively connecting the plurality of nodes corresponding to each commodity with the plurality of nodes corresponding to other commodities to form edges, wherein the weight of each edge is the co-occurrence score of the two nodes connected with the edge, and a knowledge graph is constructed.

In a further embodiment, calculating a recommendation score among the commodities according to a co-occurrence score among a plurality of nodes corresponding to the commodities in the knowledge graph and a plurality of nodes corresponding to other commodities includes:

and determining the co-occurrence scores of a plurality of nodes corresponding to each commodity and a plurality of nodes corresponding to other commodities according to the knowledge graph aiming at each commodity, and calculating the sum of the corresponding co-occurrence scores to be divided by the number of the nodes corresponding to the commodities to obtain the recommendation scores among the commodities.

In a further embodiment, before obtaining the commodity pictures of the full quantity of commodities, the method further comprises the following steps:

the method comprises the steps of obtaining commodity titles of a plurality of commodities, classifying each commodity title according to a commodity category, and obtaining a commodity title corresponding to each commodity category;

performing word segmentation on each commodity title, counting word frequency and inverse text frequency indexes of each word element under the corresponding commodity category, and calculating a keyword score;

and screening out the lemmas corresponding to the keyword scores meeting the preset conditions as commodity labels, and constructing a commodity label set.

acquiring commodity pictures of a plurality of commodities and each commodity label in a preset commodity label set to form a plurality of image-text data pairs as training samples, and according to whether the commodity pictures and the commodity labels in the image-text data pairs of the training samples are matched with corresponding labeling supervision labels or not;

inputting the training sample into a graph-text matching model, extracting deep semantic information corresponding to the commodity picture and the commodity label in each graph-text data pair, and obtaining corresponding image characteristic information and text characteristic information;

mapping image characteristic information and text characteristic information corresponding to each image-text data pair to the same multi-modal space, and calculating the similarity between the image characteristic information and the text characteristic information corresponding to each image-text data pair;

and determining a loss value of the similarity by adopting the supervision label of the training sample, updating the weight of the image-text matching model when the loss value does not reach a preset threshold value, and continuously calling other training samples to carry out iterative training until the model converges.

In a further embodiment, screening out, for each commodity, a recommended commodity corresponding to a recommendation score that meets a preset condition includes:

and sorting each commodity according to the recommendation scores between the commodity and other commodities, and screening out a plurality of recommended commodities which are sorted in the front.

On the other hand, the commodity recommendation device adapted to one of the purposes of the present application includes a commodity marking module, a map building module, a score calculating module and a commodity screening module, wherein the commodity marking module is configured to obtain commodity pictures of a full quantity of commodities, and determine a plurality of commodity labels corresponding to each commodity by using a preset image-text matching model, where the commodity labels belong to members of a preset commodity label set; the map building module is used for determining the co-occurrence score between every two nodes by taking each commodity label as a node, and building a knowledge map; the score calculating module is used for calculating a recommendation score among the commodities according to the co-occurrence scores among the nodes corresponding to the commodities in the knowledge graph and the nodes corresponding to other commodities; and the commodity screening module is used for screening out recommended commodities corresponding to the recommendation scores meeting the preset conditions for each commodity.

In a further embodiment, the commodity screening module includes: the characteristic extraction submodule is used for acquiring commodity pictures of a full quantity of commodities, extracting image characteristic information of the commodity pictures by an image encoder in a preset image-text matching model, and extracting text characteristic information corresponding to each commodity label in a preset commodity label set by a text encoder in the preset image-text matching model; and the label determining submodule is used for calculating the similarity between the image characteristic information corresponding to the commodity picture of each commodity and the text characteristic information corresponding to each commodity label aiming at each commodity, and screening out a plurality of commodity labels with the similarity larger than a preset threshold value.

In a further embodiment, the atlas construction module includes: the co-occurrence score determining submodule is used for taking each commodity label as a node, if the two nodes are the same, the co-occurrence score is a preset value, otherwise, the number of commodities is calculated as the co-occurrence times when every two nodes belong to the same commodity at the same time, the co-occurrence times are normalized, and the co-occurrence score between every two nodes is obtained; and the knowledge graph constructing sub-module is used for connecting the plurality of nodes corresponding to each commodity with the plurality of nodes corresponding to other commodities to form edges, and the weight of each edge is the co-occurrence score of the two nodes connected with the edge to construct the knowledge graph.

In a further embodiment, the score calculating module includes: and the recommendation score calculation submodule is used for determining the co-occurrence scores between the plurality of nodes corresponding to each commodity and the plurality of nodes corresponding to other commodities according to the knowledge graph aiming at each commodity, calculating the sum of the plurality of corresponding co-occurrence scores and dividing the sum by the number of the nodes corresponding to the commodities to obtain the recommendation scores among the commodities.

In a further embodiment, before the article marking module, the method further includes: the category title module is used for acquiring the commodity titles of a plurality of commodities, classifying each commodity title according to the commodity category and acquiring the commodity title corresponding to each commodity category; the keyword score calculation module is used for segmenting each commodity title, counting word frequency and inverse text frequency indexes of each word element under the corresponding commodity category and calculating keyword scores; and the label screening module is used for screening out the lexical elements corresponding to the keyword scores meeting the preset conditions as commodity labels to construct a commodity label set.

In a further embodiment, before the article marking module, the method further includes: the system comprises a sample construction and labeling module, a label analysis module and a label analysis module, wherein the sample construction and labeling module is used for acquiring commodity pictures of a plurality of commodities to form a plurality of picture and text data pairs with each commodity label in a preset commodity label set as training samples, and whether the commodity pictures and the commodity labels are matched with corresponding labeling monitoring labels in the picture and text data pairs of the training samples is determined; the feature extraction module is used for inputting the training samples into the image-text matching model, extracting deep semantic information corresponding to the commodity images and the commodity labels in each image-text data pair, and obtaining corresponding image feature information and text feature information; the similarity calculation module is used for mapping the image characteristic information and the text characteristic information corresponding to each image-text data pair to the same multi-modal space and calculating the similarity between the image characteristic information and the text characteristic information corresponding to each image-text data pair; and the iterative training module is used for determining the loss value of the similarity by adopting the supervision label of the training sample, updating the weight of the image-text matching model when the loss value does not reach a preset threshold value, and continuously calling other training samples to carry out iterative training until the model converges.

In a further embodiment, the commodity screening module includes: and the sequencing optimization submodule is used for sequencing each commodity according to the recommendation scores between each commodity and other commodities and screening out a plurality of recommended commodities which are sequenced in the front.

In yet another aspect, a computer device adapted for one of the purposes of the present application includes a central processing unit and a memory, the central processing unit being configured to invoke execution of a computer program stored in the memory to perform the steps of the merchandise recommendation method described in the present application.

In still another aspect, a computer-readable storage medium is provided, which stores a computer program implemented according to the method for recommending items of merchandise in the form of computer-readable instructions, and when the computer program is called by a computer, executes the steps included in the method.

The technical solution of the present application has various advantages, including but not limited to the following aspects:

the commodity recommendation method and the commodity recommendation system are used for uniformly providing a commodity recommendation solution for an online shop in an independent site of an e-commerce platform, a plurality of commodity labels corresponding to commodities are determined by adopting an image-text matching model through commodity pictures based on total commodities in the online shop, the commodity labels are used as nodes, the co-occurrence score between every two nodes is determined, a knowledge graph is constructed, the co-occurrence score between a plurality of nodes corresponding to the commodities and other commodities is calculated according to the co-occurrence score between the nodes corresponding to the commodities, and accordingly recommended commodities corresponding to the recommended score meeting preset conditions are screened out. On one hand, the image-text matching model is suitable for multi-mode data processing, a plurality of commodity labels matched with commodity pictures of commodities are determined, the method is very simple, convenient and efficient, manual intervention is not needed, on the other hand, the corresponding commodities are accurately represented in a fine-grained mode through the commodity labels, the recommendation score calculated based on the co-occurrence score between each commodity label corresponding to every two commodities is accurate enough, and accordingly the recommendation effect of the selected recommended commodities is good.

Drawings

The above and/or additional aspects and advantages of the present application will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:

fig. 1 is a schematic flowchart of an exemplary embodiment of a product recommendation method according to the present application;

fig. 2 is a schematic flowchart illustrating a process of determining a plurality of article tags corresponding to each article in an embodiment of the present application;

FIG. 3 is a schematic flow chart of construction of a knowledge graph in an embodiment of the present application;

FIG. 4 is a schematic flow chart of pre-constructing a commodity tag set in an embodiment of the present application;

FIG. 5 is a diagram illustrating a training process of a graph-text matching model according to an embodiment of the present application;

FIG. 6 is a functional block diagram of the merchandise recommendation device of the present application;

fig. 7 is a schematic structural diagram of a computer device used in the present application.

Detailed Description

Reference will now be made in detail to the embodiments of the present application, examples of which are illustrated in the accompanying drawings, wherein like reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the drawings are exemplary only for the purpose of explaining the present application and are not to be construed as limiting the present application.

As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Further, "connected" or "coupled" as used herein may include wirelessly connected or wirelessly coupled. As used herein, the term "and/or" includes all or any element and all combinations of one or more of the associated listed items.

It will be understood by those within the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

As will be appreciated by those skilled in the art, "client," "terminal," and "terminal device" as used herein include both devices that are wireless signal receivers, which are devices having only wireless signal receivers without transmit capability, and devices that are receive and transmit hardware, which have receive and transmit hardware capable of two-way communication over a two-way communication link. Such a device may include: cellular or other communication devices such as personal computers, tablets, etc. having single or multi-line displays or cellular or other communication devices without multi-line displays; PCS (personal communications System), which may combine voice, data processing, facsimile and/or data communications capabilities; a PDA (personal digital assistant), which may include a radio frequency receiver, a pager, internet/intranet access, web browser, notepad, calendar, and/or GPS (global positioning System) receiver; a conventional laptop and/or palmtop computer or other appliance having and/or including a radio frequency receiver. As used herein, a "client," "terminal device" can be portable, transportable, installed in a vehicle (aeronautical, maritime, and/or land-based), or situated and/or configured to operate locally and/or in a distributed fashion at any other location(s) on earth and/or in space. The "client", "terminal device" used herein may also be a communication terminal, a web terminal, and a music/video playing terminal, and may be, for example, a PDA, an MI D (Mob I e intemet Dev I ce, mobile internet device), and/or a mobile phone with music/video playing function, and may also be a smart tv, a set-top box, and other devices.

The hardware referred to by the names "server", "client", "service node", etc. is essentially an electronic device with the performance of a personal computer, and is a hardware device having necessary components disclosed by the von neumann principle such as a central processing unit (including an arithmetic unit and a controller), a memory, an input device, an output device, etc., a computer program is stored in the memory, and the central processing unit calls a program stored in an external memory into the internal memory to run, executes instructions in the program, and interacts with the input and output devices, thereby completing a specific function.

It should be noted that the concept of "server" as referred to in this application can be extended to the case of a server cluster. According to the network deployment principle understood by those skilled in the art, the servers should be logically divided, and in physical space, the servers may be independent from each other but can be called through an interface, or may be integrated into one physical computer or a set of computer clusters. Those skilled in the art will appreciate this variation and should not be so limited as to restrict the implementation of the network deployment of the present application.

One or more technical features of the present application, unless expressly specified otherwise, may be deployed to a server for implementation by a client remotely invoking an online service interface provided by a capture server for access, or may be deployed directly and run on the client for access.

Unless specified in clear text, the neural network model referred to or possibly referred to in the application can be deployed in a remote server and used for remote call at a client, and can also be deployed in a client with qualified equipment capability for direct call.

Various data referred to in the present application may be stored in a server remotely or in a local terminal device unless specified in the clear text, as long as the data is suitable for being called by the technical solution of the present application.

The person skilled in the art will know this: although the various methods of the present application are described based on the same concept so as to be common to each other, they may be independently performed unless otherwise specified. In the same way, for each embodiment disclosed in the present application, it is proposed based on the same inventive concept, and therefore, concepts of the same expression and concepts of which expressions are different but are appropriately changed only for convenience should be equally understood.

Unless expressly stated otherwise, the technical features of the embodiments disclosed in the present application may be cross-linked to form a new embodiment, so long as the combination does not depart from the spirit of the present application and can satisfy the requirements of the prior art or solve the disadvantages of the prior art. Those skilled in the art will appreciate variations therefrom.

The commodity recommendation method can be programmed into a computer program product and deployed in a client or a server to run, for example, in an exemplary application scenario of the present application, the commodity recommendation method can be deployed and implemented in a server of an e-commerce platform, so that the method can be executed by accessing an open interface after the computer program product runs and performing human-computer interaction with a process of the computer program product through a graphical user interface.

Referring to fig. 1, in an exemplary embodiment, a method for recommending a commodity according to the present application includes the following steps:

step S1100, acquiring commodity pictures of a full quantity of commodities, and determining a plurality of commodity labels corresponding to each commodity by adopting a preset image-text matching model, wherein the commodity labels belong to members of a preset commodity label set;

according to the technical scheme, the operating environment of the e-commerce platform is used as the application environment of the e-commerce platform, and the e-commerce platform can be an e-commerce platform with an open independent site service, typically, for example, a cross-border e-commerce platform. Such a platform allows the e-commerce platform to serve a large number of such independent sites by configuring each merchant's online stores as individual independent sites, due to the need to take into account the network environment between locations around the world and the independence between merchants.

The merchant of the independent site can configure commodity information corresponding to each commodity on the online shop shelf to submit the commodity information to the server, so that the server can shelf the corresponding commodity to display the commodity to the consumer after receiving the commodity information. The commodity information includes, but is not limited to, text information and picture information of the commodity, and the text information may include a commodity title, a commodity detail text, a commodity category and the like of the commodity. The commodity information includes, but is not limited to, different types of data such as pictures, texts, and the like. The pictures in the commodity information comprise pictures uploaded for the commodities when the merchants of online shops put on shelves, and the commodities can be displayed from the whole and/or different sides, and the pictures comprise a commodity main graph, a commodity detail graph and the like. The text in the commodity information comprises any one or more of a commodity title, a commodity detail text, a commodity category and the like. In the present application, it is recommended to use the product main graph of the product as a product picture to be subsequently processed.

The image-Text matching model comprises an image encoder and a Text encoder, wherein the image encoder can adopt a model suitable for extracting image features, the recommended selection type is a ViT (Vis ion Transformer) model, and can also adopt any one of other models such as a CNN (compact neural network) model, a deep convolution model Effi cientNet, a Densenet model and a Resnet, the Text encoder can adopt a model suitable for extracting Text features in the NLP field, the recommended selection type is a BERT model, and can also adopt any one of other models such as a Text Transfomer, a RoBERTA, an XLM-RoBERTA and an MPNet. The image-text matching model can be trained in advance to be convergent, the capability of determining the similarity between the image features corresponding to the extracted images and the text features corresponding to the extracted texts is learned, the specific training process is further disclosed by the subsequent part of the embodiment, and the step is not required to be pressed down temporarily. The graphics and text matching model recommends adopting a CLI P (Contrast live Language-Image Pre-tra in i ng) model.

Generally, each independent site can create and maintain a commodity database thereof to store commodity information corresponding to each commodity in the online shop, and a corresponding data interface can be packaged in advance for accessing the commodity information in the commodity database and carrying out operations such as addition, deletion, check, modification and the like on the commodity information. The data interface may be flexibly implemented by those skilled in the art. Therefore, the data interface is called to access commodity information of the total commodities in the commodity database, and commodity pictures of each commodity are obtained from the commodity information. Further, by adopting the preset image-text matching model, the image encoder extracts the image characteristic information of the commodity image, and the text encoder extracts the text characteristic information corresponding to each commodity label in the preset commodity label set. And aiming at each commodity, calculating the inner product between the image characteristic information corresponding to the commodity picture and the text characteristic information corresponding to each commodity label as the similarity, and screening out a plurality of commodity labels with the similarity being greater than a preset threshold value. The image characteristic information is image characteristics of the commodity picture extracted by vectorization, and the text characteristic information is text characteristics of the commodity label extracted by vectorization.

The commodity label is a text describing at least one commodity attribute, the commodity attribute comprises a brand, a material, a specification, a color, a weight, a function, a selling point, an effect and the like, the commodity label can be extracted from a plurality of commodity titles, the commodity label set is further constructed in advance by the plurality of commodity labels for calling, the specific implementation is further disclosed by the following partial embodiments, and the step is not shown once and is not shown after being pressed down.

S1200, determining a co-occurrence score between every two nodes by taking each commodity label as a node, and constructing a knowledge graph;

taking each commodity label as a node, if two nodes are the same, directly setting the co-occurrence score between the two nodes as 1, otherwise, calculating the number of commodities as co-occurrence times when every two nodes belong to the same commodity, for example, a node a and a node b belong to a commodity a, a commodity b and a commodity c at the same time, and the corresponding co-occurrence times are that the number of commodities is 3, normalizing the co-occurrence times to make the value range thereof belong to [0-1], and obtaining the co-occurrence score between every two nodes.

Further, aiming at each commodity, each corresponding node is respectively connected with a plurality of nodes corresponding to other commodities, an edge is formed between every two connected nodes, the weight of each edge is the co-occurrence score of the two nodes connected with the edge, and a knowledge graph is constructed.

Step S1300, calculating recommendation scores among the commodities according to co-occurrence scores among a plurality of nodes corresponding to the commodities and a plurality of nodes corresponding to other commodities in the knowledge graph;

determining, for each commodity, a co-occurrence score between a plurality of nodes corresponding to each commodity and a plurality of nodes corresponding to other commodities according to a knowledge graph, and calculating a sum of a plurality of corresponding co-occurrence scores divided by the number of nodes corresponding to the commodity to obtain a recommendation score between the commodities, where an exemplary formula is as follows: taking two commodities (commodity a and commodity b) as an example,

wherein: and (2) recommending scores between two commodities of the scor, wherein a numerator is the sum of co-occurrence scores between a plurality of nodes corresponding to the commodity a and a plurality of nodes corresponding to the commodity b, and a denominator is the number of the nodes of the commodity a.

And S1400, screening out recommended commodities corresponding to the recommendation scores meeting the preset conditions for each commodity.

In one implementation, each commodity is sorted according to the order of recommendation scores between the commodity and other commodities from high to low, and a plurality of commodities which are sorted in the front are screened out as recommended commodities, wherein the specific number of the recommended commodities can be set by a person skilled in the art as required.

In another embodiment, for each commodity, a commodity corresponding to a recommendation score greater than a preset threshold is screened out as a recommended commodity, and the preset threshold may be set by a person skilled in the art as needed.

As can be known from the exemplary embodiments of the present application, the technical solution of the present application has various advantages, including but not limited to the following:

Referring to fig. 2, in a further embodiment, the step S1100 of obtaining the commodity pictures of the total quantity of commodities and determining a plurality of commodity labels corresponding to each commodity by using a preset image-text matching model includes the following steps:

step S1110, obtaining commodity pictures of a total quantity of commodities, extracting image feature information of the commodity pictures by an image encoder in a preset image-text matching model, and extracting text feature information corresponding to each commodity label in a preset commodity label set by a text encoder in the preset image-text matching model;

the graphics context matching model and the corresponding model selection of the image encoder and the text encoder in the graphics context matching model can be disclosed with reference to the relevant part of step S1100.

In one embodiment, viT is used as an image encoder in the image-text matching model, and Bert is used as a text encoder in the image-text matching model, so that the robustness and generalization capability of the image-text matching model can be ensured. And extracting corresponding characteristics based on the forward reasoning functions of the image encoder and the text encoder, so that the image characteristic information of the commodity picture and the text characteristic information of the commodity label can be extracted.

And calling the data interface to access commodity information of the total commodities in the commodity database, and obtaining commodity pictures of each commodity from the commodity information.

The image characteristic information is image characteristics of the commodity picture extracted by vectorization, and the text characteristic information is text characteristics of the commodity label extracted by vectorization.

Step S1120, for each commodity, calculating a similarity between the image feature information corresponding to the commodity picture and the text feature information corresponding to each commodity label, and screening out a plurality of commodity labels with the similarity greater than a preset threshold.

As can be understood by those skilled in the art, by performing normalization operation on the image feature information of the commodity picture and the text feature information of the commodity label, so that the image feature information and the text feature information are unified to the same dimension, and the similarity between the image feature information and the text feature information can be calculated.

The similarity calculation can be implemented by any one of large-scale vector retrieval engines such as Fa i ss, E l ast i cSearch, M i l vus and the like, and can also be calculated by any one of ready-made algorithms such as cosine similarity, inner product, manhattan distance, euclidean distance and the like.

In the embodiment, the image characteristic information and the text characteristic information corresponding to the commodity picture and the commodity label can be accurately extracted through the image-text matching model, so that the commodity picture and the commodity label can be determined to be matched based on the similarity between the image characteristic information and the text characteristic information of uniform dimensionality, and the method is very efficient and convenient.

Referring to fig. 3, in a further embodiment, in step S1200, taking each commodity label as a node, determining a co-occurrence score between every two nodes, and constructing a knowledge graph, includes the following steps:

step S1210, taking each commodity label as a node, if two nodes are the same, taking the co-occurrence score as a preset value, otherwise, calculating the number of commodities as the co-occurrence times when every two nodes belong to the same commodity at the same time, normalizing the co-occurrence times, and obtaining the co-occurrence score between every two nodes;

taking each commodity label as a node, if two nodes are the same, the co-occurrence score between the two nodes is a preset value, the preset value can be 1, otherwise, the number of commodities when every two nodes belong to the same commodity is calculated as the co-occurrence frequency, for example, a node a and a node b belong to a commodity a, a commodity b and a commodity c at the same time, the corresponding co-occurrence frequency is 3, the co-occurrence frequency is normalized, the value range thereof belongs to [0-1], and the co-occurrence score between every two nodes is obtained.

The normalization can be performed by one skilled in the art by any of truncation, binning, logarithmic transformation, min-max normalization, center normalization, etc., as desired.

Step S1220, connecting the plurality of nodes corresponding to each commodity with the plurality of nodes corresponding to other commodities to form an edge, where the weight of each edge is the co-occurrence score of two nodes connected to the edge, and thus constructing a knowledge graph.

Aiming at each commodity, each corresponding node is respectively connected with a plurality of nodes corresponding to other commodities, and an edge is formed between every two connected nodes.

In this embodiment, each commodity label is taken as a node, a co-occurrence score between every two nodes is determined, a plurality of nodes corresponding to each commodity are respectively connected with a plurality of nodes corresponding to other commodities to form an edge, and a knowledge graph is constructed, so that the association between each commodity can be accurately represented in a fine-grained manner through the quantized co-occurrence scores between the plurality of nodes corresponding to each commodity and the plurality of nodes corresponding to other commodities.

Referring to fig. 4, in a further embodiment, before obtaining the product pictures of the full amount of products in step S1100, the method further includes the following steps:

step S1010, obtaining commodity titles of a plurality of commodities, classifying each commodity title according to the commodity category, and obtaining the commodity title corresponding to each commodity category;

generally, the e-commerce platform sets a commodity category system for serving each independent station, where the commodity category system includes multiple levels of commodity categories, each upper level commodity category includes one or more lower levels of commodity categories, exemplarily, the upper level commodity category is clothing, the corresponding lower level commodity category is jacket, under-coat, hat, scarf, waistband, and sock, the next lower level commodity category corresponding to the lower level commodity category is short-sleeved T-shirt, long-sleeved T-shirt, sweater, and coat, and the next lower level commodity category corresponding to the lower level commodity category is shorts/pants, jeans, casual pants/pants, and sports pants, and accordingly, the independent stations of the commodity category system are applied, so that the commodities of the independent stations have the corresponding commodity categories, and it can be understood that the commodity category of the commodity is generally composed of one or multiple levels of level commodity categories, exemplarily, short-sleeved T-shirt, clothing-under-jeans.

And calling the corresponding prepackaged data interfaces of the plurality of independent sites, accessing the commodity information of the total amount of commodities in the commodity database of each independent site, and acquiring the commodity title and the commodity category of each commodity from the commodity database, so that the abundant commodity titles are acquired, and the commodity labels extracted from the commodities are ensured to be abundant to a certain extent. Further, the product titles corresponding to the product categories are obtained by classifying the product categories to which the products corresponding to the product titles belong.

Step S1020, segmenting each commodity title, counting word frequencies and inverse text frequency indexes of each word element under corresponding commodity categories, and calculating keyword scores;

and preprocessing each commodity title, wherein the preprocessing can be any one or more of punctuation removal operation and stop word removal operation, a word segmentation algorithm is adopted to segment each commodity title to obtain a plurality of word elements corresponding to each commodity title, and the word segmentation algorithm can be any one of j i eba and N-gram.

And counting the frequency of each word element in all the commodity titles under the corresponding commodity category, namely word frequency, aiming at each word element corresponding to each commodity title under each commodity category, counting the number of the commodity titles of each word element under the corresponding commodity category, calculating the total number of the commodity titles under the corresponding commodity category divided by the number of the commodity titles, and taking the logarithm of the calculation result as an inverse text frequency index. And further, calculating the word frequency corresponding to each word element and multiplying the word frequency by the inverse text frequency index to obtain a keyword score.

And step S1030, screening out the word elements corresponding to the keyword scores meeting the preset conditions as commodity labels, and constructing a commodity label set.

And aiming at each commodity category, sequencing each word element corresponding to each commodity title under each commodity category in the order of the keyword scores from high to low, and screening a plurality of word elements which are sequenced at the top and correspond to each commodity category as commodity labels, wherein the quantity of the commodity labels can be flexibly set by a person skilled in the art.

In the embodiment, the keyword scores of the word elements corresponding to the commodity titles of the commodity categories are calculated, so that the word elements with higher keyword scores are screened out and used as the commodity labels to construct the commodity label set, the commodity labels can be ensured to be abundant, the method is very simple and efficient, and manual intervention is not needed.

Referring to fig. 5, in a further embodiment, before obtaining the product pictures of the full amount of products in step S1100, the method further includes the following steps:

s1000, acquiring commodity pictures of a plurality of commodities, forming a plurality of image-text data pairs with each commodity label in a preset commodity label set respectively as training samples, and marking supervision labels according to whether the commodity pictures and the commodity labels in the image-text data pairs of the training samples are matched with the corresponding marking supervision labels or not;

in one embodiment, the training samples may be batched, each batch containing multiple pairs of teletext data, each time a batch of training samples is input to the teletext matching model for training. And according to the image-text data of the training sample, matching the commodity picture and the commodity label, marking the supervision label as 1, and if not, marking the supervision label as 0.

Step S1001, inputting the training samples into a graph-text matching model, extracting deep semantic information corresponding to commodity pictures and commodity labels in each graph-text data pair, and obtaining corresponding image characteristic information and text characteristic information;

a plurality of image-text data pairs are input into the image-text matching model, and in one embodiment, viT is used as an image encoder in the image-text matching model, and Bert is used as a text encoder in the image-text matching model, so that the robustness and generalization capability of the image-text matching model can be ensured. And extracting corresponding features, namely the deep semantic information based on the forward reasoning functions of the image encoder and the text encoder, so that the image feature information of the commodity picture and the text feature information of the commodity label in each image-text data pair can be extracted.

Step S1002, mapping image characteristic information and text characteristic information corresponding to each image-text data pair to the same multi-modal space, and calculating the similarity between the image characteristic information and the text characteristic information corresponding to each image-text data pair;

because the image characteristic information and the text characteristic information are data of different modes, the image characteristic information and the text characteristic information corresponding to each image-text data pair are subjected to linear mapping and mapped to the same multi-mode space to obtain the image characteristic information and the text characteristic information of the same dimension, so that the similarity between the image characteristic information and the text characteristic information corresponding to each image-text data pair is calculated based on the same dimension, and the similarity can be calculated by adopting a cosine similarity calculation method.

And S1003, determining a loss value of the similarity by adopting a supervision label of the training sample, updating the weight of the image-text matching model when the loss value does not reach a preset threshold value, and continuously calling other training samples to carry out iterative training until the model converges.

Calling a preset cross entropy loss function, wherein the preset cross entropy loss function can be flexibly set by a person skilled in the art according to prior knowledge or experimental experience, calculating a cross entropy loss value of the similarity of each corresponding image-text data pair based on a supervision label of the training sample, and when the loss value reaches a preset threshold value, indicating that the image-text matching model is trained to a convergence state, so that model training can be stopped; and when the loss value does not reach the preset threshold value, indicating that the model is not converged, performing gradient updating on the model according to the loss value, usually correcting the weight parameters of each link of the model through back propagation to further approximate the model to be converged, and then continuously calling other batches of training samples to perform iterative training on the model until the model is trained to be in a convergence state.

In the embodiment, the image-text matching model is trained to be convergent through supervision, so that the image-text matching model can accurately determine the similarity between the commodity picture and the commodity label according to the commodity picture and the commodity label, and the high-quality target advertisement text can be constructed according to the corresponding word extraction of the optimized probability distribution.

Referring to fig. 6, a commodity recommendation device adapted to one of the purposes of the present application is a functional embodiment of the commodity recommendation method of the present application, and the commodity recommendation device includes a commodity marking module 1100, a map building module 1200, a score calculating module 1300, and a commodity screening module 1400, where the commodity marking module 1100 is configured to obtain commodity pictures of a full quantity of commodities, and determine a plurality of commodity labels corresponding to each commodity by using a preset image-text matching model, where the commodity labels belong to members of a preset commodity label set; the map construction module 1200 is configured to determine a co-occurrence score between every two nodes by using each commodity label as a node, and construct a knowledge map; the score calculating module 1300 is configured to calculate a recommendation score among the commodities according to a co-occurrence score among the plurality of nodes corresponding to the commodities in the knowledge graph and the plurality of nodes corresponding to other commodities; the commodity screening module 1400 is configured to screen, for each commodity, a recommended commodity corresponding to a recommendation score that meets a preset condition.

In a further embodiment, the commodity screening module 1400 includes: the feature extraction sub-module is used for acquiring commodity pictures of a total quantity of commodities, extracting image feature information of the commodity pictures by an image encoder in a preset image-text matching model, and extracting text feature information corresponding to each commodity label in a preset commodity label set by a text encoder in the preset image-text matching model; and the label determining submodule is used for calculating the similarity between the image characteristic information corresponding to the commodity picture of each commodity and the text characteristic information corresponding to each commodity label aiming at each commodity, and screening out a plurality of commodity labels with the similarity larger than a preset threshold value.

In a further embodiment, the map building module 1200 includes: the co-occurrence score determining submodule is used for taking each commodity label as a node, if the two nodes are the same, the co-occurrence score is a preset value, otherwise, the number of commodities is calculated as the co-occurrence times when every two nodes belong to the same commodity at the same time, the co-occurrence times are normalized, and the co-occurrence score between every two nodes is obtained; and the knowledge graph constructing submodule is used for connecting the nodes corresponding to the commodities with the nodes corresponding to other commodities to form edges, and the weight of each edge is the co-occurrence score of the two nodes connected with the edge to construct the knowledge graph.

In a further embodiment, the score calculating module 1300 includes: and the recommendation score calculation submodule is used for determining the co-occurrence scores between the plurality of nodes corresponding to each commodity and the plurality of nodes corresponding to other commodities according to the knowledge graph aiming at each commodity, calculating the sum of the plurality of corresponding co-occurrence scores and dividing the sum by the number of the nodes corresponding to the commodities to obtain the recommendation scores among the commodities.

In a further embodiment, before the article marking module 1100, the method further includes: the category title module is used for acquiring the commodity titles of a plurality of commodities, classifying each commodity title according to the commodity category and acquiring the commodity title corresponding to each commodity category; the keyword score calculation module is used for segmenting each commodity title, counting word frequency and inverse text frequency indexes of each word element under the corresponding commodity category and calculating keyword scores; and the label screening module is used for screening out the lemmas corresponding to the keyword scores meeting the preset conditions as commodity labels to construct a commodity label set.

In a further embodiment, before the article marking module 1100, the method further includes: the system comprises a sample construction and labeling module, a label analysis module and a label analysis module, wherein the sample construction and labeling module is used for acquiring commodity pictures of a plurality of commodities to form a plurality of picture and text data pairs with each commodity label in a preset commodity label set as training samples, and whether the commodity pictures and the commodity labels are matched with corresponding labeling monitoring labels in the picture and text data pairs of the training samples is determined; the feature extraction module is used for inputting the training samples into the image-text matching model, extracting deep semantic information corresponding to the commodity images and the commodity labels in each image-text data pair, and obtaining corresponding image feature information and text feature information; the similarity calculation module is used for mapping the image characteristic information and the text characteristic information corresponding to each image-text data pair to the same multi-modal space and calculating the similarity between the image characteristic information and the text characteristic information corresponding to each image-text data pair; and the iterative training module is used for determining the loss value of the similarity by adopting the supervision label of the training sample, updating the weight of the image-text matching model when the loss value does not reach a preset threshold value, and continuously calling other training samples to carry out iterative training until the model converges.

In a further embodiment, the commodity screening module 1400 includes: and the sequencing optimization submodule is used for sequencing each commodity according to the recommendation scores between each commodity and other commodities, and screening out a plurality of recommended commodities which are sequenced in the front.

In order to solve the technical problem, an embodiment of the present application further provides a computer device. Fig. 7 is a schematic diagram of the internal structure of the computer device. The computer device includes a processor, a computer-readable storage medium, a memory, and a network interface connected by a system bus. The computer-readable storage medium of the computer device stores an operating system, a database and computer-readable instructions, the database can store control information sequences, and the computer-readable instructions, when executed by the processor, can cause the processor to implement a commodity recommendation method. The processor of the computer device is used for providing calculation and control capability and supporting the operation of the whole computer device. The memory of the computer device may have stored therein computer readable instructions, which, when executed by the processor, may cause the processor to perform the merchandise recommendation method of the present application. The network interface of the computer device is used for connecting and communicating with the terminal. Those skilled in the art will appreciate that the architecture shown in fig. 7 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.

In this embodiment, the processor is configured to execute specific functions of each module and its sub-module in fig. 6, and the memory stores program codes and various data required for executing the modules or sub-modules. The network interface is used for data transmission to and from a user terminal or a server. The memory in this embodiment stores program codes and data necessary for executing all modules/submodules in the product recommendation device of the present application, and the server can call the program codes and data of the server to execute the functions of all the submodules.

The present application also provides a storage medium storing computer-readable instructions, which when executed by one or more processors, cause the one or more processors to perform the steps of the method of recommending items of any of the embodiments of the present application.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments of the present application can be implemented by a computer program, which can be stored in a computer-readable storage medium, and when the computer program is executed, the processes of the embodiments of the methods can be included. The storage medium may be a magnetic disk, an optical disk, a Read-only Memory (ROM) or other computer readable storage medium, or a Random Access Memory (RAM).

In summary, the cross-border e-commerce platform based on the independent site provides a commodity recommendation solution and can improve service experience.

Those of skill in the art will appreciate that the various operations, methods, steps in the processes, acts, or solutions discussed in this application can be interchanged, modified, combined, or eliminated. Further, other steps, measures, or schemes in various operations, methods, or flows that have been discussed in this application can be alternated, altered, rearranged, broken down, combined, or deleted. Further, the steps, measures, and schemes in the various operations, methods, and flows disclosed in the present application in the prior art can also be alternated, modified, rearranged, decomposed, combined, or deleted.

The foregoing is only a partial embodiment of the present application, and it should be noted that, for those skilled in the art, several modifications and decorations can be made without departing from the principle of the present application, and these modifications and decorations should also be regarded as the protection scope of the present application.

Claims

1. A method for recommending an article, comprising:

the method comprises the steps of obtaining commodity pictures of a total quantity of commodities, and determining a plurality of commodity labels corresponding to each commodity by adopting a preset image-text matching model, wherein the commodity labels belong to members of a preset commodity label set;

2. The commodity recommendation method according to claim 1, wherein commodity pictures of a whole quantity of commodities are obtained, and a plurality of commodity labels corresponding to each commodity are determined by adopting a preset picture-text matching model, and the method comprises the following steps:

and calculating the similarity between the image characteristic information corresponding to the commodity picture and the text characteristic information corresponding to each commodity label aiming at each commodity, and screening out a plurality of commodity labels with the similarity larger than a preset threshold value.

3. The commodity recommendation method according to claim 1, wherein co-occurrence scores between every two nodes are determined by taking each commodity label as a node, and a knowledge graph is constructed, and the method comprises the following steps:

4. The commodity recommendation method according to claim 1, wherein calculating the recommendation score between each commodity according to the co-occurrence scores between the plurality of nodes corresponding to each commodity and the plurality of nodes corresponding to other commodities in the knowledge graph includes:

5. The product recommendation method according to claim 1, further comprising, before acquiring product pictures of a total number of products, the steps of:

6. The product recommendation method according to claim 1, further comprising, before acquiring the product pictures of the entire amount of products, the steps of:

7. The commodity recommendation method according to claim 1, wherein screening out, for each commodity, a recommended commodity corresponding to a recommendation score that satisfies a preset condition includes:

8. An article recommendation device, comprising:

the commodity marking module is used for acquiring commodity pictures of a whole quantity of commodities, and determining a plurality of commodity labels corresponding to each commodity by adopting a preset image-text matching model, wherein the commodity labels belong to members of a preset commodity label set;

the map building module is used for determining the co-occurrence score between every two nodes by taking each commodity label as a node, and building a knowledge map;

the score calculation module is used for calculating a recommendation score among the commodities according to the co-occurrence scores among the nodes corresponding to the commodities in the knowledge graph and the nodes corresponding to other commodities;

and the commodity screening module is used for screening out recommended commodities corresponding to the recommendation scores meeting the preset conditions for each commodity.

9. A computer device comprising a central processor and a memory, characterized in that the central processor is adapted to invoke execution of a computer program stored in the memory to perform the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that it stores, in the form of computer-readable instructions, a computer program implemented according to the method of any one of claims 1 to 7, which, when invoked by a computer, performs the steps comprised by the corresponding method.