CN116823404A

CN116823404A - Commodity combination recommendation method, device, equipment and medium thereof

Info

Publication number: CN116823404A
Application number: CN202310864293.7A
Authority: CN
Inventors: 冯一丁
Original assignee: Guangzhou Shangyan Network Technology Co ltd
Current assignee: Guangzhou Shangyan Network Technology Co ltd
Priority date: 2023-07-13
Filing date: 2023-07-13
Publication date: 2023-09-29

Abstract

The application relates to a commodity combination recommendation method, a device, equipment and a medium thereof in the technical field of electronic commerce, wherein the method comprises the following steps: and constructing a corresponding commodity combination set, a commodity combination set and a brand combination set based on a user purchase data set corresponding to each online store by adopting a correlation analysis algorithm, recalling candidate commodity combinations in the commodity combination set of the online store according to commodity identifications, commodity categories and commodity brands of target commodities in the online store, recalling target commodity combinations and target brand combinations in the commodity combination set and the brand combination set of all online stores, forming candidate commodity combinations according to commodity categories and commodity brands of commodity combinations corresponding to the target commodities in the combinations, correspondingly recalling first commodities and second commodities and the target commodities, and adopting a purchase prediction model to predict purchase rates corresponding to the candidate commodity combinations containing the target commodities, thereby optimizing the recommended commodity combinations. Whereby a variety of commodity combinations can be recommended.

Description

Commodity combination recommendation method, device, equipment and medium thereof

Technical Field

The present application relates to the field of electronic commerce technologies, and in particular, to a method for recommending a combination of commodities, and a corresponding apparatus, computer device, and computer readable storage medium thereof.

Background

The commodity combination promotion refers to the binding sales of a plurality of commodities to provide greater value and attraction to consumers so that consumers can purchase the plurality of commodities at a lower price or with more added value, thereby encouraging consumers to purchase more commodities and further increasing sales amount and sales commodity quantity.

In the conventional technology, closely related commodities are obtained by analysis based on user purchase data of online shops, and the obtained commodity combination is recommended to a consumer user, however, the online shops which are just established and the online shops which are soon established are seriously lack of user purchase data, so that corresponding commodity combinations cannot be obtained by analysis, and even if the online shops have a certain amount of user purchase data, the obtained commodity combination is not examined for the recommended effect, and the effect is likely to be unexpected.

In view of the shortcomings of the conventional technology, the inventor conducts research in the related field for a long time, and develops a new way for solving the problem in the field of electronic commerce.

Disclosure of Invention

It is therefore a primary object of the present application to solve at least one of the above problems and provide a commodity combination recommendation method, and a corresponding apparatus, computer device, and computer-readable storage medium thereof.

In order to meet the purposes of the application, the application adopts the following technical scheme:

the application provides a commodity combination recommendation method which is suitable for one of the purposes of the application, and comprises the following steps:

constructing a commodity combination set, a class combination set and a brand combination set corresponding to each online store by adopting a correlation analysis algorithm according to a user purchase data set corresponding to each online store, wherein the user purchase data set comprises user purchase data generated when a user purchases a commodity of the corresponding online store, and comprises a user identifier, a commodity class and a commodity brand;

according to commodity identification, commodity class and commodity brand of target commodity in the target online store, recall candidate commodity combination from commodity combination set of target online store, and recall target commodity combination and target brand combination from class combination set, brand combination set of all online stores;

according to the target product combination, the target brand combination recalls a first product corresponding to the product type of the target product combination, forms a candidate product combination with the target product, and recalls a second product corresponding to the product brand of the target product combination, and forms a candidate product combination with the target product;

Predicting all candidate commodity combinations containing target commodities by adopting a preset purchase prediction model, determining the purchase rate corresponding to each candidate commodity combination, and screening out recommended commodity combinations with the purchase rate meeting preset conditions.

In a further embodiment, according to the user purchase data set corresponding to each online store, a commodity combination set, a category combination set and a brand combination set corresponding to each online store are constructed by adopting a correlation analysis algorithm, and the method comprises the following steps:

acquiring user purchase data sets of each online store, clustering corresponding dimensions of commodity identifications, commodity categories and commodity brands by taking users as clustering units, and determining each data set corresponding to each dimension;

based on the data sets, adopting an association analysis algorithm to respectively determine a combination set of corresponding dimensions, wherein the combination set comprises a commodity combination set, a class combination set and a brand combination set;

determining the support degree and the confidence degree of each combination in the commodity combination set, the class combination set and the brand combination set, wherein the support degree represents the frequency of occurrence of the corresponding combination, and the confidence degree represents the relationship tightness degree of the corresponding combination;

and eliminating combinations of the commodity combination set, the class combination set and the combination of which the support and/or the confidence in the brand combination set do not meet preset conditions.

In a further embodiment, recall candidate commodity combinations from the commodity combination set of the on-line stores and recall target commodity combinations and target brand combinations from the commodity combination set of all on-line stores according to commodity identifications, commodity classes and commodity brands of target commodities in the on-line stores, comprising the steps of:

the candidate commodity combination containing the commodity identification is recalled from the commodity combination of the target online store according to the commodity identification of the target commodity in the target online store;

recall the candidate class combination of commodity class containing the target commodity from the class combination set of all online stores, calculate the corresponding sorting score of each candidate class combination according to the support and/or confidence matching weight of the candidate class combination corresponding to the corresponding class combination set, screen out the target class combination with the sorting score meeting the preset condition, the weight is the similarity between the online store corresponding to the class combination set of the candidate class combination and the target online store;

and recalling candidate brand class combinations containing the commodity brands of the target commodity from the brand combination sets of all online stores, calculating the sorting score corresponding to each candidate brand combination according to the support degree and/or confidence matching weight of the candidate brand combination corresponding to the corresponding brand combination set, and screening out the target brand combination with the sorting score meeting the preset condition, wherein the weight is the similarity between the online store corresponding to the brand combination set of the candidate brand combination and the target online store.

In a further embodiment, before the candidate commodity combination containing the commodity identification is recalled from the commodity combination of the on-line shop according to the commodity identification of the target commodity in the on-line shop, the method comprises the following steps:

acquiring all commodity categories corresponding to each online store, and constructing corresponding category distribution;

and calculating the similarity between the online stores according to the class distribution corresponding to each online store.

In a further embodiment, predicting all candidate commodity combinations including the target commodity by using a preset purchase prediction model, and determining a purchase rate corresponding to each candidate commodity combination, including the following steps:

acquiring commodity information of each commodity contained in each candidate commodity combination in all candidate commodity combinations containing target commodities, wherein the commodity information comprises commodity description texts and commodity pictures;

extracting image-text fusion information corresponding to commodity information of each commodity contained in the candidate commodity combination by adopting a preset purchase prediction model, and splicing to form a combination characteristic representation;

and determining the purchase rate of the candidate commodity combination based on the combination characteristic representation.

In a further embodiment, recalling a first commodity corresponding to a commodity class of the commodity class combination of the target commodity, forming a candidate commodity combination with the target commodity, and recalling a second commodity corresponding to a commodity brand of the commodity class combination of the target commodity, forming a candidate commodity combination with the target commodity, according to the target class combination, the method comprises the following steps:

According to the target product combination, recalling a plurality of first candidate products corresponding to the product products of the product combination of the target product, determining sales results corresponding to each first candidate product, screening out first products with sales results meeting preset conditions, and forming a candidate product combination with the target product, wherein the sales results are determined according to any one or more of click rate, conversion rate and sales volume of the first candidate products;

and according to the target brand combination recall, a plurality of second candidate commodities corresponding to commodity brands of the commodity brand combination of the target commodity, determining sales results corresponding to each second candidate commodity, screening out the second commodities with sales results meeting preset conditions, and forming a candidate commodity combination with the target commodity, wherein the sales results are determined according to any one or more of click rate, conversion rate and sales volume of the second candidate commodity.

In a further embodiment, predicting all candidate commodity combinations including the target commodity by using a preset purchase prediction model, and before determining the purchase rate corresponding to each candidate commodity combination, the method includes the following steps:

acquiring a single training sample and a supervision tag thereof from a prepared training set, wherein the training sample is commodity information of all commodities contained in a commodity combination, and the supervision tag characterizes whether the commodity combination of the training sample is purchased by a user or not;

Extracting image-text fusion information corresponding to each commodity in a training sample by adopting a preset purchase prediction model, and splicing to form a combined characteristic representation;

determining a predicted outcome of the candidate commodity combination based on the combination feature representation;

and determining a loss value of the predicted purchase rate by adopting a supervision tag of the training sample, updating the weight of the purchase prediction model when the loss value does not reach a preset threshold, and continuously calling other training samples to perform iterative training until the purchase prediction model converges.

On the other hand, the commodity combination recommending device provided by the application is suitable for one of the purposes of the application, and comprises a combination set constructing module, a combination recall module, a candidate recall module and a purchase rate predicting module, wherein the combination set constructing module is used for constructing a commodity combination set, a class combination set and a brand combination set corresponding to each online store by adopting a correlation analysis algorithm according to a user purchase data set corresponding to each online store, and the user purchase data set comprises user purchase data generated when a user purchases a commodity of the corresponding online store, wherein the user purchase data set comprises a user identifier, a commodity class and a commodity brand; the combination recall module is used for recalling candidate commodity combinations from the commodity combination sets of the online stores according to the commodity identification, commodity class and commodity brand of the target commodity in the online stores, and recalling target commodity combinations and target brand combinations from the commodity combination sets, the brand combination sets of all the online stores; a candidate recall module, configured to recall, according to the target item combination, a first item corresponding to a item category of the target item combination, a candidate item combination with a target item, and a second item corresponding to a item brand of the target item combination, and a candidate item combination with a target item; and the purchase rate prediction module is used for predicting all candidate commodity combinations containing the target commodity by adopting a preset purchase prediction model, determining the purchase rate corresponding to each candidate commodity combination, and screening out recommended commodity combinations with the purchase rate meeting preset conditions.

In a further embodiment, the combined set construction module includes: the multidimensional clustering sub-module is used for acquiring user purchase data sets of each online store, clustering corresponding dimensions of commodity identifications, commodity categories and commodity brands by taking users as clustering units, and determining each data set corresponding to each dimension; the association analysis sub-module is used for respectively determining combination sets of corresponding dimensions based on the data sets by adopting an association analysis algorithm, wherein the combination sets comprise a commodity combination set, a category combination set and a brand combination set; the numerical value determining submodule is used for determining the support degree and the confidence degree of each combination in the commodity combination set, the class combination set and the brand combination set, wherein the support degree represents the frequency of occurrence of the corresponding combination, and the confidence degree represents the relationship tightness degree of the corresponding combination; and the combination eliminating sub-module is used for eliminating combinations of the commodity combination set, the class combination set and the brand combination set, wherein the support degree and/or the confidence degree of the combination set do not accord with preset conditions.

In a further embodiment, the combined recall module comprises: the commodity combination recall sub-module is used for intensively recalling candidate commodity combinations containing the commodity identifications from the commodity combinations of the target online shops according to the commodity identifications of the target commodities in the target online shops; the first sorting recall sub-module is used for recalling candidate class combinations containing commodity classes of the target commodity from class combinations of all online shops, calculating sorting scores corresponding to each candidate class combination according to support and/or confidence matching weights of the candidate class combinations corresponding to the corresponding class combination sets, screening out target class combinations with sorting scores meeting preset conditions, wherein the weights are the similarity between the online shops corresponding to the class combination sets where the candidate class combinations are located and the target online shops; and the second sorting recall sub-module is used for recalling candidate brand combinations containing commodity brands of the target commodity from the brand combination sets of all online shops, calculating sorting scores corresponding to each candidate brand combination according to the support degree and/or confidence matching weight of the candidate brand combination corresponding to the corresponding brand combination set, screening out target brand combinations with sorting scores meeting preset conditions, and the weight is the similarity between the online shops corresponding to the brand combination set where the candidate brand combination is located and the target online shops.

In a further embodiment, prior to the merchandise assembly recall sub-module, comprising: the product distribution construction submodule is used for acquiring all product products corresponding to each online store and constructing corresponding product distribution; and the similarity calculation sub-module is used for calculating the similarity between the online stores according to the category distribution corresponding to each online store.

In a further embodiment, the purchase rate predicting module includes: the information acquisition sub-module is used for acquiring commodity information of each commodity contained in each candidate commodity combination in all candidate commodity combinations containing target commodities, wherein the commodity information comprises commodity description texts and commodity pictures; the first feature extraction submodule is used for extracting image-text fusion information corresponding to commodity information of each commodity contained in the candidate commodity combination by adopting a preset purchase prediction model to splice, so as to form a combination feature representation; and the purchase rate determination submodule is used for determining the purchase rate of the candidate commodity combination based on the combination characteristic representation.

In a further embodiment, the candidate recall module comprises: the first candidate recall sub-module is used for recalling a plurality of first candidate commodities corresponding to commodity classes of commodity class combination of the target commodity according to the target commodity class combination, determining sales results corresponding to each first candidate commodity, screening out first commodities with sales results meeting preset conditions, and forming a candidate commodity combination with the target commodity, wherein the sales results are determined according to any one or more of click rate, conversion rate and sales volume of the first candidate commodity; and the second candidate recall sub-module is used for recalling a plurality of second candidate commodities corresponding to commodity brands of the target commodity according to the target brand combination, determining sales results corresponding to each second candidate commodity, screening out the second commodities with sales results meeting preset conditions, and forming a candidate commodity combination with the target commodity, wherein the sales results are determined according to any one or more of the click rate, the conversion rate and the sales quantity of the second candidate commodities.

In a further embodiment, before the purchase rate predicting module, the purchase rate predicting module includes: the training preparation sub-module is used for acquiring a single training sample and a supervision label thereof from a prepared training set, wherein the training sample is commodity information of all commodities contained in the commodity combination, and the supervision label characterizes whether the commodity combination of the training sample is purchased by a user or not; the second feature extraction submodule is used for extracting image-text fusion information corresponding to each commodity in the training sample by adopting a preset purchase prediction model to splice, so as to form a combined feature representation; the model prediction sub-module is used for determining a prediction result of the candidate commodity combination based on the combination characteristic representation; and the iterative training sub-module is used for determining a loss value of the predicted purchase rate by adopting a supervision label of the training sample, updating the weight of the purchase prediction model when the loss value does not reach a preset threshold value, and continuously calling other training samples to perform iterative training until the purchase prediction model converges.

In yet another aspect, a computer device adapted to one of the objects of the present application comprises a central processor and a memory, said central processor being adapted to invoke the steps of running a computer program stored in said memory for performing the product combination recommendation method according to the present application.

In yet another aspect, a computer readable storage medium adapted to another object of the present application stores a computer program implemented according to the commodity combination recommendation method in the form of computer readable instructions, which when invoked by a computer, performs the steps included in the method.

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

the application constructs a corresponding commodity combination set, a commodity combination set and a brand combination set based on a user purchase data set corresponding to each online store by adopting a correlation analysis algorithm, recalls candidate commodity combinations in the commodity combination set of the online stores according to commodity identifications, commodity categories and commodity brands of target commodities in the online stores, recalls target commodity combinations and target brand combinations in the commodity combination set and the brand combination set of all online stores according to commodity categories and commodity brands of the commodity combinations corresponding to the target commodities, recalls the first commodity and the second commodity respectively and forms the candidate commodity combinations corresponding to the target commodities, and adopts a purchase prediction model to predict purchase rates corresponding to the candidate commodity combinations containing the target commodities, thereby optimizing the recommended commodity combinations. On one hand, the problem of cold start of commodity combination recommendation can be solved, so that rich commodity combinations can be recommended for online shops which seriously lack user purchase data, and the commodity combinations comprise commodity combinations related to commodity categories and commodity brand various dimension combinations. More abundant commodity combinations can be recommended for online shops with a certain amount of user purchase data, including commodity combinations related to commodity categories, commodity brands and various dimension combinations of commodities per se. On the other hand, the recommended commodity combination is optimized according to the purchase rate of the candidate commodity combination, so that the recommended commodity combination is expected to attract users, the users are stimulated to purchase the commodity combination, and the commodity combination sales are promoted.

Drawings

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

FIG. 1 is a flow chart of an exemplary embodiment of a method for combined recommendation of goods according to the present application;

FIG. 2 is a flow chart of removing combinations of merchandise sets, brand sets, and combinations of items that do not meet a predetermined condition;

FIG. 3 is a schematic flow chart of recall of candidate commodity combinations, target class combinations, and target brand combinations in an embodiment of the present application;

FIG. 4 is a flow chart of determining similarity between online stores in an embodiment of the application;

FIG. 5 is a flowchart illustrating a method for determining a purchase rate of a candidate combination of goods using a purchase prediction model according to an embodiment of the application;

FIG. 6 is a schematic flow chart of a candidate merchandise combination recall in an embodiment of the application;

FIG. 7 is a flow chart of a training process for purchasing a predictive model in accordance with an embodiment of the application;

FIG. 8 is a schematic block diagram of a product combination recommendation device of the present application;

fig. 9 is a schematic structural diagram of a computer device used in the present application.

Detailed Description

Embodiments of the present application are described in detail below, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the drawings are illustrative only and are not to be construed as limiting the application.

As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless expressly stated otherwise, as understood by those skilled in the art. 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. The term "and/or" as used herein includes all or any element and all combination of one or more of the associated listed items.

It will be understood by those skilled in the art that 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 unless defined otherwise. 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 used herein, "client," "terminal device," and "terminal device" are understood by those skilled in the art to include both devices that include only wireless signal receivers without transmitting capabilities and devices that include receiving and transmitting hardware capable of two-way communication over a two-way communication link. Such a device may include: a cellular or other communication device such as a personal computer, tablet, or the like, having a single-line display or a multi-line display or a cellular or other communication device without a multi-line display; a PCS (Personal Communications Service, personal communication system) that may combine voice, data processing, facsimile and/or data communication capabilities; a PDA (Personal Digital Assistant ) that can include a radio frequency receiver, 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 that has and/or includes a radio frequency receiver. As used herein, "client," "terminal device" may be portable, transportable, installed in a vehicle (aeronautical, maritime, and/or land-based), or adapted and/or configured to operate locally and/or in a distributed fashion, at any other location(s) on earth and/or in space. As used herein, a "client," "terminal device," or "terminal device" may also be a communication terminal, an internet terminal, or a music/video playing terminal, for example, a PDA, a MID (Mobile Internet Device ), and/or a mobile phone with music/video playing function, or may also be a device such as a smart tv, a set top box, or the like.

The application refers to hardware such as a server, a client, a service node, and the like, which essentially is an electronic device with personal computer and other functions, and is a hardware device with necessary components disclosed by von neumann principles such as a central processing unit (including an arithmetic unit and a controller), a memory, an input device, an output device, and the like, wherein a computer program is stored in the memory, and the central processing unit calls the program stored in the memory to run, executes instructions in the program, and interacts with the input and output devices, thereby completing specific functions.

It should be noted that the concept of the present application, called "server", is equally applicable to the case of server clusters. The servers should be logically partitioned, physically separate from each other but interface-callable, or integrated into a physical computer or group of computers, according to network deployment principles understood by those skilled in the art. Those skilled in the art will appreciate this variation and should not be construed as limiting the implementation of the network deployment approach of the present application.

One or more technical features of the present application, unless specified in the clear, may be deployed either on a server for implementation and the client remotely invokes an online service interface provided by the acquisition server for implementation of the access, or may be deployed and run directly on the client for implementation of the access.

The neural network model cited or possibly cited in the application can be deployed on a remote server and can be used for implementing remote call on a client, or can be deployed on a client with sufficient equipment capability for direct call, unless specified by plaintext, and in some embodiments, when the neural network model runs on the client, the corresponding intelligence can be obtained through migration learning so as to reduce the requirement on the running resources of the hardware of the client and avoid excessively occupying the running resources of the hardware of the client.

The various data related to the present application, unless specified in the plain text, may be stored either remotely in a server or in a local terminal device, as long as it is suitable for being invoked by the technical solution of the present application.

Those skilled in the art will appreciate that: although the various methods of the present application are described based on the same concepts so as to be common to each other, the methods may be performed independently of each other unless specifically indicated otherwise. Similarly, for the various embodiments disclosed herein, all concepts described herein are presented based on the same general inventive concept, and thus, concepts described herein with respect to the same general inventive concept, and concepts that are merely convenient and appropriately modified, although different, should be interpreted as equivalents.

The various embodiments of the present application to be disclosed herein, unless the plain text indicates a mutually exclusive relationship with each other, the technical features related to the various embodiments may be cross-combined to flexibly construct a new embodiment as long as such combination does not depart from the inventive spirit of the present application and can satisfy the needs in the art or solve the deficiencies in the prior art. This variant will be known to the person skilled in the art.

The commodity combination recommendation method of the application can be programmed into a computer program product and deployed in a client or a server for operation, for example, in the exemplary application scenario of the application, the commodity combination recommendation method can be deployed in a server of an electronic commerce platform, and thus the method can be executed by accessing an interface opened after the computer program product is operated and performing man-machine interaction with a process of the computer program product through a graphical user interface.

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

step S1100, constructing a commodity combination set, a class combination set and a brand combination set corresponding to each online store by adopting a correlation analysis algorithm according to a user purchase data set corresponding to each online store, wherein the user purchase data set comprises user purchase data generated when a user purchases a commodity of the corresponding online store, and comprises a user identifier, a commodity class and a commodity brand;

The association analysis algorithm can be implemented by using an Apriori algorithm and an FP Growth algorithm, and one skilled in the art can choose one implementation according to requirements. Compared with the Apriori algorithm, the FP Growth algorithm has lower time complexity, is easy to process in parallel, occupies smaller memory space and is more suitable for processing a sparse data set, so that the FP Growth algorithm is recommended to be used for realizing the key analysis algorithm of the application.

The user can select the commodity in the online store, add the selected commodity to the shopping cart and place an order for the purchase of the selected commodity to perform the purchase of the commodity in the online store. In general, when a user adds a selected commodity to a shopping cart, a control for indicating the addition of the selected commodity to the shopping cart needs to be touched on a page displaying the selected commodity, and when touched, user purchase data is generated and stored for calling. In addition, when the user orders the selected commodity to purchase, a control for representing the purchase is also required to be touched on the page displaying the selected commodity, and after the control is touched, user purchase data is generated and stored for calling.

The user identification is an identifier, such as a user ID, for uniquely identifying and identifying an individual user, and the merchandise identification is an identifier, such as a merchandise ID, for uniquely identifying and identifying a merchandise. The commodity class is obtained by classifying according to characteristics such as properties, types, purposes and functions of the commodity, for example, the short sleeve POLO shirt for the jacket, the corresponding classification method can be realized by adopting a line classification method, a surface classification method and the like, and the commodity class can be selected and realized by a person skilled in the art as required. The brand of merchandise is a brand name created by a business or corporation, such as apple, hua Cheng, naak, etc.

Constructing a commodity combination set, a class combination set and a brand combination set corresponding to each online store, specifically taking one online store as an example, clustering commodity identifications, commodity classes and commodity brand dimensions according to user purchase data sets corresponding to the online stores, respectively based on the same user identifications, determining each data set corresponding to each dimension, clustering commodity identification dimensions based on the same user identifications by using an exemplary example, and determining a first data set, for example: { user identification 1: (commodity identification 1, commodity identification 2), user identification 2: (article identification 2, article identification 3, article identification 4), user identification 3 (article identification 1, article identification 3, article identification 5) > clustering commodity class dimensions based on the same user identification to determine a second data set, for example: { user identification 1: (commodity class 1, commodity class 2), user identification 2: (commodity class 2, commodity class 3, commodity class 4), user identification 3 (commodity class 1, commodity class 3, commodity class 5); clustering commodity class dimensions based on the same user identification to determine a second data set, for example: { user identification 1: (commodity class 1, commodity class 2), user identification 2: (commodity class 2, commodity class 3, commodity class 4), user identification 3 (commodity class 1, commodity class 3, commodity class 5); clustering the brand dimensions of the goods based on the same user identification, and determining a third data set, for example: { user identification 1: (commodity brand 1, commodity brand 2), user identification 2: (brand 2, brand 3, brand 4), user identification 3 (brand 1, brand 3, brand 5).

Further, based on the data sets, a combination set of corresponding dimensions is determined by adopting a correlation analysis algorithm, including a commodity combination set, a class combination set and a brand combination set, taking the commodity combination set as an example, in one embodiment, the correlation analysis algorithm implemented by adopting an Apriori algorithm is used, based on the data set of commodity identification dimensions, each user identification is used as a transaction, each commodity identification corresponding to the user identification is used as an item, a transaction set containing all items (commodity identifications) is generated, the transaction set is scanned, the frequency of each item is counted, and the frequency is that the corresponding item appears in several transactions. According to a preset first frequency threshold value, a frequent single item set is determined, wherein the frequent single item set comprises single items with the frequency greater than or equal to the first frequency threshold value. Starting from the frequent single item set, iteratively generating a frequent K item set (K > 1), the specific process comprising: obtaining candidate K item sets through connection and pruning operation of frequent (K-1) item sets; scanning the transaction set, and counting the frequency of each candidate K item set in the transaction set, wherein the frequency is that all items in the corresponding candidate K item set appear in several transactions together; and determining a frequent K item set according to a preset second frequency threshold, wherein the frequent K item set comprises candidate K item sets with the frequency being more than or equal to the second frequency threshold. The iteration is repeated until no new frequent K-term sets can be generated. And according to the frequent K item sets, determining all possible association rules, namely commodity combinations, through permutation and combination, and summarizing all commodity combinations to form a commodity combination set. The first frequency threshold and the second frequency threshold can be set by a person skilled in the art as required. Similarly, the brand combination set and the category combination set can be correspondingly determined based on the commodity brand dimension and the commodity category dimension data set by adopting the association analysis algorithm realized by the Apriori algorithm.

Step S1200, recall candidate commodity combinations from the commodity combination sets of the online stores according to the commodity identification, commodity class and commodity brand of the target commodity in the online stores, and recall target commodity combinations and target brand combinations from the commodity combination sets, the brand combination sets of all online stores;

the target online store may be any online store, and the target commodity may be any commodity in the target store.

The commodity database is generally constructed for storing commodity information of all commodities in the online store, wherein each commodity is associated with a commodity identification of the corresponding commodity, the commodity information comprises commodity description texts and commodity pictures, the commodity description texts comprise commodity brands, commodity classes, product parameters, commodity titles, commodity detail texts and the like, and the commodity pictures comprise commodity main diagrams, commodity detail diagrams and the like. Accordingly, the commodity class and the commodity brand of the commodity identification of the related target commodity are obtained from the commodity database of the target online store.

And according to the commodity identifications of the target commodities, the commodity combination containing the commodity identifications is recalled from the commodity combination of the target online shops in a centralized manner, and commodities corresponding to the commodity identifications contained in the commodity combination are combined to form candidate commodity combinations, wherein the target commodities are necessarily contained. And recalling the brand combinations of all online shops according to the brand of the target commodity to intensively contain the brand combinations of the commodity brand as target brand combinations. And recalling the commodity combination of the online store according to the commodity combination of the target commodity, wherein the commodity combination is intensively contained as the target commodity combination.

Step S1300, recall a first commodity corresponding to the commodity class of the commodity class combination of the target commodity according to the target commodity class combination, and recall a second commodity corresponding to the commodity brand of the commodity class combination of the target commodity, and form a candidate commodity combination with the target commodity;

and recalling the commodity with the commodity description text in the commodity database as a first commodity according to the commodity in the commodity combination with the target commodity. And combining each first commodity with the target commodity to form a candidate commodity combination.

And recalling the commodities with the commodity brands in the commodity description text from a commodity database as second commodities according to the commodity brands in the target brand combination and the commodity brand combination of the target commodities. And combining each second commodity with the target commodity to form a candidate commodity combination.

And S1400, predicting all candidate commodity combinations containing the target commodity by adopting a preset purchase prediction model, determining the purchase rate corresponding to each candidate commodity combination, and screening out recommended commodity combinations with the purchase rate meeting preset conditions.

The purchase prediction model is trained in advance until convergence, and the capability of the purchase rate of the commodity combination is determined based on the commodity information of each commodity contained in the commodity combination. The training process of the purchase prediction model is further disclosed by the following partial embodiments, and the step is temporarily pressed against the table.

The purchase rate represents the probability that the corresponding candidate commodity combination is purchased by the user, and it is easy to understand that factors affecting the purchase rate of the candidate commodity combination are related to commodity description texts and commodity pictures of the candidate commodity combination containing various commodities, and accordingly the purchase prediction model of the application is correspondingly modeled. The purchase prediction model comprises an image encoder, a text encoder and a classifier, wherein the image encoder can adopt a model suitable for extracting image characteristics, the recommended model is a ViT (Vision Transformer) model, and any other model such as a CNN model, a depth convolution model EfficientNet, denseNet, resnet and the like can also be adopted. The text encoder can adopt a model suitable for extracting text characteristics in the NLP field, for example, the Bert model is a better neural network model capable of processing text time sequence information so far, and can be suitable for being responsible for text extraction work in the application, and the Electrora model can obtain the effect equal to or similar to the Bert model with lower parameter, so that the text encoder is recommended to be used. The classifier is suitable for a bi-classification task, which may be MLP (feed forward neural network) or FC (fully connected layer).

And acquiring commodity information associated with commodity identifications from a commodity database according to the commodity identifications of the commodities contained in each candidate commodity combination for all candidate commodity combinations containing the target commodity. Inputting commodity information of each commodity contained in the candidate commodity combination into a purchase prediction model, extracting deep semantic information corresponding to commodity pictures in commodity information of each commodity by an image encoder of the purchase prediction model to obtain corresponding picture feature vectors, extracting deep semantic information corresponding to commodity description texts in commodity information of each commodity by a text encoder of the purchase prediction model to obtain corresponding text feature vectors, and splicing the picture feature vectors and the text feature vectors of the same commodity to obtain picture-text fusion vectors corresponding to each commodity. And splicing the image-text fusion vectors corresponding to the commodities to form a combined characteristic representation. And mapping the combination characteristic representation to a preset dichotomy class by a classifier of the purchase prediction model, wherein the dichotomy class comprises a positive class representing that the commodity combination is purchased by a user and a negative class representing that the commodity combination is not purchased by the user, and determining the classification probability mapped to the positive class as the purchase rate.

As can be appreciated 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 aspects:

Referring to fig. 2, in a further embodiment, step S1100, according to a user purchase data set corresponding to each online store, constructs a commodity combination set, a category combination set and a brand combination set corresponding to each online store by adopting a correlation analysis algorithm, including the following steps:

step S1110, acquiring user purchase data sets of each online store, and clustering corresponding dimensions of commodity identifications, commodity categories and commodity brands by taking users as clustering units to determine each data set corresponding to each dimension;

Step S1120, respectively determining a combination set of corresponding dimensions based on the data sets by adopting a correlation analysis algorithm, wherein the combination set comprises a commodity combination set, a category combination set and a brand combination set;

taking the example of determining a commodity combination set, in one embodiment, an association analysis algorithm implemented by using an FP Growth algorithm is adopted, each user identifier is taken as a transaction based on a data set of commodity identifier dimensions, each commodity identifier corresponding to the user identifier is taken as an item, a transaction set containing all items (commodity identifiers) is generated, the transaction set is scanned, and the frequency of each item is counted, wherein the frequency is the frequency of the corresponding item appearing in several transactions. And determining all the items contained in the transaction set, wherein the items are larger than or equal to a preset third frequency threshold value, as frequent items, and constructing the frequent item set according to descending order of frequencies corresponding to the frequent items. And scanning the transaction set again according to the frequent single set, reconstructing each transaction in the transaction set to obtain the reconstructed transaction set, wherein the reconstructed transaction set comprises the steps of eliminating the single items which are not in the frequent single set in each item corresponding to the transaction, and sequentially arranging the items corresponding to the transaction according to the sequence in the frequent single set. And constructing a corresponding FP Tree according to the reconstructed transaction set, wherein the FPTree comprises a root node and a leaf node, the root node is null, and the leaf node marks a corresponding single item and frequency thereof. And determining a condition pattern base corresponding to each item contained in the frequent item set according to the FPTree, constructing a corresponding condition FPTree based on the condition pattern base, determining a corresponding association rule, namely a commodity combination based on the condition FPTree, and summarizing commodity combinations corresponding to each item to form a commodity combination set. The third frequency threshold may be set as desired by one skilled in the art. Similarly, the brand combination set and the category combination set can be correspondingly determined based on the commodity brand dimension and the commodity category dimension data set by adopting the correlation analysis algorithm realized by the FPGrow algorithm.

Step S1130, determining the support degree and the confidence degree of each combination in the commodity combination set, the class combination set and the brand combination set, wherein the support degree represents the frequency of occurrence of the corresponding combination, and the confidence degree represents the relationship tightness degree of the corresponding combination;

those skilled in the art will appreciate that the association analysis algorithm implemented by using the FP Growth algorithm or the Apriori algorithm obtains a commodity combination set, a category combination set, and a brand combination set, and also obtains a frequency number corresponding to each of these combinations, where the frequency number is all items in the corresponding combination, and appears in several transactions in the corresponding transaction set. Further, the frequency number corresponding to each combination in the commodity combination set, the class combination set and the brand combination set is calculated, and the ratio of the frequency number to the total number of the transactions in the corresponding transaction set is used as the support degree.

In addition, the correlation analysis algorithm implemented by the FP Growth algorithm or the Apriori algorithm is used for obtaining a commodity combination set, a category combination set and a brand combination set, and can also be used for obtaining the frequency number corresponding to each combination in the combination sets and the frequency number of partial items contained in each combination, wherein the partial items contain at least one item in the corresponding combination, the frequency number of the partial items is all items in the partial items, and the partial items appear in several transactions in the corresponding transaction set. Further, the frequency number corresponding to each combination in the commodity combination set, the category combination set and the brand combination set is calculated, and the ratio of the frequency number to the frequency number of the partial items in the corresponding combination is used as the confidence.

Step S1140, eliminating combinations of the commodity combination set, the class combination set, and combinations of which the support and/or confidence in the brand combination set do not meet the preset conditions.

The minimum support for measuring the support and the minimum confidence for measuring the confidence may be set, and may be set by those skilled in the art as needed. And removing the combination of which the support degree is lower than the minimum support degree in the commodity combination set, the class combination set and the brand combination set according to the support degree and the confidence degree corresponding to each combination in the commodity combination set, the class combination set and the brand combination set, or removing the combination of which the support degree is lower than the minimum confidence degree in the commodity combination set, the class combination set and the brand combination set, or removing the combination of which the support degree is lower than the minimum support degree and the confidence degree in the commodity combination set, the class combination set and the brand combination set, or removing the combination of which the support degree is lower than the minimum confidence degree, which can be realized by one skilled in the art according to the requirements.

In this embodiment, the support and the confidence corresponding to each combination in the commodity combination set, the class combination set and the brand combination set are determined, so that the combination in which the support and/or the confidence does not meet the preset conditions is removed, and the compactness of the combination in the commodity combination set, the class combination set and the brand combination set, namely, the elements contained in the combination are closely related.

Referring to fig. 3, in a further embodiment, step S1200, recall candidate commodity combinations from the commodity combinations of the online stores according to the commodity identification, commodity class, commodity brand of the target commodity in the online stores, and recall target commodity combinations, target brand combinations from the commodity combination sets, brand combination sets of all online stores, includes the following steps:

step S1210, recalling candidate commodity combinations containing the commodity identifications from commodity combinations of the target online stores according to the commodity identifications of the target commodities in the target online stores;

and acquiring a commodity combination set of the on-line shops, recalling commodity combinations containing target commodity identifications in the on-line shops, and combining commodities corresponding to the commodity identifications contained in the commodity combinations to form candidate commodity combinations, wherein the target commodity is necessarily contained.

Step S1220, recall the candidate class combination containing commodity class of the target commodity from the class combination set of all online stores, calculate the sorting score corresponding to each candidate class combination according to the support and/or confidence matching weight of the candidate class combination corresponding to the corresponding class combination set, screen out the target class combination with the sorting score meeting the preset condition, wherein the weight is the similarity between the online store corresponding to the class combination set of the candidate class combination and the target online store;

It will be appreciated that for the same candidate class combination, the similarity between different online stores and target online stores affects the reference accuracy of the support and/or confidence of that candidate class combination in the class combination set for the corresponding online store.

And calculating a ranking score corresponding to each candidate class combination, taking a single candidate class combination as an example, obtaining a class combination set of online stores containing the candidate class combination, and taking the support degree and/or the confidence degree of the candidate class combination in each class combination set and the similarity between the online store corresponding to each class combination set and the online store of the target as weights. And multiplying the support degree of the corresponding candidate class combination by the corresponding weight for each class combination set, or multiplying the confidence degree by the corresponding weight, or multiplying the support degree and the confidence degree by the corresponding weight and then summing the support degree and the confidence degree, respectively, thereby calculating the score corresponding to the candidate class combination in each class combination set, and realizing the method according to the need of a person skilled in the art. Further, after adding the corresponding scores of the candidate class combinations in each class combination set, dividing the sum by the total number of the class combination sets, and calculating the ranking score of the candidate class combinations.

And sorting N candidate class combinations with the top sorting degree from high to low according to the sorting scores corresponding to each candidate class combination, and screening out the N candidate class combinations with the top sorting degree as target class combinations, wherein N can be set by a person skilled in the art according to requirements.

Step S1230, recall the candidate product class combination containing the commodity brand of the target commodity from the brand combination set of all online stores, calculate the sorting score corresponding to each candidate brand combination according to the support degree and/or confidence matching weight of the candidate brand combination corresponding to the corresponding brand combination set, and screen out the target brand combination with the sorting score meeting the preset condition, wherein the weight is the similarity between the online store corresponding to the brand combination set of the candidate brand combination and the target online store.

It will be appreciated that for the same candidate brand combination, the similarity between different online stores and target online stores affects the reference accuracy of the support and/or confidence of that candidate brand combination in the brand combination set of the corresponding online stores.

And calculating a ranking score corresponding to each candidate brand combination, taking a single candidate brand combination as an example, obtaining a brand combination set of online stores containing the candidate brand combination, and taking the support degree and/or the confidence degree of the candidate brand combination in each brand combination set and the similarity between the online store corresponding to each brand combination set and the online store of the target as weights. And multiplying the support degree of the corresponding candidate brand combination by the corresponding weight for each brand combination set, or multiplying the confidence degree by the corresponding weight, or multiplying the support degree and the confidence degree by the corresponding weight and then summing the support degree and the confidence degree, respectively, so as to calculate the score corresponding to the candidate brand combination in each brand combination set, and one skilled in the art can realize the realization according to the needs. Further, after adding the corresponding scores of the candidate brand combinations in each brand combination set, dividing the total number of the brand combination sets, and calculating the ranking scores of the candidate brand combinations.

And sorting N candidate brand combinations with the top sorting order according to the descending order of the sorting scores corresponding to each candidate brand combination, wherein N can be set by a person skilled in the art as required by taking the N candidate brand combinations with the top sorting order as target brand combinations.

In this embodiment, the target product class combination and the target brand combination, in which the corresponding ranking scores meet the preset conditions, are respectively screened out from the product class, the candidate product class combination and the candidate brand combination of the product brand corresponding to the product class combination set and the brand combination set of the online store. The sorting scores according to which the target brand combinations are screened can accurately represent the compactness of the candidate brand combinations and the candidate brand combinations, so that the reliability of the target brand combinations can be ensured.

Referring to fig. 4, in a further embodiment, step S1210, before recalling a candidate commodity combination including a commodity identifier from a commodity combination of a target commodity in a target online store according to the commodity identifier of the target commodity, includes the following steps:

Step S1201, obtaining all commodity categories corresponding to each online store, and constructing corresponding category distribution;

and according to the commodity identifications of all commodities corresponding to each online store, acquiring commodity categories in commodity description texts associated with each commodity identification from a commodity database, and filtering repeated commodity categories, thereby acquiring all commodity categories corresponding to each online store, and further summarizing and forming the category distribution.

Step S1202, calculating the similarity between the online stores according to the category distribution corresponding to each online store.

In One embodiment, one-Hot Encoding is used to encode the category distribution of each online store to obtain a corresponding category encoded representation that is a vectorized representation of the category distribution. And determining the vector distance between class code representations corresponding to each two online stores by adopting a vector distance algorithm, and taking the vector distance as the similarity between the online stores. The vector distance algorithm can be any one of cosine similarity algorithm, euclidean distance algorithm, pearson correlation coefficient algorithm, jacquard coefficient algorithm and the like.

In the recommended embodiment, the similarity between online stores is determined, taking two online stores as an example, a preset text similarity model can be adopted, class distribution corresponding to the two online stores is respectively used as input of two branches in the model, corresponding vectorization representations are respectively extracted, and further, a classifier in the text similarity model is adopted to map the two vectorization representations to a preset binary class, the binary class comprises a positive class representing similarity between the two online stores and a negative class representing dissimilarity between the two online stores, and the classification probability mapped to the positive class is determined as the similarity between the two online stores.

The text similarity model is trained in advance until convergence, and the similarity capability of the two online stores is determined according to the input category distribution corresponding to the two stores. The text similarity model can be a double-tower model, structurally comprises two branches and a classifier, the two branches are identical text feature extraction models, the text feature extraction models adopt models which are suitable for extracting text features in the NLP field, for example, the Bert model is a superior neural network model which can process text time sequence information so far, the model can be suitable for being responsible for text feature extraction work, and the electric model can obtain the effect which is equal to or similar to that of the Bert model with lower parameter quantity, so that the model is recommended to be used. The classifier is suitable for a bi-classification task, which may be MLP (feed forward neural network) or FC (fully connected layer).

The training process of the text similarity model comprises the steps of obtaining a single training sample in a prepared training set and a supervision label thereof, wherein the training sample comprises class distribution of two online stores, and the supervision label is used for correspondingly labeling the training sample according to whether the two online stores of the training sample are similar or not in an artificial judgment mode. Those skilled in the art can flexibly adapt labels based on the classifier disclosed and modeled herein. The labeling exemplary example is that two online stores of the training sample are similar, and a supervision label of the training sample is labeled as [1,0]; the two online stores of the training sample are dissimilar, and the supervision labels of the training sample are labeled [0,1]. And inputting the single training sample into a text similarity model, respectively extracting vectorization representations of class distribution corresponding to two stores of the training sample by two branches in the text similarity model, further mapping the two vectorization representations to preset binary classes by adopting a classifier of the text similarity model, and predicting classification probabilities corresponding to positive and negative classes to serve as prediction results. Invoking a preset cross entropy loss function, which can be flexibly set by a person skilled in the art according to priori knowledge or experimental experience, calculating a cross entropy loss value of the prediction result based on a supervision label according to the training sample, and when the loss value reaches a preset threshold value, indicating that the model is trained to a convergence state, so that model training can be terminated; when the loss value does not reach the preset threshold value, the model is indicated to be not converged, gradient update is carried out on the model according to the loss value, the model is further approximated to convergence by generally back-propagating weight parameters of all links of the correction model, and then iterative training is carried out on the model by continuously calling other training samples in the training set until the model is trained to a convergence state. The construction of the training set may be prepared in advance by one skilled in the art based on the disclosure herein.

Referring to fig. 5, in a further embodiment, step S1400 of predicting all candidate commodity combinations including the target commodity by using a preset purchase prediction model, determines a purchase rate corresponding to each candidate commodity combination, and includes the following steps:

step S1410, acquiring commodity information of each commodity contained in each candidate commodity combination in all candidate commodity combinations containing target commodities, wherein the commodity information comprises commodity description texts and commodity pictures;

and acquiring commodity information associated with commodity identifications from a commodity database according to the commodity identifications of the commodities contained in each candidate commodity combination for all candidate commodity combinations containing the target commodity.

Step S1420, extracting image-text fusion information corresponding to commodity information of each commodity contained in the candidate commodity combination by adopting a preset purchase prediction model, and splicing to form a combination characteristic representation;

Inputting commodity information of each commodity contained in the candidate commodity combination into a purchase prediction model, extracting deep semantic information corresponding to commodity pictures in commodity information of each commodity by an image encoder of the purchase prediction model to obtain corresponding picture feature vectors, extracting deep semantic information corresponding to commodity description texts in commodity information of each commodity by a text encoder of the purchase prediction model to obtain corresponding text feature vectors, and splicing the picture feature vectors and the text feature vectors of the same commodity to obtain picture-text fusion vectors corresponding to each commodity. And splicing the image-text fusion vectors corresponding to the commodities to form a combined characteristic representation.

Step S1430, determining the purchase rate of the candidate commodity combination based on the combination characteristic representation.

Further, the combined feature representation is mapped to a preset dichotomy class by a classifier of the purchase prediction model, the dichotomy class comprises a positive class representing that the commodity combination is purchased by the user and a negative class representing that the commodity combination is not purchased by the user, and the classification probability mapped to the positive class is determined as the purchase rate.

In this embodiment, the purchase rate of the candidate commodity combination is determined based on commodity information of each commodity included in the candidate commodity combination by using a purchase rate prediction model. Because of factors affecting the purchase rate of the candidate commodity combination, the commodity description text and the commodity picture of each commodity contained in the candidate commodity combination are related, and the purchase prediction model is correspondingly modeled according to the commodity description text and the commodity picture, so that the purchase rate of the candidate commodity combination is accurately determined by the model.

Referring to fig. 6, in a further embodiment, step S1300 of recalling a first commodity corresponding to a commodity class of the commodity class combination of the target commodity according to the target commodity class combination, recalling a second commodity corresponding to a commodity brand of the commodity class combination of the target commodity, and forming a candidate commodity combination with the target commodity includes the steps of:

step 1310, recalling a plurality of first candidate commodities corresponding to commodity classes of the commodity class combination of the target commodity according to the target commodity class combination, determining sales results corresponding to each first candidate commodity, screening out first commodities with sales results meeting preset conditions, and forming a candidate commodity combination with the target commodity, wherein the sales results are determined according to any one or more of click rate, conversion rate and sales volume of the first candidate commodity;

And recalling the commodity with the commodity class in the commodity description text from a commodity database as a first candidate commodity according to the commodity class combined with the commodity class of the target commodity in the target commodity combination.

For each first candidate commodity, taking a single first candidate commodity as an example, the ratio of the number of times of exposure to the first candidate commodity can be calculated as the click rate by collecting the number of times of the first candidate commodity clicked by the user after the exposure and the corresponding number of times of exposure in the last period of time, such as in the last month, the last three months, the last half year and the last year. The period of time may be set as desired by one skilled in the art.

The ratio of the number of times of the first candidate commodity to the number of times of the latter can be calculated as the conversion rate by collecting the number of times of any one of purchasing, adding shopping carts, praying, forwarding and collecting actions of the user on the first candidate commodity after the first candidate commodity is clicked by the user in the last period of time, such as the last month, the last three months, the last half year and the last year, and the corresponding number of times of clicking. The period of time may be set as desired by one skilled in the art.

The sales of the first candidate good may be collected over a recent period of time, such as over a period of about one month, over a period of about three months, over a period of about half a year, over a period of about one year. The period of time may be set as desired by one skilled in the art.

In one embodiment, any one of click rate, conversion rate and sales volume of the first candidate commodities is determined as sales results of the first candidate commodities, of which sales results are larger than a preset threshold, are screened out to serve as first commodities, and then each first commodity is combined with the target commodity to form a candidate commodity combination. The preset threshold may be set as desired by one skilled in the art.

In another embodiment, determining any multiple of click rate, conversion rate and sales volume of the first candidate commodity, and presetting corresponding weights according to different preset reference values corresponding to sales effects of each item, wherein the sum of the weights is 1, and determining the click rate, the conversion rate and the sales volume as 0.3, 0.5 and 0.2 according to an exemplary example; click rate and conversion rate were determined, and weights were 0.4 and 0.6, respectively. And multiplying each determined item by a corresponding weight, and then adding the multiplied items, so that the calculated score is determined as the sales effect of the first candidate commodity. Further, sorting is carried out according to the order of sales results of the first candidate commodities from high to low, N first candidate commodities with the top sorting are screened out to serve as first commodities, and then each first commodity is further combined with the target commodity to form a candidate commodity combination. The weights may be set as desired by one of ordinary skill in the art based on the disclosure herein. The N can be set as desired by a person skilled in the art.

Step S1320, according to the target brand combination recall, a plurality of second candidate commodities corresponding to commodity brands of the commodity brand combination of the target commodity, determining sales results corresponding to each second candidate commodity, screening out second commodities with sales results meeting preset conditions, and forming a candidate commodity combination with the target commodity, wherein the sales results are determined according to any one or more of click rate, conversion rate and sales volume of the second candidate commodity.

And recalling the commodities with the commodity brands in the commodity description text from a commodity database according to the commodity brands in the target brand combination and the commodity brand combination of the target commodity, wherein the commodities with the commodity brands are used as second candidate commodities.

For each second candidate commodity, taking a single second candidate commodity as an example, the ratio of the number of times of exposure to the second candidate commodity can be calculated as the click rate by collecting the number of times of the second candidate commodity clicked by the user after the exposure and the corresponding number of times of exposure in the last period of time, for example, in the last month, the last three months, the last half year and the last year. The period of time may be set as desired by one skilled in the art.

The ratio of the number of times of the former to the latter can be calculated as the conversion rate by collecting the number of times of any one of purchasing, adding shopping carts, praying, forwarding and collecting actions of the user on the second candidate commodity after the second candidate commodity is clicked by the user in the last period of time, such as in the last month, the last three months, the last half year and the last year, and the corresponding number of times of clicking. The period of time may be set as desired by one skilled in the art.

The sales of the second candidate good may be collected over a recent period of time, such as over a period of about one month, over a period of about three months, over a period of about half a year, over a period of about one year. The period of time may be set as desired by one skilled in the art.

In one embodiment, any one of the click rate, the conversion rate and the sales volume of the second candidate commodities is determined as the sales results of the second candidate commodities, the second candidate commodities with the sales results greater than a preset threshold value in all the second candidate commodities are screened out to serve as the second commodities, and then each second commodity is combined with the target commodity to form a candidate commodity combination. The preset threshold may be set as desired by one skilled in the art.

In another embodiment, determining any multiple of click rate, conversion rate and sales volume of the second candidate commodity, and presetting corresponding weights according to different preset reference values corresponding to sales effects of each item, wherein the sum of the weights is 1, and determining the click rate, the conversion rate and the sales volume as 0.3, 0.5 and 0.2 according to an exemplary example; click rate and conversion rate were determined, and weights were 0.4 and 0.6, respectively. And multiplying each determined item by a corresponding weight, and then adding the multiplied items, so that the calculated score is determined as the sales effect of the second candidate commodity. Further, sorting is carried out according to the order of sales success of each second candidate commodity from high to low, N second candidate commodities with the top sorting are screened out to be used as second commodities, and each second commodity is further combined with the target commodity to form a candidate commodity combination. The weights may be set as desired by one of ordinary skill in the art based on the disclosure herein. The N can be set as desired by a person skilled in the art.

In this embodiment, a plurality of first candidate commodities corresponding to commodity classes of the commodity class combination of the target commodity and a plurality of second candidate commodities corresponding to commodity brands of the commodity class combination of the target commodity are recalled respectively according to the target class combination and the target brand combination, so that sales results corresponding to each first candidate commodity and each second candidate commodity are determined, and therefore the first commodity and the second commodity, the sales results of which meet preset conditions, are screened out to form the candidate commodity combination with the target commodity. Because sales success rate can accurately represent the effect obtained by selling candidate commodities, on one hand, the first commodity and the second commodity obtained by screening according to the sales success rate can be guaranteed to be hopefully attracted to users, and the users are stimulated to purchase. On the other hand, it is useful to promote sales of a combination of products including the first product and a combination of products including the second product.

Referring to fig. 7, in a further embodiment, step S1400 of predicting all candidate commodity combinations including the target commodity by using a preset purchase prediction model, before determining the purchase rate corresponding to each candidate commodity combination, includes the following steps:

Step 2400, acquiring a single training sample and a supervision tag thereof from a prepared training set, wherein the training sample is commodity information of all commodities contained in a commodity combination, and the supervision tag characterizes whether the commodity combination of the training sample is purchased by a user or not;

multiple commodity combinations can be collected from the commodity database, pushed to the user, and the commodity combinations purchased by the user and the commodity combinations not purchased by the user are determined. Further, for each commodity combination, commodity information associated with commodity identifications is obtained from a commodity database according to commodity identifications of commodities contained in the commodity combination, so that commodity information of each commodity contained in each commodity combination is obtained and is respectively used as a single training sample, and for each training sample, a corresponding supervision label is marked according to whether the commodity combination of the training sample is purchased by a user or not, so that the supervision label corresponding to each training sample is obtained. Those skilled in the art can flexibly adapt labels based on the classifier disclosed and modeled herein. The labeling of the exemplary example is that the commodity combination of the training sample is purchased by a user, and the supervision label of the training sample is [1,0]; the commodity combination of the training sample is not purchased by the user, and the supervision label of the training sample is labeled as [0,1].

And summarizing all training samples and supervision labels thereof to form a training set.

Step S2410, extracting image-text fusion information rows corresponding to all commodities in a training sample by adopting a preset purchase prediction model, and splicing to form a combined characteristic representation;

the purchase prediction model comprises an image encoder, a text encoder and a classifier, wherein the image encoder can adopt a model suitable for extracting image characteristics, the recommended model is a ViT (Vision Transformer) model, and any other model such as a CNN model, a depth convolution model EfficientNet, denseNet, resnet and the like can also be adopted. The text encoder can adopt a model suitable for extracting text characteristics in the NLP field, for example, the Bert model is a better neural network model capable of processing text time sequence information so far, and can be suitable for being responsible for text extraction work in the application, and the Electrora model can obtain the effect equal to or similar to the Bert model with lower parameter, so that the text encoder is recommended to be used. The classifier is suitable for a bi-classification task, which may be MLP (feed forward neural network) or FC (fully connected layer).

Inputting the training sample into a purchase prediction model, extracting deep semantic information corresponding to commodity pictures in commodity information of each commodity in the training sample by an image encoder of the purchase prediction model to obtain corresponding picture feature vectors, extracting deep semantic information corresponding to commodity description texts in commodity information of each commodity in the training sample by a text encoder of the purchase prediction model to obtain corresponding text feature vectors, and splicing the picture feature vectors and the text feature vectors of the same commodity to obtain picture-text fusion vectors corresponding to each commodity in the training sample. And splicing the image-text fusion vectors corresponding to the commodities to form a combined characteristic representation.

Step S2420, determining a prediction result of the candidate commodity combination based on the combination characteristic representation;

further, mapping the combination feature representation to preset dichotomy categories by a classifier of the purchase prediction model, wherein the dichotomy categories comprise positive categories representing commodity combinations purchased by users and negative categories representing commodity combinations not purchased by users, and predicting classification probabilities corresponding to the positive categories and the negative categories as prediction results.

And step S2430, determining a loss value of the predicted purchase rate by adopting a supervision tag of the training sample, updating the weight of the purchase prediction model when the loss value does not reach a preset threshold, and continuously calling other training samples to perform iterative training until the purchase prediction model converges.

Invoking a preset cross entropy loss function, which can be flexibly set by a person skilled in the art according to priori knowledge or experimental experience, calculating a cross entropy loss value of the predicted purchase rate based on a supervision label according to the training sample, and when the loss value reaches a preset threshold value, indicating that the purchase prediction model is trained to a convergence state, so that the purchase prediction model training can be terminated; and when the loss value does not reach the preset threshold value, indicating that the purchase prediction model is not converged, then carrying out gradient update on the purchase prediction model according to the loss value, correcting the weight parameters of each link of the purchase prediction model through back propagation to enable the purchase prediction model to further approach convergence, and then continuously calling other training samples in the training set to carry out iterative training on the purchase prediction model until the purchase prediction model is trained to a convergence state.

In this embodiment, a training process of purchasing a prediction model is disclosed, so that after the model is trained to be converged, the purchasing rate capability of the commodity combination is determined based on the obtained commodity information of each commodity contained in the commodity combination. The purchase rate of the combination of goods is quickly and accurately determined by employing a purchase prediction model trained to converge.

Referring to fig. 8, a commodity combination recommending apparatus provided in accordance with one of the purposes of the present application is a functional implementation of the commodity combination recommending method of the present application, and in another aspect, the commodity combination recommending apparatus provided in accordance with one of the purposes of the present application includes a combination set construction module 1100, a combination recall module 1200, a candidate recall module 1300, and a purchase rate prediction module 1400, where the combination set construction module 1100 is configured to construct, by using a correlation analysis algorithm, a commodity combination set, a class combination set, and a brand combination set corresponding to each online store according to a user purchase data set corresponding to each online store, where the user purchase data set includes user purchase data generated by purchasing a commodity of the corresponding online store by a user, and includes a user identifier, a commodity class, and a commodity brand; the combination recall module 1200 is configured to recall candidate commodity combinations from the commodity combination set of the online stores according to the commodity identification, commodity class, commodity brand of the target commodity in the online stores, and recall target commodity combinations and target brand combinations from the commodity combination set, the brand combination set of all online stores; a candidate recall module 1300, configured to recall, according to the target product combination, a first product corresponding to a product class of the target product combination, a candidate product combination with a target product, and a second product corresponding to a product brand of the target product combination, and a candidate product combination with a target product; the purchase rate prediction module 1400 is configured to predict all candidate commodity combinations including the target commodity by using a preset purchase prediction model, determine a purchase rate corresponding to each candidate commodity combination, and screen out recommended commodity combinations whose purchase rate meets a preset condition.

In a further embodiment, the combined set construction module 1100 includes: the multidimensional clustering sub-module is used for acquiring user purchase data sets of each online store, clustering corresponding dimensions of commodity identifications, commodity categories and commodity brands by taking users as clustering units, and determining each data set corresponding to each dimension; the association analysis sub-module is used for respectively determining combination sets of corresponding dimensions based on the data sets by adopting an association analysis algorithm, wherein the combination sets comprise a commodity combination set, a category combination set and a brand combination set; the numerical value determining submodule is used for determining the support degree and the confidence degree of each combination in the commodity combination set, the class combination set and the brand combination set, wherein the support degree represents the frequency of occurrence of the corresponding combination, and the confidence degree represents the relationship tightness degree of the corresponding combination; and the combination eliminating sub-module is used for eliminating combinations of the commodity combination set, the class combination set and the brand combination set, wherein the support degree and/or the confidence degree of the combination set do not accord with preset conditions.

In a further embodiment, the combined recall module 1200 includes: the commodity combination recall sub-module is used for intensively recalling candidate commodity combinations containing the commodity identifications from the commodity combinations of the target online shops according to the commodity identifications of the target commodities in the target online shops; the first sorting recall sub-module is used for recalling candidate class combinations containing commodity classes of the target commodity from class combinations of all online shops, calculating sorting scores corresponding to each candidate class combination according to support and/or confidence matching weights of the candidate class combinations corresponding to the corresponding class combination sets, screening out target class combinations with sorting scores meeting preset conditions, wherein the weights are the similarity between the online shops corresponding to the class combination sets where the candidate class combinations are located and the target online shops; and the second sorting recall sub-module is used for recalling candidate brand combinations containing commodity brands of the target commodity from the brand combination sets of all online shops, calculating sorting scores corresponding to each candidate brand combination according to the support degree and/or confidence matching weight of the candidate brand combination corresponding to the corresponding brand combination set, screening out target brand combinations with sorting scores meeting preset conditions, and the weight is the similarity between the online shops corresponding to the brand combination set where the candidate brand combination is located and the target online shops.

In a further embodiment, the purchase rate predicting module 1400 includes: the information acquisition sub-module is used for acquiring commodity information of each commodity contained in each candidate commodity combination in all candidate commodity combinations containing target commodities, wherein the commodity information comprises commodity description texts and commodity pictures; the first feature extraction submodule is used for extracting image-text fusion information corresponding to commodity information of each commodity contained in the candidate commodity combination by adopting a preset purchase prediction model to splice, so as to form a combination feature representation; and the purchase rate determination submodule is used for determining the purchase rate of the candidate commodity combination based on the combination characteristic representation.

In a further embodiment, the candidate recall module 1300 includes: the first candidate recall sub-module is used for recalling a plurality of first candidate commodities corresponding to commodity classes of commodity class combination of the target commodity according to the target commodity class combination, determining sales results corresponding to each first candidate commodity, screening out first commodities with sales results meeting preset conditions, and forming a candidate commodity combination with the target commodity, wherein the sales results are determined according to any one or more of click rate, conversion rate and sales volume of the first candidate commodity; and the second candidate recall sub-module is used for recalling a plurality of second candidate commodities corresponding to commodity brands of the target commodity according to the target brand combination, determining sales results corresponding to each second candidate commodity, screening out the second commodities with sales results meeting preset conditions, and forming a candidate commodity combination with the target commodity, wherein the sales results are determined according to any one or more of the click rate, the conversion rate and the sales quantity of the second candidate commodities.

In a further embodiment, before the purchase rate predicting module 1400, the method further includes: the training preparation sub-module is used for acquiring a single training sample and a supervision label thereof from a prepared training set, wherein the training sample is commodity information of all commodities contained in the commodity combination, and the supervision label characterizes whether the commodity combination of the training sample is purchased by a user or not; the second feature extraction submodule is used for extracting image-text fusion information corresponding to each commodity in the training sample by adopting a preset purchase prediction model to splice, so as to form a combined feature representation; the model prediction sub-module is used for determining a prediction result of the candidate commodity combination based on the combination characteristic representation; and the iterative training sub-module is used for determining a loss value of the predicted purchase rate by adopting a supervision label of the training sample, updating the weight of the purchase prediction model when the loss value does not reach a preset threshold value, and continuously calling other training samples to perform iterative training until the purchase prediction model converges.

In order to solve the technical problems, the embodiment of the application also provides computer equipment. As shown in fig. 9, the internal structure of the computer device is schematically shown. 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 a control information sequence, and when the computer readable instructions are executed by a processor, the processor can realize a commodity combination recommendation method. The processor of the computer device is used to provide computing and control capabilities, supporting the operation of the entire computer device. The memory of the computer device may have stored therein computer readable instructions that, when executed by the processor, cause the processor to perform the merchandise combination recommendation method of the application. The network interface of the computer device is for communicating with a terminal connection. It will be appreciated by persons skilled in the art that the architecture shown in fig. 9 is merely a block diagram of some of the architecture relevant to the present inventive arrangements and is not limiting as to the computer device to which the present inventive arrangements are applicable, and that a particular computer device may include more or fewer components than shown, or may combine some of the components, or have a different arrangement of components.

The processor in this embodiment is configured to execute specific functions of each module and its sub-module in fig. 8, and the memory stores program codes and various data required for executing the above modules or sub-modules. The network interface is used for data transmission between the user terminal or the server. The memory in the present embodiment stores program codes and data required for executing all modules/sub-modules in the commodity combination recommendation apparatus according to the present application, and the server can call the program codes and data of the server to execute the functions of all sub-modules.

The present application also provides a storage medium storing computer readable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the merchandise combination recommendation method of any one of the embodiments of the application.

Those skilled in the art will appreciate that all or part of the processes implementing the methods of the above embodiments of the present application may be implemented by a computer program for instructing relevant hardware, where the computer program may be stored on a computer readable storage medium, where the program, when executed, may include processes implementing the embodiments of the methods described above. The storage medium may be a computer readable storage medium such as a magnetic disk, an optical disk, a Read-Only Memory (ROM), or a random access Memory (Random Access Memory, RAM).

In summary, the application can solve the problem of cold start of commodity combination recommendation, and recommend rich commodity combinations, and the recommended commodity combinations are expected to achieve the expected recommendation effect.

Those of skill in the art will appreciate that the various operations, methods, steps in the flow, acts, schemes, and alternatives discussed in the present application may be alternated, altered, combined, or eliminated. Further, other steps, means, or steps in a process having various operations, methods, or procedures discussed herein may be alternated, altered, rearranged, disassembled, combined, or eliminated. Further, steps, measures, schemes in the prior art with various operations, methods, flows disclosed in the present application may also be alternated, altered, rearranged, decomposed, combined, or deleted.

The foregoing is only a partial embodiment of the present application, and it should be noted that it will be apparent to those skilled in the art that modifications and adaptations can be made without departing from the principles of the present application, and such modifications and adaptations are intended to be comprehended within the scope of the present application.

Claims

1. The commodity combination recommending method is characterized by comprising the following steps of:

2. The commodity combination recommendation method according to claim 1, wherein the commodity combination set, the class combination set, and the brand combination set corresponding to each online store are constructed by using a correlation analysis algorithm according to the user purchase data set corresponding to each online store, comprising the steps of:

3. The commodity combination recommendation method according to claim 1, wherein recall candidate commodity combinations from commodity combinations of the on-line stores and recall target commodity combinations, target brand combinations from a group of commodity combinations set, a group of brand combinations set of all on-line stores according to commodity identifications, commodity categories, commodity brands of target commodities in the on-line stores, comprising the steps of:

4. The commodity combination recommendation method according to claim 3, wherein before recalling a candidate commodity combination including a commodity identification from a commodity combination of a target commodity in a target online store according to the commodity identification of the target commodity, comprising the steps of:

5. The commodity combination recommendation method according to claim 4, wherein predicting all candidate commodity combinations including the target commodity by using a preset purchase prediction model, and determining a purchase rate corresponding to each candidate commodity combination, comprises the steps of:

6. The commodity combination recommendation method according to claim 1, wherein a first commodity corresponding to a commodity class of the commodity class combination of the target commodity is recalled from the target commodity class combination, a candidate commodity combination is constituted with the target commodity, and a second commodity corresponding to a commodity brand of the commodity combination of the target commodity is recalled, and a candidate commodity combination is constituted with the target commodity, comprising the steps of:

7. The commodity combination recommendation method according to claim 1, wherein the method comprises the steps of, before predicting all candidate commodity combinations including the target commodity by using a preset purchase prediction model and determining a purchase rate corresponding to each candidate commodity combination:

8. A commodity combination recommendation device, comprising:

the combination set construction module is used for constructing a commodity combination set, a class combination set and a brand combination set corresponding to each online store by adopting a correlation analysis algorithm according to a user purchase data set corresponding to each online store, wherein the user purchase data set comprises user purchase data generated when a user purchases a commodity of the corresponding online store, and comprises a user identifier, a commodity class and a commodity brand;

The combination recall module is used for recalling candidate commodity combinations from the commodity combination sets of the online stores according to the commodity identification, commodity class and commodity brand of the target commodity in the online stores, and recalling target commodity combinations and target brand combinations from the commodity combination sets, the brand combination sets of all the online stores;

a candidate recall module, configured to recall, according to the target item combination, a first item corresponding to a item category of the target item combination, a candidate item combination with a target item, and a second item corresponding to a item brand of the target item combination, and a candidate item combination with a target item;

and the purchase rate prediction module is used for predicting all candidate commodity combinations containing the target commodity by adopting a preset purchase prediction model, determining the purchase rate corresponding to each candidate commodity combination, and screening out recommended commodity combinations with the purchase rate meeting preset conditions.

9. A computer device comprising a central processor and a memory, characterized in that the central processor is arranged to invoke a computer program stored in the memory for performing the steps of the method according to any 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.