CN115564531A

CN115564531A - Commodity combination recommendation method and device, equipment, medium and product thereof

Info

Publication number: CN115564531A
Application number: CN202211281377.XA
Authority: CN
Inventors: 李露
Original assignee: Guangzhou Huaduo Network Technology Co Ltd
Current assignee: Guangzhou Huaduo Network Technology Co Ltd
Priority date: 2022-10-19
Filing date: 2022-10-19
Publication date: 2023-01-03

Abstract

The application relates to a commodity combination recommendation method and a device, equipment, medium and product thereof, wherein the method comprises the following steps: acquiring a tag set of a basic commodity, wherein the tag set comprises a plurality of attribute items, and each attribute item comprises an attribute name and an attribute tag; recalling all label combination templates carrying at least one attribute item in the label set from a template library, wherein the label combination templates comprise a plurality of attribute items; recalling all the commodities matched with each label combination template from the mapping relation library, and combining all the commodities corresponding to each label combination template with the basic commodities respectively to construct a plurality of commodity combinations; at least one best combination of items is determined from the plurality of combinations of items such that when one item in the best combination of items is accessed, other items in the best combination of items are pushed. The method and the device can determine the optimal commodity combination for the basic commodities with the help of the label combination template.

Description

Commodity combination recommendation method and device, equipment, medium and product thereof

Technical Field

The application relates to an e-commerce information technology, in particular to a commodity combination recommendation method and device, equipment, medium and product thereof.

Background

In the E-commerce platform, the commodity combination is a means for matching and selling commodities, and has important significance for promoting the sales volume of merchants. The current technologies for constructing commodity combinations are mainly as follows:

(1) A random matching method: for example, the commodities with better sales are selected to be randomly collocated under the condition of meeting the inventory. This approach is not suitable for independent station merchants with cold starts or low sales volumes, and the randomly generated combinations of goods do not necessarily meet the needs of the user.

(2) The method based on the predefined rule comprises the following steps: some combination rules are preset based on commodity characteristics, for example, combination and collocation are performed according to predefined association rules (such as supplier information and product attribute information), and commodity combinations are produced according to rules defined by package activities.

(3) The data modeling-based method comprises the following steps: for example, a commodity association network is constructed according to commodity combinations, sales promotion commodities are selected and combined by an information gain method, the commodity combination problem is used as a multi-objective optimization problem to be modeled, commodity combinations are carried out based on model prediction results, a probability parameter prediction model is constructed, and combined commodities are selected according to joint probabilities.

(4) The mining method based on the order data comprises the following steps: for example, a weighted frequent sequence is constructed by using order data, frequent item set mining is performed, similar commodities are replaced according to commodity combinations in the order data, new combinations are produced, commodity class combinations are mined from a multi-commodity order, and commodity combinations are constructed according to the commodity class combinations.

Therefore, the commodity combination technology adopted at present is not suitable for cold start scenes, or has poor universality or can not meet the requirements of users, and has inherent defects. In the independent station-based e-commerce platform, each independent station is independent on data and independent from each other, and the problems of cold start and data sparsity are particularly prominent, so that other feasible combination modes need to be explored.

Disclosure of Invention

An object of the present application is to solve the above problems and provide a method for recommending a combination of products, a corresponding apparatus, a device, a non-volatile readable storage medium, and a computer program product.

According to an aspect of the present application, there is provided a commodity combination recommendation method including the steps of:

acquiring a tag set of a basic commodity, wherein the tag set comprises a plurality of attribute items, and each attribute item comprises an attribute name and an attribute tag;

recalling all label combination templates carrying at least one attribute item in the label set from a template library, wherein the label combination templates comprise a plurality of attribute items;

recalling all the commodities matched with each label combination template from the mapping relation library, and combining all the commodities corresponding to each label combination template with the basic commodities respectively to construct a plurality of commodity combinations;

at least one best combination of items is determined from the plurality of combinations of items such that when one item in the best combination of items is accessed, other items in the best combination of items are pushed.

Optionally, before obtaining the tag set of the base product, the method includes:

generating an attribute item of a commodity according to commodity information of the commodity in a commodity library of an online shop by adopting a preset commodity labeling model;

combining attribute items marked by a plurality of different commodities in each commodity subset based on the commodity subsets with access behavior incidence relations in the co-occurrence commodity set to construct a plurality of label combinations corresponding to the commodity subsets;

and counting the co-occurrence frequency of each label combination in all label combinations generated by the co-occurrence commodity set, and configuring the partial label combinations with relatively higher co-occurrence frequency as the label combination templates in the template library.

Optionally, before the subset of commodities having an access behavior association relationship in the co-occurrence commodity set, the method includes:

extracting commodity information of commodities purchased in the same order from a preset commodity access behavior data set to form a first commodity subset in a co-occurrence commodity set;

commodity information of commodities purchased in the same conversation period is extracted from a preset commodity access behavior data set to form a second commodity subset in the co-occurrence commodity set;

and extracting commodity information of commodities purchased by the same user in the same period from a preset commodity access behavior data set to form a third commodity subset in the co-occurrence commodity set.

Optionally, recalling all the commodities matched with each tag combination template from the mapping relation library, and combining all the commodities corresponding to each tag combination template with the base commodity to construct a plurality of commodity combinations, including:

for each label combination template, inquiring the commodities carrying all attribute items in the label combination template from the mapping relation library to obtain a corresponding commodity list;

and adding the basic commodities into the commodity list corresponding to each label combination template, so that each commodity list forms a commodity combination.

Optionally, determining at least one best combination of commodities from the plurality of combinations of commodities includes:

counting the total template hit number of the label combination templates of the basic commodities hit by each commodity combination;

determining part of commodity combinations with relatively high total hit numbers of the templates as candidate commodity combinations;

and acquiring external evaluation indexes of the candidate commodity combinations, and determining the optimal commodity combination according to the external evaluation indexes.

Optionally, the obtaining of external evaluation indexes of each candidate commodity combination and the determining of the best commodity combination according to the external evaluation indexes includes:

responding to the accessed event of the commodity detail page of the basic commodity, and randomly acquiring a candidate commodity combination of the basic commodity as a target commodity combination;

acquiring commodity information of other commodities except the basic commodity in the target commodity combination, and loading the commodity information into the commodity detail page;

monitoring visited events of other commodities in the target commodity combination, counting the conversion rate of the whole target commodity combination according to the visited events, and taking the conversion rate as an external evaluation index;

and determining part of candidate commodity combinations with relatively high external evaluation indexes as optimal commodity combinations.

Optionally, after determining at least one best combination of commodities from the plurality of combinations of commodities, the method includes:

responding to the event that the commodity detail page of the basic commodity is accessed, and determining one optimal commodity combination corresponding to the basic commodity;

and acquiring commodity information of other commodities except the basic commodity in the optimal commodity combination, and loading the commodity information into the commodity detail page for display.

According to another aspect of the present application, there is provided a combination merchandise recommendation device including:

the system comprises a tag set acquisition module, a tag set acquisition module and a tag matching module, wherein the tag set acquisition module is used for acquiring a tag set of a basic commodity, the tag set comprises a plurality of attribute items, and each attribute item comprises an attribute name and an attribute tag;

a template recalling module, configured to recall all tag combination templates carrying at least one attribute item in the tag set from a template library, where the tag combination template includes multiple attribute items;

the combined recall module is set to recall all the commodities matched with each label combined template from the mapping relation library, and all the commodities corresponding to each label combined template are combined with the basic commodities to construct a plurality of commodity combinations respectively;

and the combination determining module is used for determining at least one optimal commodity combination from the plurality of commodity combinations so as to push other commodities in the optimal commodity combination when one commodity in the optimal commodity combination is accessed.

According to another aspect of the present application, there is provided a product combination recommendation device, including a central processing unit and a memory, where the central processing unit is configured to call and run a computer program stored in the memory to perform the steps of the product combination recommendation method described in the present application.

According to another aspect of the present application, a non-transitory readable storage medium is provided, which stores a computer program implemented according to the product portfolio recommendation method in the form of computer readable instructions, and when the computer program is called by a computer, the steps included in the method are executed.

According to another aspect of the present application, there is provided a computer program product comprising computer programs/instructions which, when executed by a processor, implement the steps of the method described in any one of the embodiments of the present application.

Compared with the prior art, the method and the device have the advantages that the pre-constructed label combination templates are recalled according to various attribute items in the label set of the basic commodity, the label combination templates are provided in advance, then a large number of commodity combinations are further recalled according to the label combination templates, the optimal commodity combination is determined preferentially on the basis of the commodity combinations, personalized customization or customization according to statistical data can be achieved with the help of the label combination templates, accordingly, the method and the device can recall other commodities matched with the basic commodity by utilizing the corresponding relation between the label combination templates and the labels carried by the commodities without depending on behavior data of the commodities, the determined optimal commodity combination can obtain a conversion effect which is more consistent with expectation, is not influenced by the sparse congenital deficiency of data under the cold starting condition of the independent station, and is suitable for the application scene of the cross-border e-commerce platform based on the independent station.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.

Fig. 1 is a schematic network architecture diagram of an application environment according to the present invention;

FIG. 2 is a flowchart illustrating an embodiment of a combined merchandise recommendation method according to the present application;

FIG. 3 is a schematic flow chart of constructing a label assembly template according to an embodiment of the present application;

FIG. 4 is a schematic flow chart of constructing a co-occurrence commodity set according to an embodiment of the present application;

FIG. 5 is a schematic flow chart illustrating the process of determining the best combination of commodities according to an embodiment of the present application;

FIG. 6 is a schematic flow chart illustrating the determination of the best combination of commodities by actual measurement in the embodiment of the present application;

fig. 7 is a schematic block diagram of a combined merchandise recommendation device according to the present application;

fig. 8 is a schematic structural diagram of a combined product recommendation device used in the present application.

Detailed Description

The models cited or possibly cited in the application comprise a traditional machine learning model or a deep learning model, and unless specified in clear text, the models can be deployed in a remote server and remotely called at a client, and can also be deployed in a client with qualified equipment capability to be directly called.

Referring to fig. 1, a network architecture adopted in an exemplary application scenario in the present application includes a terminal device 80, an independent station server 81, and an application server 82, where the application server 82 may be configured to deploy a commodity combination recommendation service, and the commodity combination recommendation service is implemented by running a computer program product implemented according to the commodity combination recommendation method in the present application and opening a corresponding interface. The independent station server 81 is called an independent station for short, and can be used for deploying and opening online stores. The user on the terminal device 80 can access a certain basic commodity in the online shop through the independent station, including adding the basic commodity to a shopping cart, accessing a commodity detail page of the commodity, and the like, and the independent station server 81 further calls commodity information of other associated commodities in the optimal commodity combination corresponding to the basic commodity, pushes the commodity information of the other associated commodities to the user in association with the access of the user to the basic commodity, so that the terminal device 80 can further display the commodity information, and the effect of commodity collocation popularization is achieved.

The application server can be a server cluster, and various services are operated through a micro-service architecture, so that the commodity combination recommendation service can be operated in one or more containers provided by the micro-service architecture and can be used for concurrently opening and servicing massive independent stations.

With reference to the above schematic description, referring to fig. 2, a method for recommending a combination of products according to the present application, in an embodiment, includes the following steps:

step S1100, a tag set of a basic commodity is obtained, wherein the tag set comprises a plurality of attribute items, and each attribute item comprises an attribute name and an attribute tag;

in the independent station-based e-commerce platform, each independent station forms a commodity library for the online stores deployed therein, and stores commodity information of commodities in the online stores in the commodity library, wherein the commodity information includes, but is not limited to, commodity titles, commodity pictures, commodity attribute data, commodity links and the like. According to the commodity information of the commodity, the commodity can be marked by adopting various means, namely, the attribute labels corresponding to one or more attribute names of the commodity are marked. For example: the attribute name is 'C l ass', and the attribute label is 'jacket'; the attribute name is "Co l or", and the attribute label is "Red", etc. Therefore, the attribute items generated by marking can be actually regarded as one kind of the product attribute data, so that after each product is marked, a tag set can be obtained, wherein the tag set includes a plurality of attribute items, and each attribute item includes an attribute name and an attribute tag corresponding to the attribute name.

In one embodiment, the commodity library may store mapping relationship data between commodities and attribute items thereof in a form of a knowledge graph, so as to improve data storage efficiency, facilitate quick invocation, and subsequently may quickly generate a template library and a mapping relationship library on the basis of the knowledge graph.

When a user implements an access action for a basic commodity, for example, the basic commodity is added to a shopping cart, enters a commodity detail page of the basic commodity, enters a settlement page of an order in which the basic commodity is located, and the like, a corresponding event is triggered and sent to an independent station, and the independent station can respond to the event to acquire a tag set of the basic commodity, so that commodity information of other commodities forming a matching sale relationship with the basic commodity is further acquired and popularized to the user, and the promotion of the commodity transaction total amount of the independent station is promoted.

Step S1200, recalling all label combination templates carrying at least one attribute item in the label set from a template library, wherein the label combination templates comprise a plurality of attribute items;

the method includes the steps that a template library is prepared, a large number of label combination templates can be stored in the template library, the label combination templates usually comprise two or more attribute items, and a plurality of attribute items in the label combination templates can have the same attribute name or different attribute names and can be flexibly combined. It is easy to understand that the tag combination template defines the collocation relationship between different attribute items, and because the attribute items belong to tag data and have a one-to-many relationship with the commodities, the recall of the commodities carrying the same attribute items in the commodity library can be realized through the tag combination template.

In one embodiment, the tag combination template can be customized manually, and the practical experience can be templated by the tag combination template customized by means of manual sales experience, so that the method is obviously helpful for ensuring the rationalization of commodity combination relationship.

In another embodiment, the tag combination template may be extracted from the accumulated product visit industry dataset by data mining. In the data mining process, co-occurrence commodities with matching purchasing relations are extracted, and a label combination template is constructed by preferred combination according to the label sets of the co-occurrence commodities, so that the obtained label combination template is expected to be matched with the purchasing behaviors of users. Because deep association can be found by data mining, matching and purchasing relationships among commodities can be more comprehensively mined, and the obtained label combination template is more scientific and practical. For example, research in the united states shows that when a male consumer purchases diapers in a shopping mall, the male consumer often purchases beer by the way, and a merchant combines and populates the beer and the diapers, so that the sales volume of the beer can be effectively increased. Similar to the relation, the method can be more easily combed out by means of data mining, and then the method can be applied to online popularization to construct a corresponding label combination template so as to obtain an unexpected matching sale effect.

For the number of attribute items included in each tag combination template, it is understood that the larger the number is, the fewer the number of items included in the product combination recalled according to the tag combination template is, and conversely, the greater the number of items included in the corresponding product combination is. When the number of the commodities in a single commodity combination is too large, garbage recommendation is easily caused, and when the number of the commodities in the single commodity combination is too small, recommendation opportunities are wasted, so that the number of the attribute items included in the tag combination template can be preset according to actual measurement or experience, for example, the number can be set to 2, 3 or 4.

On the basis of the template library, each attribute item in the label set of the basic commodity is used as a query item, and all label combination templates containing the attribute item are queried from the template library, so that all label combination templates corresponding to each attribute item in the label set of the basic commodity can be obtained.

In one embodiment, the template library may establish an index table corresponding to each attribute item, and store all the tag combination templates carrying the attribute item corresponding to the index feature in association with the index feature, so that all the tag combination templates carrying the attribute item may be called quickly according to each attribute item. And the method can be flexibly implemented by a person skilled in the art as long as all the label combination templates containing the attribute items are conveniently and quickly called for the target attribute items.

Step S1300, recalling all the commodities matched with each label combination template from the mapping relation library, and combining all the commodities corresponding to each label combination template with the basic commodities to construct a plurality of commodity combinations respectively;

similarly, in order to conveniently recall the commodities according to the tag combination templates, a mapping relation library may be established in advance, and mapping relation data between each tag combination template and all commodities carrying all attribute items of the tag combination template is stored in the mapping relation library, so that each tag combination template may correspondingly obtain a commodity list, where each commodity carries each attribute item in the tag combination template. The plurality of label combination templates correspondingly obtain a plurality of commodity lists. On the basis of each product list, the base product is added thereto to constitute a product combination, thereby obtaining a plurality of product combinations.

In an embodiment, the mapping relation library may establish an index table corresponding to each tag combination template, and store all the commodities (specifically, ids of the commodities) carrying the tag combination template corresponding to the index feature in association with the index feature by using the tag combination template as the index feature, so that all the commodities carrying the tag combination model can be called quickly according to each tag combination template. And the like, the method can be flexibly implemented by a person skilled in the art as long as all goods containing the label combination template are conveniently and quickly called out for the target label combination template.

For example, if M attribute items are included in the tag set according to the basic product, and if N tag combination templates can be recalled for each attribute item, then M × N product combinations can be finally obtained based on the basic product. Therefore, the actual recalled commodity combinations are numerous, and the goal of recall is expected from the viewpoint of data retrieval.

Step S1400, determining at least one optimal combination of commodities from the plurality of commodities combinations, so as to push other commodities in the optimal combination of commodities when one commodity in the optimal combination of commodities is accessed.

Through the processes, a large number of commodity combinations are generated, the excessive commodity combinations are relatively disordered for collocation and popularization, and the accuracy is not high, so that from the data retrieval point of view, the multiple commodity combinations determined in advance can be sequenced and preferred by means of various existing sequencing means, and partial commodity combinations with popularization advantages are determined to be used as the optimal commodity combinations, so that data retrieval accuracy is realized. Then, the optimal commodity combination is configured to be the optimal commodity combination suitable for the basic commodity in the collocation promotion scene, and then the optimal commodity combination is matched with the basic commodity according to any one of the optimal commodity combinations to promote other commodities in the basic commodity to the user.

In one embodiment, the advertisement performance data obtained by each commodity combination on the e-commerce advertisement platform, including but not limited to CTR (click through rate) and/or CVR (conversion rate), may be obtained in units of each commodity combination, the advertisement performance data of all commodities in each commodity combination is counted, and the advertisement performance data is quantized into an external evaluation index corresponding to the uniqueness of the commodity combination through counting, and the optimal commodity combination is determined according to the external evaluation index.

In another embodiment, when any user accesses the basic product, one product combination may be randomly determined, and then other products in the product combination are promoted to the user according to the product combination, and then advertisement performance data of the promoted product combination itself is obtained (when each promotion is clicked, the product combination is considered to be effective for secondary promotion), and the same principles include, but are not limited to CTR (click through rate) and/or CVR (conversion rate), and the like, so that the advertisement performance data of each product combination may be obtained through practical tests, and then quantized into an external evaluation index corresponding to the uniqueness of the product combination, and the optimal product combination is determined according to the external evaluation index.

In another embodiment, the number of template hits of the N tag combination templates for each commodity combination may be queried in a reverse direction, and when the number of template hits of a commodity combination is larger, it may be indirectly indicated that the commodities in the commodity combination are more similar, and the commodities are expected to be bought in association, so that the number of template hits may be determined as the external evaluation index, and the optimal commodity combination may be determined preferentially according to the external evaluation index.

The process of determining one or more optimal combinations of commodities for a plurality of combinations of commodities based on the external evaluation index can be flexibly implemented. For example, only one commodity combination with the highest external evaluation index may be determined as the optimal commodity combination, a predetermined number of commodity combinations with the highest external evaluation index may be determined as the optimal commodity combination after the commodity combinations are sorted from the largest to the smallest external evaluation index, and one or more commodity combinations with the external evaluation index larger than a preset threshold value may be selected as the optimal commodity combination. And so on, as may be flexibly implemented by those skilled in the art.

After the optimal commodity combination of the basic commodities is determined according to any embodiment, the optimal commodity combination can provide basic data for collocation and promotion of the basic commodities, when a user accesses the basic commodities, any optimal commodity combination of the basic commodities is obtained, other commodities except the basic commodities are determined in the optimal commodity combination, commodity information of the other commodities is obtained and pushed to the user, and the promotion purpose can be achieved.

In another embodiment, each best product combination can be configured as a best product combination suitable for each product therein, so that when each product in the best product combination is accessed, other products except the product can be determined from the best product combination for collocation promotion.

According to the embodiments, the pre-constructed label combination templates are recalled according to various attribute items in the label set of the basic commodity, the label combination templates are provided in advance, then a large number of commodity combinations are recalled according to the label combination templates, the optimal commodity combination is determined preferentially on the basis of the commodity combination, and personalized customization or customization according to statistical data can be achieved with the help of the label combination templates.

On the basis of any embodiment of the present application, please refer to fig. 3, before obtaining the tag set of the base product, the method includes:

step S2100, generating an attribute item of a commodity according to commodity information of the commodity in a commodity library of an online shop by adopting a preset commodity labeling model;

any commodity labeling model suitable for labeling commodities in a commodity library of an online shop can be adopted for labeling the commodity library of the independent station, so that attribute items of all the commodities are generated, and each attribute item comprises an attribute name and an attribute label.

The commodity labeling model can be realized by adopting a deep learning model, the commodity information of the commodity is taken as input, deep semantic information in the input information is extracted, the feature representation of the commodity information is obtained, then classification mapping is executed according to the feature representation, the classification probability of each attribute item mapped to a preset commodity label system by the feature representation is obtained, all the attribute items with the classification probability reaching a preset threshold value can form a label set of the commodity, and the commodity is labeled by adopting the label set.

The commodity label system is generally suitable for unified planning of each independent station, and is divided into a plurality of attribute names, each attribute name can be assigned by adopting a single or a plurality of attribute labels in a corresponding attribute label set, and each attribute label is associated with the attribute name to which the attribute label belongs, so that an attribute item is formed.

The commodity information as the input information of the commodity labeling model may be monomodal information such as a commodity picture or a commodity title, or multimodal information such as a commodity picture or a commodity title, depending on the specific structure of the commodity labeling model.

In one embodiment, the commodity labeling model includes a text encoder followed by a multi-classifier, the text encoder employs a pre-training model such as LSTM, bert, and the like, which is suitable for performing feature representation on a text, the multi-classifier sets a plurality of corresponding classifications corresponding to the total amount of attribute items of a commodity label system, and the commodity labeling model employs a commodity title as a corresponding training sample to perform iterative training until convergence, so that the commodity labeling model can be used for predicting each attribute item corresponding to a commodity based on the commodity title of the commodity to form a label set of the commodity.

In another embodiment, the text encoder in the previous embodiment is replaced by an image encoder, which may be a convolutional neural Network (CNN, convo l ut i ona l neural Network), a residual error Network (Res idua l Network, resNet), or the like, and similarly, the commodity picture of the commodity is used as a training sample to train it to converge, so that it is suitable for predicting its label set according to the commodity picture of the commodity.

In another embodiment, the product labeling model may include the image encoder and the text encoder in the two previous embodiments, which are respectively used to extract the product image of the same product and the image feature representation and the text feature representation of the product title, and then the image feature representation and the text feature representation are spliced into the comprehensive feature information through a splicing layer, and then the comprehensive feature information is input into the classifier to perform classification mapping, so as to predict the label set of the product. In contrast, when the model is used, the product picture and the product title of the product are simultaneously input to provide information of two modes, and the model predicts the tag set of the product according to the multi-mode information.

Besides the above various embodiments, the commodity labeling model may be implemented by using any other architecture, as long as the finally obtained commodity labeling model is suitable for marking the corresponding commodity according to the commodity information.

When labeling a corresponding tag set on a certain commodity, in one embodiment, the tag set is added to a commodity library of an independent station and is stored in association with the commodity. In another embodiment, when the knowledge graph is used to store the mapping relationship data between the commodities and the attribute items, the mapping relationship between the commodities and the corresponding attribute items in the knowledge graph may be established according to the mapping relationship data between the commodities and the attribute items in the tag set thereof.

Step S2200, based on the commodity subsets with access behavior incidence relation in the co-occurrence commodity set, combining the attribute items marked by a plurality of different commodities in each commodity subset to construct a plurality of label combinations corresponding to the commodity subset;

a co-occurrence commodity set is prepared, and the co-occurrence commodity set can be obtained through data cleaning from a commodity access behavior data set prepared in advance. The co-occurrence merchandise set may include a plurality of merchandise subsets, each of which may be associated with access behaviors of the same nature, e.g., merchandise items a and merchandise items B are purchased in the same order, during the same session, in the same user cycle, etc. It can be seen that the subset of items, in effect characterizing the co-occurrence relationship information between items therein, is a representation of the relationship between different items having associated purchase facts at the time of actual sale. Generally, different items in the same subset of items will be identical or related with high probability in some attribute item. According to such features, each subset of items can be utilized to form a tag combination.

Specifically, for a plurality of commodities in each commodity subset, a tag set of each commodity is called, then, one attribute item of one of the commodities is combined with one or more attribute items of one or more other commodities to form a corresponding tag combination, and a plurality of tag combinations can be obtained for the same commodity subset by changing a combination mode. Since there are multiple subsets of co-occurring merchandise, a large number of combinations of labels can be obtained. Of course, the total number of attribute items included in each tag combination, for example, 2, 3, or 4, may be appropriately controlled in consideration of the rationality of the tag combination template customization.

Step S2300, counting the co-occurrence frequency of each label combination in all label combinations generated by the co-occurrence commodity set, and configuring a part of label combinations with relatively high co-occurrence frequency as the label combination templates in the template library.

As mentioned above, each subset of items may obtain a plurality of tag combinations, and the same tag combination may exist in each subset of items, and the greater the number of the tag combinations hit in the subset of items, the more closely the association degree of the tag combination is. The number of the subsets of the goods hit by the same tag combination, that is, the co-occurrence frequency between the attribute items in the tag combination. Therefore, the number of the commodity subsets hit by each label combination can be counted in the range of all the label combinations corresponding to all the commodity subsets of the co-occurrence commodity set, so that the co-occurrence frequency of each label combination is obtained, and finally the co-occurrence frequency list is obtained. On the basis of the co-occurrence frequency list, all label combinations are sorted according to the co-occurrence frequency, and then a part of label combinations with the maximum co-occurrence frequency are taken as label combination templates and stored in a preset template library.

According to the embodiments, data mining is performed by referring to a co-occurrence commodity set, a series of label combination templates are automatically generated, the label combination templates comprise a plurality of attribute items, the potential relationship of associated purchase is represented among commodities carrying the attribute items, valuable information can be mined without manual intervention, and the mined label combination templates are based on the attribute items and do not need to rely on commodity historical data, so that the requirement of a cold start scene can be met, and even if the commodity historical data of an independent station is scarce, the label combination templates can still be relied on to realize combination recommendation among commodities with collocation sales potential.

On the basis of any embodiment of the present application, please refer to fig. 4, before the subset of commodities having an access behavior association relationship in a co-occurrence commodity set, the method includes:

step S3100, extracting commodity information of commodities purchased in the same order from a preset commodity access behavior data set to form a first commodity subset in a co-occurrence commodity set;

the commodity access behavior data set can be commodity sales data or commodity advertisement data generated by a third-party e-commerce platform or a third-party independent station, wherein the commodity sales data or the commodity advertisement data comprise ordering information of a user, and therefore a plurality of commodity subsets can be extracted from the commodity access behavior data set.

The first commodity subset refers to commodities purchased in the same order, the number of the commodities in the order is considered by taking the order as a unit, when the number of the commodities is larger than 1, the user is indicated to purchase more than two commodities in the order, the two commodities are related in purchasing behavior, all commodities contained in the order can be constructed into the first commodity subset, commodity information of related commodities is stored in the first commodity subset, and the first commodity subset can be used for constructing a label combination.

Step S3200, extracting commodity information of commodities purchased in the same conversation period from a preset commodity access behavior data set to form a second commodity subset in a co-occurrence commodity set;

the second commodity subset in the commodity access behavior data set refers to commodities purchased in the same session, and the duration of the session can be flexibly set. Specifically, taking 5 minutes as an example, all orders of the user within the 5-minute session period are acquired, two or more items included in all the orders are configured as a second item subset, and the item information of the relevant items is stored therein, so that the second item subset can be used for configuring the label combination.

Step S3300, product information of products purchased by the same user in the same period is extracted from the preset product access behavior data set, and a third product subset in the co-occurrence product set is formed.

Similarly, the third subset of the items in the item access behavior data set refers to items purchased by the same user in the same period, and the period can also be flexibly set, for example, to be one week. It will be appreciated that the third subset of merchandise and the second subset of merchandise belong to the same nature, but are different from each other in the duration of the time period, and represent the long-term and short-term purchases of the user, respectively. In any event, a third subset of articles may be obtained in which article information for the associated articles is stored for use in constructing the tag combination.

According to the embodiment, the data cleaning is carried out by utilizing the existing commodity access behavior data set, so that the co-occurrence commodity set which can more accurately express the associated purchasing relationship information can be obtained, the tag combination template is conveniently and efficiently excavated, the excavation cost is low, and the excavation effect is high.

On the basis of any embodiment of the present application, recalling all the commodities matched with each tag combination template from the mapping relation library, and combining all the commodities corresponding to each tag combination template with the base commodity to construct a plurality of commodity combinations respectively, including:

step 1310, for each label combination template, querying the commodities carrying all attribute items in the label combination template from the mapping relation library to obtain a corresponding commodity list;

as described above, the mapping relation database stores mapping relation data of each tag combination template and all the commodities hitting the tag combination template, so that after all the tag combination templates corresponding to the basic commodity are determined, the mapping relation database is queried for each tag combination template, and a commodity list corresponding to each tag combination template can be obtained, and all the commodities hitting the tag combination template in the tag set are stored in the commodity list.

In one embodiment, to make the number of commodities in the commodity list more reasonable, the whole commodity list may be further filtered, for example, a commodity list in which only a single commodity or the number of commodities exceeds a preset threshold may be filtered, and a commodity list including two or more commodities and the total number of commodities being not higher than the preset threshold may be retained.

And S1320, adding the basic commodities to the commodity list corresponding to each label combination template, and enabling each commodity list to form a commodity combination.

The basic commodities can be added to the commodity list correspondingly obtained by each label combination template, so that the commodity list becomes a commodity combination, namely the commodity combination comprises the basic commodities and the associated commodities having potential purchasing relations. Therefore, as can be understood, the label combination templates can obtain corresponding product combinations, and the product combinations are generated based on the basic product.

According to the embodiment, the mapping relation library can provide a data source with efficient access for quickly generating the commodity combination of the basic commodity, a plurality of commodity combinations corresponding to the basic commodity are quickly generated, the operation on the original data of the commodity library every time is avoided, and the overall response efficiency of commodity combination recommendation can be improved.

Based on any embodiment of the present application, referring to fig. 5, determining at least one best combination from the plurality of combinations of commodities includes:

step 1410, counting the total template hit number of the label combination templates of the basic commodities hit by each commodity combination;

according to the above embodiment, each tag combination template obtained from a basic product can obtain a product combination, but a plurality of tag combination templates may be hit in the same product combination, so that the total number of all tag combination templates hit in the basic product in each product combination can be counted, thereby obtaining the total number of template hits corresponding to each product combination.

Step S1420, determining partial commodity combinations with relatively high total hit number of the templates as candidate commodity combinations;

it is understood that the higher the total number of template hits of a product combination, the more easily the product combination can be recalled by using a plurality of tag combination templates, and the more closely the association between the products in the product combination. Therefore, the commodity combinations of the basic commodities can be sorted according to the total number of the template hits, and then the commodity combinations with the relatively higher total number of the template hits are selected to be determined as candidate commodity combinations. The specific selection mode can be that a plurality of names are selected from the first ranked number of the total hit numbers of the templates from big to small, or the total hit number of the templates is higher than a preset threshold value.

And S1430, obtaining external evaluation indexes of each candidate commodity combination, and determining the optimal commodity combination according to the external evaluation indexes.

With regard to specific ways of obtaining the external evaluation index of each candidate product combination, reference may be made to various embodiments disclosed hereinbefore. In short, the candidate commodity combinations are screened by using the external evaluation indexes, and one or more optimal commodity combinations can be determined. The best commodity combination can determine the basis for selecting other commodities for popularization when the basic commodity is recommended by the commodity combination.

According to the embodiment, on the basis of the total number of matched basic commodity combinations, the template hit total number of each commodity combination is combined, so that part of commodity combinations with low practical value can be filtered, the optimal commodity combination is determined, and the determined optimal commodity combination can accurately contain associated commodities having potential purchasing relations with the basic commodity.

On the basis of any embodiment of the present application, please refer to fig. 6, obtaining external evaluation indexes of each candidate commodity combination, and determining an optimal commodity combination according to the external evaluation indexes, including:

step S1431, responding to an accessed event of the commodity detail page of the basic commodity, and randomly acquiring a candidate commodity combination of the basic commodity as a target commodity combination;

after all the commodity combinations are matched for the basic commodities, the commodity combinations can be used as candidate commodity combinations, and the best commodity combination is determined by actually measuring the candidate commodity combinations. This approach is particularly suitable for use with a cold start stand-alone station.

And when the user of the independent station visits the commodity detail page of the basic commodity in the online shop of the independent station, triggering an accessed event of the basic commodity to be sent to the independent station, responding to the event by the independent station, and randomly determining one of the candidate commodity combinations determined for the basic commodity as a target commodity combination according to actual measurement requirements.

Step S1432, obtaining the commodity information of other commodities except the basic commodity in the target commodity combination, and loading the commodity information into the commodity detail page;

after the target commodity combination is determined, the independent station further determines other commodities in the target commodity combination except the basic commodity, then calls commodity information of the other commodities from a commodity library, wherein the commodity information comprises commodity pictures, commodity prices, commodity links and other commodity display blocks used for packaging the other commodities, and then pushes the packaged commodity display blocks to the terminal equipment of the user, so that the packaged commodity display blocks are loaded and displayed in a commodity detail page of the basic commodity after being analyzed to realize exposure.

Step S1433, monitoring accessed events of other commodities in the target commodity combination, counting the conversion rate of the whole target commodity combination according to the accessed events, and taking the conversion rate as an external evaluation index;

after the user receives the display information of the commodity display blocks of the other commodities, the user may have a desire to purchase any other commodity, click the corresponding other commodity, and then perform ordering operation for the other commodities, thereby completing one conversion.

The independent station may monitor, in the background, the visited events triggered by the user after the other commodities are promoted, specifically, the visited events may be ordering events for the other commodities, and each visited event may be resolved into a primary conversion event for a target commodity combination where the other commodity is located. Therefore, by randomly selecting different candidate commodity combinations for actual measurement, data corresponding to the conversion events of each candidate commodity combination can be obtained, the exposure times and the conversion times of each candidate commodity combination are counted according to the data, then the conversion rate of each candidate commodity combination can be obtained by dividing the conversion times of each candidate commodity combination by the exposure times, and the conversion rate can be used as an external evaluation index of the corresponding candidate commodity combination.

Step S1434 is to determine a partial candidate commodity combination with a relatively high external evaluation index as an optimal commodity combination.

After the external evaluation index is determined, all candidate commodity combinations of the base commodities may be screened according to the external evaluation index to determine one or more optimal commodity combinations, for example, only one unique candidate commodity combination with the highest external evaluation index may be determined as the optimal commodity combination, after the external evaluation index ranks the candidate commodity combinations from large to small, a predetermined number of candidate commodity combinations ranked in the front may be determined as the optimal commodity combination, and one or more candidate commodity combinations with the external evaluation index larger than a preset threshold may be screened as the optimal commodity combination. And so on, as may be flexibly implemented by those skilled in the art.

According to the embodiments, for the independent station with sparse data and needing cold start, multiple candidate commodity combinations determined for a certain basic commodity can be associated with the basic commodity to be pushed to a user for actual measurement under the condition of not depending on external reference data, the conversion rate of each candidate commodity combination is counted by behavior data generated by the actual measurement of the user to serve as an external evaluation index, an optimal commodity combination is selected from all candidate commodity combinations according to the external evaluation index, the process is dynamically performed, the obtained actual measurement results are more and more obtained along with the extension of the running time of the independent station, the behavior characteristics of the user of the station can be reflected in a personalized mode, the acceptance degree of each candidate commodity combination can be effectively reflected, the finally determined optimal commodity combination can reflect the collocation and sales potential association relation among different commodities more accurately, and the commodity combination recommendation can be realized in this mode, and the total commodity transaction amount of a shop can be increased.

Based on any embodiment of the present application, after determining at least one best combination from the plurality of combinations, the method includes:

s1500, responding to the event that the commodity detail page of the basic commodity is accessed, and determining the optimal commodity combination corresponding to the basic commodity;

when the best combinations of the commodities are matched for the basic commodities, the best combinations of the commodities can be used in the commodity combination recommendation service of the independent station.

When a user of the independent station visits a commodity detail page of the basic commodity in an online shop of the independent station, an accessed event of the basic commodity is triggered to be sent to the independent station, the independent station responds to the event, one of a plurality of optimal commodity combinations determined for the basic commodity can be randomly determined according to the requirement of commodity combination recommendation, and accordingly combination recommendation of other commodities is conducted.

And S1600, acquiring the commodity information of other commodities except the basic commodity in the optimal commodity combination, and loading the commodity information into the commodity detail page for display.

After the optimal commodity combination used for implementing commodity recommendation is determined, the independent station further determines other commodities in the optimal commodity combination except the basic commodity, then the commodity information of the other commodities, including commodity pictures, commodity prices, commodity links and other commodity display blocks used for packaging the other commodities, is called from the commodity library, and then the packaged commodity display blocks are pushed to the terminal equipment of the user, so that the packaged commodity display blocks are loaded and displayed in the commodity detail page of the basic commodity after being analyzed to realize exposure.

After the user receives the display information of the commodity display blocks of the other commodities, the user may have a desire to purchase any other commodity, so that the corresponding other commodity is clicked, then ordering operation for the other commodities is performed, one-time conversion is completed, the effect of commodity collocation popularization is achieved, and the purpose of commodity collocation sale is achieved.

According to the embodiment, the best commodity combination can be determined for each commodity of the independent station, and then other commodities with potential purchase association relations are matched for the commodity according to the best commodity combination for popularization, so that the commodity combination recommendation strategy of the independent station is better, cold start can be realized, and higher commodity sales expectation can be obtained.

It can be further understood that, for an independent station-based e-commerce platform, in order to provide the commodity combination recommendation service of the present application to a large number of independent stations under the platform, a computer program product implemented according to the commodity combination recommendation method of the present application may be deployed in an application server, and open a corresponding service interface for each independent station, and each independent station calls the service interface to use the commodity combination recommendation service, so that a large-scale economic effect is obtained, and a service experience of the whole e-commerce platform is provided.

Referring to fig. 7, an apparatus for recommending a combination of commodities, provided according to an aspect of the present application, includes a tag set obtaining module 1100, a template recall module 1200, a combination recall module 1300, and a combination determining module 1400, where the tag set obtaining module 1100 is configured to obtain a tag set of a basic commodity, the tag set includes a plurality of attribute items, and each attribute item includes an attribute name and an attribute tag; the template recalling module 1200 is configured to recall all the tag combination templates carrying at least one attribute item in the tag set from a template library, where the tag combination templates include a plurality of attribute items; the combined recall module 1300 is configured to recall all the commodities matched with each label combination template from the mapping relation library, and combine all the commodities corresponding to each label combination template with the basic commodity to construct a plurality of commodity combinations; the combination determination module 1400 is configured to determine at least one best combination of the plurality of combinations of commodities, so that when one commodity in the best combination of commodities is accessed, other commodities in the best combination of commodities are pushed.

On the basis of any embodiment of the present application, the tag set obtaining module 1100 includes: the marking processing module is used for generating an attribute item of the commodity according to the commodity information of the commodity in the commodity library of the online shop by adopting a preset commodity marking model; the tag group construction module is arranged for combining attribute items marked by a plurality of different commodities in each commodity subset based on the commodity subsets with access behavior incidence relation in the co-occurrence commodity set and constructing a plurality of tag combinations corresponding to the commodity subsets; and the statistical optimization module is arranged for counting the co-occurrence frequency of each label combination in all label combinations generated by the co-occurrence commodity set, and configuring the part of the label combinations with relatively high co-occurrence frequency as the label combination templates in the template library.

On the basis of any embodiment of the application, a module is constructed before the tag group, and the module comprises: the first acquisition module is used for extracting commodity information of commodities purchased in the same order from a preset commodity access behavior data set to form a first commodity subset in a co-occurrence commodity set; the second acquisition module is used for extracting commodity information of commodities purchased in the same conversation period from a preset commodity access behavior data set to form a second commodity subset in the co-occurrence commodity set; and the third acquisition module is used for extracting commodity information of commodities purchased by the same user in the same period from the preset commodity access behavior data set to form a third commodity subset in the co-occurrence commodity set.

On the basis of any embodiment of the present application, the combined recall module 1300 includes: the query processing unit is arranged for querying the commodities carrying all attribute items in the label combination template from the mapping relation library aiming at each label combination template to obtain a corresponding commodity list; and the combination processing unit is used for adding the basic commodities to the commodity lists corresponding to the label combination templates to enable the commodity lists to form commodity combinations.

On the basis of any embodiment of the present application, the combination determination module 1400 includes: a total counting unit configured to count a template hit total of tag combination templates of the base commodity hit by each commodity combination; a candidate preference unit configured to determine a partial commodity combination with a relatively high total number of hits in the templates as a candidate commodity combination; and the index optimization unit is arranged to acquire external evaluation indexes of the candidate commodity combinations and determine the optimal commodity combination according to the external evaluation indexes.

On the basis of any embodiment of the present application, the index optimizing unit includes: the random recommendation subunit is set to respond to the accessed event of the commodity detail page of the basic commodity and randomly acquire a candidate commodity combination of the basic commodity as a target commodity combination; a recommended loading subunit, configured to acquire commodity information of other commodities except the basic commodity in the target commodity combination, and load the commodity information into the commodity detail page; the index generation subunit is configured to monitor accessed events of other commodities in the target commodity combination, count the conversion rate of the whole target commodity combination according to the accessed events, and use the conversion rate as an external evaluation index; an optimum determination subunit configured to determine, as an optimum product combination, a partial candidate product combination for which the external evaluation index is relatively high.

On the basis of any embodiment of the present application, the combination determining module 1400 includes: the combination recommending module is used for responding to the event that the commodity detail page of the basic commodity is accessed and determining the optimal commodity combination corresponding to the basic commodity; and the recommendation loading module is used for acquiring the commodity information of other commodities except the basic commodity in the optimal commodity combination, and loading the commodity information into the commodity detail page for display.

Another embodiment of the present application further provides a combined commodity recommending apparatus. As shown in fig. 8, the internal structure of the product combination recommendation device is schematically illustrated. The commodity combination recommending device comprises a processor, a computer readable storage medium, a storage and a network interface which are connected through a system bus. The computer-readable non-volatile storage medium of the commodity combination recommendation device stores an operating system, a database and computer-readable instructions, the database can store information sequences, and the computer-readable instructions can enable a processor to realize a commodity combination recommendation method when being executed by the processor.

The processor of the commodity combination recommendation device is used for providing calculation and control capacity and supporting the operation of the whole commodity combination recommendation device. The memory of the commodity combination recommendation device can store computer readable instructions, and when the computer readable instructions are executed by the processor, the processor can execute the commodity combination recommendation method. The network interface of the commodity combination recommending device is used for being connected and communicated with the terminal.

It will be understood by those skilled in the art that the configuration shown in fig. 8 is a block diagram of only a portion of the configuration associated with the present application, and does not constitute a limitation on the merchandise combination recommendation device to which the present application is applied, and a particular merchandise combination recommendation device may include more or less components than those shown in the drawings, or may combine certain components, or have a different arrangement of components.

In this embodiment, the processor is configured to execute specific functions of each module in fig. 7, and the memory stores program codes and various data required for executing the modules or sub-modules. The network interface is used for realizing data transmission between user terminals or servers. The non-volatile readable storage medium in the present embodiment stores program codes and data necessary for executing all modules in the product combination recommendation device of the present application, and the server can call the program codes and data of the server to execute the functions of all modules.

The present application also provides a non-transitory readable storage medium storing computer-readable instructions, which when executed by one or more processors, cause the one or more processors to perform the steps of the method for combined recommendation of an item of any of the embodiments of the present application.

The present application also provides a computer program product comprising computer programs/instructions which, when executed by one or more processors, implement the steps of the method according to any embodiment of the present application.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments of the present application may be implemented by hardware related to instructions of a computer program, which may be stored in a non-volatile readable storage medium, and when executed, may include the processes of the embodiments of the methods as described above. The storage medium may be a magnetic disk, an optical disk, a Read-only Memory (ROM), or a Random Access Memory (RAM).

In summary, the application can determine the optimal commodity combination for the basic commodities according to the corresponding relation of the templates and the commodities about the labels with the help of the label combination templates, realizes collocation and sales, obtains a conversion effect more in line with expectation, improves the sales total of shops on the e-commerce platform line, is not influenced by the sparse congenital deficiency of data under the cold start condition of the independent station, and is suitable for the cross-border e-commerce platform application scene based on the independent station.

Claims

1. A combined commodity recommendation method is characterized by comprising the following steps:

recalling all the commodities matched with each label combination template from the mapping relation library, and combining all the commodities corresponding to each label combination template with the basic commodities to construct a plurality of commodity combinations respectively;

2. The combination recommendation method for commodities as claimed in claim 1, wherein before obtaining the tag set of the base commodity, the method comprises:

combining attribute items marked by a plurality of different commodities in each commodity subset based on commodity subsets with access behavior association relations in a co-occurrence commodity set to construct a plurality of label combinations corresponding to the commodity subsets;

and counting the co-occurrence frequency of each label combination in all label combinations generated by the co-occurrence commodity set, and configuring the partial label combinations with relatively high co-occurrence frequency as the label combination templates in the template library.

3. The commodity combination recommendation method according to claim 2, wherein before the commodity subset having the access behavior association relationship in the co-occurrence commodity set, the method comprises:

4. The method of claim 1, wherein the recalling all the commodities matched with each tag combination template from the mapping relation library, and combining all the commodities corresponding to each tag combination template with the base commodity to construct a plurality of commodity combinations respectively comprises:

and adding the basic commodities to the commodity lists corresponding to the label combination templates to enable the commodity lists to form commodity combinations.

5. The merchandise combination recommendation method according to claim 1, wherein determining at least one best merchandise combination from the plurality of merchandise combinations comprises:

determining part of commodity combinations with relatively high total hit number of the templates as candidate commodity combinations;

6. The commodity combination recommendation method according to claim 5, wherein the step of obtaining an external evaluation index of each candidate commodity combination and determining an optimal commodity combination according to the external evaluation index comprises:

7. The product combination recommendation method according to any one of claims 1 to 6, wherein determining at least one optimal product combination from the plurality of product combinations comprises:

8. An article combination recommendation device, comprising:

the system comprises a tag set acquisition module, a tag set processing module and a tag matching module, wherein the tag set acquisition module is used for acquiring a tag set of a basic commodity, the tag set comprises a plurality of attribute items, and each attribute item comprises an attribute name and an attribute tag;

9. A merchandise portfolio recommendation device comprising a central processor and a memory, wherein the central processor is configured to invoke execution of a computer program stored in the memory to perform the steps of the method of any one of claims 1 to 7.

10. A non-transitory readable storage medium, storing a computer program implemented according to the method of any one of claims 1 to 7 in the form of computer readable instructions, the computer program, when invoked by a computer, performing the steps comprised by the corresponding method.