KR20190114703A - Method for recommending items and server using the same - Google Patents

Abstract

The present invention relates to a product recommendation method and a server using the same. According to an embodiment of the present invention, the product recommendation method comprises: a product selection step of receiving selection input for a first product from a user among a plurality of registration products; a candidate product extraction step of extracting a plurality of candidate products related to the first product by using a correlation model in which correlation among the registration products is modeled; and a recommendation product setting step of setting a recommendation product for the first product by comparing a preference vector of the user and an expression vector of the candidate products.

Description

Method for recommending items and server using the same}

The present application relates to a product recommendation method and a server using the same, and a product recommendation method capable of providing a customized product recommendation service to a user during online shopping, and a server using the same.

An online shopping mall or an internet shopping mall refers to a virtual shop made to buy and sell goods using the Internet. Unlike an offline shopping mall, an online shopping mall has no time and space constraints, and consumers can purchase products at a relatively low price anytime and anywhere. The operator of the online shopping mall can provide product information suitable for the consumer's taste without restriction of commercial rights. The operator of an online shopping mall may directly sell a product to a consumer or may connect a seller and a consumer to relay a sale. The online shopping market is growing rapidly in recent years due to the development of information and communication technology and the convenience of product purchase.

On the other hand, online shopping malls recommend products in a variety of ways to facilitate the purchase of customers. Specifically, two analysis methods are adopted for product recommendation. There are two types of analysis methods, one for analyzing individual customers' online activities and another for analyzing online activities. An example of the first method is to refer to a purchase history of a customer and recommend a product related to a product purchased in the past to generate a buying activity. Examples of the second method include recommending a product having a high sales volume, recommending a product by analyzing comments on the product, or recommending a product by referring to a purchase history of another customer.

The present application is to provide a product recommendation method that can provide a personalized product recommendation service and a server using the same.

The present application is to provide a product recommendation method and a server using the same that can present a recommendation product associated with a specific product in consideration of the user's taste or style when the user selects a specific product.

A product recommendation method according to an embodiment of the present invention includes a product selection step of receiving a selection input for a first product from a user among a plurality of registered products; A candidate product extraction step of extracting a plurality of candidate products associated with the first product by using an association model modeling the association between the registered products; And a recommendation product setting step of setting a recommendation product for the first product by comparing the taste vector of the user and the expression vector of the candidate products.

Server according to an embodiment of the present invention, the selection input detection unit for receiving a selection input for the first product from the plurality of registered products from the user; A candidate product extraction unit configured to extract a plurality of candidate products associated with the first product by using an association model modeling the association between the registered products; And a recommendation product setting unit configured to set a recommendation product for the first product by comparing the taste vector of the user with the expression vector of the candidate products.

Server according to another embodiment of the present invention, a processor; And a memory coupled to the processor, wherein the memory includes one or more modules configured to be executed by the processor, wherein the one or more modules select input for a first product from among a plurality of registered products. Is extracted from a user, a plurality of candidate products associated with the first product are extracted using an association model modeling the association between the registered products, and the taste vector of the user is compared with the expression vector of the candidate products. In this case, the command may include a command for setting a recommendation product for the first product.

In addition, the solution of the said subject does not enumerate all the characteristics of this invention. Various features of the present invention and the advantages and effects thereof may be understood in more detail with reference to the following specific embodiments.

According to the product recommendation method and the server using the same according to an embodiment of the present invention, since a personalized product recommendation service can be provided, reliability of the recommended product and user convenience can be improved.

According to a product recommendation method and a server using the same according to an embodiment of the present invention, a candidate product is extracted by using a Meta-Prod2Vec technique, and then a recommended product is set using a user's taste vector extracted by using a topic model. In addition, accurate product recommendations based on user preferences and preferred styles are possible.

However, the product recommendation method and the effect that can be achieved by the server using the same according to the embodiments of the present invention are not limited to those mentioned above, and other effects not mentioned are described in the following description. It will be clearly understood by those skilled in the art.

1 is a schematic diagram showing a product recommendation system according to an embodiment of the present invention.
2 and 3 is a block diagram showing a server according to an embodiment of the present invention.
4 is a schematic diagram showing a product list and a product recommendation area according to an embodiment of the present invention.
5 and 6 are schematic diagrams illustrating a preference style vector setting using a topic model according to an embodiment of the present invention.
7 is a schematic diagram showing a reading list according to an embodiment of the present invention.
Figure 8 is a schematic diagram showing the association between products using meta-prod2vec according to an embodiment of the present invention.
9 is a schematic diagram illustrating a product recommendation scenario according to an embodiment of the present invention.
10 is a flowchart illustrating a product recommendation method of a server according to an exemplary embodiment of the present invention.

Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings, and the same or similar components are denoted by the same reference numerals regardless of the reference numerals, and redundant description thereof will be omitted. The suffixes "module" and "unit" for components used in the following description are given or used in consideration of ease of specification, and do not have distinct meanings or roles from each other. In other words, the term 'part' used in the present invention refers to a hardware component such as software, FPGA or ASIC, and 'part' plays a role. But wealth is not limited to software or hardware. The 'unit' may be configured to be in an addressable storage medium or may be configured to play one or more processors. Thus, as an example, a 'part' may include components such as software components, object-oriented software components, class components, and task components, processes, functions, properties, procedures, Subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays and variables. The functionality provided within the components and 'parts' may be combined into a smaller number of components and 'parts' or further separated into additional components and 'parts'.

In addition, in describing the embodiments disclosed herein, when it is determined that the detailed description of the related known technology may obscure the gist of the embodiments disclosed herein, the detailed description thereof will be omitted. In addition, the accompanying drawings are intended to facilitate understanding of the embodiments disclosed herein, but are not limited to the technical spirit disclosed herein by the accompanying drawings, all changes included in the spirit and scope of the present invention. It should be understood to include equivalents and substitutes.

1 is a schematic diagram showing a product recommendation system according to an embodiment of the present invention.

Referring to FIG. 1, a product recommendation system according to an embodiment of the present invention may include a user terminal 1 and a server 100.

Hereinafter, a product recommendation system according to an embodiment of the present invention will be described with reference to FIG. 1.

The user terminal 1 is a terminal device used by a user, and the user terminal 1 may communicate with the server 100 or the like using a wired or wireless communication network. Here, the user terminal 1 may include a user equipment (UE), a mobile station (MS), a mobile subscriber station (MSS), a subscriber station (SS), an advanced mobile station (AMS), a wireless terminal (WT), and a machine- MTC. It may be replaced by terms such as a type communication (M2M) device, a machine-to-machine (M2M) device, a device-to-device (D2D) device, and a station (STA).

The user terminal 1 may perform voice or data communication through a wired or wireless communication network, and the user terminal 1 executes operations by executing various programs, a memory for storing a browser, a program, and a protocol for transmitting and receiving information. It may be provided with a microprocessor for controlling.

The user terminal 1 according to an embodiment of the present invention may be implemented in various forms. For example, the user terminal 1 may be a smart phone, a tablet PC, a notebook computer, a personal digital assistant (PDA), a portable multimedia player (PMP), an MP3 player, a wearable device (eg, a wearable device). For example, a mobile terminal to which a wireless communication technology such as a smartwatch, a smart glass, or a head mounted display (HMD) is applied, or a digital device that is difficult to carry such as a PC, an IPTV, a smart TV, etc. Can be.

The user terminal 1 and the server 100 may be connected to each other via a communication network. The communication network may include a wired network and a wireless network. Specifically, the communication network may include various networks such as a local area network (LAN), a metropolitan area network (MAN), and a wide area network (WAN). It may include. The communication network may also include a known World Wide Web (WWW). However, the communication network according to the present invention is not limited to the networks listed above, and may include at least one of a known wireless data network, a known telephone network, and a known wired / wireless television network.

The user terminal 1 may provide a service to a user based on the information provided from the server 100. For example, when the server 100 is a web server, the user terminal 1 may provide a web service based on content provided from the server 100. In addition, according to an embodiment, the user may use the user terminal 1 to access a shopping website provided by the server 100 or to execute a shopping-only application, through which an online shopping service, such as purchasing a product, to the user. It is also possible to provide.

In this case, the user terminal 1 may display the user interface on the display unit of the user terminal 1, and as shown in FIG. 4A, a list of a plurality of items may be displayed on the display unit. . When the user applies a selection input for a specific product, as illustrated in FIG. 4B, product information including an image or a detailed description of the selected product may be displayed. In this case, the server 100 may extract a recommendation product related to the product selected by the user and provide it to the user terminal 1. In this case, the user terminal 1 may display a recommendation product in the product recommendation region r of the display unit to provide a product recommendation service.

The server 100 is a service providing server, and may provide a product recommendation service or an online shopping service including a product recommendation service to the user terminal 1. In the server 100, a plurality of products may be registered in order to sell products through online shopping. In addition, the server 100 may store the product information corresponding to each registered product, the stored product information may include a variety of information such as the image of the product, product name, price, manufacturer, vendor, detailed description of the product have.

On the other hand, the server 100 can determine the user's style or taste through the history information about the products that the user has viewed or purchased in the past, and can set the recommended product according to the identified user's style or taste. That is, when the user selects a specific product, among the plurality of products associated with the product, a product corresponding to the user's style or taste may be determined and provided to the user terminal 1. Thereafter, the user terminal 1 may display the recommended product and provide it to the user.

That is, the server 100 according to an embodiment of the present invention may provide a customized product recommendation service for each user in consideration of a user's style, etc., rather than simply recommending a product similar to the product selected by the user. have. Hereinafter, a server according to an embodiment of the present invention will be described in detail.

2 is a block diagram illustrating a server according to an exemplary embodiment of the present invention.

Referring to FIG. 2, a server according to an embodiment of the present invention includes a 100, a taste analyzer 110, a selection input receiver 120, a candidate product extractor 130, and a recommendation product setter 140. can do.

The taste analyzer 110 may generate a taste vector corresponding to a user's taste or style. In detail, the taste analyzing unit 110 may identify the taste or style of the user and generate a taste vector corresponding to the taste or style using the history information storing the list of a plurality of registered products that the user has viewed in the past.

When the user performs online shopping, the user may view various registered products and receive product information such as an image or a detailed description of the product in order to purchase the product. In this case, the list of registered products viewed by the user may be stored as history information.

Since the user generally checks the product information by reading the registered products according to the user's preferred style or taste, the list of registered products stored in the history information may reflect the user's preferred style. Therefore, the taste analyzer 110 may generate a taste vector indicating the taste or style of the user by using the user's history information.

According to an embodiment, a user may generate a preference style vector of a user using a topic model, and set the generated preference style vector as a taste vector.

However, in order to generate a user's preferred style vector using the topic model, it is necessary to first train the topic model. That is, the topic model is trained, which styles each registered product corresponds to, and then, based on the styles of each registered product viewed or purchased by the user, a preferred style vector corresponding to the user's preferred style is obtained. Can be generated. Here, the topic model may be generated in advance by using the server 100 or a separate device.

Topic models are generally developed statistically to discover abstract topics in a document set. In some embodiments, a topic model may be formed by using a latent rich allocation (LDA) technique. Here, the LDA technique is an example of an unsupervised learning algorithm that can classify which subjects exist in each document using a plurality of document sets.

Referring to FIG. 5A, the product information for each registered product may include text information formed of text, such as a name t1 of the product or a detailed description t2 of the product. Here, it is assumed that all text information describing one product represents one style. That is, the product information of the registered products can all be seen as expressing the same style. Subsequently, a topic model may be generated by applying the LDA technique to the extracted text information. In this case, the topic model may derive a style corresponding to each registered product as a topic based on the text information.

Fig. 5 (b) illustrates the formation of a topic model for a registered product, where α is a parameter of k-dimensional Dirichlet distribution, θ is a k-dimensional vector and θ ⁱ is a document in the i-th subject. Β is the matrix parameter of size kx V, β _ij is the probability that the ith subject produces the j word, Z _n is the k-dimensional vector, and Z _n ⁱ is the word ω _n is the i theme Probability distribution belonging to Here, N is the number of words in one product information, K is the number of styles (topics) included in one product information, M is the number of product information.

That is, the text information may be extracted from the product information of the registered product, and the topic model may be generated by learning the text information about the plurality of registered products by the LDA technique. According to the topic model, a distribution of words included in a specific style, a style corresponding to a specific product, and the like can be derived.

As described above, by using the topic model, it is possible to distinguish which style each product corresponds to, and to display each product as a product style vector corresponding to the corresponding style.

After the topic model is generated, the taste analyzer 110 may apply the history information of the user to the topic model to extract a preference style vector corresponding to the user. Specifically, product information of registered products that the user has viewed in the past may be extracted from the user's history information. Here, text information formed from text such as a detailed description may be extracted, and the corresponding preference style vector may be extracted by inputting the text information into the topic model.

That is, as shown in Fig. 6 (a), the word set of the user A is extracted by extracting the text information included in the product name, the detailed description of the product, etc. from the product information included in the history information h1 clicked by the user A. Can be formed. Then, by inputting the word sets into the topic model, a preference style vector for user A can be obtained. 6 (a) shows, as an example, a 10-dimensional preferred style vector v1 for a user, and the k-th preferred style vector is a probability distribution of the user belonging to the k-th style, and the probability of belonging to the style is high. It is shown in darker colors. Although shown here in 10 dimensions for convenience, the size of the preferred style vector may be adjusted as needed, and preferably may be configured to about 200 dimensions. In addition, according to the topic model, style expressions e1 and e2 displaying words representing a user's preferred style may also be generated.

On the other hand, as shown in Figure 6 (b), when compared to the user A, the user B is different from the previously clicked history information (h2), the word set included in the text information extracted from the history information is different do. Accordingly, the preference style vectors v1 and v2 extracted through the learned topic model also differ from the style expressions e1 and e2. In other words, user A prefers a style that corresponds to words such as "open and close", "ivory", "white", while user B prefers a style that corresponds to words such as "race" and "sexy". Can be determined. Accordingly, the user may determine a preference style of each user by using a topic model, which may be represented as a preference style vector (v1, v2). In addition, according to an exemplary embodiment, after extracting the product style vectors of the products included in the user's history information, it may be set as the user's preferred style vector using the extracted product style vectors.

In addition, according to an embodiment, the taste analysis unit 110 extracts a corresponding characteristic vector by applying each registered product included in the user's history information to the association model, and user characteristics corresponding to the average of the extracted characteristic vectors. It is also possible to set the vector as a taste vector. However, since the characteristic vector of the registered products will be described later, a detailed description thereof will be omitted.

The selection input receiver 120 may receive a selection input applied by the user. The server 100 may provide a list of registered products to the user terminal 1 according to a user's request, and the user may select a first product he / she wants to view from among a plurality of registered products. In this case, the selection input receiver 120 may receive a selection input for the first product selected by the user from the user terminal 1, and include an image or a detailed description of the first product in response to the received selection input. Product information can be displayed. Here, the selection input applied by the user terminal 1 may be an input such that the user clicks or touches the registered product, and according to an embodiment, an interaction such as “steaming” or “adding a shopping cart” may be performed. It may include.

The candidate product extraction unit 130 may extract a plurality of candidate products associated with the first product by using an association model that models the association between the registered products. In order to provide a recommendation product for the first product selected by the user, it is necessary to extract candidate products associated with the first product. To this end, the candidate product extraction unit 130 may extract candidate products associated with the first product by using an association model formed through machine learning.

That is, the candidate product extractor 130 may extract the feature vectors corresponding to the first product and the registered product, respectively, by using the association model, and calculate the similarity between each feature vector to determine the candidate product. Here, the association model may be formed by prod2vec or meta-prod2vec. Meta-Prod2Vec is a method for embedding products, similar to word2vec, which embeds words in a vector space. A model that can be trained to have similar values for shared product eigenvectors or metavectors in context to be. In the present invention, the products that are viewed in the same context model the association between the products through the assumption that they are related to the set of products.

In this case, the association model may collect and view a list of reading lists of a plurality of registered products viewed by a user in one shopping session, and model and learn using the collected samples. As shown in FIG. 7, the user can sequentially view and confirm a plurality of products in a single session (s1, s2) during online shopping. As such, the products in the same shopping context are associated with each other. It can be seen as a commodity set. For example, when a user wants to purchase a "blouse", the user can check and sequentially view the items corresponding to his or her style by clicking among the various "blouses". That is, since the set of products clicked by the user in relation to the "blouse" corresponds to a product that is related to each other, the association between the products may be extracted based on the product.

However, instead of extracting relevant products using only selected products before and after the products clicked by the user, the meta information of each product and the user information on the user who clicked the product are considered. It is also possible to extract the association between them. Product specific information, meta information, and user information may be set in one product, and product unique information is an identifier for distinguishing each product, and meta information is an image of a product, a product name, a price of a product, a brand of a product, and an attribute. Etc. may be variously set. In addition, the user information may include a gender, age, and the like of the user.

According to Meta-Prod2Vec, product specific information, meta information, and user information of each product may be displayed in a vector in space according to the association between the products. Referring to FIG. 8, a product specific vector (P _n ), a meta vector (M _n ), and a user information vector (Un) corresponding to each product may be set, and the product unique vector is embedded as a unique vector for the product. However, since the meta vector or the user information vector may have the same information or the same user may be the same for other products, the embedding vector may be shared between the products.

The candidate product extraction unit 130 uses the product inherent vectors P _A and P _B and the meta vectors M _A and M _B for each product A and B extracted by the trained Meta-Prod2Vec model. Can generate a feature vector for an individual product. According to an exemplary embodiment, the characteristic vector may be generated by using a metavector as a characteristic vector or by joining a product-specific vector and a metavector. Here, the server 100 may generate and store a characteristic vector for each registered product in advance. In the case of the product specific vector, embedding may be difficult when there is no corresponding product information in the learning data, and thus it may be supplemented using a technique such as zero padding.

The candidate product extractor 130 may calculate similarity between the feature vectors of the registered products and the feature vector of the selected first product. Thereafter, the candidate product extracting unit 130 may extract registered products having a high similarity with the characteristic vector of the first product, and set the candidate products to be viewed together with the first product. Here, the candidate product may be set by extracting a predetermined number of registered products according to the order of high similarity, and according to the exemplary embodiment, all registered products having a similarity or more may be set as candidate products. The similarity between characteristic vectors can be obtained by calculating cosine similarity, and any other similarity can be applied as long as the similarity between vectors can be calculated.

Meanwhile, candidate products for the first product may be calculated and extracted by the candidate product extraction unit 130 when the user selects the first product, or may be determined in advance by calculating in advance. Therefore, when the user clicks on a specific product and selects it as the first product while browsing the registered products, the candidate product extractor 130 may extract candidate products associated with the first product.

The recommendation product setting unit 140 may set a recommendation product for the first product by comparing the taste vector of the user and the expression vector of the candidate products. That is, among the candidate products extracted by the candidate product extraction unit 130, the recommended product may be set in consideration of the user's taste or style. In detail, the recommendation product setting unit 140 may finally determine a recommendation product to be viewed together with the first product selected by the user through a calculation between a user's taste vector and an expression vector representing characteristics of the candidate product.

Here, the taste vector may be a user's preferred style vector extracted from the topic model or an average value of the feature vectors based on Meta-prod2vec or prod2vec corresponding to the registered product viewed by the user. In addition, the expression vector of the candidate products may be a product style vector of each registered products extracted using a topic model, or may be a feature vector based on Meta-prod2vec. Here, the similarity can be obtained by calculating cosine similarity, and any other thing can be applied as long as the similarity between the vectors can be calculated. According to an embodiment, an additional space for joining two spaces may be used for calculation between a vector using meta-prod2vec and a vector using a topic model.

The recommendation product setting unit 140 may calculate the similarity between the taste vector for the user and the expression vector for each candidate product, and set candidate products having high similarity as the recommendation product. High similarity between the user's taste vector and the expression vector of the candidate product means that the candidate product is similar to the user's preferred style. Therefore, the candidate product setting unit 140 may extract products that are close to the user's preferred style from the plurality of candidate products and set them as recommended products, and display the recommended products on the user terminal 1. In this case, the user may be recommended as a recommendation product a product similar to his or her taste or style, thereby improving the convenience and reliability of the user.

In detail, the recommendation product setting unit 140 may extract a preset number of candidate products according to a high degree of similarity with the taste vector, and set them as recommended products. In some embodiments, the candidate products having a similarity or more than a predetermined value may be selected. It can be set to correspond to the recommended product. In addition, it is also possible to set the priority according to the order of high similarity, and to provide a recommendation product to the user in the order of high priority.

As another embodiment, the recommendation product setting unit 140 may set the recommendation product by further considering the similarity between the first product and the candidate products. That is, after calculating the first similarity between the expression vector of the first product and the expression vector of the candidate product, and the second similarity between the taste vector and the expression vector of the candidate product, respectively, the first similarity and the second similarity are respectively calculated. The weighted value may be added and the recommended products may be set according to the size of the summed result. That is, a predetermined number of candidate products may be set as a recommendation product in order of increasing result value, or all candidate products having a predetermined value or more may be set as recommendation products. Here, in order to obtain an accurate recommendation result, weights for the first similarity and the second similarity may be variously set according to embodiments.

9 is a schematic diagram showing product recommendation according to an embodiment of the present invention. Referring to FIG. 9A, user A and user B may access the server 100 and click to select the same first product. Here, the first product corresponds to the "flower blouse". Here, in the case of providing a recommendation product by considering only the association between the first product and the registered products, the same recommendation products may be provided regardless of the taste or preference style of the user A and the user B.

On the other hand, in the server 100 according to an embodiment of the present invention, the recommendation product may be provided to the user in consideration of the user's preferred style. That is, referring to Figs. 9B and 9D, user A has a preferred style corresponding to keywords such as “open and close”, “white”, and thus, a plurality of candidates corresponding to “flower blouse”. Among the products, a white flower blouse can be recommended. On the other hand, referring to Figs. 9 (c) and 9 (e), since the user B has a preferred style corresponding to keywords such as "lace" and "sexy", the plurality of corresponding "flower blouse" Among the candidates, it is possible to offer an intense red flower blouse as a recommendation. As such, according to an embodiment of the present invention, it is possible to provide a recommendation product to the users in consideration of the preference style of each user.

On the other hand, as shown in Figure 3, the server 100 according to an embodiment of the present invention, may include a physical configuration such as a processor 10, a memory 40, the memory 40, One or more modules may be included that are configured to be executed by the processor 10. Specifically, the one or more modules may include a taste analysis module, a selection input receiving module, a candidate product extraction module, a recommendation product setting module, and the like.

The processor 10 may execute various software programs and an instruction set stored in the memory 40 to perform various functions and to process data. The peripheral interface unit 30 may connect an input / output peripheral device of the computer device to the processor 10 and the memory 40, and the memory controller 20 may include the memory 10 of the processor 10 or a component of the computer device. In the case of accessing, the function of controlling the memory access may be performed. In some embodiments, the processor 10, the memory controller 20, and the peripheral interface unit 30 may be implemented on a single chip or may be implemented as separate chips.

The memory 40 may include fast random access memory, one or more magnetic disk storage devices, nonvolatile memory such as a flash memory device, and the like. In addition, the memory 40 may further include a storage device located away from the processor 10, a network-attached storage device accessed through a communication network such as the Internet, and the like.

As shown in FIG. 3, the server 100 according to an exemplary embodiment of the present invention includes a taste analysis module, a selection input receiving module, a candidate product extracting module, and an application program corresponding to an application program in a memory 40. The recommendation product setting module may be included. Here, each module may be stored in the memory 40 as a set of instructions for performing the above-described function.

Accordingly, the server 100 according to an embodiment of the present invention may allow the processor 10 to access the memory 40 to execute instructions corresponding to each module. However, since the taste analysis module, the selection input receiving module, the candidate product extraction module, and the recommendation product setting module correspond to the taste analysis unit, the selection input receiving unit, the candidate product extraction unit, and the recommendation product setting unit, respectively, detailed description thereof will be omitted here. .

10 is a flowchart illustrating a product recommendation method according to an embodiment of the present invention.

Referring to FIG. 10, a product recommendation method according to an embodiment of the present invention may include a taste analysis step (S10), a product selection step (S20), a candidate product extraction step (S30), and a recommended product setting step (S40). Can be.

Hereinafter, a product recommendation method according to an embodiment of the present invention will be described with reference to FIG. 10.

In the taste analysis step S10, the taste vector of the user may be generated by using history information in which a list of a plurality of registered products that the user has viewed in the past is stored. That is, the user may grasp the user's taste or style using the history information storing the list of a plurality of registered products that the user has viewed in the past and generate a taste vector corresponding thereto.

According to an embodiment, the topic model may be used to generate a preference style vector corresponding to the user's preference style. That is, the product information of the registered products that the user has viewed in the past may be extracted from the history information of the user, and text information included in the product information may be applied to the topic model to extract a corresponding preference style vector. Thereafter, the extracted preferred style vector may be set as a user's taste vector.

Meanwhile, according to another exemplary embodiment, it is also possible to generate a user characteristic vector corresponding to the user's preferred style using the association model. That is, first, each registered product included in the user's history information may be applied to the association model to extract a feature vector corresponding to each registered product. Thereafter, the average of the extracted feature vectors can be obtained and set as a user feature vector, and the user feature vector can be used as a taste vector for the user.

In the product selection step S20, a selection input for the first product may be received from the user from a list of a plurality of registered products provided to the user. The server may provide a list of registered products to the user terminal according to a user's request, and the user may select a first product he or she wishes to view from among a plurality of registered products. In this case, a selection input for the first product selected by the user may be received from the user terminal, and product information including an image or a detailed description of the first product may be displayed in response to the received selection input. Here, the selection input applied by the user terminal may be an input such that the user clicks or touches the registered product, and according to an embodiment, may include an interaction such as “steaming” or “adding a shopping cart”. have.

In the candidate product output step (S30), a plurality of candidate products associated with the first product may be extracted using an association model that models the association between the registered products. In order to provide a recommendation product for the first product selected by the user, it is necessary to extract candidate products associated with the first product. To this end, a candidate product may be determined by extracting feature vectors corresponding to the first product and the registered product, respectively, and calculating similarity between the feature vectors using the association model. Here, the association model may be formed by prod2vec or meta-prod2vec.

In detail, in the candidate product output step S30, first, a feature vector corresponding to the first product and the registered product may be extracted using the association model. In this case, the association model may be a model that collects a list of reading lists for a plurality of registered products viewed by a user in one shopping session, and models the sample. According to an embodiment, the association model may apply the Prod2Vec to the view list to embed the registered products as a vector, and set the vector corresponding to the embedded registered products as the characteristic vector.

According to another embodiment, the association model may further include each metadata set in the registered product in the reading list, and embed the registered product and the metadata as vectors by applying meta-prod2vec to the reading list. In this case, the product eigenvector corresponding to the registered product and the metavector corresponding to the metadata may be combined and set as the characteristic vector. According to an embodiment, it is also possible to generate the feature vector further including a user vector for the user.

Here, the similarity between the characteristic vectors may be calculated by comparing the characteristic vectors of the first product and the registered products, and the candidate product may be selected according to the similarity. That is, the feature vectors of the registered product corresponding to the characteristic vector of the registered product whose similarity is equal to or more than a preset value or the preset number of registered products selected according to the order of high similarity are extracted, and the registered products corresponding to the extracted characteristic vector are extracted. Can be set as a candidate product.

In the recommendation product setting step S40, a recommendation product for the first product may be set by comparing the taste vector of the user and the expression vector of the candidate products. That is, it is possible to finally determine a recommendation product to be viewed together with the first product selected by the user by calculating between the taste vector of the user and the expression vector representing the characteristics of the candidate product.

Here, the taste vector may be a user's preferred style vector extracted from the topic model or an average value of the feature vectors based on Meta-prod2vec or prod2vec corresponding to the registered product viewed by the user. In addition, the expression vector of the candidate products may be a product style vector of each registered products extracted using a topic model, or may be a feature vector based on Meta-prod2vec.

In the recommendation product setting step S40, the similarity between the taste vector for the user and the expression vector for each candidate product may be calculated, and candidate products having high similarity may be set as the recommendation product. Here, the high similarity between the taste vector of the user and the expression vector of the candidate product means that the candidate product is similar to the user's preferred style. Accordingly, products close to the user's preferred style may be extracted from the plurality of candidate products and set as recommended products, and the recommended products may be displayed on the user terminal. In this case, the user may be recommended as a recommendation product a product similar to his or her taste or style, thereby improving the convenience and reliability of the user.

Specifically, in the recommendation product setting step (S40), a predetermined number of candidate products may be extracted and set as recommended products according to the order of high similarity with the taste vector, and in some embodiments, candidate products having a similarity or more than a predetermined value may be selected. It is also possible to set all to correspond to the recommended products. In addition, there may exist embodiments in which priorities are set according to the order of high similarity and the recommendation products are provided to the user in the order of high priority.

In another embodiment, in the recommendation product setting step (S40), the recommendation product may be set by further considering the similarity between the first product and the candidate product. That is, after calculating the first similarity between the expression vector of the first product and the expression vector of the candidate product, and the second similarity between the taste vector and the expression vector of the candidate product, respectively, the first similarity and the second similarity are respectively calculated. It can be added by applying weights. Thereafter, the recommended products may be set according to the sum of the sum of the result values. That is, a predetermined number of candidate products may be set as a recommendation product in order of increasing result value, or all candidate products having a predetermined value or more may be set as recommendation products. Here, in order to obtain an accurate recommendation result, weights for the first similarity and the second similarity may be variously set according to embodiments.

The present invention described above can be embodied as computer readable codes on a medium in which a program is recorded. The computer readable medium may be to continuously store a computer executable program or temporarily store the program for execution or download. In addition, the medium may be a variety of recording means or storage means in the form of a single or several hardware combined, not limited to a medium directly connected to any computer system, it may be distributed on the network. Examples of the medium include magnetic media such as hard disks, floppy disks and magnetic tape, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, And ROM, RAM, flash memory, and the like, configured to store program instructions. In addition, examples of another medium may include a recording medium or a storage medium managed by an app store that distributes an application, a site that supplies or distributes various software, a server, or the like. Accordingly, the above detailed description should not be construed as limiting in all aspects and should be considered as illustrative. The scope of the invention should be determined by reasonable interpretation of the appended claims, and all changes within the equivalent scope of the invention are included in the scope of the invention.

The present invention is not limited by the above-described embodiment and the accompanying drawings. It will be apparent to those skilled in the art that the present invention may be substituted, modified, and changed in accordance with the present invention without departing from the spirit of the present invention.

1: user terminal 100: server
110: taste analysis unit 120: selection input detection unit
130: candidate product extraction unit 140: recommended product setting unit
S10: taste analysis step S20: product selection step
S30: candidate product extraction step S40: recommendation product setting unit

Claims (16)

A product selection step of receiving a selection input for a first product from a plurality of registered products;
A candidate product extraction step of extracting a plurality of candidate products associated with the first product by using an association model modeling the association between the registered products; And
And a recommendation product setting step of setting a recommendation product for the first product by comparing the taste vector of the user and the expression vector of the candidate products.
The method of claim 1, wherein the extracting of the candidate product is performed.
Extracting characteristic vectors corresponding to the first product and the registered product, respectively, using the association model; And
And comparing the characteristic vectors of the first product and the registered products, calculating similarity between the characteristic vectors, and selecting the candidate product according to the similarity.
The method of claim 2, wherein the association model is
A method for recommending a product, characterized in that a user collects a list of reading lists for a plurality of registered products viewed in one shopping session, and models the sample.
The method of claim 3, wherein the association model is
A product recommendation method according to claim 1, wherein the registered products are embedded as vectors by applying Prod2Vec to the reading list, and a vector corresponding to the embedded registered products is set as the characteristic vector.
The method of claim 4, wherein the association model is
The metadata further includes respective metadata set for the registered product in the view list, and applies the meta-prod2vec to the view list to embed the registered product and the metadata as vectors, respectively, and a product specific vector corresponding to the registered product. And a meta vector corresponding to the metadata and setting the characteristic vector as the characteristic vector.
The method of claim 2, wherein selecting the candidate product
Extracts feature vectors of a registered product whose similarity is greater than or equal to a predetermined value or a predetermined number of feature vectors of the registered products selected according to the order of high similarity, and sets registered products corresponding to the extracted feature vector as the candidate products. Product recommendation method characterized in that.
The method of claim 1,
And a taste analysis step of generating a taste vector of the user by using history information storing a list of a plurality of registered products that the user has viewed in the past.
The method of claim 7, wherein the taste analysis step
Collecting product information describing the registered products stored in the history information, generating the user's preferred style vector by applying the product information to a topic model, and converting the preferred style vector into the taste vector. The product recommendation method characterized by the above-mentioned.
The method of claim 7, wherein the taste analysis step
Each registered product included in the history information is applied to the association model to extract a corresponding characteristic vector, a user characteristic vector corresponding to an average of the extracted characteristic vector is calculated, and the user characteristic vector is used as the taste vector. A product recommendation method comprising: setting.
According to claim 1, wherein the recommendation product setting step
And a feature vector extracted by applying the candidate products to the association model is set as an expression vector of the candidate products.
According to claim 1, wherein the recommendation product setting step
And extracting the words included in the product information of the candidate products and applying the collected words to the topic model to set the extracted product style vector as the expression vector of the candidate products.
According to claim 1, wherein the recommendation product setting step
And calculating a similarity between the taste vector and the expression vector, and setting the candidate product having the similarity more than a predetermined value or a predetermined number of candidate products selected according to the order of high similarity as the recommendation product. Recommended way.
According to claim 1, wherein the recommendation product setting step
Calculate a first similarity between the expression vector of the first product and the expression vector of the candidate product, and a second similarity between the taste vector and the expression vector of the candidate product, and calculate the first similarity and the second similarity. The recommendation method of the product, characterized in that for setting the recommended product according to the magnitude of the sum value by applying a weight for the.
A computer program, coupled with hardware, stored on a medium to execute the product recommendation method of any one of claims 1 to 13.
A selection input detecting unit configured to receive a selection input for a first product among a plurality of registered products from a user;
A candidate product extraction unit configured to extract a plurality of candidate products associated with the first product by using an association model modeling the association between the registered products; And
And a recommendation product setting unit configured to set a recommendation product for the first product by comparing the taste vector of the user with the expression vector of the candidate products.
A processor; And
Including a memory coupled to the processor,
The memory includes one or more modules configured to be executed by the processor,
The one or more modules,
Receive a selection input for the first product from the plurality of registered products from the user,
Extracting a plurality of candidate products associated with the first product by using an association model modeling the association between the registered products;
Comparing the taste vector of the user with the expression vector of the candidate products and setting a recommendation product for the first product;
Server containing instructions.

Images