KR102619084B1 - Method for recommending products and service server using the same - Google Patents

Abstract

This application relates to a product recommendation method and a service server using the same. The product recommendation method according to an embodiment of the present invention includes search log data storing the search history performed by users to search for products, and the attributes of each product. Collecting product information including; Creating a commerce knowledge graph structure using each node created based on the search log data and product information and edges between the nodes; calculating the correlation between the nodes using a weight set in a connection path connecting each node included in the commerce knowledge graph structure; and providing recommended products corresponding to the search query entered by the user using the correlation.

Description

Product recommendation method and service server using the same {Method for recommending products and service server using the same}

This application relates to a product recommendation method that can provide a product recommendation service to users when shopping online and a service server that uses the same.

An online shopping mall or internet shopping mall refers to a virtual store that allows products to be bought and sold using the Internet. Unlike offline shopping malls, online shopping malls have no time or space restrictions, so consumers can purchase products at relatively low prices anytime, anywhere. From the online shopping mall operator's perspective, they can provide product information tailored to consumers' tastes without restrictions on commercial areas. Operators of online shopping malls can sell products directly to consumers or connect sellers and consumers to relay sales. Online shopping has been rapidly growing in market size in recent years due to the advancement of information and communication technology and the convenience of purchasing products.

Meanwhile, online shopping malls recommend products in various ways to encourage customers' purchasing activities. Specifically, two major analysis methods are used to recommend products: one that analyzes the online activities of individual customers and the other that analyzes the online activities of other customers. An example of the first method is to refer to the customer's individual purchase history and recommend products related to products purchased in the past to encourage purchase activity. Examples of the second method include recommending products with high sales volume, recommending products by analyzing comments about products, or recommending products by referring to the purchase history of other customers.

This application seeks to provide a product recommendation method that can perform product recommendation using a commerce knowledge graph structure and a service server that uses the same.

This application seeks to provide a product recommendation method and a service server using the method that can flexibly respond to the purpose of each product recommendation service and provide an effective product recommendation service.

A product recommendation method of a service server according to an embodiment of the present invention includes the steps of collecting search log data storing search history performed by users to search for products, and product information including the attributes of each product; Creating a commerce knowledge graph structure using each node set based on the search log data and product information and edges between the nodes; calculating the correlation between the nodes using a weight set in a connection path connecting each node included in the commerce knowledge graph structure; and providing recommended products corresponding to the search query entered by the user using the correlation.

A product recommendation method of a service server providing an integrated search service according to an embodiment of the present invention includes the steps of receiving a user's search query from an integrated search web page; extracting recommended products corresponding to the search query using a commerce knowledge graph structure; and displaying the recommended products in a setting area of the integrated search web page.

The service server according to an embodiment of the present invention includes a collection unit that collects search log data that stores search history performed by users to search for products, and product information including the attributes of each product; a graph structure generator that generates a commerce knowledge graph structure using each node set based on the search log data and product information and edges between the nodes; a calculation unit that calculates the correlation between the nodes using a weight set in a connection path connecting each node included in the commerce knowledge graph structure; and an output unit that provides recommended products corresponding to the search query entered by the user using the correlation.

Additionally, the means for solving the above problems do not enumerate all the features of the present invention. The various features of the present invention and the resulting advantages and effects can be understood in more detail by referring to the specific embodiments below.

Since the product recommendation method according to an embodiment of the present invention and the service server using the same use a commerce knowledge-based graph structure, it is possible to provide an effective product recommendation service by flexibly responding to the purpose of each product recommendation service.

The product recommendation method according to an embodiment of the present invention and the service server using the same continuously update the commerce knowledge-based graph structure, enabling product recommendation that reflects the latest trends.

The product recommendation method according to an embodiment of the present invention and the service server using the same extract correlations by considering multidimensional relationships between nodes in the commerce knowledge-based graph structure, thereby improving the quality of product recommendation.

However, the effects that can be achieved by the product recommendation method and the service 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 included in the product recommendation method according to the embodiments of the present invention. It will be clearly understandable to those with ordinary knowledge in the technical field.

Figure 1 is a schematic diagram showing a product recommendation system according to an embodiment of the present invention.
Figure 2 is a block diagram showing a service server according to an embodiment of the present invention.
Figure 3 is an exemplary diagram showing a node according to an embodiment of the present invention.
Figures 4 and 5 are exemplary diagrams showing edges between nodes according to an embodiment of the present invention.
Figure 6 is an exemplary diagram showing a knowledge-based graph structure according to an embodiment of the present invention.
Figures 7 to 9 are exemplary diagrams showing a product recommendation service using a knowledge-based graph structure according to an embodiment of the present invention.
Figure 10 is a flowchart showing a product recommendation method according to an embodiment of the present invention.
Figure 11 is a flowchart showing a product recommendation method according to another embodiment of the present invention.

Hereinafter, embodiments disclosed in the present specification will be described in detail with reference to the attached drawings. However, identical or similar components will be assigned the same reference numbers regardless of reference numerals, and duplicate descriptions thereof will be omitted. The suffixes “module” and “part” for components used in the following description are given or used interchangeably only for the ease of preparing the specification, and do not have distinct meanings or roles in themselves. That is, the term 'unit' used in the present invention refers to a hardware component such as software, FPGA, or ASIC, and the 'unit' performs certain roles. However, 'wealth' is not limited to software or hardware. The 'part' may be configured to reside on an addressable storage medium and may be configured to run on one or more processors. Therefore, as an example, 'part' refers to components such as software components, object-oriented software components, class components and task components, processes, functions, properties, procedures, Includes 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 may be further separated into additional components and 'parts'.

Additionally, in describing the embodiments disclosed in this specification, if it is determined that detailed descriptions of related known technologies may obscure the gist of the embodiments disclosed in this specification, the detailed descriptions will be omitted. In addition, the attached drawings are only for easy understanding of the embodiments disclosed in this specification, and the technical idea disclosed in this specification is not limited by the attached drawings, and all changes included in the spirit and technical scope of the present invention are not limited. , should be understood to include equivalents or substitutes.

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

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

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

The user terminal 1 is a terminal device used by a user, and the user terminal 1 can communicate with the service server 100 and the like using a wired or wireless communication network. The user terminal 1 may be equipped with a browser for transmitting and receiving information, a memory for storing programs and protocols, and a microprocessor for calculating and controlling various programs.

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 laptop, a personal digital assistant (PDA), a portable multimedia player (PMP), an MP3 player, or a wearable device (e.g. For example, mobile terminals that apply wireless communication technology, such as watch-type terminals (smartwatch), smart glass, and HMD (head mounted display), or digital devices that are difficult to carry, such as PCs, IPTVs, and smart TVs. It can be.

The user terminal 1 and the service server 100 may be connected to each other through a communication network. Communication networks may include wired networks and wireless networks, and specifically, various networks such as local area networks (LANs), metropolitan area networks (MANs), and wide area networks (WANs). It can be included. Additionally, the communication network may include the 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 the user based on information provided from the service server 100. As an example, the user can use the user terminal 1 to access a shopping website provided by the service server 100 or run a shopping-specific application, through which online shopping services such as purchasing products can be provided to the user. there is. In this case, the user terminal 1 can display a list of a plurality of products on the display unit, and when the user authorizes a selection input for a specific product, a product information page containing an image or detailed description of the selected product etc. can be displayed. Depending on the embodiment, it is also possible to provide a product recommendation service to the user by having the service server 100 extract recommended products related to the selected product and then display the recommended products on the display unit of the user terminal 1. . Here, products may include services in addition to objects in physical form, and programs, files, and services implemented in digital form may also be included in the product.

Additionally, the service server 100 may operate an integrated search site or an application dedicated to integrated search, and the user terminal 1 may provide an integrated search service through communication with the service server 100. Here, the integrated search service is a service that provides search results in various categories such as news, blog or cafe posts, language dictionaries, and websites in response to the search query entered by the user. Depending on the embodiment, search in the shopping category. It is also possible to provide results. In this case, the user can be provided with online shopping services and product recommendation services based on the search results of the shopping category.

The service server 100 is a service providing server and can provide a product recommendation service, an online shopping service including a product recommendation service, an integrated search service, etc. to the user terminal 1. In order to sell products through online shopping, a plurality of products may be registered in the service server 100, and product information corresponding to each registered product may be stored in the service server 100. Stored product information may include various information such as product image, product name, price, manufacturer, seller, and detailed description of the product.

Meanwhile, the service server 100 may utilize a commerce knowledge graph structure to provide various product recommendation services. In other words, a graph structure can be built by expressing the search activities performed by the user to search for products and the product information of the products as nodes and edges connecting the nodes, and this can be used to construct heterogeneous nodes. You can calculate the correlation between them. Here, the service server 100 can extract recommended products related to a specific product based on the correlation between each node, so it is possible to provide a product recommendation service to the user terminals 1. Hereinafter, a service server according to an embodiment of the present invention will be described in detail.

Figure 2 is a block diagram showing the service server 100 according to an embodiment of the present invention.

Referring to Figure 2, the service server 100 according to an embodiment of the present invention may include a collection unit 110, a graph structure generation unit 120, an operation unit 130, and an output unit 140. .

The collection unit 110 may collect search log data that stores search history performed by users to search for products, and product information including the attributes of each product.

That is, a plurality of user terminals 1 connected to the service server 100 can perform various search activities for product searches, etc., and the resulting search history, etc. can be stored as search log data. Accordingly, the collection unit 110 can collect and store search log data of each user terminal 1. Here, the search log data includes a search log containing the history of search queries entered by the user, a product selection log of products selected by the user through a click or touch from the list of products appearing from the search results, and a log of products entered by the user within one session. It may include a sequence log, which is a series of search queries or a list of selected products.

In the case of product information, it may be input by a seller, etc. when registering a product on the service server 100, and the collection unit 110 can extract product information from the seller's input. Here, product information may include product image, product name, price, manufacturer, seller, detailed description of the product, etc.

Depending on the embodiment, the collection unit 110 may receive search log data and product information in an unstructured data format, then perform data purification and store the data. Data purification is refining each search log data and product information to facilitate creating a commerce knowledge graph structure. Each node and the edges between nodes can be set based on the search log data and product information. .

Referring to Figure 3, each search query, product, product attribute, category, seller, and brand included in the search log data and product information. ), manufacturer, series, etc. can be created as nodes.

An edge connects nodes, and can be set differently depending on the relationship between nodes. In other words, using edges, it is possible to display one of the predefined relationships between each node.

Specifically, referring to Figure 4(a), the edge "adj-view" can define relationships between search queries used within the same session. Here, it is also possible to display the order of each search query used within the same session using the order of each node connected through "adj-view". For example, if the user enters “sofa cushion” as a search query and then enters “cushion” again, “adj-view” can be displayed by connecting “sofa cushion” and “cushion” in that order.

Additionally, a weight may be set for each edge, and the weight may be set differently depending on the type of edge. In the case of "adj-view", the weight can be set based on the number of times the search query has been used, and in order to maintain the up-to-dateness of the commerce knowledge graph structure, the weight can be set to reflect more recently used search queries. . For example, in the same session, the search queries "one piece" and "long dress" are entered sequentially as search queries: 20 times on March 15th, 15 times on March 16th, and 15 times on March 17th. It can happen 10 times. In this case, the weight for “adj-view” between the search queries “one piece” and “long dress” is set to (20)*(0.98)*(0.98)+(15)*(0.98)+(10). You can. In other words, based on March 17th, the weight can be set so that the closer it is to March 17th, the more the number of uses on that date is reflected.

Figure 4(b) shows an edge “co-view”, where “co-view” can define a relationship between search queries used to search for the same product in different sessions. For example, when users click on a specific cushion A, they can check the search query used to search for that cushion A. At this time, users entered "triangle cushion", "triangle backrest cushion", and "cushion" respectively. You can. In other words, “triangle cushion”, “triangle backrest cushion”, and “cushion” each correspond to different search queries, but the user selected the same cushion A from the search results according to the search query, so “triangle cushion”, “triangle backrest” “Cushion” and “cushion” can be seen as having a similar relationship with each other. Therefore, “triangular cushion”, “triangular backrest cushion”, and “cushion” can be displayed by connecting them to “co-view”, respectively.

Here, the weight for “co-view” can be set based on the number of times each search query is used to select a product. For example, if product B was clicked 10 times per day with the search query "one piece" and 15 times with the search query "long dress", the middle value is (10*15)/(total number of times product B was clicked per day). You can calculate with the result value. Afterwards, in addition to product B, intermediate result values are calculated for all other products that were searched for as "one piece" and "long dress" and clicked on in the same way, and the intermediate result values are added together to determine "one piece" and "long dress". You can obtain daily weights for “co-view”. Afterwards, the weights can be set by summing the daily weights. At this time, in order to maintain the up-to-dateness of the commerce knowledge graph structure, the weights can be set to reflect more recent daily weights.

Figure 4(c) shows the edge "specialize", and "specialize" may represent the relationship between search queries constituting the upper-lower concept. For example, “padding” and “long padding” can each be used as a search query, and “long padding” can be viewed as a type of “padding.” In other words, "padding" is a higher concept of "long padding", and "long padding" is a lower concept of "padding", so "padding" and "long padding" are connected to each other as "specialize" to form a higher-lower concept. It can be indicated that it corresponds to . Here, the distinction between high-level and low-level concepts can be learned using the transE model, etc., and depending on the embodiment, the L1 distance between each node learned from the transE model can be set as a weight.

Figure 5(a) shows an edge "click", and "click" may represent the relationship between a search query and product information of a product selected by a user's click or touch, etc. among the search results based on the search query. For example, a user can enter “padding” as a search query and select product A from the list of products that appear as search results. In this case, “padding” and product A can be connected with “click”, and additionally, the product category, seller, manufacturer, etc., which are product information of product A, can be connected and displayed with “click”. At this time, the weight of “click” can be set based on the number of times the user selects the product by clicking or touching. Here, in order to reflect recency, the number of past selections may be reflected less than the recent selection number. For example, if specific product A is selected 20 times on March 15th, 15 times on March 16th, and 10 times on March 17th, about the “click” between the specific product A and each product information. (20)*(0.98)*(0.98)+(15)*(0.98)+(10) can be set as the weight.

Figure 5(b) shows the edge "query-is", where "query-is" may represent the relationship between a search query and product information included in the search query. For example, if the search query is “Company A one-piece,” “Company A” corresponds to the manufacturer in the product information, and “Dress” corresponds to the product category. Therefore, “query-is” can be used to indicate that the search query includes the manufacturer and category of the product. Here, the weight of “query-is” can be set to 1 for applicable items among product information included in the search query, and 0 for non-applicable items.

However, the edges are not limited to the five types shown in Figures 4 and 5, and various additions and changes are possible depending on the embodiment. Additionally, the weight for each edge can be set in various ways depending on the embodiment.

The graph structure generator 120 may generate a commerce knowledge graph structure using each node and the edges between nodes. For example, each node is displayed from each search query and product information collected by the collection unit 120, and “co-view”, “adj-view”, “click”, and “query” are used between these nodes. You can create a commerce knowledge graph structure by connecting to edges such as "-is". Here, a weight may be assigned between each edge, and the weight may reflect frequency attributes such as the number of each search query used by users or the number of clicks on a specific product. Figure 6 is an exemplary diagram showing the structure of a commerce knowledge graph according to an embodiment of the present invention.

The graph structure generator 120 may update nodes, edges, weights, etc. of the commerce knowledge graph structure when a setting event occurs. The setting event may be the arrival of a certain time interval (for example, every 24 hours) or an update request from the service server 100 operator. In addition, when various setting events occur, the commerce knowledge graph structure can be updated. In other words, since shopping trends and user shopping behavior can change from time to time, the commerce knowledge graph can be continuously updated to reflect this.

The calculation unit 130 can calculate the correlation between each node included in the commerce knowledge graph structure. Specifically, the calculation unit 130 can extract a weight set on an edge included in a connection path connecting each node, and use the weight to calculate the correlation between each node. Depending on the embodiment, it is also possible to calculate the correlation using a correlation extraction algorithm. Here, the correlation extraction algorithm can quantitatively express the correlation between nodes by calculating an edge score that reflects the weight between each node.

Specifically, the calculation unit 130 may utilize inverse log-weighted similarity, personalized page-rank, personalized katz centrality, Dijsktra similarity, etc. as a correlation extraction algorithm.

Inverse log-weighted similarity may extract common neighboring nodes between target nodes for which correlation is to be calculated and measure the correlation between target nodes depending on how many common neighboring nodes there are. Here, weight can be assigned to the case where the common neighbor node is a low-degree neighbor with few connection paths to other nodes, rather than to the case where the common neighbor node is a high-degree neighbor with many connection paths to other nodes. In other words, in the case of a high-degree neighbor, there is a high possibility of being accidentally connected to the target nodes, so the closer the neighboring nodes are to low-degree neighbors, the higher the correlation can be determined.

Personalized page-rank corresponds to a method of measuring the correlation between nodes using an edge score extracted by considering all connection paths between nodes and the weight of each connection path.

Personalized katz centrality is a method of measuring the correlation between nodes using edge scores extracted by considering all connection paths between nodes and the weight of each connection path. As the distance between nodes on the connection path to be obtained increases, a penalty is applied. It corresponds to the method of granting.

Dijsktra similarity is a method of measuring the correlation between the nodes by calculating the highest probability that two given nodes will occur together using Dijkstra's algorithm. In other words, the weight between nodes can be set to the probability of two nodes occurring together, and the edge score can be calculated by multiplying the weights existing on the connection path between nodes.

Meanwhile, the calculation unit 130 may include a plurality of calculation modules, as shown in FIG. 2, and it is also possible to create different associations between nodes using each calculation module. In other words, there may be a plurality of product recommendation services that the service server 100 wishes to provide, and each product recommendation service may extract recommended products in a different way. Accordingly, each calculation module may be designed to create associations between each node differently depending on the purpose of the individual product recommendation service. To this end, each calculation module can apply different weights and correlation extraction algorithms to the commerce knowledge graph structure.

For example, when using personalized page-rank and inverse log-weighted similarity to extract search queries that are close to the "one piece" search query included in the commerce knowledge graph structure, the results can appear as shown in Table 1 below.

ranking personalized page-rank Inverse log-weighted similarity One winter dress Winter girls dress 2 Girl's dress princess dress 3 one piece Girls coat 4 Wedding guest dress princess clothes 5 Knit dress Girl's sweatshirt 6 Winter girls dress Girl's Asha dress 7 Long dress children's dress

Here, the results of inverse log-weighted similarity can be interpreted as giving users more surprises than the results of personalized page-rank. In other words, inverse log-weighted similarity measures similarity based on common neighboring nodes and assumes that overlapping low-degree neighbors are more similar than high-degree neighbors. In this way, even in the case of the same commerce knowledge graph structure, different results are shown depending on each correlation extraction algorithm, so an effective product recommendation service can be provided by selectively applying the correlation extraction algorithm depending on the type of product recommendation service.

The output unit 140 can use the correlation between each node to provide recommended products corresponding to the search query entered by the user. For example, within the commerce knowledge graph structure, the nodes of products most closely related to the node corresponding to the search query entered by the user can be extracted and provided as recommended products. In addition, depending on the embodiment, recommended products are provided by extracting keywords or tags most relevant to the search query within the commerce knowledge graph structure and then extracting products corresponding to the keywords or tags. It is also possible.

On the other hand, when a plurality of calculation modules are included in the calculation unit 130, the correlation calculated by each calculation module may be different, and the output unit 140 can extract different recommended products accordingly. Therefore, , the output unit 140 is capable of providing a plurality of different product recommendation services by utilizing each calculation module. For example, the product recommendation service provided by the output unit 140 may include a responsive recommendation service, a trend recommendation service, and a gift recommendation service.

Responsive recommendation service provides search results according to the search query entered by the user, and then, when the user selects a specific product and returns to the search results, product recommendation displays recommended products in a setting area such as the bottom of the search result. Applies to service. Specifically, referring to Figure 7(a), the user can enter "padding" as a search query in the integrated search page, and in this case, the service server 100 provides a list of products corresponding to "padding" on the search results page. It can be provided within the shopping category. Afterwards, the user can select and confirm a specific product from the displayed product list and then return to the search results page. At this time, the service server 100 may display and provide recommended products related to the selected specific product, as shown in FIG. 7(b).

The trend recommendation service provides search results according to the search query entered by the user, compares the search query with preset trend tags, and extracts trend tags determined to be related to the search query. Afterwards, products containing the corresponding trend tag can be extracted from the product information and provided as recommended products. That is, referring to Figure 8(a), the user can enter "DYYYYXX padding" as a search query in the integrated search page, and in this case, the service server 100 provides a search query corresponding to "DYYYYXX padding" in the shopping category. A product list can be provided within the search results page. At this time, the service server 100 can distinguish that “DYYYYXX” corresponds to the manufacturer and “padding” corresponds to the category, and compares them with the trend tags to determine trend tags that are highly correlated with “DYYYYXX” and “padding”. It can be extracted. Thereafter, as shown in Figure 8(b), products including the corresponding trend tag within a setting area such as the bottom of the search results page can be displayed as recommended products and provided to the user.

The gift recommendation service is a service that recommends products that users frequently purchase as gifts on holidays such as Chuseok or anniversaries such as Valentine's Day and Parents' Day. Specifically, as shown in Figure 9(a), the user may enter a search query such as "Chuseok gift", and in this case, the service server 100 creates a trend tag corresponding to a sub-relationship with "Chuseok gift". can be extracted. Afterwards, as shown in Figure 9(b), products including trend tags corresponding to subordinate relationships can be provided as recommended products.

In addition, the output unit 140 can provide various types of product recommendation services, etc., and operation modules for providing each product recommendation service can be designated differently. In other words, each calculation module can be designed and applied to correspond to each product recommendation service.

Figure 10 is a flowchart showing a product recommendation method according to an embodiment of the present invention. Here, each step may be performed by a service server according to an embodiment of the present invention.

Referring to Figure 10, the service server can collect search log data that stores search history performed by users to search for products, and product information including the attributes of each product (S110). Here, the search log data includes a search log containing the history of search queries entered by the user, a product selection log of products selected by the user through a click or touch from the list of products appearing from the search results, and a log of products entered by the user within one session. It may include a sequence log, which is a series of search queries or a list of selected products.

In the case of product information, it may be entered by a seller, etc. when registering a product on a service server, and the service server may extract product information from the seller's input. Here, product information may include product image, product name, price, manufacturer, seller, detailed description of the product, etc.

Depending on the embodiment, it is possible for the service server to receive search log data and product information in an unstructured data format, then clean the data and store it. Data purification is refining each search log data and product information to facilitate creating a commerce knowledge graph structure. Each node and the edges between nodes can be set based on the search log data and product information.

Here, the service server includes each search query, product, product attribute, category, seller, and brand included in the search log data and product information. , manufacturer, series, etc. can be created as nodes.

An edge connects nodes, and can be set differently depending on the relationship between nodes. In other words, using edges, it is possible to display one of the predefined relationships between each node. Specifically, the edge includes "adj-view", which represents the relationship between search queries used within the same session, "co-view", which represents the relationship between search queries used to search for the same product in different sessions, “specialized,” which represents the relationship between search queries that constitute a parent-child concept; “click,” which represents the relationship between a search query and the product information of a product selected by a user’s click or touch among the search results based on the search query; , “query-is”, etc., which indicates the relationship between a search query and product information included in the search query. However, the edge is not limited to this, and various additions and changes are possible depending on the embodiment.

The service server can create a commerce knowledge graph structure using each node and the edges between nodes set based on search log data and product information (S120). Here, the commerce knowledge graph structure can update nodes, edges, and edge weights when a setting event occurs. The configuration event may be the arrival of a certain time interval or an update request from the service server operator. Since shopping trends and user shopping behavior can change from moment to moment, the commerce knowledge graph can be continuously updated to reflect this.

The service server can calculate the correlation between nodes using the weight set in the connection path connecting each node included in the commerce knowledge graph structure (S130). Specifically, the service server can extract the weights of edges included in the connection path, and use the weights to calculate the correlation between each node. Depending on the embodiment, a correlation extraction algorithm may be applied to calculate the correlation between nodes. Here, the correlation extraction algorithm can quantitatively express the correlation between nodes by calculating an edge score that reflects the weight of each edge.

Specifically, the service server can utilize inverse log-weighted similarity, personalized page-rank, personalized katz centrality, and Dijsktra similarity as correlation extraction algorithms.

Meanwhile, the service server may include a plurality of computation modules, and each computation module can be used to create different associations between nodes. In other words, there may be a plurality of product recommendation services provided by the service server, and each product recommendation service may extract recommended products in a different way. Accordingly, each calculation module may be designed to create associations between each node differently depending on the purpose of the individual product recommendation service. To this end, each calculation module can apply different weights and correlation extraction algorithms to the commerce knowledge graph structure.

The service server can use the correlation to provide recommended products corresponding to the search query entered by the user (S140). Here, when multiple calculation modules are included in the service server, different recommended products can be extracted according to the correlation calculated by each calculation module, and by utilizing this, it is possible to provide multiple different product recommendation services. do. For example, product recommendation services provided by the service server may include responsive recommendation services, trend recommendation services, and gift recommendation services. However, it is not limited to this and the service server can provide various types of product recommendation services.

Figure 11 is a flowchart showing a product recommendation method according to another embodiment of the present invention. Here, each step may be performed by a service server according to an embodiment of the present invention, and the service server may provide an integrated search service.

Referring to Figure 11, the service server can receive the user's search query from the integrated search web page (S210). That is, the service server can provide an integrated search web page to the user terminal and receive a search query input by the user from the user terminal. The service server may extract search results corresponding to the user's search query from a plurality of preset categories and provide them to the user terminal. Here, the service server can also provide search results for the shopping category. In the case of the shopping category, search results for products corresponding to the search query entered by the user can be displayed and provided as a product list.

The service server can extract recommended products corresponding to the search query using the commerce knowledge graph structure (S220). The service server can provide a plurality of product recommendation services in response to the user's search query. In this case, recommended products according to each product recommendation service can be extracted using the commerce knowledge graph structure. Here, each product recommendation service may extract each recommended product by applying different weights and correlation extraction algorithms to the commerce knowledge graph structure. For example, product recommendation services provided by the service server may include responsive recommendation services, trend recommendation services, gift recommendation services, etc.

The service server may display recommended products within the settings area of the integrated search web page (S230). The service server may display recommended products differently for each product recommendation service. For example, in the case of a responsive recommendation service, after providing the user with search results corresponding to the search query, when the user selects and confirms a specific product and returns to the search results, it is recommended within the setting area such as the bottom of the search results. Products can be displayed. On the other hand, in the case of trend recommendation services or gift recommendation services, search results corresponding to the user's search query can be provided. Furthermore, in the case of a gift recommendation service, it is also possible to display recommended products in the settings area located at the top of the search results when providing search results.

The above-described present invention can be implemented as computer-readable code on a program-recorded medium. A computer-readable medium may continuously store a computer-executable program or temporarily store it for execution or download. In addition, the medium may be a variety of recording or storage means in the form of a single or several pieces of hardware combined. It is not limited to a medium directly connected to a computer system and may be distributed over a network. Examples of media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, And there may be something configured to store program instructions, including ROM, RAM, flash memory, etc. Additionally, examples of other media include recording or storage media managed by app stores that distribute applications, sites or servers that supply or distribute various other software, etc. Accordingly, the above detailed description should not be construed as restrictive in all respects and should be considered illustrative. The scope of the present invention should be determined by reasonable interpretation of the appended claims, and all changes within the equivalent scope of the present invention are included in the scope of the present invention.

The present invention is not limited to the above-described embodiments and attached drawings. For those skilled in the art to which the present invention pertains, it will be clear that components according to the present invention can be replaced, modified, and changed without departing from the technical spirit of the present invention.

1: User terminal 100: Service server
110: Collection unit 120: Graph structure generation unit
130: calculation unit 140: output unit

Claims (12)

In the product recommendation method of the service server,
Collecting search log data storing search history performed by users for product search and product information including the attributes of each product;
Creating a commerce knowledge graph structure using each node set based on the search log data and product information and edges between the nodes;
calculating the correlation between the nodes using a weight set in a connection path connecting each node included in the commerce knowledge graph structure; and
It includes providing recommended products corresponding to the search query entered by the user using the correlation,
The step of creating the commerce knowledge graph structure is
A product recommendation method of a service server, characterized in that at least one of search queries included in the search log data and at least one of a category, seller, and manufacturer of a product included in the product information are generated as the node. .
The method of claim 1, wherein the step of calculating the correlation is
A product recommendation method of a service server, characterized in that the correlation between the nodes is calculated by applying a weight set to an edge in the connection path to a preset correlation extraction algorithm.
The method of claim 2, wherein the step of calculating the correlation is
A product recommendation method of a service server, characterized in that the correlation is generated by using at least one of inverse log-weighted similarity, personalized page-rank, personalized katz centrality, and Dijsktra similarity as the correlation extraction algorithm.
delete The method of claim 1, wherein the edge is
It is set according to predefined relationships between the nodes,
The predefined relationship is
Relationship between search queries used within the same session, relationship between search queries used to search the same product in different sessions, relationship between search queries forming a parent-child concept, search query and corresponding search A product recommendation method of a service server, comprising at least one of a relationship between product information of a product selected among search results by a query, and a relationship between a search query and product information included in the search query.
The method of claim 1, wherein the step of calculating the correlation is
A product recommendation method of a service server, characterized in that each correlation between the nodes is generated differently using a plurality of calculation modules.
The method of claim 6, wherein the step of providing the recommended products is
A product recommendation method of a service server, characterized in that it provides a plurality of product recommendation services using each recommended product extracted according to the correlation calculated by the plurality of calculation modules.
The method of claim 7, wherein the calculation module is
A product recommendation method of a service server, characterized in that the weight and correlation extraction algorithm are individually set according to each product recommendation service.
The method of claim 1, wherein the commerce knowledge graph structure is
A product recommendation method of a service server, characterized in that the node, edge, and weight are updated when a setting event occurs.
In the product recommendation method of a service server providing an integrated search service,
Receiving a user's search query from an integrated search web page;
extracting recommended products corresponding to the search query using a commerce knowledge graph structure; and
Comprising the step of displaying the recommended products in a setting area of the integrated search web page,
The commerce knowledge graph structure is
A service that provides an integrated search service, characterized in that each node includes at least one of the search queries included in the search log data, and at least one of the product category, seller, and manufacturer included in the product information. Server product recommendation method.
A computer program combined with hardware and stored in a medium to perform the product recommendation method of any one of claims 1 to 3 and 5 to 10.
a collection unit that collects search log data that stores search history performed by users to search for products, and product information including the attributes of each product;
a graph structure generator that generates a commerce knowledge graph structure using each node set based on the search log data and product information and edges between the nodes;
a calculation unit that calculates the correlation between the nodes using a weight set in a connection path connecting each node included in the commerce knowledge graph structure; and
An output unit that provides recommended products corresponding to the search query entered by the user using the correlation,
The graph structure creation unit
A service server that generates at least one of search queries included in the search log data and at least one of a category, seller, and manufacturer of a product included in the product information as the node.

Images