KR102246538B1 - Item recommendation method and apparatus using graph database - Google Patents

Abstract

One aspect of the present invention discloses a product recommendation method in an electronic commerce system. The method includes collecting order history data of a plurality of customers from a plurality of shopping malls based on e-commerce (the order history data includes data on at least one product ordered by a customer), the collected order Converting the historical data into graph-type data composed of at least three node layers (the at least three node layers are a first node layer associated with a product, a second node layer associated with an order, and a customer). Including an associated third node layer), storing the graph type data in a database, and on the stored graph type data, at least one node of the first node layer in response to a query for a similar product, and And determining and returning a similarity based on the number of connection paths between at least two nodes among the at least one third node layer.

Description

Product recommendation method and device using graph database {ITEM RECOMMENDATION METHOD AND APPARATUS USING GRAPH DATABASE}

The present invention relates to a product recommendation method, and more particularly, to a product recommendation method for efficient personalized recommendation in an online shopping mall service based on an e-commerce transaction.

With the recent spread of the Internet, purchases through online shopping malls are becoming more active, and a variety of personalized product recommendation methods are being studied in order to support customers' purchases efficiently. Using this personalized product recommendation method, shopping mall operators try to increase the purchase probability of buyers and increase the sales of their stores.

In these studies, the conventional personalized product recommendation method used a method of extracting a product with a high probability of purchase by analyzing customer preferences in advance in an online shopping mall. Through this, a method to inspire buyers to purchase in the form of changing the arrangement of products with high purchase possibility or displaying them on a part of the screen has been used. In the conventional method, the purchase possibility is calculated in consideration of factors such as a user's previous purchase history, customer attributes, and product similarities.

However, in the case of clothing, since multiple sets of similar products are not purchased, there is a high possibility of purchasing products in different categories or alternative categories rather than purchasing similar products at the same time. For example, rather than buying two sets of very similar tops at once, it is more likely to buy a combination of tops and bottoms.

That is, the conventional personalized product recommendation method has a large macroscopic correlation as described above, but does not reflect the phenomenon of excluding each other microscopically, and accordingly, there is a problem in that it does not meet the expectations of buyers who receive product recommendation.

An object according to an aspect of the present invention for solving the above-described problem is a product in an e-commerce system that performs product recommendation by converting historical data into a graph form and extracting an indirect connection structure between products through analysis using graphs. It is to provide a recommended method and device.

A product recommendation method in an e-commerce system according to an aspect of the present invention for achieving the above object comprises the steps of collecting order history data of a plurality of customers from a plurality of shopping malls based on e-commerce (the order history data is Including data on at least one product ordered by a customer), converting the collected order history data into graph-type data composed of at least three node layers (the at least three nodes The layer includes a first node layer associated with a product, a second node layer associated with an order, and a third node layer associated with a customer), storing the graph type data in a database, and on the stored graph type data , In response to a query for similar products, determining and returning the similarity based on the number of connection paths between at least two nodes among at least one node of the first node layer and at least one node of the third node layer. It may include.

In the graph-type data, at least one second node of the second node layer is connected to the first node of the first node layer, and a third node of the third node layer is connected to the at least one second node. As connected, it can be generated in the form of a radial graph.

Extracted by extracting connection paths within a first distance range among at least one connection paths to which at least two nodes can be connected among nodes of the at least one first node layer and nodes of at least one third node layer The higher the number of connection paths, the higher the degree of similarity can be determined.

A distance between two directly connected nodes is defined as a distance of one hop, and the first distance range may have a range exceeding a distance of two hops indicating direct correlation.

The first distance range may have a range excluding a connection path having a distance greater than or equal to the second distance.

At least two of the at least one first node layer nodes and at least one third node layer nodes include (i) a first product node and a second product node of the first node layer, (ii) It may include one of a first customer node and a second customer node of the third node layer, and (iii) a first product node of the first node layer and a first customer node of the third node layer.

In response to a query for a second product node having a similarity to the first product node of (i), the first product node-the first product node is included in the first order node of the second node hierarchy-is associated with A connection path between the first order node and a product included in another second order node among at least one node of the second node hierarchy of the third customer node of the third node hierarchy may be considered.

In response to a query for a second customer node similar to the first customer node in (ii), the second customer is determined to be a customer associated with a third product node of the first node hierarchy associated with the first customer node. However, the third order node of the second customer-the third order node is a node of the second node hierarchy, and is an order node including the third product node-in which a product node other than the third product is selected. I can.

In response to a query for a first customer node similar to the first product node or a first product node similar to the first customer node in (iii), a second node hierarchy including the first product node The connection path with the customer associated with the third product node that is bundled within the third order node of and the connection path with at least one of the bundled products within the fourth order node of the second node hierarchy associated with the first customer node Can be considered.

The product of the first node layer is categorized and defined together with a product category identifier, and in response to a query for a similar product for the first product of the first node layer, a second product is directly or indirectly connected to the first product. A product of a second product category having a similar correspondence to the first product category identifier of the first product among products associated with at least one of the second node of the node hierarchy and the third node of the third node hierarchy. Bay can be considered as a connection path.

The product of the first node is categorized and defined together with a product category identifier, and a second node layer directly or indirectly connected to the first product in response to a query for a similar product for the first product of the first node Based on the first product category identifier of the first product among products associated with at least one of the second node of the second node and the third node of the third node hierarchy, the product of the second product category having a similar correspondence Connection paths can be excluded.

The product of the first node is categorized and defined together with a product category identifier, and a second node layer directly or indirectly connected to the first product in response to a query for a similar product for the first product of the first node Based on the first product category identifier of the first product among products associated with at least one of the second node of the second node and the third node of the third node hierarchy, the product of the third product category having a corresponding relationship with the first product category The connection path of can be considered.

In the step of determining and returning the similarity based on the number of connection paths, in calculating the number of connection paths, a distance between two directly connected nodes is defined as a distance of one hop, and the at least one second The number of connection paths may be adjusted in consideration of a distance of a connection path between at least two nodes among nodes of the first node layer and the nodes of the at least one third node layer.

Between at least one node of the first node layer and at least two nodes of the at least one third node layer, the distance of the first connection path is n _1, and the distance of the second connection path is n ₂ -n ₁ and n _{When 2} is a natural number-, the number of first connection paths may _{be adjusted to 1/n 1} and calculated, and the number of second connection paths may be adjusted to _{1/n 2 to be calculated.}

The node of the first node layer is defined as p, and the node of the third node layer is defined as m, so that a query for similar products and a connection path between nodes corresponding thereto is defined as one of m and p. And can be created as a query by defining it as an index number for a specific node hierarchy identifier.

The product recommendation method further includes the step of learning an artificial intelligence model for improving the accuracy of the similarity determination, wherein the first order history data and the similarity determination including a first node serving as a criterion for a query for the similar product The artificial intelligence model may be trained by generating second order history data including a second node related to a product in which an actual order was placed among similar products returned by the training data set.

When calculating the number of connection paths, only the connection paths within the first distance range are calculated, but based on the learning data set, the artificial intelligence model can be trained while changing the first distance range, which is a critical range for calculating the connection paths. .

Based on the training data set, the artificial intelligence model may be trained while changing whether the class of product category division and the use of a similar category corresponding to the class or a category by combination are changed.

The product recommendation method may further include visualizing the data in the form of a graph.

In order to achieve the above object, a product recommendation device in an electronic commerce system according to an aspect of the present invention includes a communication unit communicating with at least one of a wired and wireless network, and a plurality of shopping malls based on electronic commerce through the communication unit. Of customers' order history data-the order history data includes data on at least one product ordered by the customer-by collecting the collected order history data at least three nodes layer- The at least three node layers include a first node layer associated with a product, a second node layer associated with an order, and a third node layer associated with a customer. In the data, in response to a query for similar products, the degree of similarity is determined and returned based on the number of connection paths between at least two nodes among at least one node of the first node layer and at least one node of the third node layer. And a memory for storing the data in the form of a graph.

A product recommendation method in an e-commerce system according to another aspect of the present invention for achieving the above object comprises the steps of collecting inquiry history data of a plurality of customers from a plurality of shopping malls based on e-commerce (the inquiry history data is The step of converting the collected inquiry history data into graph-type data composed of at least three node hierarchies (including data on the page viewed in a specific session and the product searched in the viewed page) ( The at least three node layers include a first node layer associated with a product, a second node layer associated with a web page, and a third node layer associated with a session), storing the graph-type data in a database, and the stored graph In response to a query for similar products, on the form of data, sorting and returning the products in the order of the highest number of views among the products searched on the web page within the range of the first distance before or after the first product node search of the first node hierarchy. It may include.

A product recommendation device in an e-commerce system according to another aspect of the present invention for achieving the above object includes a communication unit that communicates with at least one of a wired and wireless network, and a plurality of electronic commerce-based shopping malls through the communication unit. Of customers' order history data-the inquiry history data includes data on the page viewed in a specific session and the product searched within the viewed page-and stores the collected order history data by at least 3 nodes (node) layer-The at least 3 node layers include a first node layer related to a product, a second node layer related to a web page, and a third node layer related to a session-Converted to graph-type data consisting of And, on the converted graph-type data, in response to a query for similar products, the order in which the number of views among the products viewed on the web page within the range of the first distance before or after the first product node search of the first node hierarchy is high. It may include a processor that is sorted and returned by and a memory that stores the data in the form of a graph.

According to a product recommendation method and apparatus using a graph database of the present invention, in product recommendation, in consideration of a phenomenon that is macroscopically related but excluded from each other, a more suitable product is recommended to a buyer, thereby enhancing the operational efficiency of a shopping mall, It has the effect of inducing the shopper's shopping activities conveniently.

1 is a conceptual diagram showing a system to which a product recommendation method using a graph database of the present invention is applied;
2 is a conceptual diagram for explaining a similarity-based product recommendation method;
3 is a graph based on a product node, an order node, and a customer node used in a product recommendation method using a graph database according to an embodiment of the present invention;
4 is a conceptual diagram for explaining direct and indirect correlations between products;
5 is a flowchart showing a product recommendation method using a graph database according to an embodiment of the present invention;
6 is a conceptual diagram showing a state in which a plurality of radial graphs centered on a product are entangled;
7 is a diagram for explaining appropriate use of the same/similar category and a combined category in a product recommendation method using a graph database according to an embodiment of the present invention;
8 is a conceptual diagram illustrating an artificial intelligence learning method related to a recommendation algorithm used in a product recommendation method using a graph database according to an embodiment of the present invention;
9 is a conceptual diagram illustrating a connection path indexing method of a product recommendation method using a graph database according to an embodiment of the present invention;
10A to 10D are diagrams showing various embodiments of searching for similar products and customers with similar tendencies as outputs by designating products and customer nodes as inputs;
11 is a diagram showing a result of implementing a product recommendation method using a graph database according to an embodiment of the present invention in a web page;
12 is a flowchart showing a product recommendation method using a graph database according to another embodiment of the present invention;
13A and 13B are diagrams showing an exemplary view of product path prediction in a product recommendation method using a graph database according to another embodiment of the present invention;
14A and 14B are diagrams illustrating exemplary product path prediction in a situation in which a next product is inquired after product inquiry in the embodiment of FIG. 13A;
15 is a conceptual diagram for explaining a situation in which a promotion is required for products with low sales according to product sales activity analysis;
16A and 16B are conceptual diagrams for explaining a matching method for providing a special promotion by matching a specific number of customer groups and a specific number of product groups when there are 5 customers and 6 products;
17A and 17B are flowcharts illustrating a method for recommending a product basket according to another embodiment of the present invention;
18 is a flowchart showing a product-based basket recommendation method according to another embodiment of the present invention;
19A to 19C are tables for specifically explaining a method of generating a matrix for each step of the basket recommendation method of FIG. 18;
20 is a flowchart showing a customer-based basket recommendation method according to another embodiment of the present invention;
21A to 21C are tables for specifically explaining a method of generating a matrix for each step of the basket recommendation method of FIG. 20;
22 is a block diagram schematically illustrating an apparatus using a product recommendation method based on a graph database and a product basket recommendation method according to an embodiment of the present invention.

In the present invention, various modifications may be made and various embodiments may be provided, and specific embodiments will be illustrated in the drawings and described in detail.

However, this is not intended to limit the present invention to a specific embodiment, it should be understood to include all changes, equivalents, and substitutes included in the spirit and scope of the present invention.

Terms such as first and second may be used to describe various elements, but the elements should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another component. For example, without departing from the scope of the present invention, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element. The term and/or includes a combination of a plurality of related listed items or any of a plurality of related listed items.

When a component is referred to as being "connected" or "connected" to another component, it is understood that it may be directly connected or connected to the other component, but other components may exist in the middle. It should be. On the other hand, when a component is referred to as being "directly connected" or "directly connected" to another component, it should be understood that there is no other component in the middle.

The terms used in the present application are only used to describe specific embodiments, and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In the present application, terms such as "comprise" or "have" are intended to designate the presence of features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, but one or more other features. It is to be understood that the presence or addition of elements or numbers, steps, actions, components, parts, or combinations thereof does not preclude in advance.

Unless otherwise defined, all terms used herein including technical or scientific terms have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related technology, and should not be interpreted as an ideal or excessively formal meaning unless explicitly defined in the present application. Does not.

Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the accompanying drawings. In describing the present invention, in order to facilitate an overall understanding, the same reference numerals are used for the same elements in the drawings, and duplicate descriptions for the same elements are omitted.

Product recommendation method using graph database

1 is a conceptual diagram showing a system to which a product recommendation method using a graph database of the present invention is applied. 1, the product recommendation system according to an embodiment of the present invention includes a plurality of shopping malls 110-1 to 110-N, a network 120, a server 130, a database 140, and a user terminal. 150 may be included.

Referring to FIG. 1, shopping malls 110-1 to 110 to N refer to online shopping malls that sell goods and/or services through the Internet. Shopping malls 110-1 to 110 to N sell products to visitors to a corresponding address on the Internet with a specific URL. The shopping mall operator operates the shopping mall by acquiring at least part of the amount of the sold product through his/her account associated with the shopping mall.

In the embodiment of FIG. 1, the shopping malls 110-1 to 110 to N are web pages associated with a specific address (eg, URL) on the Internet, a server handling information related to the shopping mall, or an operator of the shopping mall, as described above. It can be implemented as a terminal of. The terminal may include a computing device such as a smartphone or a PC.

According to an embodiment of the present invention, a plurality of shopping malls 110-1 to 110 -N may interwork with the server 130 through the network 120. Here, the server 130 is a device that obtains information from a plurality of shopping malls 110-1 to 110-N and recommends a product in response to a product recommendation query. The term “device” or “product recommendation device” may be interpreted as referring to the server 130 of FIG. 1. The server 130 may be a server (not shown) that distributes and manages a platform that supports opening and operating a shopping mall, or may be a server interlocking with the server. Therefore, the server 130 not only handles information related to one shopping mall 110-1, but also handles information related to a plurality of shopping malls 110-1 to 110-N, and provides various information based on big data. Can analyze and learn.

The server 130 may be implemented as a computing device. The computing device referred to herein may be implemented as a server-class computer terminal. The computing device may include an input device of a typical computer terminal, a display device, a networking device, a hard disk, a memory for storing a program, a processor for executing a program stored in the memory, and the like. However, it does not necessarily have to be implemented as a server-class computer terminal.

The network 120 is a communication network connecting the shopping malls 110-1 to 110-N and the server 130, and may include a wired network and/or a wireless network (Internet).

The server 130 is the order history data and/or product from the shopping mall operation server (not shown) that generally manages the information of the shopping malls 110-1 to 110-N and/or the shopping malls 110-1 to 110-N. Acquire inquiry data. Then, the obtained data is converted into a graph form having a plurality of node hierarchies. The data converted into a graph form is stored in the database 140 external to the server 130. However, the database 140 does not necessarily have to be outside the server 130. It may exist inside.

The server 130 responds to a product recommendation request from the user terminal 150 (which may be referred to as a "recommended product search request", a "query", etc.), using a graph composed of three node hierarchies to provide similar products from a specific product. Process queries for search. At this time, on the converted graph data, the connection path between the product and the product, the customer and the customer, or the product and the customer is extracted, and the degree of similarity between the extracted connection paths may be determined by considering the distance and/or number of the extracted connection paths. . In addition, when product search data is used, a query for similar products is processed by analyzing the inquiry history during a certain step before or after product search on a graph. The part related to the technology for processing queries for similar products will be described in more detail below.

In another embodiment of the present invention, the server 130 uses the product based on the existence of an index (which may include the degree of similarity using the graph database above) indicating the purchaseability between an individual customer and an individual product. The basket can be optimized. That is, if the product basket recommendation method according to an embodiment of the present invention is used, it is possible to appropriately promote a plurality of product groups optimized for at least one customer (in multiple cases, a customer group).

The user terminal 150 may be a terminal of a purchaser of the shopping mall 110-1. The purchaser may transmit a product recommendation request to the server 130 or a separate shopping mall management server (not shown) using the user terminal 150. If there is a separate shopping mall management server, the shopping mall management server transmits the received product recommendation request to the server 130. The product recommendation request may include a query for similar products. Furthermore, it may include queries for similar customers. Then, when the processing of the query is completed using a graph-based data analysis method and/or a basket recommendation method, the processed data is provided to the user terminal 150 so that the buyer checks it and selects a product based on this. To be able to. The user terminal 150 may provide a product recommendation request to the server 130 using the network 120 (a specific web page in the network).

According to another embodiment of the present invention, the terminal or user terminal 150 associated with the shopping malls 110-1 to 110-N may download an application provided from the server 130 and perform automatic generation through the application. . That is, in the terminal or user terminal 150 associated with the shopping malls 110-1 to 110-N, when a user inputs a query for a similar product or similar customer, a graph-based product recommendation process or a product basket recommendation process is performed in the terminal. Can be done in Without a separate server device, a processor (not shown) of the terminal executes the downloaded application, and it is possible to recommend an optimal product through data transformation and analysis of the transformed data.

2 is a conceptual diagram illustrating a product recommendation method based on similarity.

Referring to FIG. 2, the device generally recommends products based on similarities between images and attributes of products. That is, first, by calculating the similarity between product images using the image analysis module, it can be utilized as a purchaseability index. In the embodiment of FIG. 2, the apparatus may analyze the image of the color and shape of the shoe product 1 and extract the shoe product 2 having the most similar color and shape among numerous product images stored in the database. Alternatively, shoe product 2 having similar attributes may be extracted based on the product attributes of shoe product 1. For example, if the color of shoe product 1 is red, the size is 250, the brand is "A brand", and the shape is "shoe", the shoe product 2 of the same or similar attribute may be searched and extracted from the database. . However, the method of using the similarity based on the image and/or product attributes, as described above, corresponds to the case of purchasing similar products at the same time, so it cannot be said that the purchase possibility is very high, and the purchaser's demand cannot be satisfied. Occurs.

According to an embodiment of the present invention, the device may extract attributes of products that have been purchased a lot based on the purchase history of a specific customer, and set a high purchaseability index of products having attributes corresponding thereto. Using this point, it is also possible to calculate the purchaseability index by categorizing the customer with the attributes of the product purchased by the customer. That is, it is possible to find customers who have a similar tendency to those who have purchased the corresponding product, and recommend a product purchased by the similar customer as a similar product.

As described above, the purchaseability index based on various methods of product image, product attribute, and/or customer attribute analysis may be used as an element that enters the blank space of the matrix in the method of recommending product baskets in the future.

3 is a graph based on product, order, and customer nodes used in a product recommendation method using a graph database according to an embodiment of the present invention.

Referring to FIG. 3, the device may collect order history data and generate a graph composed of at least three node hierarchies. The order history data basically includes information that "a specific customer A has ordered product C by using order B in a specific shopping mall". In other words, it is data that includes the relationship between customers, orders, and products.

The device generates a radial graph by forming a product into a first node layer, an order into a second node layer, and a customer into a third node layer based on each order history data. In the embodiment of FIG. 3, product A 310 is included in six orders 320 (orders A to F), and each order is obtained from a different customer 330. In this case, the device connects the order A (320) to the order F using a connecting line, centering on the product A (310), and connects the customer A (330) requesting the order A (320) to the order A (320). Connect each so that a radial graph is created from the product. In other words, the product-order connection line represents the product included in the order, and the order-customer connection line represents the customer who placed the order. In this case, a first node layer (product layer) of each node layer of the graph may be indicated in red color, a second node layer (order layer) may be indicated in green color, and a third node layer (customer layer) may be indicated in blue color. The generated graph can be visualized and displayed through a display device.

As shown in the embodiment of FIG. 3, the device may display a plurality of orders including one product 310 (product A) and a customer who has placed the order. At this time, it can be seen that between the customers 330 included in the graph of FIG. 3, there is one path connected by a length of distance 4 (customer → order → product → order → customer). Therefore, it can be seen that these customers show similarity in purchasing tendency compared to those who do not. The meaning of the distance 4 will be described in more detail with reference to FIG. 4 below.

4 is a conceptual diagram illustrating direct correlation and indirect correlation between products.

Referring to FIG. 4A, there are not many people who order two very similar products (which may belong to a top) at the same time. That is, in the case of a connection path having a high direct correlation and a distance of 2, it is preferable to determine that the similarity is low.

However, referring to (b) of FIG. 4, when viewed through a different product (bottom (product 3)), a number of customers who purchase with a combination of (product 1- product 3) and (product 2- product 3) May exist. That is, the initial direct correlation (distance 2) (refer to the embodiment (a) of Fig. 4) shows a low degree of similarity, but when the distance becomes distant (e.g., distance 4), indirect correlation can be observed, It is desirable that this indirect correlation be implemented as a high indicator in terms of purchasing feasibility. In other words, the number of paths to which two products (product 1 and product 2) can be connected at a predetermined distance (distance 4 in the embodiment of FIG. 4B) may be an index indicating similarity.

Returning to FIG. 3 again, in view of the above correlation, customers connected to the distance 4 may have a relationship of “other customers who have purchased a product purchased by a specific customer”. For a specific customer, by looking for the relationship "customer who has purchased the same product" as above, a path through which a specific customer and another customer are connected through different products can be extracted. According to an embodiment of the present invention, the device may be determined as a customer with the largest number of connected paths, and as a customer B having the most similar purchase propensity to a specific customer (customer A).

Accordingly, the device may recommend a product that a specific customer A has purchased, but a similar customer B has not purchased, to the customer B based on the graph database. Similarly, a product purchased by customer B but not purchased by customer A may be recommended to A. That is, the device can connect the customer A and the customer B via the common product A 310, and analyzes the similarity in consideration of the length of the connection path and the number of connection paths.

5 is a flowchart illustrating a product recommendation method using a graph database according to an embodiment of the present invention.

Referring to FIG. 5, a server (which may be referred to as a "device") collects order history data from an online shopping mall or a server managing the shopping mall (S510). At this time, the acquired data includes information on the product, the order including the product, and the relationship between the customer, who executed the order.

The device converts the collected order history data into data in the form of a graph (S512). The graph consists of three node hierarchies, and the first node layer is connected to the product, which is the second node layer, through a connecting line, and the order is connected to the customer, which is the third node layer, through a connecting line, and the first node layer is connected. It can have a radial graph shape centered on it. Then, the device stores the data converted in the form of a graph in the database (S514).

The client terminal inputs a recommended product search request (S516). The input request information is transmitted to the device (server) side. In this case, the recommended product search request may be divided into (i) a product-oriented search request and (ii) a customer-oriented search request, and input. This will be described in more detail through FIGS. 10A to 10D.

When a request to search for a recommended product is input, the device analyzes the correlation centering on the target product or the target customer using the converted data in the form of a graph (S518). Then, the target product or target customer is placed at the center, and a connection path to other products or other customers is created based on the analyzed correlation (S520). At this time, it is determined whether the connection path exists within a certain distance range based on the distance of the generated connection path (S522). As described above, it is preferable to have a distance larger than the distance 2 representing a direct correlation (e.g., a=2), and as the distance increases, the route to be connected increases proportionally. It is preferable to exclude the connection path (eg, b=10). The values a and b that define the distance range can be set through user settings and can be arbitrarily changed. In addition, the values of a and b can be changed in the form of finding an optimal solution through machine learning. A connection path having a distance less than a predetermined range or a distance exceeding the range is distributed (S524).

According to an embodiment of the present invention, the device may adjust the number of connection paths in consideration of the distance of connection paths between two nodes. As described above, the distance between two directly connected nodes is defined as a distance of 1 hop. At this time, when the distance of the first connection path is n ₁ (n ₁ is a natural number) and the distance of the second connection path is n ₂ (n ₂ is a natural number), the number of the first connection paths is adjusted to _{1/n 1} Can be calculated. In addition, the number of second connection paths may be calculated by adjusting to _{1/n 2.} That is, except for the direct correlation of distance 2, since nodes connected by a short distance may have high similarity, it is possible to adjust the number of connection paths by dividing by the corresponding distance considering this more importantly. For example, the number of connection paths between nodes of the distance 4 is considered to be 1/4, and may be considered larger (importantly) than 1/8, which is the number of connection paths of the distance 8.

The device calculates the number of connection paths existing within the set distance range based on the above method (S526). Then, the product or customer having the most number of connection paths is extracted (S528). At this time, by sorting based on the number of connection paths, the top few products or customers may be extracted as products or customers with high similarity. The high degree of similarity means that the propensity to purchase is similar. In particular, in the case of high customer-to-customer similarity, it is highly likely that the target customer will purchase the remaining products minus the common product within the first order containing the common product through a common product from the viewpoint of the customer determined to be similar. Because it can be judged, these products can be derived as recommended products.

In this way, information on products and customers resulting from the similarity degree, and other information on the remaining products in the first order, excluding common products based on the corresponding customer, is provided to the client terminal. The terminal may display the provided information to the purchaser (S530).

6 is a conceptual diagram showing a state in which a plurality of radial graphs centered on a product are entangled.

Referring to FIG. 6, a plurality of radial graphs 610, 620, 630, and 640 may be entangled. In the apparatus, a first radial graph 610 centered on product A and a second radial graph 620 centered on product B may be connected to each other via order A or customer A. These two radial graphs 610 and 620 may be visualized and displayed in two dimensions on a display screen. Further, the first radial graph 610 and the third radial graph 630 centered on the product C may be connected to each other via order B or customer B, and at this time, the second radial graph 620 and the third radial graph 630 The graph 630 may also be connected to each other via order C or customer C. In this way, when three or more radial graphs 620 are connected, it is difficult to visualize and express the correlation between each of the radial graphs in two dimensions.

Thus, in an embodiment of the present invention, the device may display radial graphs in three dimensions. That is, if the first and second radial graphs 610 and 620 are displayed in the xy plane, the first and third radial graphs 610 and 630 may be displayed with a first angle in the z-axis direction, and the second The second and fourth radial graphs 620 and 640 may be displayed with an angle. Accordingly, it is possible to pull the graphs through the zoom-in/zoom-out function of the user interface, and to see the connection relationship between a plurality of radial graphs by adjusting the three-dimensional angle using the panning function. Can be visualized. In addition, when a specific node is selected, the selected node is brought to the center of the screen, so that a plurality of radial graphs connected to the center node can be displayed in all directions.

According to another embodiment of the present invention, it is not necessary to take the type of the relationship as "customer-order-product-order-customer", but the relationship of "product-order-customer-order-product" centering on the customer. Can be specified. In other words, it is okay to take various standards of relationship.

7 is a diagram for explaining appropriate use of the same/similar category and a combined category in a product recommendation method using a graph database according to an embodiment of the present invention.

Referring to FIG. 7, the device may utilize product category information when determining the degree of similarity and when extracting similar products/customers and then derives recommended products. For example, in searching for similar products, similar products can only be found within the same category (e.g., tops), or in different categories (category by competition or combination) (e.g., tops and bottoms, bottoms and shoes, etc.) You can try to find a product that corresponds to ).

In particular, by classifying the class of the product category, the first class is divided into the largest category such as tops, bottoms, shoes, accessories, etc., and the same/similar category and the combined category are classified, and the second class is within the first class. It can be defined by subdividing it in. Each product category is defined through an identifier, and a similarity/combination relationship between the identifiers is coded so that it is possible to identify what kind of relationship each product category is. In the case of the second class, for example, it can be divided into a shirt, a southern shirt, a T-shirt, a hood, and a vest, within a top. The third class is defined as a category obtained by subdividing the second class.

According to an embodiment of the present invention, assuming that class classification is up to the third class, it is possible to derive similar products by using the same/similar product category in all classes from the first to the third class. In this case, for the "T-shirt", the "Southern" product category may be selected as a "top" and similar product category as the "T-shirt", and products belonging to the "Southern" product category within a specific order are derived as similar products. Can be.

Alternatively, similar products may be derived using the combined category in the first class. In this case, a "bottom" category is selected for the "T-shirt", so that a product belonging to the bottom product category may be derived as a similar product of the T-shirt. In the case of T-shirts, since there are many situations where they are purchased in combination with bottoms, this setting may be appropriate.

In another embodiment, the first class may use the same/similar category, but the second class may use a combined category. In this case, for the "T-shirt", as the same/similar category in the first class, a category of "top" may be selected, and in the second class, a "jumper" category combined with the "T-shirt" may be selected. That is, within a specific order, products belonging to the product category of "jumper" may be derived as similar products.

In this way, in deriving a similar product using a graph, determining whether to use the same/similar product category or a combined product category is an important factor to be considered because the resulting value is different. In addition, the resulting value varies depending on which class uses the similar category and the combined category. The device can define these settings through user settings and allow the user to arbitrarily change these settings. Or, big data-based artificial intelligence can be used to optimally adjust these areas. A method of using the artificial intelligence model will be described in more detail with reference to FIG. 8.

8 is a conceptual diagram illustrating an artificial intelligence learning method related to a recommendation algorithm used in a product recommendation method using a graph database according to an embodiment of the present invention.

Referring to FIG. 8, it can be said that it is very important to determine whether to use "same/similar category" or "combined category" in a divided specific class because the result value is completely different. Accordingly, the device may train an artificial intelligence model using an artificial intelligence learning module (not shown), and then set whether to use a similar or combined category in a specific class based on the learned artificial intelligence model.

According to an embodiment of the present invention, the artificial intelligence model used in the device includes a deep learning convolutional neural network model (CNN). However, the present invention is not limited thereto, and may include any one of a deep neural network (DNN), a recurrent neural network (RNN), and a long short term memory network (LSTM).

The device generates order data 1 and order data 2 as a training data set. At this time, the two order data are generated by the order data prior to the selection process (order data subject to the similarity search request) and the actual order after product recommendation according to graph-based similar product derivation according to an embodiment of the present invention. It has a relationship of order data. That is, a customer who ordered order data 1 receives a recommendation of a plurality of products using a graph-based similar product derivation method according to an embodiment of the present invention, and combines order data 2 formed by selecting one of the recommended data. Create a learning data set. At this time, in deriving graph-based similar products for product recommendation, similar products are derived while changing whether to use a similar or combined category in the first class and whether to use a similar or combined category in the second class, Based on the order data after derivation, artificial intelligence is learned in the direction of placing an order within the product derived to the highest level through the process of deriving similar products based on the graph.

In addition, while changing the threshold for the optimal distance range (values a and b in Fig. 5), the optimal threshold value is found by accumulating the result value based on the training data set of order data 1 and order data 2. Let's do it.

At this time, the node layer used as an intermediary between the order data 1 and the order data 2 forming the learning data does not necessarily have to be the first node layer “product”, but may be the third node layer “customer”. Also, the second node layer may be the reference.

9 is a conceptual diagram illustrating a connection path indexing method of a product recommendation method using a graph database according to an embodiment of the present invention.

Referring to FIG. 9, the device identifies products and products, customers and customers, or a connection path between products and customers by indexing data. In this case, the connection path may be composed of a node layer identifier and an index number. The node layer identifier is composed of a combination of the identifier of the target node that is the center of the search and the identifier of the final destination node. In this case, the first node layer (product node) may be identified as p, and the third target node layer (customer node) may be identified as m. After identifying the node hierarchy of the starting node and the destination node in this way, the indexing number is attached afterwards. The indexing number is related to a query and may have a special meaning. For example, if an index of "0100" is attached to the pp node hierarchy, it means a query for customers who purchased the product to find another product purchased, and the path associated with this is a group of customers who ordered the first product (departure node). It means another product ordered. A description of the query according to the meaning of the index directly followed by the node layer identifier will be described in more detail with reference to FIG. 10.

In this way, each connection path is identified in the form of "-1" while the index is attached. That is, the first path associated with the pp0100 query may be dataized as "pp0100-1". After that, the number at the end of the "-" is increased by 1 to identify the route. For example, if there are a plurality of other products ordered in a bundle by a customer who ordered the first product, it may be defined as pp0100-1, pp0100-2, pp0100-3, and the like.

In the same way, the connection path from the first customer to the second customer is expressed as mm0100-1, mm0100-2, or mm0101-1, mm0101-2, and the like. The connection path from the first product to the first customer is expressed as pm0100-1, pm0100-2, and the like. The connection path from the first customer to the first product is expressed as mp0100-1, mp0100-2, and the like.

In this way, the connection path is converted into data and the device can hold it, and the number of connection paths within a specific distance range is calculated using this.

Additionally, the device may be controlled to enable conversion in the form of a list of connection paths in the implementation of the graph of FIG. 6. That is, while visualizing a plurality of radial graphs, if there is a request to convert information related to a connection path between a specific departure node and a specific arrival node into a list form, the connection path index between the departure node and the destination node is converted into a list. Can be displayed. In addition, the distance of the connection path of the first index (eg, pp0100-1) may also be displayed in the list. Accordingly, when a connection route having a specific index is clicked, the corresponding route is displayed in a highlighted state on the graph, and through which route is connected from the departure node to the destination node, it is visualized so that it can be seen intuitively.

In addition, in the case of pp0100-1 and pp0101, it may be duplicated and indexed.To prevent this, when extracting all connection paths for pp0100, it is extracted as pp0100-1, and all connection paths for pp0101 are extracted. In this case, it is extracted as pp0101-1 so that it is not duplicated, and when all routes are extracted, the duplicated pp0100-1 can be excluded and identified as pp0101-1 representing a more specific connection route.

10A to 10D are diagrams illustrating various embodiments of searching for similar products and customers with similar tendencies as outputs by designating products and customer nodes as inputs.

The device generates a query to find similar products and customers by designating a product or customer node as an input.

Referring to FIG. 10A, a query for finding products having similarity to a specific product is generated. A customer who ordered a product can generate a pp0100 query that shows other products ordered in a bundle. In addition, the connection path can be derived by using the pp0101 query indicating products in the same category among other products ordered in a bundle by a customer who ordered a product.

In addition, a pp0200 query indicating a product purchased by a customer who has ordered a product in another order and a pp0201 query indicating a product in the same category in the other order may be used.

Referring to FIG. 10B, there is a query for finding a customer with similarity to a specific customer as follows. An mm0100 query indicating a second customer who has ordered the same product as the first customer can be used. Further, the mm0101 query may be used to find a route to a second customer who has ordered the second product of the same category, not the product that is not exactly the same as the first product ordered by the first customer.

Referring to FIG. 10C, in relation to a query for finding a customer with similarity to a specific product, a query pm0000 indicating a path to a customer who has purchased a product in a bundle with a corresponding product, and among the products purchased in a bundle, the corresponding product You can use the pm0100 query indicating the path of the customer who purchased the product in the same category as and.

Referring to FIG. 10D, in relation to a query for finding a product similar to a specific customer, an mp0000 query indicating a path to products purchased in a bundle, such as a product ordered by a corresponding customer, and among the products purchased in the bundle, the corresponding You can use the mp0001 query indicating the path to the product and the product in the same category.

According to an embodiment of the present invention, in determining the similarity between a product and a product, the above query exists, and the connection path for the query is also expressed as ppXXXX-X. In particular, although ppXXXX-X may be used as the connection path between products and products, a combination of the pmXXXX-X path, which represents the path related to the similarity between product and customer, and the mpXXXX-X path, represents the connection path related to the similarity between the customer and the product. You can use That is, it is also possible to determine the similarity between the product and the product using pm0000-1 + mp0000-1. In addition, it is possible to determine the similarity between a product and a product by using three or more connection paths related to similarity, such as using pm0000-1 + mm0100-1 + mp0000-1.

According to an embodiment of the present invention, when using any of the paths of (i) ppXXXX-X and (ii) pmXXXX-X + mmXXXX-X + mpXXXX-X, also prior to FIG. 8 The accuracy can be improved by learning while changing the parameters (including weights) associated with the path using the artificial intelligence learning model of.

11 is a view showing a result of implementing a product recommendation method using a graph database according to an embodiment of the present invention in a web page.

Referring to FIG. 11, the device enables the above algorithm to be executed by accessing it using a REST API. The device designates which logic (relationship) to apply using a URL (Uniform Resource Locator). In the embodiment of FIG. 11, pp0100 is designated to derive other products purchased by members who have purchased the product.

In the URL path, a query parameter may be attached following a query such as pp0100, which may include a product number, a number of data to be returned, and a member ID. In other words, only customer or product information can be used according to the recommendation logic. Additional parameters such as product code and customer number are transmitted in a URL encoding method.

In response to these queries, as the query output, the number of data (n), product or member (node), importance (hits: number of paths retrieved), and cryptographic queries (Cypher query for debugging purposes) will be provided. I can.

The device may return the number of products of the top n (n=5, ie, 5) and the number of paths for each product in the order of high similarity, that is, the number of connection paths.

In the case of using the product purchase history, the time to grasp the customer's preference may be delayed, and in the case of purchase, it is difficult to see statistical relevance due to insufficient data, so that the user recommends a product by using the product search history. Can be considered.

12 is a flowchart illustrating a product recommendation method using a graph database according to another embodiment of the present invention.

Referring to FIG. 12, an apparatus (server) collects inquiry history data related to product inquiry from a server that manages shopping malls and/or shopping malls (S1210). This can be achieved by extracting the user's product information inquiry history from the web log file. The inquiry history can be converted into sequential product codes and converted into data. For example, after viewing product 1, inquiring product 2, and inquiring product 4, the inquiry history data is generated by converting the product code into a product code by considering product 1 and logging out from the logout history.

Then, the device converts the inquiry history data into a graph form (S1212) and stores the data in the database (S1214). At this time, the structure of the graph may have a structure of “session-page-product” so as to correspond to the preceding “product-order-customer”.

The client (user terminal) may input a request to search for a recommended product (S1216). This can be accomplished using a URL, as described in FIG. 11.

The device extracts the searched product within the range of x web pages before or after searching for a specific product (S1218). The x value can be determined by user settings and can be arbitrarily changed.

The device may recommend to the buyer in the order of the products with the highest number of views among the searched products (S1220). The client may receive and output the recommended product information (S1222).

13A and 13B are diagrams illustrating exemplary product path prediction in a product recommendation method using a graph database according to another embodiment of the present invention.

Referring to FIGS. 13A and 13B, when the number of recommended products n is 10 and there is a request for recommended products, the device generates a graph formed by at least three node hierarchies. This may include a web page 1320 (a second node layer) searched by a user centering on the session 1330, which is a first node layer, and a product 1310 (third node layer) searched on the web page. The node layer may be changed so that the product is the first node layer, the page is the second node layer, and the session is the third node layer.

The device connects the web page 1320 associated with the specific session 1330 and configures the product 1310 searched in the web page 1320 to be connected to the web page 1320. In addition, it is preferable to construct a graph by using an arrow (→) instead of a connecting line (-) formed in a straight line to identify the time-series order by separating the preceding operation and the subsequent operation.

After constructing the graph, the device searches and recommends 10 products searched within the range of 3 pages in the order of the highest number of views after product 24522 (1310) search. Contents related to the following three web pages will be described with reference to FIGS. 14A and 14B.

14A and 14B are diagrams illustrating exemplary product path prediction in a situation in which a next product is inquired after product inquiry in the embodiment of FIG. 13A.

14A and 14B, the device uses a web page 1422 after a customer inquires a product 24522 (not shown) and then a product 25799 (1410) through a web page 1420. The fact of inquiring the product 25622 (1412) can be confirmed based on the converted graph data. In this way, after a specific product, the clients of FIGS. 13A to 13B are searched for and returned 10 products in the order of the highest number of views within the range of pages (which may be 3 pages in this embodiment). Can process requests from However, in this case, the three pages may be set to be a page on which product search has been performed. That is, it is preferable to exclude a web page without a product search from calculating the number of page strokes, calculate three pages based on the page on which the product search has been performed, and recommend a higher product that has been viewed a lot within the page range.

According to another embodiment of the present invention, it is not necessary to recommend a product by limiting only to the page, but by limiting the session, it is possible to recommend a product based on the number of views of products made within three sessions. Furthermore, by using the category identifier of the product recommended due to the high number of views, it is possible to recommend only products in a contending category or a product in a similar category. The use of similar or contention categories and the use of classes can also be optimized using the artificial intelligence model of FIG. 8. The learning data set at this time may be a set of inquiry data 1 that is a target of a search request and inquiry data 2 related to a product that has been actually purchased through a graph-based product recommendation process according to an embodiment of the present invention.

In an embodiment of the present invention, the apparatus may mix and use a first graph database based on product purchase history data and a second graph database based on product inquiry history data. For example, after searching for the top 10 products with many views using the second graph database, the degree of similarity between the 10 recommendation candidate products and the requested product may be calculated using the first graph database. Thus, it is possible to change the order of recommendation of 10 recommendation candidates based on products with high similarity using the first graph database. The opposite is also possible.

In addition, a weight of 1 is given to the recommended product candidate calculated based on the first graph database, and a weight 2 is assigned to the recommended product candidate calculated based on the second graph database to finalize the product with the highest similarity based on the weight. You can consider how to select a candidate.

Product basket recommendation method

15 is a conceptual diagram illustrating a situation in which a promotion is required for products whose sales are poor according to product sales activity analysis.

Referring to FIG. 15, the apparatus performs promotion for some of 10 products (products with an activity lower than 0.2) (products in lagger stage) with low sales by performing an analysis of the sales volume of the product and the sales activity of the product. May be required. That is, when four out of 10 low-selling products are selected to be promoted, the product basket recommendation method according to an embodiment of the present invention may be used.

16A and 16B are conceptual diagrams illustrating a matching method for providing a special promotion by matching a specific number of customer groups and a specific number of product groups when there are 5 customers and 6 products.

Referring to FIG. 16A, there are 5 customers and 6 products. When 3 products are to be promoted by 3 customers, a decision to optimize saleability is required. For this, the device can perform an optimized promotion through two methods.

(1) The first method is a method of matching multiple product groups and multiple customer groups by first selecting the best product group for 5 out of 6 products, and secondly selecting the best customer for the selected product (see Fig. 18). .

(2) The second method is a method of matching multiple product groups and multiple customer groups by first selecting the best customer group for 6 products to 5 customers, and secondly selecting the best product for the selected customers (Fig. 20). Reference).

The apparatus performs a determination to optimize the saleability using at least one of the above two methods, and by the determination, the promotion may proceed in a state as shown in FIG. 16B.

Referring to FIG. 16B, the device promotes a package consisting of product 2, product 3, and product 6 to customer 1, customer 3, and customer 4. Promotion can be conducted in the form of offering a special discount only to the customer of the product included in the package. Alternatively, a product of a corresponding package may be displayed in a pop-up manner on the first page of a shopping mall, or packaged products may be arranged in some of the most conspicuous areas to induce sales.

17A and 17B are flowcharts illustrating a method of recommending a product basket according to another embodiment of the present invention.

Referring to FIGS. 17A and 17B, the apparatus follows the following procedure when selecting and promoting a customer to be recommended for each of a plurality of products (eg, 10 products).

First, the device acquires product and customer data that are candidates for promotion (S1710). In the embodiment of FIG. 17B, it is assumed that there are 10 product candidates and about 100 customer candidates, and data for this is obtained. At this time, in the customer candidate, the purchaser who has previously purchased a specific product is excluded or data is set to 0 and collected. It is preferable that the collected data take the form of a matrix. At this time, products are formed in rows and customers are formed in columns. Of course, the opposite is okay. In addition, in the blank column of the matrix, an index related to the individual product and the purchaseability of each customer is input. As an index related to the purchaseability, an index calculated based on a product image analysis, a product attribute analysis, and a tendency according to a customer purchase history may be used. Alternatively, a similarity degree based on a graph database according to an embodiment of the present invention may be used as the index. In the embodiment of FIG. 17B, the degree of similarity calculated as the number of connection routes using a graph database is used as the purchaseability index.

After acquiring data and generating a matrix, the device calculates the total number of connection paths for each product (S1720), and selects products with a higher ranking in the number of connection paths (S1730). The number of products to be selected depends on the number of products to be promoted. This is a situation in which a user (eg, a shopping mall operator) can select through a user interface. Alternatively, all products having a number of connection paths greater than or equal to the threshold value may be selected without setting a specific number of products as a reference value. For example, products having a connection path greater than the number of 700 connection paths may be selected without limiting the number. A small number of connection paths may be judged to be less relevant. In this case, it is possible to select one with a smaller number of connection paths and process it in the direction of offering more discounts. In the embodiment of Fig. 17B, the device selects the top four products with a high number of connection paths.

The device reorganizes the matrix into a new matrix by leaving only the top four selected products (which can be changed through user settings), deleting the remaining products (S1740). Then, for the remaining 4 products, 6 customers with a high total of connection paths are selected as target customers for the promotion. The device promotes the selected four product packages to six customers of the promotion target customer soldiers (S1750).

18 is a flowchart illustrating a product-based basket recommendation method according to another embodiment of the present invention. This relates to method (1) of FIG. 16A.

Referring to FIG. 18, after acquiring data on a product to be a promotion candidate and a purchaseability index associated with a customer, the device generates a matrix in which the product is a column and the customer is a row (S1810). In the blanks of the matrix, the index of purchaseability between product and customer is entered.

Then, the device calculates the sum of purchase possibilities for each row (S1820). Then, the products are ranked in the order of the highest total purchase possibilities, and the products with the highest total purchase possibilities are selected as many as the number of products to be recommended by promotion, and the remaining products are deleted to reconstruct the matrix (S1830).

After the product is selected, the device calculates the individual subtotal for each row (S1840). Then, after ranking the subtotals for each customer in a large order, a customer with a higher purchaseability subtotal ranking as many as the number of customers to be promoted is selected and the remaining customers are excluded (S1850).

When the selection of the customer group and the product group is completed, the device forms the product group as a package targeting the customer group and performs promotion (S1860).

19A to 19C are tables for explaining in detail a method of generating a matrix for each step of the basket recommendation method of FIG. 18.

Referring to FIG. 19A, the apparatus generates a matrix by configuring customers as rows and products as columns based on purchaseability indicators for customers 1 to 5 and products 1 to 6.

Referring to FIG. 19B, the device ranks a subtotal for each product through the summation of rows. According to the embodiment of FIG. 19B, product 2 is ranked first with a value of 69, product 3 is ranked second with a value of 55, and product 6 is ranked third with a value of 53.

Referring to FIG. 19C, the top three products (product 2, product 3, and product 6) are selected, and a subtotal for each customer is calculated by performing a summation of columns. Then, rank them. According to the embodiment of FIG. 19C, customer 1 is ranked first with a value of 48, customer 3 is ranked second with a value of 43, and customer 4 is ranked third with a value of 33.

Accordingly, the device selects customer 1, customer 3, and customer 4 as target customers for promotion, and selects product 2, product 3, and product 6 as promotion target products to perform the promotion.

20 is a flowchart illustrating a method for recommending a customer-based basket according to another embodiment of the present invention. This relates to method (2) of FIG. 16A.

Referring to FIG. 20, after acquiring data on a product to be a promotion candidate and a purchaseability index associated with a customer, the device generates a matrix in which the product is a column and the customer is a row (S2010). In the blanks of the matrix, the index of purchaseability between product and customer is entered.

Then, the device calculates the sum of purchase possibilities for each column (S2020). Then, the customers are ranked in the order of the highest total purchase probability, and as many as the number of customers to be recommended, the customers with the highest total purchase probability are selected, and the remaining customers are deleted to reconstruct the matrix (S2030).

After the customer is selected, the device calculates a subtotal for each product for each row (S2040). Then, after ranking the subtotals for each product in a large order, a product having a higher purchaseability subtotal ranking as many as the number of products to be promoted is selected and the remaining products are excluded (S2050).

When the selection of the customer group and the product group is completed, the device forms the product group as a package targeting the customer group and performs promotion (S2060).

21A to 21C are tables for explaining in detail a method of generating a matrix for each step of the basket recommendation method of FIG. 20.

Referring to FIG. 21A, the apparatus generates a matrix by configuring customers as rows and products as columns based on purchaseability indicators for customers 1 to 5 and products 1 to 6.

Referring to FIG. 21B, the device ranks by calculating a subtotal for each customer through the summation of columns. According to the embodiment of FIG. 21B, customer 1 is ranked first with a value of 77, customer 3 is ranked second with a value of 73, and customer 4 is ranked third with a value of 72.

Referring to FIG. 21C, the top three customers (customer 1, customer 3, and customer 4) are selected, and a subtotal for each product is calculated by adding up the rows. Then, rank them. According to the embodiment of FIG. 21C, product 2 is ranked first with a value of 47, product 3 is ranked second with a value of 41, and product 5 is ranked third with a value of 37.

Accordingly, the device selects customer 1, customer 3, and customer 4 as target customers for promotion, and selects product 2, product 3, and product 5 as promotion target products to perform the promotion.

According to the product-based recommendation method according to the exemplary embodiment of FIG. 17 and the customer-based recommendation method according to the exemplary embodiment of FIG. 19, the promotional customer is the same, but the product is different. As such, depending on which method you use, the results can be completely different.

22 is a block diagram schematically illustrating an apparatus using a product recommendation method based on a graph database and a product basket recommendation method according to an embodiment of the present invention. As shown in FIG. 22, the product recommendation device according to an embodiment of the present invention may include a communication unit 2210, a display unit 2220, a processor 2230, a memory 2240, and an input unit 2250. .

Referring to FIG. 22, the communication unit 2210 is a component for communication with other devices by being connected to a network. The communication unit 2210 may be implemented with an antenna, a communication related chip, or the like.

The display unit 2220 is a component that displays information input/output through the communication unit 2210 and information processed through the processor 2230. This can be implemented with one of a monitor, a TV and other panels.

The processor 2230 is a component that executes a graph-based similarity determination and product basket recommendation algorithm according to an embodiment of the present invention. This can be implemented with a microprocessor. The processor 2230 converts the data in graph form based on the order history data and inquiry history data input through the communication unit 2210, and uses the data in graph form converted in response to a query for similar products from a client. As a result, information on similar products and similar customers is returned. At this time, products and/or customers may be sorted and returned in the order of the highest similarity based on the number of connection paths from the first node to the second node (which may be limited to having a certain range of distances). .

Also, the processor 2230 may sort and return products in the order of the most searched products in pages within a certain range before or after the first product search. The returned result data may be displayed through the display unit 2220 or may be provided to a terminal requested through the communication unit 2210.

The processor 2230 obtains graph-based similarity information or other similarity index-based information on purchaseability indexes of individual products and individual customers, and generates a matrix in which the customer and product are each column and row. Then, a partial sum of rows and columns of the generated matrix is calculated, and the calculated values are ranked to generate an optimal product package for a specific customer group, so that appropriate promotion can be performed.

According to an embodiment of the present invention, the processor 2230 includes an artificial intelligence model learning module (not shown) and an artificial intelligence model execution module (not shown), and provides an artificial intelligence model for increasing the accuracy of graph-based similarity determination. It is possible to improve the accuracy of actual product recommendation by training and executing the learned artificial intelligence model.

The memory 2240 is a storage for storing instructions of a program executed by the processor 2230. The memory 2240 stores information received or transmitted through the communication unit 2210. In addition, at least some of the information processed by the processor 2230 may be stored.

The input unit 2250 is a component that receives a user's input, and may be implemented as a keyboard, a mouse, a touch panel, or the like. Various parameters related to optimal product recommendation may be directly input through the input unit 2250.

The system or device described above may be implemented as a hardware component, a software component, and/or a combination of a hardware component and a software component. For example, the systems, devices, and components described in the embodiments are, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA). ), a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications executed on the operating system. Further, the processing device may access, store, manipulate, process, and generate data in response to the execution of software. For the convenience of understanding, although it is sometimes described that one processing device is used, one of ordinary skill in the art, the processing device is a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that it may include. For example, the processing device may include a plurality of processors or one processor and one controller. In addition, other processing configurations are possible, such as a parallel processor.

The software may include a computer program, code, instructions, or a combination of one or more of these, configuring the processing unit to operate as desired or processed independently or collectively. You can command the device. Software and/or data may be interpreted by a processing device or, to provide instructions or data to a processing device, of any type of machine, component, physical device, virtual equipment, computer storage medium or device. , Or may be permanently or temporarily embodyed in a transmitted signal wave. The software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

The method according to the embodiments may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiment, or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -A hardware device specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those produced by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware device described above may be configured to operate as one or more software modules to perform the operation of the embodiment, and vice versa.

As described above, although the embodiments have been described by the limited embodiments and drawings, various modifications and variations are possible from the above description to those of ordinary skill in the art. For example, the described techniques are performed in a different order from the described method, and/or components such as systems, structures, devices, circuits, etc. described are combined or combined in a form different from the described method, or other components Alternatively, even if substituted or substituted by an equivalent, an appropriate result can be achieved.

Therefore, other implementations, other embodiments, and those equivalent to the claims also fall within the scope of the claims to be described later.

Claims (22)

In the product recommendation method in the e-commerce system,
Collecting order history data of a plurality of customers from a plurality of shopping malls based on e-commerce, the order history data including data on at least one product ordered by the customer;
Converting the collected order history data into graph-type data consisting of at least three node layers, the at least three node layers being a first node layer associated with a product, a second node associated with an order Including a layer and a third node layer associated with the customer;
Storing the data in the form of a graph in a database; And
Based on the number of connection paths between at least two nodes of at least one first node layer and at least one third node layer node on the stored graph-type data in response to a query for similar products Including the step of determining and returning the similarity,
Further comprising the step of training an artificial intelligence model to improve the accuracy of the similarity determination,
The second order history data including first order history data including a first node serving as a criterion for querying the similar product and a second node related to the product in which an actual order was placed among the similar products returned by the similarity determination. Generated as a training data set to train the artificial intelligence model,
When calculating the number of connection paths, only the connection paths within the range of the first distance are calculated,
Based on the training data set, the artificial intelligence model is trained while changing the first distance range, which is a critical range for calculating a connection path,
A product recommendation method in an e-commerce system in which an artificial intelligence model is trained while changing whether a class of product category division and a similar category corresponding to the class is used based on the learning data set or whether to use a category by combining. The method of claim 1, wherein the data in the form of a graph
At least one second node of the second node layer is connected around the first node of the first node layer; And,
A product recommendation method in an e-commerce system that is generated in the form of a radial graph as a third node of a third node hierarchy is connected to the at least one second node. The method of claim 1,
Extracted by extracting connection paths within a first distance range among at least one connection paths to which at least two nodes can be connected among nodes of the at least one first node layer and nodes of at least one third node layer A product recommendation method in an e-commerce system that judges that the degree of similarity is high as the number of connection paths increases. The method of claim 3,
The distance between two directly connected nodes is defined as the distance of 1 hop,
The product recommendation method in an e-commerce system having a range in which the first distance range exceeds a 2-hop distance indicating a direct correlation. The method of claim 4,
The first distance range has a range excluding a connection path having a distance greater than or equal to the second distance. The method of claim 1,
At least two of the nodes of the at least one first node layer and the at least one third node layer are:
(i) a first product node and a second product node of the first node hierarchy;
(ii) a first customer node and a second customer node of the third node hierarchy; And
(iii) a first product node of the first node layer and a first customer node of the third node layer;
Product recommendation method in an e-commerce system comprising one of. The method of claim 6, wherein in response to a query for a second product node having similarity to the first product node of (i),
The first among nodes of at least one second node hierarchy of a third customer node of a third node hierarchy associated with the first product node-the first product node is included in the first order node of the second node hierarchy- A product recommendation method in an e-commerce system that considers a connection path between an order node and a product included in another second order node. The method of claim 6, wherein in response to a query for a second customer node similar to the first customer node in (ii),
The second customer is determined as a customer associated with a third product node of a first node hierarchy associated with the first customer node,
The third order node of the second customer-the third order node is a node of the second node hierarchy, and is an order node including the third product node-adopting a product node other than the third product in the former How to recommend products in a commerce system. The method of claim 6, wherein in response to a query for a first customer node similar to the first product node of (iii) or a first product node similar to the first customer node,
A connection path with a customer associated with a third product node purchased in a bundle within a third order node of a second node hierarchy including the first product node; And
A product recommendation method in an e-commerce system, considering a connection path with at least one of bundled products in a fourth order node of a second node hierarchy associated with the first customer node. The method of claim 1,
The products of the first node layer are categorized and defined together with a product category identifier,
In response to a query for a similar product for a first product in the first node hierarchy,
Similar to the first product category identifier of the first product among products associated with at least one of the second node of the second node layer and the third node of the third node layer directly or indirectly connected to the first product. A product recommendation method in an e-commerce system that considers only products of the second product category having a corresponding relationship of as a connection path. The method of claim 1,
The products of the first node are categorized and defined together with a product category identifier,
In response to a query for a similar product for the first product of the first node,
Similar to the first product category identifier of the first product among products associated with at least one of the second node of the second node layer and the third node of the third node layer directly or indirectly connected to the first product. A product recommendation method in an e-commerce system, excluding a connection path with a product of a second product category having a corresponding relationship of. The method of claim 1,
The products of the first node are categorized and defined together with a product category identifier,
In response to a query for a similar product for the first product of the first node,
Based on the first product category identifier of the first product among products associated with at least one of the second node of the second node layer and the third node of the third node layer directly or indirectly connected to the first product, the combination thereof A product recommendation method in an e-commerce system that considers a connection path with a product of a third product category having a corresponding relationship by. The method of claim 1,
In the step of determining and returning the similarity based on the number of connection paths, in calculating the number of connection paths,
The distance between two directly connected nodes is defined as the distance of 1 hop,
E-commerce system comprising the step of adjusting the number of connection paths in consideration of a distance of a connection path between at least two nodes among the at least one first node layer nodes and the at least one third node layer nodes Product recommendation method The method of claim 12,
Between at least one node of the first node layer and at least two nodes of the at least one third node layer, the distance of the first connection path is n _1, and the distance of the second connection path is n ₂ -n ₁ and n _{When 2} is a natural number-
The number of the first connection path is calculated by adjusting to _{1/n 1,}
A product recommendation method in an e-commerce system, wherein the number of the second connection paths _{is adjusted and calculated as 1/n 2.} The method of claim 1,
The node of the first node hierarchy is defined as p,
The node of the third node hierarchy is defined as m
A product in an e-commerce system in which a query for similar products and the connection path between nodes corresponding thereto are defined as a node hierarchy identifier defined as one of m and p and an index number for a specific node hierarchy identifier, and created as a query. Recommended way. delete delete delete The method of claim 1,
Product recommendation method in an e-commerce system further comprising the step of visualizing the data in the form of a graph. In the product recommendation device in an e-commerce system,
A communication unit that communicates with at least one of wired and wireless networks;
By collecting order history data of a plurality of customers from a plurality of shopping malls based on e-commerce through the communication unit-the order history data includes data on at least one product ordered by the customer-and the collected order history Data is composed of at least three node layers, the at least three node layers comprising a first node layer associated with a product, a second node layer associated with an order, and a third node layer associated with a customer. At least two nodes of at least one node of the first node layer and at least one node of the third node layer in response to a query for similar products on the converted graph type data, and on the converted graph type data A processor that determines and returns a degree of similarity based on the number of connection paths therebetween; And
Including a memory for storing the data in the form of the graph,
The processor trains an artificial intelligence model to improve the accuracy of the similarity determination,
First order history data including first order history data including a first node serving as a criterion for querying the similar products, and second order history data including a second node related to the actual ordered product among similar products returned by the similarity determination. The artificial intelligence model is trained by generating it as a training data set,
When calculating the number of connection paths, only the connection paths within the range of the first distance are calculated,
Based on the training data set, the artificial intelligence model is trained while changing the first distance range, which is a critical range for calculating a connection path,
A product recommendation device in an e-commerce system for learning an artificial intelligence model while changing whether a class of product category division and a similar category corresponding to the class is used or whether a category is used by combining based on the learning data set. In the product recommendation method in the e-commerce system,
Collecting inquiry history data of a plurality of customers from a plurality of shopping malls based on an e-commerce transaction, the inquiry history data including a page searched in a specific session and data on a product searched within the searched page;
Converting the collected inquiry history data into graph-type data consisting of at least three node hierarchies, wherein the at least three node hierarchies are a first node layer associated with a product, a second node layer associated with a web page, and a session A third node layer associated with;
Storing the data in the form of a graph in a database; And
On the stored graph-type data, in response to a query for similar products, the first product node of the first node hierarchy is sorted in the order of the highest number of views among products searched on the web page within the range of the first distance before or after the first product node search. Including the step of returning,
Further comprising the step of training an artificial intelligence model to improve the accuracy of the similarity determination,
The first inquiry history data including the first node that is the criterion for querying the similar product and the second inquiry history data including the second node related to the product in which an actual order was placed among the similar products returned by the similarity determination. Generated as a training data set to train the artificial intelligence model,
When calculating the number of connection paths, only the connection paths within the range of the first distance are calculated,
Based on the training data set, the artificial intelligence model is trained while changing the first distance range, which is a critical range for calculating a connection path,
A product recommendation method in an e-commerce system in which an artificial intelligence model is trained while changing whether a class of product category division and a similar category corresponding to the class is used based on the learning data set or whether to use a category by combining. In the product recommendation device in an e-commerce system,
A communication unit that communicates with at least one of wired and wireless networks;
Inquiry history data of a plurality of customers from a plurality of shopping malls based on e-commerce through the communication unit-The inquiry history data includes a page searched in a specific session and data on products searched within the searched page. By collecting the collected inquiry history data at least three node layers-the at least three node layers are a first node layer associated with a product, a second node layer associated with a web page, and a third layer associated with a session. Including a node hierarchy-Converts to graph-type data composed of, and on the converted graph-type data, in response to a query for similar products, the first distance before or after the inquiry of the first product node of the first node hierarchy A processor that sorts and returns the products in the order of the number of views among the products searched on the web page within the range of; And
Including a memory for storing the data in the form of the graph,
The processor trains an artificial intelligence model to improve the accuracy of the similarity determination,
The first inquiry history data including the first node that is the criterion for querying the similar product and the second inquiry history data including a second node related to the product in which an actual order was placed among the similar products returned by the similarity determination. The artificial intelligence model is trained by generating it as a training data set,
When calculating the number of connection paths, only the connection paths within the range of the first distance are calculated,
Based on the training data set, an artificial intelligence model is trained while changing the first distance range, which is a critical range for calculating a connection route,
A product recommendation device in an e-commerce system for learning an artificial intelligence model while changing whether a class of product category division and a similar category corresponding to the class is used or whether a category is used by combining based on the learning data set.

Images