KR20220005328A - Apparatus and method for recommending product based on user behavior pattern and for forecasting user preference for recommended product - Google Patents

Abstract

Disclosed are an apparatus for recommending a product based on a user's behavior pattern and predicting the user's preference for the recommended product, which enhances variety of products recommended to a user, and a method thereof. According to one embodiment of the present invention, the method comprises the following steps: collecting a user's purchase history information and purchase behavior information; acquiring product information for a plurality of candidate recommended products; and determining a recommended product among the plurality of candidate recommended products on the basis of the purchase history information, the purchase behavior information, and the product information. The recommended product determination step determines a recommended product including each extracted product group on the basis of a plurality of different recommendation algorithms to comprehensively reflect the user's purchasing behavior characteristics associated with a plurality of preset customer types.

Description

Apparatus and method for recommending products based on user behavior patterns and predicting user preference for recommended products

The present application relates to an apparatus and method for recommending a product based on a user behavior pattern and predicting a user's preference for the recommended product. In particular, the present application relates to an apparatus and method for recommending a product based on a user behavior pattern, and an apparatus and method for predicting a user's preference for a recommended product.

Online recommendation is a technique that is already widely used. Recommendation algorithms are used not only in familiar services such as showing items of interest to buyers when shopping online or providing a list of related articles when reading news articles, but also in a wide range of places, such as notifying users of online friends that users know on social media. .

The effect of such a recommendation system is more pronounced when the characteristics of the industry (field) to which the recommendation system is applied are reflected. Conventional techniques for a recommendation system in the fashion field that have been made so far have mainly focused on accurately deriving a product having a high similarity with a specific product, such as a product purchased by a user. This method has the disadvantage that it can increase customer fatigue by continuously recommending only products similar to purchased products in the past. There is an aspect that may decrease the quality of recommendations by reducing the range of choices. Furthermore, there is an opinion that the application of a uniform recommendation system that does not consider the diversity of recommendation items can reduce overall consumption of customers.

The background technology of the present application is disclosed in Korean Patent Publication No. 10-2113739.

The present application is to solve the problems of the prior art described above, and the diversity of recommended products provided to the user through hybrid filtering based on different recommendation algorithms according to various customer types linked to the purchasing behavior characteristics of the user. It aims to provide a recommendation system that can be improved.

The present application is to solve the problems of the prior art described above, and considers the purchase history of the user and the characteristics of purchase behavior before purchasing the product, but provides recommended products derived in consideration of various aspects so that the user can The purpose of the present invention is to provide a product recommendation device and method based on user behavior patterns that not only provide cognitive satisfaction by acquiring information about products that have not been touched, but also lead to actual purchases.

The present application is to solve the problems of the prior art described above, and by constructing a preference prediction model by selecting key indicators indicating the user's preference for the product from the vast amount of purchasing behavior information collected from a large number of customers, An object of the present invention is to provide an apparatus and method for predicting a user's preference for a recommended product for which validity can be determined.

However, the technical problems to be achieved by the embodiments of the present application are not limited to the technical problems described above, and other technical problems may exist.

As a technical means for achieving the above technical problem, the product recommendation method based on a user behavior pattern according to an embodiment of the present application includes the steps of collecting the user's purchase history information and purchase behavior information; It may include obtaining product information and determining a recommended product among the plurality of recommended candidate products based on the purchase history information, the purchase behavior information, and the product information.

In addition, in the determining of the recommended product, the recommended product including a product group each derived based on a plurality of different recommendation algorithms to comprehensively reflect the purchasing behavior characteristics of the user associated with a plurality of preset customer types. can decide

In addition, the customer type is a first customer type that reflects a first purchasing behavior characteristic, which is a tendency to purchase products as a gift, and purchases products of a different category from the previously purchased products. The second type of customer that reflects the characteristics of the second purchasing behavior, which is the tendency to It may include at least one of the fourth customer types reflecting the fourth purchasing behavior characteristic, which is a tendency to purchase the purchased product or the product of high interest of others.

Also, the recommendation algorithm corresponding to the first customer type may be a demographic filtering algorithm.

Also, the recommendation algorithm corresponding to the second customer type may be an Apriori algorithm.

Also, the recommendation algorithm corresponding to the third customer type may be an Item-based collaborative filtering algorithm.

Also, the recommendation algorithm corresponding to the fourth customer type may be a user-based collaborative filtering algorithm.

In addition, the determining of the recommended product may include determining a first product group including _{n 1 recommended products based on the demographic filtering algorithm, n 2 based on the Apriori algorithm} Determining a second product group including n recommended products, determining a third product group including _{n 3} recommended products based on the item-based collaborative filtering algorithm, the user-based collaboration _{Determining a fourth product group including n 4} recommended products based on a user-based collaborative filtering algorithm and including the first product group, the second product group, the third product group, and the fourth product group It may include generating a list of recommended products for the recommended products.

Also, n ₁ , n ₂ , n _{3 ,} and n ₄ may be the same number.

Also, the n ₁ , n ₂ , n _{3 ,} and n ₄ may be determined based on the purchasing behavior characteristic of the user derived based on the purchase history information and the purchase behavior information of the user.

In addition, the generating of the recommended product list may include determining the ranking of each recommended product in the recommended product list based on the user's preference prediction result for each of the recommended products in the recommended product list. .

On the other hand, the user's preference prediction method for a recommended product according to an embodiment of the present application includes the steps of preparing learning data including purchase history information and purchase behavior information collected for a first period for a plurality of customers, the plurality of preparing target data including purchase behavior information collected for a second period after the first period for a customer of , based on the learning data, the target data, and purchase history information and purchase behavior information of a predetermined user constructing a preference prediction model for predicting the preference for the recommended product provided to the predetermined user, collecting the purchase history information and purchase behavior information of the target user, and identifying the recommended product provided to the target user and predicting the target user's preference for the identified recommended product through the preference prediction model based on the target user's purchase history information and the target user's purchase behavior information.

In addition, the purchase behavior information may include a user's access history to a predetermined product sold through an online channel.

In addition, the predicting of the preference may include predicting whether the target user will access the recommended product.

In addition, the constructing of the preference prediction model may include extracting a preset feature from the purchase history information and the purchase behavior information, and constructing a preference prediction model for predicting the preference based on the feature.

In addition, the features include features for each of the purchase path stages based on the purchase funnel theory including the Aware stage, the Interest stage, the Desire stage, and the Action stage. can do.

In addition, the step of identifying the recommended product may include identifying the recommended product determined based on the product recommendation method based on the user behavior pattern according to an embodiment of the present application.

On the other hand, the product recommendation apparatus based on a user behavior pattern according to an embodiment of the present application obtains a user information collection unit for collecting purchase history information and purchase behavior information of a user and product information for a plurality of recommended candidate products, and a recommender configured to determine a recommended product from among the plurality of recommended candidate products based on purchase history information, the purchase behavior information, and the product information.

In addition, the recommendation unit may determine the recommended product including each derived product group based on a plurality of different recommendation algorithms to comprehensively reflect the purchasing behavior characteristics of the user associated with a plurality of preset customer types.

On the other hand, the user's preference prediction apparatus for a recommended product according to an embodiment of the present application prepares learning data including purchase history information and purchase behavior information collected for a first period for a plurality of customers, and A data collection unit for preparing target data including purchase behavior information collected for a second period after the first period for a customer, the learning data, the target data, and purchase history information and purchase behavior information of a predetermined user A model generator for constructing a preference prediction model for predicting the preference for the recommended product provided to the predetermined user based on the collected purchase history information and purchase behavior information for the target user, and the recommended product provided to the target user and a predictor configured to predict the target user's preference for the identified recommended product through the preference prediction model based on the target user's purchase history information and the target user's purchase behavior information.

The above-described problem solving means are merely exemplary, and should not be construed as limiting the present application. In addition to the exemplary embodiments described above, additional embodiments may exist in the drawings and detailed description.

According to the above-described problem solving means of the present application, it is possible to improve the diversity of recommended products provided to users through hybrid filtering based on different recommendation algorithms according to various customer types linked to the purchasing behavior characteristics of users. A recommendation system can be provided.

According to the above-described problem solving means of the present application, the user's purchase history and purchasing behavior characteristics until purchasing the product are taken into consideration, but the recommended products derived in consideration of various aspects are comprehensively provided to provide products that the user has not encountered before. It is possible to provide a product recommendation device and method based on user behavior patterns that not only provide cognitive satisfaction by obtaining information about the product, but also lead to actual purchase.

According to the above-described problem solving means of the present application, it is possible to determine the effectiveness of the recommended product provided to the user by selecting a key indicator indicating the user's preference for the product from the vast amount of purchase behavior information collected from a large number of customers and building a preference prediction model. It is possible to provide an apparatus and method for predicting a user's preference for a recommended product.

However, the effects obtainable herein are not limited to the above-described effects, and other effects may exist.

1 is a schematic configuration diagram of a recommendation system according to an embodiment of the present application.
2 is a graph showing a comparison of product recommendation ranges by individual recommendation algorithms and mixed recommendation algorithms corresponding to customer types.
3 is a diagram for explaining features in consideration of a purchase path stage based on a purchase funnel theory used when constructing a preference prediction model.
4 is a graph showing the comparison of the performance according to the detailed model type of the preference prediction model.
5 is a schematic configuration diagram of an apparatus for recommending a product based on a user behavior pattern according to an embodiment of the present application.
6 is a schematic configuration diagram of an apparatus for predicting a user's preference for a recommended product according to an embodiment of the present application.
7 is an operation flowchart of a product recommendation method based on a user behavior pattern according to an embodiment of the present application.
8 is a detailed operation flowchart of a process of determining a recommended product by comprehensively reflecting the characteristics of a user's purchase behavior.
9 is an operation flowchart of a method for predicting a user's preference for a recommended product according to an embodiment of the present application.

Hereinafter, embodiments of the present application will be described in detail with reference to the accompanying drawings so that those of ordinary skill in the art can easily implement them. However, the present application may be implemented in several different forms and is not limited to the embodiments described herein. And in order to clearly explain the present application in the drawings, parts irrelevant to the description are omitted, and similar reference numerals are attached to similar parts throughout the specification.

Throughout this specification, when a part is "connected" with another part, it is not only "directly connected" but also "electrically connected" or "indirectly connected" with another element interposed therebetween. "Including cases where

Throughout this specification, when a member is positioned “on”, “on”, “on”, “on”, “under”, “under”, or “under” another member, this means that a member is positioned on the other member. It includes not only the case where they are in contact, but also the case where another member exists between two members.

Throughout this specification, when a part "includes" a certain component, it means that other components may be further included, rather than excluding other components, unless otherwise stated.

The present application relates to an apparatus and method for recommending a product based on a user behavior pattern and predicting a user's preference for the recommended product. In particular, the present application relates to an apparatus and method for recommending a product based on a user behavior pattern, and an apparatus and method for predicting a user's preference for a recommended product.

1 is a schematic configuration diagram of a recommendation system according to an embodiment of the present application.

Referring to FIG. 1 , a recommendation system 10 according to an embodiment of the present application is a product recommendation device 100 based on a user behavior pattern according to an embodiment of the present application (hereinafter referred to as a 'product recommendation device 100 '). ), a user's preference prediction apparatus 200 for a recommended product according to an embodiment of the present application (hereinafter referred to as a 'preference prediction apparatus 200'), a user terminal 300 and a log DB 400 may include.

The product recommendation device 100 , the preference prediction device 200 , the user terminal 300 , and the log DB 400 may communicate with each other through the network 20 . The network 20 refers to a connection structure in which information exchange is possible between each node, such as terminals and servers, and an example of such a network 20 includes a 3rd Generation Partnership Project (3GPP) network, a long-term LTE (LTE) network. Term Evolution) network, 5G network, WIMAX (World Interoperability for Microwave Access) network, Internet, LAN (Local Area Network), Wireless LAN (Wireless Local Area Network), WAN (Wide Area Network), PAN (Personal Area) Network), a wifi network, a Bluetooth network, a satellite broadcasting network, an analog broadcasting network, a Digital Multimedia Broadcasting (DMB) network, etc. are included, but are not limited thereto.

The user terminal 300 is, for example, a smart phone, a smart pad, a tablet PC, and the like and a PCS (Personal Communication System), GSM (Global System for Mobile communication), PDC (Personal Digital Cellular), PHS ( Personal Handyphone System), PDA (Personal Digital Assistant), IMT (International Mobile Telecommunication)-2000, CDMA (Code Division Multiple Access)-2000, W-CDMA (W-Code Division Multiple Access), Wibro (Wireless Broadband Internet) terminals The same may be any type of wireless communication device.

For example, the product recommendation apparatus 100 may be mounted on the user terminal 300 to provide a product recommendation service provided by the recommendation system 10 . As another example, the product recommendation apparatus 100 may be a product recommendation server 100 provided separately from the user terminal 300 to provide a product recommendation service. Similarly, the preference prediction apparatus 200 is to be mounted on the user terminal 300 to predict and provide the user's preference of the user terminal 300 for the product associated with the product recommendation service provided by the recommendation system 10 . can As another example, the preference prediction apparatus 200 may be a preference prediction server 200 provided separately from the user terminal 300 . In addition, according to the embodiment of the present application, the product recommendation apparatus 100 and the preference prediction apparatus 200 perform detailed functions (eg, It can be understood as functionally dividing the product recommendation function and the preference prediction function for the recommended product.

In addition, in the description of the embodiments of the present application, 'products' may broadly include various goods and tangible/intangible services with real things that can be exchanged (traded) online or offline or gifted (transferred) to other users. . In particular, 'product' in the description of the embodiment of the present application mainly refers to fashion-related items (eg, clothes, clothes, shoes, bags, accessories, etc.) provided to customers related to the field of fashion. However, it is preferable to be interpreted as not limited thereto.

In addition, in the description of the embodiment of the present application, the log DB 400 may refer to a database storing purchase history information, purchase behavior information, metadata for each recommended candidate product, etc. collected for a plurality of customers. In addition, the log DB 400 can collect the collected purchase history information, purchase behavior information, and the like for each user (customer), for example, the user's purchase history information and purchase behavior information of the user terminal 300 . etc. can be kept.

The product recommendation device 100 may collect the user's purchase history information and purchase behavior information. According to an embodiment of the present application, the product recommendation apparatus 100 may collect the purchase history information and purchase behavior information of the user of the user terminal 300 from the user terminal 300 . As another example, the product recommendation apparatus 100 may operate to collect purchase history information and purchase behavior information for a predetermined user from the log DB 400 .

In the description of the embodiment of the present application, purchase history information may include type information, quantity information, price information, brand information, and the like of a product actually purchased by a user (customer). In addition, the purchase history information may include temporal information (eg, purchase date, product receipt date, etc.) on the point in time when the purchase action of the user (customer) occurs. In addition, the purchase history information may include information (eg, refund information, return information, etc.) about a history in which a corresponding purchase action is withdrawn after being purchased by a user (customer).

In addition, in the description of the embodiments of the present application, purchase behavior information includes online or offline behavior history information (eg, offline store visit history, online channel access history, It may include the history of saving/deleting items of interest in the online channel (wish list), shopping cart details, etc. In addition, the purchase behavior information is a record of accessing information about a given product even if the user (customer) has not actually purchased it. (log) information (eg, a history of accessing a brand page in an online channel, a history of accessing a product information page in an online channel, a click history of a product image, etc.) In other words, the purchase behavior information is an online channel It may include a history of a user's access to a predetermined product sold through . For example, the online channel may include an online shopping mall.

The product recommendation apparatus 100 may obtain product information on a plurality of recommended candidate products. For example, the plurality of recommended candidate products may be set to include items handled (selling) in a predetermined online channel accessed through the user terminal 300 .

As another example, the plurality of recommended candidate products is a predetermined product type determined based on a user selection input applied through the user terminal 300 among items handled (selling) in a predetermined online channel connected through the user terminal 300 . It may be set to include a plurality of items belonging to For example, when the user of the user terminal 300 applies a user selection input corresponding to 'bag', the product recommendation apparatus 100 handles (sales) in an online channel accessed through the user terminal 300 A product belonging to the 'bag' category may be selected from among the items and determined as a plurality of recommended candidate products.

As another example, the product recommendation apparatus 100 may obtain product information on a plurality of recommended candidate products separately secured in advance in the log DB 400 irrespective of an online channel accessed through the user terminal 300 .

Here, the product information is, for example, type information, price information, characteristic information, standard information, brand information, manufacturer information, manufacturing date information, country of origin information, material information, producer information, license information, etc. of each recommended candidate product. It can broadly include a variety of relevant information determined by the type of

Also, the product recommendation apparatus 100 may determine a recommended product from among a plurality of recommended candidate products based on the collected user's purchase history information, purchase behavior information, and collected product information for each of the plurality of recommended candidate products.

In this regard, the product recommendation apparatus 100 according to an embodiment of the present application selects each derived product group based on a plurality of different recommendation algorithms to comprehensively reflect the purchasing behavior characteristics of a user associated with a plurality of preset customer types. You can decide which recommended products to include. In other words, the product recommendation device 100 is a product recommendation product including product groups each derived based on at least two different recommendation algorithms so as to provide users with various recommended products in which various purchasing behavior characteristics are comprehensively considered. information can be provided.

Specifically, the plurality of customer types preset herein are associated with a first customer type reflecting a first purchasing behavior characteristic, which is a tendency to purchase a product for gifting to others, and a pre-purchased product. The second type of customer reflects the second purchasing behavior characteristic, which is the tendency to purchase products in a different category, and the third purchase behavior characteristic, which reflects the tendency to repurchase previously purchased products or products with high similarity to the products already owned. It may include at least one of the third customer type and the fourth customer type reflecting the fourth purchasing behavior characteristic, which is a tendency to purchase a product purchased by another person or a product in which another person is interested.

In particular, as an example for better understanding, the first customer type may correspond to a customer who purchases for a gift to someone other than himself/herself, and the second customer type provides additional products to coordinate with previously purchased products. Customers can be responded to, and the third customer type can correspond to customers who continuously purchase products similar to existing purchased products, and the fourth customer type is customers who are highly influenced by purchasing trends. can be matched.

For example, the aforementioned first customer type may be referred to as a Gift type, the second customer type may be referred to as a Coordinator type, the third customer type may be referred to as a carry-over type, and the fourth customer type may be referred to as a Trend setter type, The name of the customer type associated with the purchasing behavior characteristic may be determined in various ways.

Also, referring to FIG. 1 , the product recommendation apparatus 100 may utilize a _{demographic filtering (DF) algorithm A 1 as a recommendation algorithm corresponding to the aforementioned first customer type.} Also, the product recommendation apparatus 100 may utilize the _{Apriori algorithm A 2} as a recommendation algorithm corresponding to the above-described second customer type. In addition, the product recommendation apparatus 100 may utilize an _{Item-based collaborative filtering (IBCF) algorithm (A 3} ) as a recommendation algorithm corresponding to the above-described third customer type. In addition, the product recommendation apparatus 100 may utilize a _{user-based collaborative filtering (UBCF) algorithm (A 4} ) as a recommendation algorithm corresponding to the above-described fourth customer type.

An operation of each of the above-described recommendation algorithms will be described as follows.

The demographic filtering algorithm (A ₁ ) obtains demographic information (eg, gender, age, occupation, region of residence, etc.) of the user to reflect the needs of the first customer type, and the demographics of the user By using the information, a product group that matches the user's profile can be derived. Specifically, according to the demographic filtering algorithm (A ₁ ), the product recommendation device 100 uses the user's gender information and age information and K-means using brand information and product information for a plurality of recommended candidate products as variables. A group corresponding to a user's profile is classified through clustering, and a product that reflects the properties of the group classified as belonging to the user (for example, the number of accesses or purchases for customers of the group during a predetermined period) A product group to be included in the recommended product can be derived by using the product with the highest ranking of , etc.).

In order to reflect the characteristics of the second customer type, the Apriori algorithm A ₂ may determine a recommended product in consideration of the user's purchase time difference. Specifically, the Apriori algorithm (A ₂ ) may operate to quantify how strongly the purchase of one product affects the purchase of another product, and to determine a recommended product in consideration of the quantified luminosity. A large amount of computation is required to consider the correlation between product purchases. In this regard, the Apriori algorithm (A ₂ ) removes patterns that do not pass a predetermined level of correlation numerical threshold, and omits the calculation itself for unnecessary patterns. This has the advantage of reducing the amount of computation and increasing the computation speed. In addition, according to one embodiment of the present application, the product recommendation device 100 is a current season (eg, a season to provide information on recommended products) so that the purchase pattern according to the season can richly reflect the characteristics of the main item. Analysis can be performed based on the purchase history information and purchase behavior information collected for the season corresponding to the . In addition, in order to quickly quantify the correlation between product purchases, a predetermined threshold value (eg, 0.4, etc.) for the minimum value of the probability of purchasing two products at the same time may be applied.

Since the collaborative filtering technique related to the item-based collaborative filtering algorithm (A ₃ ) and the user-based collaborative filtering algorithm (A ₄ ) is obvious to those skilled in the art, a detailed description thereof will be omitted.

According to an embodiment of the present application, the product recommendation device 100 selects customers of a predetermined top percentage (eg, top 20%) based on sales in order to reduce the amount of computation when applying the _{item-based collaborative filtering algorithm (A 3 ).} By setting it as a population for analysis, a product with a high degree of similarity between products can be provided as a recommended product. In addition, according to an embodiment of the present application, the product recommendation apparatus 100 applies a dimension reduction associated with a customer's purchase pattern when applying the _{user-based collaborative filtering algorithm (A 4 ), and for calculating the similarity between customers, Jaccard (Jaccard)} ) technique can be used, but is not limited thereto.

Also, the product recommendation apparatus 100 may determine a first product group including _{n 1} recommended products based on a _{demographic filtering algorithm A 1 .} Also, the product recommendation apparatus 100 may determine a second product group including _{n 2} recommended products based on the _{Apriori algorithm A 2 .} Also, the product recommendation apparatus 100 may determine a third product group including _{n 3} recommended products based on an _{item-based collaborative filtering algorithm (A 3 ).} Also, the product recommendation apparatus 100 may determine a fourth product group including n ₄ recommended products based on a user-based collaborative filtering algorithm (A _{4 ).}

Also, the product recommendation apparatus 100 may generate a recommended product list for recommended products including the determined first product group, second product group, third product group, and fourth product group. In other words, the product recommendation apparatus 100 may generate a list of recommended products based on hybrid filtering to enhance diversity of recommended products while reflecting all the advantages of an algorithm that reflects each customer characteristic.

2 is a graph showing a comparison of product recommendation ranges by individual recommendation algorithms and mixed recommendation algorithms corresponding to customer types.

Referring to FIG. 2 , according to the hybrid filtering, it can be seen that the prediction coverage is high at a level of 78% compared to the case where each recommendation algorithm is individually applied. Here, 'Prediction coverage' is the size of the overall item set compared to the size of the recommended item set (eg, the number of products included in the recommended product in the present application) (eg, the number of products of the recommended candidate product in the present application) This is a concept used as an index to evaluate the diversity of recommendation algorithms. As such, according to the hybrid filtering of the product recommendation apparatus 100 according to an embodiment of the present application, it can be confirmed that the inherent disadvantages of the individual product recommendation algorithm can be supplemented and diversity can be maximized.

Also, the product recommendation apparatus 100 may provide the generated recommended product list to the user terminal 300 .

According to an embodiment of the present application, with respect to the number of recommended products included in the first to fourth product groups, n ₁ , n ₂ , n _{3 ,} and n ₄ may be equal (same number). For example, the product recommendation apparatus 100 may be implemented to provide a list of 20 recommended products including a first product group to a fourth product group each including 5 recommended products, but is not limited thereto. As another example, the number of recommended products included in each of the first to fourth product groups may be changed based on a customizing input for a recommendation method applied to the user terminal 300 .

As another example, according to an embodiment of the present application, the product recommendation device 100 sets the number of recommended products (in other words, n ₁ , n ₂ , n ₃ and n ₄ ) included in the first to fourth product groups to the user. It can be determined based on the purchase behavior characteristics of the user derived based on the purchase history information and purchase behavior information. For example, when it is determined that the user's purchasing behavior characteristics match the first customer type rather than the fourth customer type, the product recommendation device 100 sets n ₁ to a _{value greater than n 4} to generate a list of recommended products. can

In addition, according to an embodiment of the present application, the product recommendation device 100 provides recommended products included in the first to fourth product groups, and the user's purchase derived based on the user's purchase history information and purchase behavior information. It is possible to determine the ranking (order) of each of the recommended products in the list of recommended products based on the behavioral characteristics. For example, if the product recommendation apparatus 100 determines that the user's purchasing behavior characteristics match the first customer type rather than the fourth customer type, the recommended product included in the first product group is the recommended product included in the fourth product group It is possible to determine the ranking of the recommended product to be exposed in advance.

As another example, according to an embodiment of the present application, the product recommendation apparatus 100 may determine the ranking of the recommended products so that the recommended products that are overlapped by two or more recommendation algorithms are exposed first compared to the other recommended products. For example, when product A is simultaneously included in the first product group and the second product group, product A may be determined as the first priority in the recommended product list.

In addition, according to an embodiment of the present application, the product recommendation apparatus 100 may determine the ranking of each recommended product in the recommended product list based on the user's preference prediction result for each of the recommended products in the generated recommended product list. For example, the product recommendation device 100 selects the recommended product list so that the recommended product predicted to have high preference based on the user's preference prediction result of each recommended product in the list is placed at the beginning of the list and is preferentially exposed to the user. can create In this regard, the product recommendation apparatus 100 may obtain a preference prediction result of a user of each recommended product from the preference prediction apparatus 200 described below.

The preference prediction apparatus 200 may prepare _{the learning data D 1} including purchase history information and purchase behavior information collected during the first period for a plurality of customers. For example, the preference prediction apparatus 200 may obtain the training data D ₁ from the log DB 400 . In particular, the learning data (D ₁ ) is prepared to include offline-based purchase history information and online-based purchase history information to richly derive the propensity and patterns of a plurality of customers reflected in multi-channel can be Illustratively, in an experimental example related to the preference prediction apparatus 200 according to an embodiment of the present application, the learning data D ₁ was prepared using the period from September 2017 to August 2018 as the first period. .

According to an embodiment of the present application, in particular, in order to fully reflect the trend of the entire season for products of industries (eg, fashion industry, etc.) in which customers' purchasing patterns change by season, it is linked with the _{learning data (D 1 )} The first period may be set to a period of one year or more.

In addition, the preference prediction apparatus 200 may prepare _{target data D 2} including purchase behavior information collected during the second period after the first period for a plurality of customers associated with the _{learning data D 1 .} have. For example, the preference prediction apparatus 200 may obtain the target data D ₂ from the log DB 400. In this regard, the target data D ₂ is a specific product sold through an online channel. may mean each of a plurality of customers' access histories (for example, whether product information is clicked, etc.) to Assigned to 1 if there is access history to the product (in other words, if there is a record of clicking on the product), and 0 if there is no access history to the product (in other words, if the product has never been clicked) It is possible to prepare the binary data that becomes the target data D _2. Illustratively, in an experimental example related to the preference prediction apparatus 200 according to an embodiment of the present application, from September 2018 to November 2018 _{The target data D 2} was prepared by using the period up to the month as the second period.

In addition, the preference prediction apparatus 200 is based on the prepared learning data (D ₁ ), the target data (D ₂ ), and the purchase history information and purchase behavior information of the predetermined user, the preference for the recommended product provided to the predetermined user. It is possible to build a preference prediction model (M) to predict.

Specifically, according to an embodiment of the present application, the preference prediction apparatus 200 extracts preset features from the purchase history information and purchase behavior information included in the _{learning data D 1 , and based on the extracted features} Thus, it is possible to build a preference prediction model that predicts the user's preference to match the target data D2.

3 is a diagram for explaining features in consideration of a purchase path stage based on a purchase funnel theory used when constructing a preference prediction model.

Referring to FIG. 3 , the preference prediction device 200 is a purchase path based on a purchase funnel theory including an awareness stage, an interest stage, a desire stage, and an action stage. You can extract features for each step.

In addition, referring to FIG. 3 , the preference prediction apparatus 200 determines whether the user's needs and the product match in relation to the awareness stage, the user's last access (click) date, and the pages accessed by the user during the past month. Four features can be extracted, including, but not limited to, the number of page views and the number of pages accessed by users in the past 3 months (Number of page views). In addition, the preference prediction device 200 in relation to the interest (Interest) stage, product and brand information with a history stored in the user's shopping carts for the past 1 month and the user's shopping carts for the past 3 months Two features including product and brand information with a stored history may be extracted, but the present invention is not limited thereto. In addition, the preference prediction apparatus 200 may extract three features including the period up to the last purchase date, the purchase amount for the past 1 month, and the purchase amount for the past 3 months in relation to the Desire stage, but is not limited. In addition, the preference prediction device 200 includes information on brands purchased in the past 1 month, brand information purchased in the past year, product information purchased in the past 1 month, and information on products purchased in the past year in relation to the action stage. Four features including product information may be extracted, but the present invention is not limited thereto.

4 is a graph showing the comparison of the performance according to the detailed model type of the preference prediction model.

4 , the preference prediction model is based on artificial intelligence models such as XGBoost (eXtreme Gradient Boosting), Random Forest (RF), K-NN, and Logistic Regression, and the preference prediction model (M) can be built In this regard, in particular, referring to FIG. 4, the prediction performance of the XBGoost-based preference prediction model (M) among the aforementioned artificial intelligence models is all in terms of Accuracy, F1-score, Precision, and Recall. It can be seen that the best performance is shown. Accordingly, according to an embodiment of the present application, the preference prediction apparatus 200 may construct the XBGoost-based preference prediction model M in consideration of prediction performance.

Also, the preference prediction apparatus 200 may collect purchase history information and purchase behavior information of the target user. Specifically, the preference prediction apparatus 200 may collect purchase history information and purchase behavior information of the user of the user terminal 300 from the user terminal 300 . As another example, the preference prediction apparatus 200 may operate to collect purchase history information and purchase behavior information for a predetermined user from the log DB 400 or the product recommendation apparatus 100 . In other words, the target user associated with the preference prediction apparatus 200 may be set as a user provided with information on the recommended product by the above-described product recommendation apparatus 100 .

Also, the preference prediction apparatus 200 may identify a recommended product provided to the target user. For example, the preference prediction apparatus 200 may receive a recommended product list including a plurality of recommended products from the product recommendation apparatus 100 . As another example, the preference prediction apparatus 200 may identify information on recommended products including at least some of items handled (selling) in a predetermined online channel.

In addition, the preference prediction apparatus 200 may predict the target user's preference for each recommended product identified through the preference prediction model M based on the collected target user's purchase history information and the target user's purchase behavior information. . Specifically, the preference prediction apparatus 200 may predict whether the target user will access each of the recommended products (eg, whether to click on the recommended product, whether to access the page for the recommended product, etc.) have.

5 is a schematic configuration diagram of an apparatus for recommending a product based on a user behavior pattern according to an embodiment of the present application.

Referring to FIG. 5 , the product recommendation apparatus 100 may include a user information collection unit 110 and a recommendation unit 120 .

The user information collection unit 110 may collect purchase history information and purchase behavior information of the user.

The recommendation unit 120 may obtain product information on a plurality of recommended candidate products. Also, the recommendation unit 120 may determine a recommended product from among a plurality of recommended candidate products based on the collected purchase history information, purchase behavior information, and product information.

In addition, according to an embodiment of the present application, the recommendation unit 120 includes a product group derived based on a plurality of different recommendation algorithms to comprehensively reflect the purchasing behavior characteristics of a user associated with a plurality of preset customer types. You can decide which products to recommend.

6 is a schematic configuration diagram of an apparatus for predicting a user's preference for a recommended product according to an embodiment of the present application.

Referring to FIG. 6 , the preference prediction apparatus 200 may include a data collection unit 210 , a model generation unit 220 , and a prediction unit 230 .

The data collection unit 210 may prepare _{the learning data D 1} including purchase history information and purchase behavior information collected for a first period for a plurality of customers. Also, the data collection unit 210 may prepare _{target data D 2} including purchase behavior information collected for a second period after the first period for a plurality of customers.

The model generator 220 predicts a preference for a recommended product provided to a predetermined user based on the _{prepared learning data D 1} , the target data D _{3 , and the purchase history information and purchase behavior information of the predetermined user} It is possible to build a preference prediction model (M).

The prediction unit 230 may collect purchase history information and purchase behavior information for the target user. Also, the prediction unit 230 may identify a recommended product provided to the target user. Also, the prediction unit 230 may predict the target user's preference for each of the recommended products identified through the preference prediction model M based on the collected target user's purchase history information and the target user's purchase behavior information.

Hereinafter, an operation flow of the present application will be briefly reviewed based on the details described above.

7 is an operation flowchart of a product recommendation method based on a user behavior pattern according to an embodiment of the present application.

The product recommendation method based on the user behavior pattern shown in FIG. 7 may be performed by the product recommendation apparatus 100 described above. Accordingly, even if omitted below, the description of the product recommendation apparatus 100 may be equally applied to the description of the product recommendation method based on the user behavior pattern.

Referring to FIG. 7 , in step S11 , the user information collection unit 110 may collect the user's purchase history information and purchase behavior information.

Next, in step S12 , the recommendation unit 120 may acquire product information on a plurality of recommended candidate products.

Next, in step S13 , the recommendation unit 120 may determine a recommended product from among a plurality of recommended candidate products based on purchase history information, purchase behavior information, and product information.

In addition, in step S13 , the recommendation unit 120 determines a recommended product including each derived product group based on a plurality of different recommendation algorithms to comprehensively reflect the purchasing behavior characteristics of the user associated with a plurality of preset customer types. can

In the above description, steps S11 to S13 may be further divided into additional steps or combined into fewer steps, according to an embodiment of the present application. In addition, some steps may be omitted if necessary, and the order between steps may be changed.

8 is a detailed operation flowchart of a process of determining a recommended product by comprehensively reflecting the characteristics of a user's purchase behavior.

The process of determining a recommended product by comprehensively reflecting the purchasing behavior characteristics of the user shown in FIG. 8 may be performed by the product recommendation apparatus 100 described above. Accordingly, even if omitted below, the description of the product recommendation apparatus 100 may be equally applied to the description of FIG. 8 .

Referring to FIG. 8 , in step S131 , the recommendation unit 120 may determine a first product group including _{n 1 recommended products based on a demographic filtering algorithm.}

Next, in step S132 , the recommendation unit 120 may determine a second product group including _{n 2 recommended products based on an Apriori algorithm.}

Next, in step S133 , the recommendation unit 120 may determine a third product group including _{n 3 recommended products based on an item-based collaborative filtering algorithm.}

Next, in step S134 , the recommendation unit 120 may determine a fourth product group including _{n 4 recommended products based on a user-based collaborative filtering algorithm.}

Next, in step S135 , the recommendation unit 120 may generate a list of recommended products for the recommended products including the first product group, the second product group, the third product group, and the fourth product group.

Next, in step S136, the recommendation unit 120 may determine the ranking of each recommended product in the recommended product list based on the user's preference prediction result for each of the recommended products in the recommended product list.

In the above description, steps S131 to S136 may be further divided into additional steps or combined into fewer steps according to an embodiment of the present application. In addition, some steps may be omitted if necessary, and the order between steps may be changed.

9 is an operation flowchart of a method for predicting a user's preference for a recommended product according to an embodiment of the present application.

The user's preference prediction method for the recommended product shown in FIG. 9 may be performed by the preference prediction apparatus 200 described above. Accordingly, even if omitted below, the description of the preference prediction apparatus 200 may be equally applied to the description of the user's preference prediction method for the recommended product.

Referring to FIG. 9 , in step S21 , the data collection unit 210 may prepare _{learning data D 1} including purchase history information and purchase behavior information collected for a first period for a plurality of customers.

Next, in step S22 , the data collection unit 210 may prepare _{target data D 2} including purchase behavior information collected for a second period after the first period for a plurality of customers.

Next, in step S23, the model generation unit 220, based on the prepared learning data, target data, and purchase history information and purchase behavior information of the predetermined user, the preference for predicting the preference for the recommended product provided to the predetermined user A predictive model (M) can be built.

Next, in step S24, the prediction unit 230 may collect purchase history information and purchase behavior information of the target user.

Next, in step S25, the prediction unit 230 may determine the recommended product provided to the target user.

Next, in step S26, the prediction unit 230, based on the purchase history information of the target user and the purchase behavior information of the target user, the target user's preference for the recommended product identified through the preference prediction model (M) predictable.

In the above description, steps S21 to S26 may be further divided into additional steps or combined into fewer steps, according to an embodiment of the present application. In addition, some steps may be omitted if necessary, and the order between steps may be changed.

The product recommendation method based on the user behavior pattern and the user's preference prediction method for the recommended product according to an embodiment of the present application may be implemented in the form of a program command 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, etc. alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the present invention, or may be known and available to those skilled in the art of computer software. Examples of the computer-readable recording medium 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 floppy disks. - includes magneto-optical media, and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

In addition, the above-described method for recommending a product based on a user behavior pattern and a method for predicting a user's preference for a recommended product may be implemented in the form of a computer program or application executed by a computer stored in a recording medium.

The foregoing description of the present application is for illustration, and those of ordinary skill in the art to which the present application pertains will understand that it can be easily modified into other specific forms without changing the technical spirit or essential features of the present application. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. For example, each component described as a single type may be implemented in a dispersed form, and likewise components described as distributed may be implemented in a combined form.

The scope of the present application is indicated by the following claims rather than the above detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts should be construed as being included in the scope of the present application.

10: Recommendation system
100: product recommendation device based on user behavior pattern
110: user information collection unit
120: recommendation
200: A device for predicting user preference for recommended products
210: data collection unit
220: model generation unit
230: prediction unit
300: user terminal
400: log DB

Claims (14)

In the product recommendation method based on user behavior pattern,
collecting the user's purchase history information and purchase behavior information;
obtaining product information for a plurality of recommended candidate products; and
determining a recommended product among the plurality of recommended candidate products based on the purchase history information, the purchase behavior information, and the product information;
including,
The step of determining the recommended product,
The product recommendation method, wherein the recommended product including each derived product group is determined based on a plurality of different recommendation algorithms to comprehensively reflect the purchasing behavior characteristics of the user associated with a plurality of preset customer types. According to claim 1,
The customer type is
The first type of customer reflects the characteristics of the first purchasing behavior, which is the tendency to purchase products for gifting to others, and the second purchasing behavior, which is related to the previously purchased product but tends to purchase a product in a different category from the previously purchased product. The second type of customer that reflects the characteristics, the third type of customer that reflects the characteristics of the third purchasing behavior, which is the tendency to repurchase previously purchased products or products with high similarity to the products already purchased, and the degree of interest in products purchased by others or others The method for recommending a product, wherein the method includes at least one of the fourth customer types reflecting the fourth purchasing behavior characteristic, which is a tendency to purchase a high product. 3. The method of claim 2,
the recommendation algorithm corresponding to the first customer type is a demographic filtering algorithm;
The recommendation algorithm corresponding to the second customer type is an Apriori algorithm,
The recommendation algorithm corresponding to the third customer type is an Item-based collaborative filtering algorithm,
Wherein the recommendation algorithm corresponding to the fourth customer type is a user-based collaborative filtering algorithm, product recommendation method. 4. The method of claim 3,
The step of determining the recommended product,
determining a first product group including _{n 1} recommended products based on the demographic filtering algorithm;
determining a second product group including _{n 2} recommended products based on the Apriori algorithm;
determining a third product group including _{n 3} recommended products based on the item-based collaborative filtering algorithm;
determining a fourth product group including _{n 4} recommended products based on the user-based collaborative filtering algorithm; and
generating a list of recommended products for the recommended products including the first product group, the second product group, the third product group, and the fourth product group;
A method of recommending a product that includes a. 5. The method of claim 4,
Wherein n ₁ , n ₂ , n ₃ and n ₄ are the same number, the product recommendation method. 5. The method of claim 4,
Wherein the n ₁ , n ₂ , n _{3 ,} and n ₄ are determined based on the purchasing behavior characteristic of the user derived based on the purchase history information and the purchase behavior information of the user. 5. The method of claim 4,
The step of generating the list of recommended products,
determining a ranking of each recommended product in the recommended product list based on the user's preference prediction result for each of the recommended products in the recommended product list;
A method of recommending a product that includes a. In a method for predicting user preference for a recommended product,
Preparing learning data including purchase history information and purchase behavior information collected during a first period for a plurality of customers;
preparing target data including purchase behavior information collected during a second period after the first period for the plurality of customers;
constructing a preference prediction model for predicting a preference for a recommended product provided to the predetermined user based on the learning data, the target data, and purchase history information and purchase behavior information of a predetermined user;
collecting purchase history information and purchase behavior information of the target user;
identifying a recommended product provided to the target user; and
Predicting the target user's preference for the identified recommended product through the preference prediction model based on the target user's purchase history information and the target user's purchase behavior information;
Including, preference prediction method. 9. The method of claim 8,
The purchase behavior information is
A method for predicting preference that includes a user's access history to a given product sold through an online channel. 10. The method of claim 9,
Predicting the preference comprises:
Predicting whether the target user will access the recommended product, preference prediction method. 11. The method of claim 10,
Building the preference prediction model comprises:
Extracting a preset feature from the purchase history information and the purchase behavior information, and constructing a preference prediction model for predicting the preference based on the feature,
The feature is
Preference prediction, which includes features for each of the purchase path stages based on the purchase funnel theory, including the Aware stage, the Interest stage, the Desire stage and the Action stage. Way. 9. The method of claim 8,
The step of identifying the recommended product is,
The method for predicting preference, which is to identify the recommended product determined based on the product recommendation method based on the user behavior pattern according to claim 1 . In the product recommendation device based on user behavior pattern,
a user information collection unit that collects the user's purchase history information and purchase behavior information; and
a recommendation unit for obtaining product information on a plurality of recommended candidate products, and determining a recommended product among the plurality of recommended candidate products based on the purchase history information, the purchase behavior information, and the product information;
including,
The recommendation is
The product recommendation apparatus, which determines the recommended product including each derived product group based on a plurality of different recommendation algorithms to comprehensively reflect the purchasing behavior characteristics of the user associated with a plurality of preset customer types. In the user's preference prediction device for recommended products,
Prepare learning data including purchase history information and purchase behavior information collected for a first period for a plurality of customers, and include purchase behavior information collected for a second period after the first period for the plurality of customers a data collection unit for preparing target data;
a model generator for constructing a preference prediction model for predicting a preference for a recommended product provided to the predetermined user based on the learning data, the target data, and purchase history information and purchase behavior information of a predetermined user; and
Collect purchase history information and purchase behavior information for a target user, identify a recommended product provided to the target user, and use the preference prediction model based on the target user's purchase history information and the target user's purchase behavior information a predictor for predicting the target user's preference for the identified recommended product;
Including, preference prediction device.

Images