KR102049777B1 - Item recommendation method and apparatus based on user behavior - Google Patents

Abstract

Obtaining information on purchase details and selection details of items of each of the plurality of users through the online shopping server; Estimating a preference of each of the plurality of users for each of the items based on the purchase history and the selection history for the items; And transmitting the information of the unpurchased item of the predetermined user to the terminal of the predetermined user based on a preference of each of the items of the predetermined user among the plurality of users. An item recommendation method by an item recommendation apparatus is disclosed.

Description

ITEM RECOMMENDATION METHOD AND APPARATUS BASED ON USER BEHAVIOR}

The present invention relates to the field of online shopping. More specifically, the present invention relates to a method and apparatus for recommending items to a user based on the user's behavior in an online environment.

In recent years, as the online shopping market has increased, the competition among e-commerce companies to attract more users has become very intense. Many e-commerce companies recommend customized items to users by understanding the user's online shopping behavior and determining the characteristics.

In order to accurately predict a user's future purchases, the user's past purchase history and explicit user feedback on that purchase were required. Explicit user feedback may include a user's rating and review of the item. Because this explicit user feedback clearly expresses the user's preferences, it is very important to predict the future purchase of the user. However, there is a problem that such explicit feedback is very small. There are still users who purchase items only offline, and users who purchase items online rarely leave a rating or review. In addition, requesting a rating or review from the user is a burden on the user, and thus may be a factor that lowers the user's desire to purchase. Thus, e-commerce companies must use a limited amount of implicit user feedback. Implicit user feedback refers to a user's past purchase history.

Most studies of predicting future purchases of users have used implicit user feedback to model user preferences. For example, Rendle's work (Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt-Thieme. 2009. BPR: Bayesian personalized ranking from implicit feedback.In Proceedings of the twenty-fifth conference on uncertainty in artificial intelligence.AUAI Press According to.), Each user will prefer a purchased item over all unpurchased items, a pair-wise running-to-ranking called Bayesian Personalized Ranking (BPR). to-rank) method. However, Rendle's research has a problem of over-interpreting the effects of unpurchased items considering the items that the user has not purchased as negative feedback. In addition, according to Rendle's research, the accuracy of the user's purchase history data may still be insufficient.

An object recommendation method and apparatus based on user behavior according to an embodiment of the present invention aims to accurately predict a user's preference for an item.

In addition, a method and apparatus for recommending items based on user behavior according to an embodiment of the present invention aims to increase profits of an operator operating an online shopping business.

In addition, a method and apparatus for recommending items based on user behavior according to an exemplary embodiment of the present invention aims to improve shopping satisfaction of a user by recommending items required by the user to the user.

In addition, an object recommendation method and apparatus based on user behavior according to an embodiment of the present invention aims to reduce the burden on a user searching for an item to purchase.

Item recommendation method according to an embodiment of the present invention,

Obtaining information on purchase details and selection details of items of each of the plurality of users through the online shopping server; Estimating a preference of each of the plurality of users for each of the items based on a purchase history and a selection history for the items; And transmitting information on an unpurchased item of the predetermined user to a terminal of the predetermined user based on a preference of each of items of the predetermined user among the plurality of users.

Item recommending apparatus according to another embodiment of the present invention,

An information obtaining unit obtaining information on purchase details and selection details of items of each of the plurality of users through the online shopping server; A preference estimator for estimating a preference of each of the plurality of users for each of the items based on a purchase history and a selection history for the items; And a communication unit configured to transmit information of the unpurchased item of the predetermined user to a terminal of the predetermined user, based on a preference of each of items of the predetermined user among the plurality of users.

The method and apparatus for recommending items based on user behavior according to an embodiment of the present invention can accurately predict users' preferences for items.

In addition, the method and apparatus for recommending items based on user behavior according to an embodiment of the present invention may increase profits of an operator operating online shopping.

In addition, the method and apparatus for recommending items based on user behavior according to an embodiment of the present invention may improve the shopping satisfaction of the user by recommending items required by the user to the user.

In addition, the method and apparatus for recommending items based on user behavior according to an embodiment of the present invention may reduce the burden on a user searching for an item to purchase.

However, the effects that can be achieved by the method and apparatus for recommending items based on user behavior according to an embodiment of the present invention are not limited to those mentioned above, and other effects that are not mentioned are not described in the following description. It will be clearly understood by those skilled in the art.

1 is an exemplary diagram illustrating an environment to which an item recommendation apparatus according to an embodiment of the present invention is applied.
2 is a flowchart illustrating an item recommendation method according to an embodiment of the present invention.
FIG. 3 is a diagram for describing purchase details and selection details for all items of a plurality of users.
4 is an exemplary diagram illustrating a first feature matrix and a second feature matrix applicable to an item recommendation method according to an embodiment of the present invention.
5 is a block diagram illustrating a configuration of an item recommending apparatus according to an embodiment of the present invention.
6 is a test result for comparing the effects of the item recommendation method and another method according to an embodiment of the present invention.

The present invention may be variously modified and have various embodiments, and specific embodiments are illustrated in the drawings and described in detail with reference to the accompanying drawings. However, this is not intended to limit the present invention to specific embodiments, it should be understood that the present invention includes all modifications, equivalents, and substitutes included in the spirit and scope of the present invention.

In describing the present invention, when it is determined that the detailed description of the related known technology may unnecessarily obscure the subject matter of the present invention, the detailed description thereof will be omitted. In addition, the numbers (eg, first, second, etc.) used in the description process of the present specification are merely identification symbols for distinguishing one component from another component.

In addition, in the present specification, when one component is referred to as "connected" or "connected" with another component, the one component may be directly connected or directly connected to the other component, but in particular It is to be understood that, unless there is an opposite substrate, it may be connected or connected via another component in the middle.

In addition, in the present specification, the components represented by '~ unit (unit)', 'module', etc., two or more components are combined into one component or two or more components for each functionalized function May be differentiated into In addition, each component to be described below may additionally perform some or all of the functions of other components in addition to the main functions of which they are responsible, and some of the main functions of each of the components may be different. Of course, it may be carried out exclusively by the component.

In addition, in the present specification, the 'item' is meant to encompass tangible and intangible goods that a user can purchase online. For example, the 'item' is not only a tangible product such as clothing, foodstuffs, home appliances, etc. And intangible goods such as movie tickets, theater tickets, performance tickets, and service contracts.

Hereinafter, exemplary embodiments according to the spirit of the present invention will be described with reference to the drawings.

1 is an exemplary diagram illustrating an environment to which an item recommendation apparatus 100 according to an embodiment of the present invention is applied.

Referring to FIG. 1, the user terminal 200 may access the online shopping server 300 to search for, select, and purchase an item through a web page. For example, the user terminal 200 may check the detailed information of the item by selecting an item displayed on the screen of the web page, and may receive an item list corresponding to the search word input by the user through the web page. In addition, the user may select an item desired by the user terminal 200 and purchase the item through a payment process.

The item recommendation apparatus 100 may obtain information such as a user's behavior in the online shopping server 300, that is, purchase of an item, selection of an item, and the like, and recommend the user-customized item to the user terminal 200. The item recommendation apparatus 100 may recommend a user customized item directly to the user terminal 200, or may recommend the user customized item through a web page provided by the online shopping server 300 to the user terminal 200. have.

Although FIG. 1 illustrates only one user terminal 200 and one online shopping server 300, this is only one embodiment, and each of the plurality of user terminals 200 is connected to the plurality of online shopping servers 300. The item recommendation apparatus 100 may obtain information on an item purchase behavior, an item selection behavior, etc. of the plurality of users by accessing the plurality of online shopping servers 300.

In addition, although FIG. 1 shows that the item recommendation apparatus 100 is physically separated from the online shopping server 300, this is only one embodiment, and the item recommendation apparatus 100 may be connected to the online shopping server 300. It can also be built in.

In the exemplary embodiment of the present invention, the user terminal 200 may be a device that may be connected to the online shopping server 300 through a network such as a desktop PC, a tablet PC, a laptop, a wearable device, as well as the illustrated smartphone. May correspond to the user terminal 200.

The user terminal 200, the online shopping server 300, and the item recommendation apparatus 100 illustrated in FIG. 1 may be connected through a network, where the network may include a wired network and a wireless network. For example, the network may include various networks such as a local area network (LAN), a metropolitan area network (MAN), and a wide area network (WAN). The network may also include a known World Wide Web (WWW). However, the network according to the embodiment of the present invention is not limited to the networks listed above, and may include at least part of a known wireless data network, a known telephone network, and a known wired / wireless television network.

In one embodiment of the present invention, as in the conventional research, implicit feedback of the user, that is, information of items purchased by the user and information of items not purchased by the user, as well as information of items selected by the user but not purchased. You can use it together to recommend custom items. Here, the user's selection means that the user selects the item link displayed on the web page by clicking, touching, or the like in order to check or pay the information of the item.

Hereinafter, an item recommendation method according to an embodiment of the present invention will be described in detail with reference to FIG. 2.

2 is a flowchart illustrating an item recommendation method according to an embodiment of the present invention.

In operation S210, the item recommendation apparatus 100 obtains information on purchase details and selection details of all items of each of the plurality of users through the online shopping server 300.

The entire item may mean all of the items sold by the online shopping server 300, or may mean an item belonging to a specific group (eg, a clothing group, a foodstuff group, or a household appliance group). The operator can select variously the entire item, which can be a recommendation candidate item for the user.

The item recommendation apparatus 100 is a purchase history and a selection history of a plurality of users, and a purchase item purchased by each of the plurality of users, a first non-purchased item selected but not purchased, and a second non-purchase not selected or purchased. Information about the item may be obtained.

FIG. 3 illustrates information on purchase details and selection details of a plurality of users as a matrix. As illustrated in FIG. 3, each user may be mapped to a row of the matrix, and each item may be mapped to a column of the matrix. . For user 1, item 1 and item M correspond to a purchased item, but item 2 corresponds to a second non-purchased item that is not selected. In addition, item 1 corresponds to the first non-purchased item that is not purchased after selection to user 2, item 2 corresponds to the purchased item, and item M corresponds to the second non-purchased item.

Referring back to FIG. 2, in step S220, the item recommendation apparatus 100 estimates the preferences of each of the plurality of users for each of the entire items, based on the purchase details and the selection details of the entire items.

According to an embodiment of the present invention, each item is assumed to be higher than the preference for the first non-purchased item and the preference for the first non-purchased item is higher than the preference for the second non-purchased item. It is possible to estimate the user's preference for. In other words, the preference of the purchased item purchased by the user is higher than the preference of the first non-purchased item not selected and then purchased, and the second non-purchased item selected by the user but not selected is not preferred. Would be higher than the preference.

A method of estimating the preference for the entire item of each of the plurality of users by the item recommendation apparatus 100 will be described with reference to FIG. 4.

4 is an exemplary diagram illustrating a first feature matrix 410 and a second feature matrix 420 applicable to an item recommendation method according to an embodiment of the present invention. The item recommendation apparatus 100 may include a plurality of users. After learning each element of the first feature matrix 410 for each of the elements and the second feature matrix 420 for the entire items, the first feature matrix 410 and the second feature matrix 420 Multiply by to estimate the preferences for each item of each user.

For example, the last learned elements a11, a12 corresponding to user 1 in the first feature matrix 410 and the last learned elements b11, b21 corresponding to item 1 in the second feature matrix 420 are multiplied (ie, a11b11 + a12b21) and user 1's preference for item 1 can be estimated.

In an embodiment of the present invention, in learning the elements of the first feature matrix 410 and the elements of the second feature matrix 420, a pair-wise solution for solving a ranking problem based on a matrix decomposition technique. matrix decomposition techniques can be used.

As described above, the preference for the purchased item is higher than the preference for the first non-purchased item, and the preference for the first non-purchased item is higher than the preference for the second non-purchased item. An element of the first feature matrix 410 such that the difference in preference for the first non-purchased item is maximized and the difference in preference for the first non-purchased item and the preference for the second non-purchased item is maximized. And elements of the second feature matrix 420.

Specifically, an element of the first feature matrix 410 corresponding to each user (an element belonging to the row corresponding to each user) and an element of the second feature matrix 420 corresponding to the purchase item of each user (each purchase item). Multiplying the elements of the first feature matrix 410 corresponding to each user and the elements of the second feature matrix 420 corresponding to the first unpurchased item of each user. The difference of the result of multiplying an element belonging to a column corresponding to one non-purchased item) is maximized, and an element of the first feature matrix 410 corresponding to each user and a second characteristic corresponding to the first non-purchased item of each user The result of multiplying the elements of the matrix 420 and the elements of the first feature matrix 410 corresponding to each user and the elements of the second feature matrix 420 corresponding to the second non-purchased item of each user ( The elements of the first feature matrix 410 and the second feature matrix 420 may be learned to maximize the difference of the result of multiplying an element belonging to a column corresponding to each second non-purchased item.

A method of learning the elements of the first feature vector and the elements of the second feature vector may be expressed by Equation 1 below.

[Equation 1]

는 사용자 u의 구매 아이템,

는 사용자 u의 제 1 미구매 아이템,

는 사용자 u의 제 2 미구매 아이템, D(·)는 거리 함수,

는 사용자 u에 대응하는 제 1 특징 행렬(410)

의 엘리먼트,

는 구매 아이템에 대응하는 제 2 특징 행렬(420)

의 엘리먼트,

는 제 1 미구매 아이템에 대응하는 제 2 특징 행렬(420)

의 엘리먼트,

는 제 2 미구매 아이템에 대응하는 제 2 특징 행렬(420)

의 엘리먼트를 나타낸다. 또한,

는 레귤러라이징 텀(regularization term)으로서, 제 1 특징 행렬(410)과 제 2 특징 행렬(420)이 학습 데이터에 과적합(over-fitting)되는 것을 방지하며,

와

는 임의의 값으로 설정되는 하이퍼파라미터이다. 구현예에 따라서, 상기 수학식 1에서

는 생략될 수도 있다. 상기 제 1 특징 행렬(410)

는 N×x행렬이고, 상기 제 2 특징 행렬(420)

는 x×M행렬이되, 상기 N는 복수의 사용자의 총 인원수, M은 상기 전체 아이템의 총 개수, x는 임의의 자연수일 수 있다.In Equation 1, u is an index of a user included in the plurality of user groups U, i is an index of a purchase item, j is an index of the first non-purchased item, k is an index of the second non-purchased item,

Is the user's purchase item,

Is the user's first unpurchased item,

Is the second unpurchased item of user u, D (·) is the distance function,

Is the first feature matrix 410 corresponding to user u.

Element of,

Is the second feature matrix 420 corresponding to the purchased item.

Element of,

Is the second feature matrix 420 corresponding to the first non-purchased item.

Element of,

Is the second feature matrix 420 corresponding to the second non-purchased item.

Represents an element of. In addition,

Is a regularization term, which prevents the first feature matrix 410 and the second feature matrix 420 from over-fitting to the training data,

Wow

Is a hyperparameter set to an arbitrary value. According to an embodiment, in Equation 1

May be omitted. The first feature matrix 410

Is an N × x matrix, and the second feature matrix 420 is

Is an x × M matrix, where N is the total number of users, M is the total number of items, and x is any natural number.

D (·) is a distance function, for example, Euclidean distance function, first-norm function, p-order norm function, dot product, cosine distance function, or Mahalanobis distance function. It may include, but is not limited to.

The item recommendation apparatus 100 sets initial values of elements of the first feature matrix 410 and elements of the second feature matrix 420, and updates the initial values so that the resultant value of Equation 1 is maximized. The initial value is updated until the result of Equation 1 converges.

When the elements of the first feature matrix 410 and the elements of the second feature matrix 420 which are finally updated are derived, the item recommendation apparatus 100 may determine the first feature matrix 410 and the second feature matrix 420. Multiplying to estimate the preference for each of the total items of each of the plurality of users.

In operation S230 of FIG. 2, the item recommendation apparatus 100 may determine a user's unpurchased item (eg, the first unpurchased item and the second non-purchased item) based on a preference of the items of the predetermined user among the plurality of users. Information of at least one of the purchased items) may be transmitted to the terminal 200 of the predetermined user. For example, information on an unpurchased item having a preference higher than a predetermined value among unpurchased items may be transmitted to the terminal 200 of a predetermined user as a recommended item.

As described above, the item recommendation apparatus 100 transmits the information of the unpurchased item directly to the user terminal 200 through a network, or through a web page provided by the online shopping server 300 to the user terminal 200. Information of the non-purchased item may be transmitted to the user terminal 200. Alternatively, the information on the unpurchased item may be transmitted to the user terminal 200 through a short message, an email, or a social network service (SNS) server. However, the method of transmitting the information of the unpurchased item to the user terminal 200 is not limited to the above example, and the information of the unpurchased item may be transmitted to the user terminal 200 in various ways that the user can recognize. .

In an embodiment of the present invention, since the first unpurchased item of the user is an item selected by the user but not purchased, the first unpurchased item is filtered by the user. As a recommended item, it may be transmitted to the user terminal 200.

5 is a block diagram showing the configuration of an item recommendation apparatus 100 according to an embodiment of the present invention.

Referring to FIG. 5, the item recommendation apparatus 100 may include a memory 510, an information acquirer 530, a preference estimator 550, and a communicator 570. The memory 510, the information acquirer 530, the preference estimator 550, and the communicator 570 may be implemented by at least one processor, and may operate according to a program stored in the memory 510.

The memory 510 stores behavior information in the online shopping server 300 of each of the plurality of users. For example, the memory 510 may include purchase item information purchased by each of the plurality of users from the online shopping server 300, information about the first non-purchased item selected but not purchased, and information on the second non-purchased item not selected. Can be mapped to identification information of each user and stored.

The information obtaining unit 530 obtains information on purchase details and selection details of all items of each of the plurality of users through the online shopping server 300 from the memory 510.

The preference estimator 550 estimates the preferences of each of the plurality of users for each of the entire items based on the purchase details and the selection details of the entire items. Since the method of estimating the preferences of the users for each of the entire items has been described above, a detailed description thereof will be omitted.

The communication unit 570 transmits the information on the unpurchased item of the predetermined user to the terminal 200 based on the preference of each of the entire items of the predetermined user among the plurality of users. The communication unit 570 transmits the information of the unpurchased item to the terminal 200 of the predetermined user through, for example, a network card that transmits the information of the unpurchased item to the terminal 200 of the predetermined user through a network, and a mobile communication network. But may include a mobile communication module, but is not limited thereto.

6 is a test result for comparing the effects of the item recommendation method and another method according to an embodiment of the present invention.

Applicants of the present invention validated the effectiveness of the item recommendation method according to one embodiment of the present invention using two actual data sets (DATASET, RecSYs2015) including purchase records and selection (click) records of the same user group. DATASET is a web portal that provides an online shopping platform, eliminating 8 purchases and less than 40 user data. The RecSYs2015 data set includes click sessions and purchase sequences extracted from the e-commerce website. Here, we assumed each session as a user, and removed data from 8 purchases and less than 10 clicks of the user. For both real data sets, a sequence of purchase records arranged in chronological order was split in half for each user, the first half split being used as training data, and the second half ) Was used as test data. We also used the user's click record from the training data to the last purchase time.

Applicant evaluated how accurately the method according to one embodiment of the invention predicts a user's future purchase among items that are not clicked. In other words, for each user, items that were not clicked by the user in the past (ie, the second unpurchased item) were considered as recommendation candidates.

The conventional method for contrast with the effect according to an embodiment of the present invention is as follows.

i) MostPoP: The basic way to recommend the most frequently purchased items.

ii) BPR: method according to Rendle's study

iii) P3S-1: A method of applying pair-wise matrix decomposition based on the assumption that the purchase item's preference is greater than that of the second non-purchased item. In this method, it is assumed that the second non-purchased item not selected by the user as the negative feedback of the user instead of the first non-purchased item not purchased by the user.

iv) P3S-2: A method according to an embodiment of the present invention, wherein a preference of a purchased item is greater than a preference of a first non-purchased item, and a preference of a first non-purchased item is large. A method using pair-wise matrix decomposition according to.

v) P3S-3: pair-wise matrix with the assumption that the preference of the purchased item is greater than the preference of the first non-purchased item and the preference of the second non-purchased item is greater than that of the first non-purchased item. Method using the decomposition technique. This method takes advantage of the assumption that a user does not prefer an item that he clicks but does not buy, rather than an item that does not even click.

Although the BPR method is designed to optimize the area under the ROC curve (AUC), we have six different ranking matrices: Precision @ 5, Recall @ 5, Mean Average Precision (MAP) and Mean Reciprocal Rank. ), Using the NDCG (Normalized Discounted Cumulative Gain) and AUC (Area under the ROC curve) to demonstrate the remarkable effects of one embodiment of the present invention. Basically, we performed a grid search to tune the hyperparameters.

The results shown in FIG. 6 show the average of the results of performing each method five times with different random seeds for initialization. Referring to Figure 6, it can be seen that the method according to an embodiment of the present invention shows superior performance compared to other methods. MostPoP method that recommends the most frequently purchased items can be seen that the performance is lower than other methods that recommend the item customized by the user, P3S-2 method according to an embodiment of the present invention consistently compared to other methods It can be seen that the excellent effect. This means that not all unpurchased items (ie, the first unpurchased item and the second unpurchased item) can be treated as negative feedback.

The P3S-1 method does not show a significant difference from the performance of the BPR method, which indicates that the first unpurchased item that is clicked but not purchased has a great impact on the accuracy of the recommendation. The P3S-3 method performs worse than the BPR method, demonstrating that the preference of the first non-purchased item that is clicked but not purchased is higher than the preference of the second non-purchased item that does not even click.

Meanwhile, the above-described embodiments of the present invention can be written as a program that can be executed in a computer, and the written program can be stored in a medium.

The medium may be to continue to store a computer executable program, or to temporarily store for execution or download. In addition, the medium may be a variety of recording means or storage means in the form of a single or several hardware combined, not limited to a medium directly connected to any computer system, it may be distributed on the network. Examples of the medium include magnetic media such as hard disks, floppy disks and magnetic tape, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, And ROM, RAM, flash memory, and the like, configured to store program instructions. In addition, examples of another medium may include a recording medium or a storage medium managed by an app store that distributes an application, a site that supplies or distributes various software, a server, or the like.

Although embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art to which the present invention pertains may realize that the present invention may be embodied in other specific forms without changing the technical spirit or essential features thereof. I can understand. Therefore, it should be understood that the embodiments described above are exemplary in all respects and not restrictive.

100: item recommendation device
200: user terminal
300: online shopping server
510: memory
530: information acquisition unit
550: preference estimator
570: communication unit

Claims (7)

Obtaining information on purchase details and selection details of items of each of the plurality of users through the online shopping server;
Estimating a preference of each of the plurality of users for each of the items by applying a pair-wise matrix decomposition technique based on purchase details and selection details for the items; And
And transmitting information of an unpurchased item of the predetermined user to a terminal of the predetermined user based on a preference of each of items of the predetermined user among the plurality of users.
Obtaining information about the purchase history and selection history,
Acquiring information about a purchase item, a first non-purchased item, and a second non-purchased item of each of the plurality of users,
The first non-purchased item includes an item selected by the user but not purchased, the second non-purchase item includes an item not selected by the user,
Estimating the preference of each of the plurality of users,
First feature for the plurality of users assuming that the purchase item has a higher preference of the user than the first non-purchased item, and the first non-purchased item has a higher preference of the user than the second non-purchased item. Learning elements of a matrix and elements of a second feature matrix for the items; And
Estimating a preference of each of a plurality of users for each of the items based on the learned first and second feature matrices. .
delete delete The method of claim 1,
The learning step,
A result of multiplying an element of a first feature matrix corresponding to each user by an element of a second feature matrix corresponding to a purchase item of each user, an element of the first feature matrix corresponding to each user, and a first non-purchase of each user To maximize the difference between the results of multiplying the elements of the second feature matrix corresponding to the item,
A result of multiplying the elements of the first feature matrix corresponding to each user by the elements of the second feature matrix corresponding to the first unpurchased item of each user, the elements of the first feature matrix corresponding to each user and the first of each user Learning the elements of the first feature matrix and the second feature matrix such that a difference in the result of multiplying the elements of the second feature matrix corresponding to the two unpurchased items is maximized.
The method of claim 1,
The step of transmitting the information of the non-purchased item to the terminal of the predetermined user,
And transmitting, to the terminal of the predetermined user, information on the second non-purchased item having a preference higher than a preset value among the second non-purchased items of the predetermined user. .
A program stored in a medium in combination with hardware for executing the method of recommending an item of any one of claims 1 to 4.
An information obtaining unit obtaining information on purchase details and selection details of items of each of the plurality of users through the online shopping server;
A preference estimating unit estimating a preference of each of the plurality of users for each of the items by applying a pair-wise matrix decomposition method based on purchase details and selection details of the items; And
It includes a communication unit for transmitting the information of the unpurchased item of the predetermined user to the terminal of the predetermined user based on the preference of each of the items of the predetermined user of the plurality of users,
The information acquisition unit
Acquire information about a purchase item, a first non-purchased item, and a second non-purchased item of each of the plurality of users, wherein the first non-purchased item includes an item selected by a user but not purchased. 2 unpurchased items include items not selected by the user,
The preference estimator
First feature for the plurality of users assuming that the purchase item has a higher preference of the user than the first non-purchased item, and the first non-purchased item has a higher preference of the user than the second non-purchased item. After learning the elements of a matrix and elements of a second feature matrix for the items, based on the learned first feature matrix and the second feature matrix, each of a plurality of users for each of the items Item recommendation apparatus, characterized in that to estimate the preference.

Images