KR20170097535A - Item recommendation method and apparatus thereof utilizing uninteresting item and apparatus - Google Patents

Abstract

The present invention relates to an item recommendation method using an uninteresting item and an item recommendation device for same, capable of solving an OCCF problem by accurately estimating preference of a user on an item. A user can express the preference on the certain item by selecting the certain item. The item recommendation device distinguishes an evaluation item which the user expresses the preference about and a non-evaluation item which the user does not express the preference about. Also, the item recommendation device recommends the item to the user by using the distinguished evaluation item and the non-evaluation item. The item recommendation device classifies at least one among the non-evaluation items into the uninteresting item and can classify the item classified by one-class setting of the evaluation item and the non-evaluation item, by binary-class setting of the evaluation item, the non-evaluation item, and the uninteresting item. The item recommendation device considers the item classified by the binary-class setting and determines the item to be recommended for the user.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for recommending an item using an item of interest,

The present invention relates to an item recommendation apparatus and an item recommendation method for recommending an item to a user.

In order for the item recommending device to recommend the item that the user prefers to the user, it is necessary to measure which item the user prefers. The item recommendation device may measure which item the user prefers, by collecting (i) the user's preference for the item, or (ii) whether the user has selected a particular item.

(i) An example of collecting the preferences the user has given for an item is to collect the user's rating for the item. The item recommendation device can know whether the user prefers or does not prefer a particular item through the high and low of the collected ratings. Furthermore, the item recommendation device may know that the user has not rated the item if the user does not rate the item.

(ii) An example of collecting whether a user has selected a particular item is to collect whether the user has clicked on a particular item or bookmarked a particular item. Bookmarking a particular item includes not only adding it to a bookmark, but also pressing a like button for a particular item or adding a particular item to a purchase list.

Generally, since the user often clicks or bookmarks a particular item rather than giving a rating to a particular item, the item recommendation device interprets the item that the user has clicked or bookmarked to determine the item that the user prefers It is necessary to calculate.

It is obvious that the item selected by the user clicking or bookmarking is an item that the user is interested in. However, an item not selected by the user not only includes items that the user does not like, but may also include items that the user has not evaluated.

Accordingly, the item recommendation apparatus should determine whether an item that the user has not selected is indeed not selected because the user is not preferred or not selected because the user has not evaluated it. If the item recommendation device can not clearly determine this, the user's preference for the item can not be accurately predicted. OCCF problem (One-Class Collaborative Filtering problem) may occur when the item recommendation device can not accurately predict the user's preference for the item.

The present invention proposes an item recommending apparatus and method for solving the OCCF problem by accurately predicting a user's preference for an item.

According to an embodiment of the present invention, there is provided a method for identifying an item classified as a one-class setting of an evaluation item in which a user expresses a preference and an unrated item in which the user does not express a preference Determining an uninteresting score of the criterion item, determining an uninteresting item out of the criterion items based on the determined uninteresting score, Classifying the item into a binary-class setting of the evaluation item, the critic item and the un-interested item, and determining an item to be recommended to the user based on the item classified by the binary-class setting The item recommendation method is provided.

According to an embodiment, the step of classifying into a binary-class setting may include classifying a certain number of critic items with a low number of non-interest scores into the non-interest items among the plurality of critic items, Item recommendation method for classifying an item as a binary-class setting is provided.

According to an exemplary embodiment, the step of classifying into the binary-class setting may include an item recommendation method of generating an augmented matrix including an element value corresponding to the non-interest item.

According to one embodiment, the step of determining an item to be recommended to the user comprises the steps of: calculating from the expansion matrix a degree of similarity indicating how similar the preferences for a plurality of items of the user are, And determining an item to be recommended to the user, the degree of similarity including a number of users who have classified a plurality of the items as the entire evaluation item, a number of users whose entirety of the plurality of items are classified as the non- An item recommendation method is provided in which a plurality of the items are all determined according to the number of users classified as criterion items.

According to one embodiment, the step of determining an item to be recommended to the user may include generating an object matrix whose element value is determined according to which of the one-class setting items classified into the one- Generating a weight matrix corresponding to the object matrix by assigning a weight to an item classified as the one-class setting; performing matrix decomposition on the object matrix using the weight matrix; Determining an item to be recommended to the user according to a result of performing matrix decomposition, wherein the weight is determined based on which of the binary-class setting items classified into the one-class setting is classified The item recommendation method is provided.

According to one embodiment, the step of generating the weight matrix is provided with an item recommendation method for determining the weight using a probability that the user does not prefer the criterion item.

According to one embodiment, the step of determining an item to be recommended to the user includes: generating a positive graph by linking the user and the evaluation item; linking the user and the un- determining a relevance score of the item to the user using the positive graph and the negative graph, and determining an item to recommend to the user according to the relevance score The item recommendation method is provided.

According to one embodiment, the step of determining an item to be recommended to the user comprises the steps of: determining a like-potential for the evaluation item; determining a dislike-potential for the un-interested item; , Determining a probability of the user's preference of the item from the determined preference potential and the non-preferential potential using BP (Belief propagation), and determining a probability of the preference of the user based on the determined probability of preference An item recommendation method is provided that includes a step of determining an item to be recommended.

According to an embodiment of the present invention, an OCCF problem can be solved by accurately predicting a user's preference for an item.

1 is a diagram for explaining an operation of recommending an item to a user by an item recommendation apparatus according to an embodiment of the present invention.
2 is a flowchart illustrating an operation of recommending an item to a user by an item recommendation apparatus according to an exemplary embodiment of the present invention.
3 is an exemplary diagram illustrating an operation of an item recommendation apparatus according to an embodiment of the present invention to classify an item classified as a one-class setting into a binary-class setting.
4 is a flowchart illustrating an operation of an item recommendation apparatus according to an exemplary embodiment of the present invention to classify an item into binary-class settings to generate an augmented matrix.
FIG. 5 is an exemplary diagram illustrating an operation for an item recommendation apparatus according to an embodiment of the present invention to determine an unattractive score of a criterion item and classify the item into a binary-class setting.
FIG. 6 is a flowchart illustrating an operation of an item recommendation apparatus according to an embodiment of the present invention to determine items to be recommended to a user using a collaborative filtering (CF) method.
FIG. 7 is a flowchart illustrating an operation of an item recommending apparatus according to an exemplary embodiment of the present invention to determine an item to be recommended to a user using a One-Class Collaborative Filtering (OCCF) method.
FIG. 8 is a flowchart illustrating an operation of an item recommendation apparatus according to an exemplary embodiment of the present invention to determine items to be recommended to a user using a random walk with restart (RWR) method.
9 is a flowchart illustrating an operation of an item recommendation apparatus according to an exemplary embodiment of the present invention to determine an item to be recommended to a user using a belief propagation (BP) method.

It is to be understood that the specific structural or functional descriptions of embodiments of the present invention disclosed herein are presented for the purpose of describing embodiments only in accordance with the concepts of the present invention, May be embodied in various forms and are not limited to the embodiments described herein.

Embodiments in accordance with the concepts of the present invention are capable of various modifications and may take various forms, so that the embodiments are illustrated in the drawings and described in detail herein. However, it is not intended to limit the embodiments according to the concepts of the present invention to the specific disclosure forms, but includes changes, equivalents, or alternatives falling within the spirit and scope of the present invention.

The terms first, second, or the like may be used to describe various elements, but the elements should not be limited by the terms. The terms may be named for the purpose of distinguishing one element from another, for example without departing from the scope of the right according to the concept of the present invention, the first element being referred to as the second element, Similarly, the second component may also be referred to as the first component.

It is to be understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, . On the other hand, when an element is referred to as being "directly connected" or "directly connected" to another element, it should be understood that there are no other elements in between. Expressions that describe the relationship between components, for example, "between" and "immediately" or "directly adjacent to" should be interpreted as well.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, the terms " comprises ", or " having ", and the like, are used to specify one or more of the features, numbers, steps, operations, elements, But do not preclude the presence or addition of steps, operations, elements, parts, or combinations thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries are to be interpreted as having a meaning consistent with the meaning of the context in the relevant art and, unless explicitly defined herein, are to be interpreted as ideal or overly formal Do not.

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. However, the scope of the patent application is not limited or limited by these embodiments. Like reference symbols in the drawings denote like elements.

1 is a diagram for explaining an operation of an item recommending apparatus 120 recommending an item to a user 110 according to an embodiment of the present invention.

Referring to FIG. 1, the user 110 may be provided with an item list 130. The user 110 may express preferences for one or more items included in the item list 130. For example, the user 110 may express a preference for an item by rating a one or more items, clicking on an item, or bookmarking an item. Referring to FIG. 1, the user 110 can express the preference for the A item, the C item, and the D item among the A items to G items included in the item list 130.

The evaluation item means an item in which the user 110 expresses the preference. Referring to FIG. 1, items A, C, and D are evaluation items. An unrated item is an item for which the user 110 does not express a preference. Referring to FIG. 1, the remaining items excluding the evaluation items-B items, E items, and G items-are critic items. Items of the item list 130 may be classified as an evaluation item or a critic item by the user 110 by expressing the preference on some of the items of the item list 130. [

Referring to FIG. 1, an item recommendation apparatus 120 according to an embodiment may be provided. The item recommendation device 120 may identify an evaluation item and a criterion item from the item list 130 received from the user 110. [ The item recommending apparatus 120 may identify an item to which a rating is given as an evaluation item and an item to which a rating is not given as a criterion item. The item recommendation device 120 can identify items that the user has clicked or bookmarked as evaluation items, and items that the user has not clicked and bookmarked as critique items.

The item recommendation apparatus 120 may classify the items included in the item list 130 into one-class settings of evaluation items and criterion items, according to the result of the discrimination. More specifically, the item recommendation device 120 assigns a value to a specific item according to whether a specific item included in the item list 130 corresponds to an evaluation item or a criterion item, thereby sorting the item into a one-class setting can do. For example, the item recommendation apparatus 120 may assign '1' to the evaluation item and 'unknown' to the criterion item.

Hereinafter, with reference to FIG. 1, it is assumed that the user 110 expresses the preference to the A item, the C item, and the D item among the items included in the item list 130. In particular, it is assumed that the user 110 expresses the preference by clicking or bookmarking the A item, the C item, and the D item. Since the user has expressed preference to the A item, the C item, and the D item, the item recommending apparatus 120 can detect that the user is interested in the A item, the C item, and the D item. In other words, the item recommendation device 120 may detect that the user is interested in the evaluation item.

In contrast, a criterion item that the user 110 has not clicked or bookmarked is not a criterion item because the user 110 did not prefer it and did not express preference, or because the user 110 did not express interest, . Referring to FIG. 1, a criterion item may be a B item, an E item or a G item, and the user 110 may not click or bookmark it because it does not prefer the B item. However, the user 110 may not click or bookmark these items because they are not interested in the E or G items.

The item recommendation apparatus 120 according to an exemplary embodiment classifies an item using a binary-class setting in which an uninteresting item having a negative class is added to the one-class setting can do. The item recommendation device 120 may estimate a criterion item that does not express preference because the user 110 does not prefer the criterion item, and may classify the estimated criterion item as a non-interest item. For example, the item recommendation device 120 may determine that the user 110 did not prefer the B item and did not express the preference, and then classify the B item as a non-interest item. As a result, some of the critic items can be classified as negative classes. The item recommendation apparatus 120 can solve the OCCF problem by utilizing the binary-class setting to determine items to be recommended to the user.

The item recommending apparatus 120 can determine an item to be recommended to the user 110 based on the items classified by the binary-class setting. The item recommending device 120 may not recommend items to the user 110 that are similar to the B and B items since the B items are classified as uninteresting items. The item recommendation device 120 can identify an item similar to the B item using the result of another user's evaluation of the item. Item recommendation device 120 may recommend at least one of the E item to the G item to the user 110 since the E item or the G item is not classified as a negative criterion item, i.e., a non-interest item.

Referring to FIG. 1, the item recommendation apparatus 120 may provide an item determined to be recommended to the user 110 to the user 110 in the form of a recommendation item list 140. Items not included in the item list 130 may be included in the recommended item list 140. 1, an H item and an M item may be included in the recommended item list 140 even though they are not included in the item list 130. [

If the criterion item is not classified as a negative class, it can be included in the recommendation item list 140. Referring to FIG. 1, the item recommendation apparatus 120 may include an F item, which is a critic item, in the recommend item list 140.

Although only one embodiment of the user 110 is shown in FIG. 1, the item recommendation device 120 may receive a plurality of item lists 130 from a plurality of users. The item recommendation apparatus 120 can determine how similar the user's preference to the item is based on the received plurality of item lists. For example, assume that a plurality of users express preferences for an A item and an F item. Item recommendation device 120 may include an F-item in recommendation item list 140 even if user 110 did not select an F-item in item list 130 because user 110 has selected item A. have.

The item recommendation device 120 classifies the items provided to the user 110 as an evaluation item, a criterion item, and a binary-class item of the non-interest item by classifying some of the criticism items that the user 110 did not express preference as a negative class Setting, and an item to be recommended to the user can be determined based on the item classified by the binary-class setting. The item recommendation apparatus 120 can more accurately determine an item to be recommended to the user 110. [

2 is a flowchart illustrating an operation of recommending an item to a user by an item recommendation apparatus according to an exemplary embodiment of the present invention. A computer-readable recording medium on which a program for executing an item recommendation method according to an embodiment is recorded may be provided. The program may include at least one of an application program, a device driver, a firmware, a middleware, a dynamic link library (DLL), and an applet storing an item recommendation method. According to an embodiment of the present invention, the item recommendation apparatus includes a processor, and the processor can perform the item recommendation method by reading the recording medium on which the item recommendation method is recorded.

Referring to FIG. 2, in step 210, the item recommendation apparatus according to one embodiment may identify an evaluation item in which the user expresses the preference and a criterion item in which the user does not express the preference. Referring to FIG. 1, the user may express a preference for one or more items of the received item list. The user may not choose a particular item because it does not like or is not interested in a particular item. The item recommendation apparatus can receive the result selected by the user.

3 is an exemplary diagram illustrating an operation of an item recommendation apparatus according to an embodiment of the present invention to classify an item classified as a one-class setting into a binary-class setting. Hereinafter, the operation performed by the item recommendation apparatus of FIG. 2 will be described in detail with reference to the example of FIG.

Referring to FIG. 3, the item recommendation apparatus may receive an item list 310 from a user. Referring to FIG. 3, the user expresses the preference for the A item, the C item, and the D item (evaluation item) and does not express the preference for the B item and E item or G item (critic item).

The item recommendation device may classify an item into a one-class setting of an evaluation item and a criterion item by assigning a value to the item depending on whether the user has evaluated the preference. Referring to FIG. 3, the item recommendation apparatus may generate item list 320 by assigning 1 to the evaluation item and 'unknown' to the criterion item.

Referring again to FIG. 2, in step 220, an item recommendation device according to one embodiment may determine an uninteresting score of a criterion item. That is, the item recommendation apparatus can distinguish items that do not express preference because the user is not interested, and items that do not express preference because the user does not prefer, among critic items. The operation in which the item recommending apparatus determines the unattractive score of the criterion item will be described later in detail.

Referring to FIG. 2, in step 230, the item recommendation apparatus according to an exemplary embodiment determines an unreconciled item among critic items based on the determined unconverted score, , A criterion item, and a binary-class setting of an unattended item. Item recommendation apparatuses can classify a certain number of critical item items having a low non-interest point value as uninterest items among critic item items. The item recommendation apparatus may change the value of the criterion item classified as the non-interest item to a value other than 'unknown'.

Referring to FIG. 3, an item list 330 is shown in which the item recommending apparatus classifies some of the critic items as unattributed items. The item recommending apparatus can classify the B item and the G item as non-interested items out of the B, E, and G items that are critic items. The item recommendation apparatus can classify the B item and the G item as un-interested items by resetting the value assigned to the B item and the G item from 'unknown' to zero. The item recommendation device may classify all or a portion of the critic item as a non-interest item. Thus, the value of some of the criterion items may be maintained as 'unknown'.

As a result, the item recommendation apparatus can generate an item list 330 classified into a binary-class setting from the item list 310 to the item list 320 classified as the one-class setting. In other words, the item recommendation apparatus can convert the binary-class setting (1, 0, 'unknown') by adding a negative class (0) to the one-class setting (1, 'unknown'). The item recommendation device can generate an augmented matrix using the result of classifying the item into a binary-class setting.

Referring again to FIG. 2, in step 240, the item recommendation device according to one embodiment may determine an item to recommend to a user based on items separated by a binary-class setting. As described above, the value assigned to the item provided to the user may be either 1, 0 or 'unknown' depending on the binary-class setting. Since the preference has been converted from the one-class setting to the binary-class setting, the item recommendation device can distinguish items that are supposed to express no preference because the user is not preferred among the criticism items. That is, the item recommendation apparatus can apply an item recommendation method used when a user gives a rating to an item, assuming that the user classifies the item classified as a binary-class setting into a rating item.

4 is a flowchart illustrating an operation of an item recommendation apparatus according to an exemplary embodiment of the present invention to classify an item into binary-class settings to generate an augmented matrix. FIG. 5 is an exemplary diagram illustrating an operation for an item recommendation apparatus according to an embodiment of the present invention to determine an unattractive score of a criterion item and classify the item into a binary-class setting. Hereinafter, the operation of the item recommending apparatus classifying items into binary-class settings will be described with reference to FIGS. 4 and 5. FIG.

Referring to FIG. 4, in step 410, an item recommendation device according to one embodiment may determine an uninteresting score of a non-favorited item. The item recommendation device can determine the degree to which the user prefers the critic item based on the similarity between the evaluation items in which the users express their preferences. The item recommendation device can determine the degree to which the user prefers the critic item by referring to the evaluation items previously expressed by the user. The item recommendation device can display the degree of preference of critic item using the non-interest score.

Referring to FIG. 5, an item list 520 generated by determining an unattractive score of a criterion item included in the item list 510 is shown. The higher the non-interest score, the higher the likelihood that the user will prefer the critic item. Referring to FIG. 5, since the non-interest score of the E item is higher than that of the F item, the item recommendation apparatus can determine that the user has determined that the user has a higher probability of preferring the E item than the F item. Also, since the non-interest score of the G item is the lowest, the probability that the user prefers the G item is lower than the probability of preferring the B item, the E item, and the F item.

Referring again to FIG. 4, in step 420, the item recommendation apparatus according to an exemplary embodiment may classify an item classified as a one-class setting into a binary-class setting using a non-interest score. The item recommendation apparatus can classify a certain number of critic items that have a low interest score as uninterest items. For example, an item recommendation device may classify a criterion item with a non-interest score of less than 0.5 as a non-interest item. Referring to FIG. 5, the item recommendation apparatus may generate an item list 530 by classifying critic items, which are determined to be non-interest items among the critic items, as non-interest items, which are lower than predetermined non-interest points. Referring to FIG. 5, by classifying a criterion item having a non-interest score lower than 0.5 as an uninterest item, a value (0) corresponding to a negative class is given to the B item and G item. The item recommendation apparatus can keep the value of a criterion item that is not classified as a non-interest item as 'unknown'.

According to one embodiment, the item recommendation device may classify a predetermined percentage of items of the criticism item as un-interested items. In this case, the item recommendation apparatus can classify the item having the lowest unconverted score as an uninterest item sequentially. Referring to FIG. 5, when 50% of the critic items are classified as uninterest items, two of the four critic items should be classified as uninterest items. Therefore, the item recommendation apparatus can classify the G item having the lowest unconverted score and the B item having the next lowest unconverted score as the uninterested item.

Referring again to FIG. 4, in step 430, an item recommendation apparatus according to an embodiment may generate an augmented matrix including element values corresponding to an unrecognized item. An expansion matrix is an element, which can contain the results of classifying items into binary-class settings. The enhancement matrix may include an element value for the entire item provided to the user divided into an evaluation item (1), a non-interest item (0), and a criterion item (unknown) according to a binary-class setting. When the item recommending apparatus determines an item to be recommended to the user, the generated expansion matrix can be used.

FIG. 6 is a flowchart illustrating an operation of an item recommendation apparatus according to an embodiment of the present invention to determine items to be recommended to a user using a collaborative filtering (CF) method. The item recommendation apparatus can perform a collaborative filtering method using an expansion matrix. The item recommendation device may utilize information that an item is evaluated by other users similar to the user (UserKNN). The item recommendation apparatus can utilize information of another user who has evaluated items similar to the user's evaluation item (ItemKNN). The item recommendation device can determine items to recommend to the user (SVD, SVD ++, PureSVD) by using latent factor generated in matrix decomposition.

Referring to FIG. 6, in step 610, an item recommendation apparatus according to an embodiment may calculate, from an expansion matrix, a degree of similarity that indicates how similar the preferences for a plurality of items of a plurality of users are. Hereinafter, an embodiment in which the item recommendation apparatus calculates the degree of similarity will be described with reference to the example of Table 1. [ Table 1 summarizes the user's preferences for item 1 and item 2 according to the binary-class setting, using the expansion matrix, when item 1 and item 2 are provided to q users through the item list.

Item 1

Item 2 One 0 Unknown Sum One a b c a + b + c 0 d e f d + e + f Unknown g h i g + h + i Sum a + d + g b + e + h c + f + i q

Referring to Table 1, a denotes the number of users who express preferences in Item 1 and Item 2. b represents the number of users who express preference in item 2 and item 1 is classified as a non-interest item by the item recommendation device. and i denotes the number of users whose items 1 and 2 are all classified as non-interest items. Since the expansion matrix can store the preferences evaluated by a plurality of users for a plurality of items according to the binary-class setting, the item recommendation apparatus can use the expansion matrix to convert the user's preferences for a plurality of items into binary- It can be counted according to the setting.

The item recommendation apparatus can classify the critic item using the negative class, thereby more accurately calculating the similarity. According to one embodiment, the item recommendation apparatus may include: a number of users (a) that classify a plurality of items into all evaluation items; a number (e) of users whose entirety of items are classified as non-interest items; The degree of similarity can be determined according to the number (i) of all users classified as criterion items. The item recommendation apparatus can determine the degree of similarity using Equation (1).

Referring to Table 1, the number of users in which a plurality of items in Equation (1) are classified as an evaluation item or a non-interest item equals to the number of users (a) All items 1 and 2 may be determined by the recommendation apparatus as the sum of the number (e) of users classified as non-interest items (a + e). The number of users who have not classified all of the plurality of items of Equation 1 as criterion items can be determined by excluding the number i of users who are all classified as criterion items in item 1 and item 2 in the total number q of users (qi) .

The item recommendation apparatus can calculate the degree of similarity of preferences for a plurality of items of a plurality of users by using the degree of similarity determined according to Equation (1). In particular, the item recommendation device may use the number of users classified as un-interested items to calculate the similarity. As a result, the item recommendation apparatus can calculate the degree of similarity among the critic items in consideration of the items that the user is unlikely to prefer. Therefore, the item recommendation apparatus can calculate the degree of similarity more accurately.

Referring to FIG. 6, in step 620, the item recommendation device according to one embodiment may determine an item to recommend to a user based on the determined similarity. The item recommendation apparatus can determine an item to be recommended to a user by performing a collaborative filtering method based on the calculated similarity. That is, the item recommendation apparatus can determine an item to be recommended to the user by extracting users having similar distributions of preferences for a plurality of items.

FIG. 7 is a flowchart illustrating an operation of an item recommending apparatus according to an exemplary embodiment of the present invention to determine an item to be recommended to a user using a One-Class Collaborative Filtering (OCCF) method. The item recommendation apparatus can determine items to be recommended to the user by using a weighted alternating least squares (WALS) algorithm. The item recommendation device may generate a weighting matrix used to decompose the object matrix and the object matrix to apply the WALS algorithm. The item recommendation apparatus can set each element value of the weight matrix according to a result of classifying each item into a binary-class setting.

Referring to FIG. 7, in step 710, an item recommendation apparatus according to an embodiment includes an object recommendation apparatus, in which an item value is determined based on which one of the items classified as the one- Lt; / RTI > That is, the element value of the i-th column and the j-th column of the object matrix can be determined according to which of the one-class settings the j-item is classified for the i-user. For example, if the j item is classified as a criterion item for the i user, the element value in column i, row j may be set to zero. When the j item is classified as the evaluation item for the i user, the element value in the i row and j column can be set to one. Table 2 shows an example of an object matrix in which element values are determined according to which one of A-to-D items is classified as one of the one-class settings for User 1 to User 3.

item
user A B C D One One 0 One 0 2 One One One 0 3 0 One 0 0

Referring to Table 2, it can be seen that the user 1 classified the A item as the evaluation item and the user 2 did not express the preference for the D item.

Referring to FIG. 7, in step 720, an item recommendation apparatus according to an embodiment may generate a weight matrix corresponding to an object matrix by assigning weights to items classified as a one-class setting. The weights can be determined according to which of the items classified in the one-class setting are classified as binary-class settings.

According to one embodiment, the item recommendation apparatus may set the weight of the item classified as the non-interest item to be equal to the weight of the evaluation item. For example, if the item recommendation apparatus sets the weight of the evaluation item to 1, the item recommendation apparatus can set the weight of the item classified as the non-interest item to 1.

The item recommending apparatus can set the weight of the criterion item to a value smaller than 1, which is the weight of the evaluation item. For example, the item recommendation device may set the weight of the criterion item to zero. The item recommendation apparatus may set the weight of the criterion item to a constant smaller than 1 and greater than 0.

The item recommendation apparatus can set the weight of the criterion item according to the number of the evaluation items expressing the user's preference. The probability that a particular user does not prefer a criterion item that does not express a preference may be higher than the probability that another user does not prefer a criterion item if the number of evaluation items expressing preference by a particular user is greater than that of another user. Accordingly, when the number of items to be evaluated by the user is large, the item recommendation apparatus can set the weight of the criterion item to a relatively high value.

The item recommendation apparatus may set the weight of the criterion item for a specific user to a higher value as the number of other users expressing the preference for the criterion item is smaller. That is, the weight may mean that the probability that the user does not prefer the critic item is high. Table 3 shows an example of the weight matrix generated by the item recommendation apparatus according to one embodiment in correspondence with the target matrix of Table 2. [

item
user A B C D One One 0.9 One 0.7 2 One One One 0.5 3 0.5 One 0 0.8

Referring to Table 2, the user 1 did not express the preference for the B item and the D item. Referring to FIG. 3, the item recommendation apparatus sets the weight of the B item to be higher than the weight of the D item. That is, the item recommendation apparatus can know that the probability that the user does not prefer the B item is higher than the probability that the user does not prefer the D item.

Referring to FIG. 7, in step 730, an item recommendation apparatus according to an embodiment may perform matrix factorization on an object matrix using a weight matrix. The item recommendation apparatus can matrix-decompose the object matrix by applying the WALS algorithm. Therefore, the item recommendation apparatus can perform matrix decomposition on the object matrix using the weight assigned to the element of the weight matrix. The item recommendation apparatus can use the weights to determine if the user will more accurately predict the preference for an item than other items.

Referring to FIG. 7, in step 740, the item recommendation apparatus according to an embodiment may determine an item to be recommended to a user according to a result of performing matrix decomposition. The item recommendation apparatus can determine an item to be recommended to a user by performing a one-class collaborative filtering method. That is, an item having a higher weight assigned to it and an item classified as a non-interest item may have a greater influence on determining an item to be recommended to a user than a criterion item assigned a relatively lower weight.

FIG. 8 is a flowchart illustrating an operation of an item recommendation apparatus according to an exemplary embodiment of the present invention to determine items to be recommended to a user using a random walk with restart (RWR) method. The item recommendation apparatus can generate a graph by making users and items nodes, and connecting nodes according to whether the user expresses the preference for the item. The item recommendation apparatus can determine an item to be recommended to the user by performing RWR on the generated graph. Since the item recommending device classifies all or a part of the critic item as a non-interested item, the item recommending device can link the user and the unrecognized item, as well as connecting the user and the evaluation item.

Referring to FIG. 8, in step 810, an item recommendation apparatus according to an embodiment may generate a positive graph by connecting a user and an evaluation item. Thus, users and items can only be linked if the user expresses their preference to that item.

Referring to FIG. 8, in step 820, an item recommendation apparatus according to an exemplary embodiment may generate a negative graph by connecting a user and an unrecognized item. That is, any item other than the item to which 'unknown' is assigned among the critic items can be connected to the user.

Referring to FIG. 8, in step 830, an item recommendation device according to an embodiment may determine the relevance score of an item for a user using a positive graph and a negative graph. The item recommendation apparatus can determine the positive score by performing the RWR on the positive graph. The item recommendation apparatus can determine the negative score by performing RWR on the negative graph. The item recommendation apparatus can determine the positive score and the negative score from the positive graph and the negative graph, respectively, according to Equation (2).

은 랭킹 벡터로써, 결정되는 포지티브 점수 또는 네거티브 점수를 의미한다. c는 재시작 확률,

은 시작 벡터,

는 정규화된 가중치 그래프를 의미한다.Referring to Equation (2)

Means a positive score or a negative score which is determined as a ranking vector. c is the restart probability,

Is a start vector,

Means a normalized weight graph.

The item recommendation device can determine the relevance score by excluding the negative score from the positive score. That is, the item recommendation apparatus can determine the relevance score in consideration of the relevance between the user and the evaluation item as well as the relevance between the user and the non-interest item.

Referring to FIG. 8, in step 840, the item recommendation apparatus according to an embodiment may determine an item to be recommended to the user according to the determined relevance score. Accordingly, the item recommendation apparatus can determine an item to be recommended to the user in consideration of the relevance between the user and the evaluation item, and the relevance between the user and the item to which the low preference (0) is assigned.

9 is a flowchart illustrating an operation of an item recommendation apparatus according to an exemplary embodiment of the present invention to determine items to be recommended to a user using a belief propagation (BP) method. The item recommendation device may determine the probability of the user's preference for the item and may use the determined probability to recommend the item to the user.

Referring to FIG. 9, in step 910, an item recommendation apparatus according to an embodiment may determine a like-potential for an evaluation item. The item recommendation apparatus can give a high probability to the preference potential of the evaluation item. For example, the item recommendation apparatus may assign the preference potential for the evaluation item to 0.9 or more.

Referring to FIG. 9, in step 920, an item recommendation apparatus according to an embodiment may determine a dislike-potential for an unreconstant item. The item recommendation apparatus can give a high probability to the non-favorable item's non-favorable potential. For example, the item recommendation device may grant a non-favorable potential for unannounced items to 0.9 or higher. Thus, the item recommendation apparatus can set the node potential of the user or the item differently depending on whether the item is classified as the evaluation item or the non-interest item.

Referring to FIG. 9, in step 930, the item recommendation apparatus according to an exemplary embodiment determines a probability of a user's item's preference from a determined preference potential and a non-preferential potential using belief propagation (BP) . Since the item recommendation apparatus has determined not only the preference potential for the evaluation item but also the non-favorable item of the non-interest item, the reliability diffusion can be applied considering the non-interest item. Therefore, the item recommendation apparatus can more accurately determine the probability of the user's preference of the item.

The item recommendation apparatus can apply reliability spreading using Equation (3).

Referring to Equation 3, m _ij (x _c ) is a message from the user (or item) v _i to the item (or user) v _j , where v _j is the probability of being in the x _c state (evaluation item state or criterion item state) Can be taken to mean the trust of v _i to. According to one embodiment, m _ij (x _c ) may refer to a message from an item (or user) v _i to a user (or item) v _j . Referring to Equation (3), m _ij (x _c ) may be generated from a product of messages received from a user (or item) v _i around the user (or user) v _i except for v _i .

는 사용자(또는 아이템) v_i의 상태가 x_d 상태(평가 아이템 상태 또는 비평가 아이템 상태)인 경우, 그 이웃 사용자 또는 이웃 아이템 v_j가 x_c 상태일 가능성을 표시할 수 있다.

는 미리 설정된 파급 행렬을 통해 결정될 수 있다. 일실시예에 따르면, 파급 행렬은 표 4와 같이 설정될 수 있다.Referring to Equation 3,

May indicate the likelihood that the neighboring user or neighbor item v _j is in the x _c state if the state of the user (or item) v _i is x _d state (evaluation item state or criterion item state).

May be determined through a predetermined spreading matrix. According to one embodiment, the spreading matrix may be set as shown in Table 4.

Rating Item Critic item Rating Item 0.5 +

0.5 -

Critic item 0.5 -

0.5 +

는 0 보다 크고 1 보다 작은 값으로 설정될 수 있다.Referring to Table 4,

May be set to a value greater than zero and less than one.

는 사용자 또는 아이템에게 할당된 초기 값으로, 사용자 v_i의 상태가 x_d 상태(평가 아이템 또는 비평가 아이템)일 확률을 나타낸다. 초기값은 단계(910) 또는 단계(920)에서 결정된 선호 포텐셜 또는 비선호 포텐셜에 따라 결정될 수 있다.Referring to Equation 3,

Is an initial value assigned to a user or item, and represents the probability that the state of the user v _i is x _d state (an item to be evaluated or a criterion item). The initial value may be determined according to the preferential potential or non-preferential potential determined in step 910 or step 920. [

The item recommendation apparatus can calculate the possibility that the user or item belongs to a specific state according to Equation (4) using m _ij (x _c ) in Equation (3).

Referring to Equation (4), k is a constant for normalizing the sum of each element of the confidence vector to be 1. Each element of the trust vector calculated according to Equation (4) can mean the possibility that the corresponding node (user or item) belongs to a particular state.

Referring to FIG. 9, in step 940, the item recommendation apparatus according to an exemplary embodiment may determine an item to be recommended to a user according to a probability of a determined preference. Since the probability of the preference is determined in consideration of the non-interest item, the item recommendation apparatus can more accurately determine an item to be recommended to the user.

The item recommending apparatus may classify at least a part of the critic items as a non-interest item, and recommend the item to the user in consideration of the non-interest item. Thus, the item recommendation apparatus can more accurately determine an item to be recommended to the user. In addition, the item recommendation apparatus can utilize a collaborative filtering method, a single-class collaborative filtering method, an RWR method, and a reliability spreading method in consideration of uninterest items. By categorizing some of the criticism items as non-interest items, the item recommendation device can solve the OCCF problem.

The apparatus described above may be implemented as a hardware component, a software component, and / or a combination of hardware components and software components. For example, the apparatus and components described in the embodiments may be implemented within a computer system, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA) , A programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. The processing device may also access, store, manipulate, process, and generate data in response to execution of the software. For ease of understanding, the processing apparatus may be described as being used singly, but those skilled in the art will recognize that the processing apparatus may have a plurality of processing elements and / As shown in FIG. For example, the processing unit may comprise a plurality of processors or one processor and one controller. Other processing configurations are also possible, such as a parallel processor.

The software may include a computer program, code, instructions, or a combination of one or more of the foregoing, and may be configured to configure the processing device to operate as desired or to process it collectively or collectively Device can be commanded. The software and / or data may be in the form of any type of machine, component, physical device, virtual equipment, computer storage media, or device , Or may be permanently or temporarily embodied in a transmitted signal wave. The software may be distributed over a networked computer system and stored or executed in a distributed manner. The software and data may be stored on one or more computer readable recording media.

The method according to an embodiment 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, and the like, alone or in combination. The program instructions to be recorded on the medium may be those specially designed and configured for the embodiments or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code 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 embodiments, and vice versa.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments. For example, it is to be understood that the techniques described may be performed in a different order than the described methods, and / or that components of the described systems, structures, devices, circuits, Lt; / RTI > or equivalents, even if it is replaced or replaced.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

110: User
120: Item recommendation device
130: Item list
140: Recommended Item List

Claims (8)

Identifying an item classified by a user as an one-class setting of an evaluation item representing the preference and an unrated item for which the user has not expressed the preference;
Determining an uninteresting score of the criterion item;
Determining an uninteresting item from among the criterion items based on the non-interest score, and setting an item separated by the one-class setting as a binary-class setting of the evaluation item, the criterion item, and the non- classifying into a binary-class setting; And
Determining an item to be recommended to the user based on the item classified by the binary-class setting;
≪ / RTI >
The method according to claim 1,
Classifying the item classified by the one-class setting into a binary-class setting by classifying a certain number of critic items having a low non-interest score among the plurality of critic items into the non-interest item, How to recommend an item to sort by setting.
The method according to claim 1,
Classifying into the binary-class setting comprises:
And generating an augmented matrix including an element value corresponding to the non-interest item.
The method of claim 3,
The step of determining an item to be recommended to the user comprises:
Calculating, from the expansion matrix, a degree of similarity that indicates how similar the preferences for a plurality of items of the plurality of users are; And
Determining an item to be recommended to the user based on the degree of similarity,
Lt; / RTI >
Preferably,
An item recommendation that is determined by the number of users who have allotted a plurality of the items to an evaluation item, the number of users whose entirety of the plurality of items is classified as a non-interest item, and the number of users whose entirety of the plurality of items is classified as a criterion item Way.
The method according to claim 1,
The step of determining an item to be recommended to the user comprises:
Generating an object matrix in which an element value is determined according to which one of the one-class setting items classified into the one-class setting is classified;
Generating a weight matrix corresponding to the object matrix by assigning weights to items classified as the one-class setting;
Performing matrix decomposition on the object matrix using the weight matrix; And
Determining an item to be recommended to the user according to a result of performing the matrix decomposition;
Lt; / RTI >
Wherein the weight is determined based on which of the binary-class settings the item classified as the one-class setting is classified as.
6. The method of claim 5,
Wherein generating the weight matrix comprises:
And determining the weight using the probability that the user does not prefer the criterion item.
The method according to claim 1,
The step of determining an item to be recommended to the user comprises:
Linking the user and the evaluation item to generate a positive graph;
Linking the user and the non-interest item to generate a negative graph;
Determining a relevance score of the item for the user using the positive graph and the negative graph; And
Determining an item to be recommended to the user according to the relevance score
≪ / RTI >
The method according to claim 1,
The step of determining an item to be recommended to the user comprises:
Determining a like-potential for the evaluation item;
Determining a dislike-potential for the un-interesting item;
Determining a probability of the user's preference of the item from the determined preference potential and the non-preferential potential using belief propagation (BP); And
Determining an item to be recommended to the user according to the determined probability of the preference;
≪ / RTI >

Images