KR101953802B1 - Method and apparatus for recommending item using implicit and explicit signed trust relationships - Google Patents

Abstract

And discloses an item recommendation method using an inner trust and an outer trust relationship. The item recommendation method using the inherent and extrinsic trust relationships according to an embodiment of the present invention is characterized in that based on evaluation information including a result of each of a plurality of users evaluating at least one of a plurality of items, Inferring a implicit trust relationship about a plurality of two-person combinations composed of two persons; Verifying the implicit trust relationship for at least one combination in which an explicit trust relationship exists among the plurality of two-person combinations based on the evaluation information; And determining an item to be recommended among the plurality of items to the recommending target person, which is one of the plurality of users, using the inferred liveness trust relationship and the verified outer trust relationship.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an item recommendation method and apparatus using a trust relationship and an implicit trust relationship,

More particularly, the present invention relates to an item recommendation method and apparatus using an implicit and explicit trust relationship, and more particularly, to a method and an apparatus for recommending an item based on an evaluation score of a user, And more particularly, to a method and apparatus for performing more accurate recommendations by reflecting the metonymy of trust and metaphors of distrust.

Trust networks are widely used to address data sparsity problems and cold-start problems in recommendation systems using collaborative filtering. In recent years, studies have been proposed that use explicit signed trust relationships (i.e., trust and distrust relationships). However, these techniques disregard the fact that users may have different / same preferences with trustworthy or distrusted objects in the trust network. Many of these techniques also deal with the transitivity of trust and distrust. However, according to other studies on the metamorphosis of relations, trust relationships can be transferred but distrust relations can not be transferred. In addition, explicitly signed trust relationships are also a rare problem in presenting user preferences in recommendation systems.

In the present invention, an implicit signed trust relationship is generated and utilized together with an explicit signed trust relationship to solve the problem of the sparseness of the trust relationship. In addition, we try to deduce a more accurate user taste by confirming similarity and dissimilarity of user taste, trust and distrust relation by calculating similarity score between users again.

In addition to these strategies, a matrix factorization model that exploits both explicit and implicitly signed trust relationships and rating information at the same time, and utilizes the metaphor of trust relationship metaphors and distrust relations, .

Related Prior Art Korean Patent Publication No. 10-2014-0046108 entitled " Product recommendation system through association analysis between products in an online shopping mall, public date: April 18, 2014) is available.

SUMMARY OF THE INVENTION The present invention seeks to provide a method and apparatus for more effectively recommending the use of inherent and extrinsic trust relationships.

The problems to be solved by the present invention are not limited to the above-mentioned problem (s), and another problem (s) not mentioned can be clearly understood by those skilled in the art from the following description.

According to an embodiment of the present invention, there is provided an item recommendation method using an inferior and extrinsic trust relationship, the item recommendation method comprising: a step of, based on evaluation information including a result of evaluating at least one of a plurality of items Inferring an implicit trust relationship for a plurality of two-person combinations comprising two of the plurality of users; Verifying the implicit trust relationship for at least one combination in which an explicit trust relationship exists among the plurality of two-person combinations based on the evaluation information; And determining an item to be recommended among the plurality of items to the recommending target person, which is one of the plurality of users, using the inferred liveness trust relationship and the verified outer trust relationship.

Advantageously, the endurance trust relationship and the implicit trust relationship can have one of a positive sign indicating trust and a negative sign indicating distrust.

Preferably, the step of determining the item to be recommended may be determined based on the transitivity of the trust relationship having the positive sign and the intransitivity of the trust relationship having the negative sign.

Preferably, the step of inferring the ICLR includes calculating similarity between members for each of the plurality of two-person combinations using the evaluation information; And inferring the ICR of each of the plurality of two-person combinations based on the predetermined threshold and the calculated similarity.

Preferably, the step of verifying the outer trust relationship includes calculating the degree of similarity between the members for each of at least one combination in which the outer trust relationship exists, using the evaluation information; And verifying the implicit trust relationship of each of the at least one combination based on the predetermined threshold and the calculated similarity.

According to another aspect of the present invention, there is provided an item recommendation apparatus using an inferior and extrinsic trust relationship, the item recommendation apparatus comprising: a plurality of users, each of which includes at least one of a plurality of items, A reasoning unit for inferring an internal trust relationship on a plurality of two-person combinations constituted by two of said plurality of users; A verification unit that verifies the external trust relationship with respect to at least one combination in which an outward trust relationship exists among the plurality of two-person combinations based on the evaluation information; And a recommendation determination unit for determining an item to be recommended among the plurality of items to the recommendation target person, which is one of the plurality of users, using the inferred ICT and the verified trust relationship.

Advantageously, the endurance trust relationship and the implicit trust relationship can have one of a positive sign indicating trust and a negative sign indicating distrust.

Preferably, the recommendation decision section may determine based on the metrics of the trust relationship having the positive sign and the nonvisibility of the trust relationship having the negative sign.

Preferably, the reasoning unit may calculate the degree of similarity between the members for each of the plurality of two-person combinations using the evaluation information, and calculate, based on the predetermined threshold and the calculated degree of similarity, I can infer my trusting relationship.

Preferably, the verification unit may calculate the degree of similarity between the members for each of at least one combination in which the external trust relationship exists, using the evaluation information, and based on the predetermined threshold and the calculated degree of similarity, The outer trust relationship of each of the one combination can be verified.

The recommendation method and apparatus according to an embodiment of the present invention mitigates the problem of scarcity of data and the problem of cold start by using the inner and outer reliability relationships and can perform more accurate recommendation.

In addition, the recommendation method and apparatus according to an embodiment of the present invention can provide more accurate recommendation by recommending the metonymity of trust and metonymy of distrust.

FIG. 1 is a flowchart illustrating an item recommendation method using an inherent and extrinsic trust relationship according to an embodiment of the present invention.
FIG. 2 is a block diagram for explaining an item recommendation apparatus using an inherent and extrinsic trust relationship according to an embodiment of the present invention.
3 is a diagram for explaining statistics of three actual life data sets.
4 is a diagram for explaining the accuracy of SimSocial and RecSSN for various types of trust relationships.
Figures 5 and 6 are diagrams for explaining the error of recommendation to all users and coldstart users for various recommendation approaches.

While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail. It should be understood, however, that the invention is not intended to be limited to the particular embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. Like reference numerals are used for like elements in describing each drawing.

The terms first, second, A, B, etc. may be used to describe various elements, but the elements should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component. And / or < / RTI > includes any combination of a plurality of related listed items or any of a plurality of related listed items.

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.

The terminology used in this application is used only to describe a specific embodiment and is not intended to limit the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In the present application, the terms "comprises" or "having" and the like are used to specify that there is a feature, a number, a step, an operation, an element, a component or a combination thereof described in the specification, But do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, 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 contextual meaning of the related art and are to be interpreted as either ideal or overly formal in the sense of the present application Do not.

Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the accompanying drawings.

In the present invention, a trust relationship does not necessarily mean a mutually trusting relationship, but can encompass both a trust relationship and a distrust relationship. Therefore, the trust relationship itself can not understand whether it implies trust or distrust. For example, a trust relationship may be a trust relationship, but it may be a distrust relationship. Likewise, an extrinsic trust relationship may be a trust relationship, but it may be a distrust relationship. Therefore, in order to distinguish the trust relationship of distrust from the trust relationship of distrust, below, it shall be described separately such as inner trust relationship of distrust, outer trust relationship of trust, trust relationship of distrust and trust relation of trust.

FIG. 1 is a flowchart illustrating an item recommendation method using an inherent and extrinsic trust relationship according to an embodiment of the present invention.

Due to the exponential growth of the Internet, information overload is becoming a major problem. Finding relevant and important information has become a tedious task for users. Various techniques have been proposed to overcome this information overload problem. Recommendation system technology, one of the most promising solutions to this problem, looks for information that may be of interest to the referral target. This information may include recommendation items, products, contents and activities called items. Also, in order to achieve the goal, the recommendation system searches for an item that is relevant to the recommendation target, and the behavior of the target user stored as data including information on ratings, clicks, and bookmarks of the item It is utilizing the history. Collaborative filtering and content-based filtering are the most commonly used techniques in recommendation systems. The present invention focuses on collaborative filtering, which identifies items to be recommended to a recommendation target person by referring to an evaluation of a preference of the recommendation target person.

However, despite the considerable success of collaborative filtering, there are still some inherent data-related problems. Collaborative filtering does not make good recommendations when most users tend to only evaluate a small number of items. Also, it happens frequently in the real world. For example, MovieLens, Epinions, and Netflix, the most commonly used datasets for evaluating a recommendation system, have a 95-98% scarcity. This problem is called a data sparsity problem. In addition, collaborative filtering does not provide effective recommendation items for users who have evaluated only a few items. This problem is called a cold-start user problem.

Several approaches have been proposed in the literature to solve this problem. Most of these approaches use additional information such as trust information, demographic information, and location information along with the user's ratings information. The present invention focuses on an approach that uses information from a trust network. In the following, this approach is referred to as a trust-aware recommendation approach. We often ask for referrals to people we trust in our daily lives (colleagues or friends), and the same concepts can be applied to referral systems. Epinions are the most widely used trust network in the literature for reviewing and rating items and establishing trust and / or distrust relationships with other users.

A trusting user in a trust network is called a trustor and a trusted user is called a trustee. Similarly, a user who creates a distrust relationship with another user is called a distruster, and a user who is distrusted by another user is called a distrustee.

In step S110, based on the evaluation information including the result of evaluating at least one of the plurality of items by each of the plurality of users, the item recommendation apparatus determines whether or not the item recommended Inferential implies a trust relationship.

Here, an implicit trust relationship is another type of relationship between two users of a trust network. For example, between two users based on evidence that they have similar preferences or belong to the same cultural group (e.g., nationality, age, hobby, etc.). Although this type of relationship is not explicitly set by the user, explicit and implicit relationships may equally be important in improving the accuracy of the recommendation.

The item recommendation apparatus can deduce whether there is an intrinsic trust relationship between the two users by using the evaluation information that is the result of evaluating a plurality of items by two users. For example, the item recommendation device may deduce that there is a continuum trust relationship if the two users have similar preferences for the same item.

At this time, when the number of items commonly evaluated by the two users is smaller than a predetermined threshold value h, the item recommendation apparatus can determine that there is no MRF relationship between the two users. That is, the two users may not be in a trust relationship, nor in a distrust relationship.

On the other hand, the evaluation information may include a result of each of a plurality of users evaluating or not evaluating a plurality of items, and the evaluation information may be stored in a database and inquired as needed. It goes without saying that the types and the numbers of the items evaluated among the plurality of items may be different for each user.

Further, a plurality of two-person combinations composed of two of the plurality of users can be determined to be a maximum _N C ₂ when a plurality of users are N persons. However, it is not so limited, and it can be determined within a range of _N C ₂ by a predetermined criterion.

In the present invention, the main reason for using the inherent trust relationship is that there are many environments in which information on the apparent trust relationship is not provided. In other words, the implicit trust relationship can contribute greatly to improving the accuracy of matrix decomposition based on the recommendation approach. The reason is that the cold-start user has little evaluation results, and the matrix decomposition uses the user's evaluation results to infer the user's preference. As a result, matrix decomposition may not infer the actual preference of the coldstart user. Therefore, if there are some of the intrinsic trustees in the case of a coldstart user, the preferences of the coldstart user can be inferred by being similar to the intrinsic trustee's basic preference.

At this time, it may be desirable to integrate the inner trust relationship even if the outer trust relationship is available, since the information about the outer trust or distrust is quite sparse. In other words, the use of additional information (persistent trust relationship) can contribute to more accurate learning of the user's feature vector.

In another embodiment, the item recommendation device may infer the interlocking trust relationship using similarity.

That is, the item recommending apparatus can infer the interlocking trust relationship using the degree of similarity through the following steps S113 and S116.

In step S113, the item recommendation apparatus calculates the similarity among the members for each of the plurality of two-person combinations using the evaluation information.

For example, the item recommendation apparatus can calculate the degree of similarity among members for each of the plurality of two-person combinations using PCC (Pearson correlation coefficient). At this time, the similarity sim (i, u) between the users i and u using the PCC can be calculated as Equation 1 below.

[Equation 1]

및

는 모든 아이템에 대한 사용자 u 및 i의 평가의 평균값이다.Here, i and u are users, I is a set of items commonly evaluated by users i and u, r _ij is a score obtained by the user i evaluating the j th item of I, and r _uj is the score of user j The second item,

And

Is the average value of the ratings of users u and i for all items.

At this time, the item recommendation apparatus can calculate a similarity score when two users commonly evaluate at least h items in order to make sim (i, u) a non-zero value. If the number of items evaluated in common between two users is less than h, it is considered that there is no inherent trust relationship (i.e., trust or distrust) between the two users.

In step S116, the item recommendation apparatus deduces the internal trust relationship of each of the plurality of two-person combinations based on the predetermined threshold and the calculated similarity.

For example, the item recommendation apparatus can infer that, when the similarity degree between two users is higher than a predetermined threshold value _s , there is an implicit trust relationship of trust between the two users. Likewise, the item recommendation apparatus can infer that there is an implicit trust relationship of distrust between the two users when the similarity degree between the two users is equal to or less than a predetermined threshold value _s .

However, in some extrinsic relationships, similarity scores are not available when the number of joint endpoints is less than h. In this case, the conventional assumption that the trustee and the trustee have similar preferences and that the unbeliever and the unbeliever have similar preferences.

In step S120, the item recommendation apparatus verifies the external trust relationship with respect to at least one combination in which there is an explicit trust relationship among the plurality of two-person combinations based on the evaluation information.

Here, an explicit trust relationship is a relationship of trust or distrust between the two users of the trust network. For example, a trust relationship or a distrust relationship between two users of a trust network may be explicitly disclosed. At this time, the item recommendation apparatus can verify the external trust relationship between the two users by using the evaluation information.

At this time, the reason for the verification of the outer trust relationship is that the user can have similar or similar preferences to the trusted and unbeliever, regardless of the trust or distrust in the trust network. Therefore, it may be necessary to confirm the similarity and non-similarity, respectively, with respect to the trustee and the unbeliever who have an outer trust relationship. The results of this study are as follows. First, the user 's preference is similar to the preference of the untrusted trust. I can reason. In other words, it is effective to infer the user 's preference from the trustee who has the most similar preference through such verification, and distance from the most unlikely unbeliever.

In another embodiment, the item recommendation apparatus may verify the implicit trust relationship using similarity.

At this time, the item recommending apparatus can verify the outer trust relationship using the similarity degree through the following steps S123 and S126.

In step S123, the item recommendation apparatus calculates the degree of similarity between the members for each of at least one combination in which an explicit trust relationship exists, using the evaluation information.

At this time, the item recommendation apparatus can calculate the degree of similarity between the members of the two-person combination using Equation (1) described above.

In step S126, the item recommendation apparatus verifies the external trust relationship of each of the at least one combination based on the predetermined threshold and the calculated similarity.

That is, when the similarity degree between the members is equal to or greater than the threshold value ( _s ), the external trust relationship of their trusts is verified with respect to the combination (trustee, trusted) . In addition, the item recommendation apparatus may be configured such that, when a non-mutual trust between members is greater than or equal to a threshold value ( _s ) with respect to a combination (unbeliever, unbeliever) It can be judged.

However, in the case of an explicit trust relationship, the similarity may become unusable when the number of joint evaluation items among members is less than h. In this case, we can follow the existing assumption that (trustee, trustee) has a similar preference, and (unbeliever, unbeliever) has a dissimilar preference.

Finally, in step S130, the item recommendation apparatus determines an item to be recommended to a recommendation target, which is one of a plurality of users, from among the plurality of items, using the inferred intrinsic trust relationship and the verified extrinsic trust relationship.

That is, the item recommendation apparatus can determine the recommendable item from a plurality of items using the inferred inherent trust relationship or using the verified explicit trust relationship.

More specifically, the item recommendation apparatus identifies the internal trust relationship and the external trust relationship existing between the users as a relationship of trust or distrust, and then selects a similar preference (trustee, trustee) , The item to be recommended can be determined based on the assumption that the trust relationship of the untrusted (unbeliever, unbeliever) has similarity.

In other embodiments, the inner trust relationship and the outer trust relationship may have a sign of a positive sign indicating trust and a sign of negative sign indicating distrust.

That is, as mentioned above, the trust relationship can encompass both the relationship of trust and distrust, and the code can be a means of determining whether there is a relationship of trust or distrust to the trust relationship. More specifically, a positive sign indicates a relationship of trust, and a negative sign indicates a relationship of distrust.

Similarly, a signed trust relationship can encompass both the relationship of trust and distrust, and an unsigned trust relationship can only mean the presence or absence of trust.

In yet another embodiment, the item recommender may determine based on the intransitivity of a trust relationship with a positive sign and the intransitivity of a trust relationship with a negative sign when determining an item to recommend.

For example, assume that user E distrusts user C and user C distrusts user F. [ Items shared by the three users C, E, and F are only one item with a score of 2, 5, and 5, respectively. According to the assumption about the trust relationship of distrust, since user E distrusts user C, users E and C should have similar preferences. Users C and F should also have similar preferences for the same reason. In this example, the assumption that a pair (unbeliever, unbeliever) is different in terms of preferences is confirmed by their assessment. Now, applying the metaphor of distrust, user E must mistrust user F, and therefore both users E and F should be treated as having different preferences. From their scores, however, we can see that users E and F are very similar in preference. Therefore, it may be desirable to use the metonymics of trust metaphors and distrust in the referral approach.

As described above, the item recommendation method using the inherent and extrinsic trust relationships according to the embodiment of the present invention alleviates the data sparseness problem and the cold start problem by using the inner and outer trust relations, There is an effect that can be performed.

FIG. 2 is a block diagram for explaining an item recommendation apparatus using an inherent and extrinsic trust relationship according to an embodiment of the present invention.

Referring to FIG. 2, an item recommendation apparatus 200 using an inferior and extrinsic trust relationship according to an embodiment of the present invention includes a reasoning unit 210, a verification unit 220, and a recommendation decision unit 230 do.

Meanwhile, the item recommendation apparatus 200 using the inherent and extrinsic trust relationships according to an embodiment of the present invention can be used for recommending items installed on a PC, a notebook, a smart phone, a tablet PC, a server, and the like.

The inference unit 210 may be configured to determine, based on the evaluation information including the result of evaluating at least one of the plurality of items, a plurality of users each having a plurality of two- Infer the relationship.

In another embodiment, the inference unit 210 may calculate the degree of similarity between members for each of a plurality of two-person combinations using the evaluation information, and based on the predetermined threshold and the calculated degree of similarity, Can be inferred.

The verification unit 220 verifies the external trust relationship with respect to at least one combination in which an outward trust relationship exists among the plurality of two-person combinations, based on the evaluation information.

In another embodiment, the verification unit 220 uses the evaluation information to calculate the degree of similarity between the members for each of at least one combination in which there is an explicit trust relationship, and based on the predetermined threshold and the calculated degree of similarity , The external trust relationship of each of the at least one combination can be verified.

Finally, the recommendation deciding section 230 decides an item to be recommended to a recommended person, which is one of a plurality of users, from among the plurality of items, using the inferred inner trust relationship and the proven outer trust relationship.

In other embodiments, the inner trust relationship and the outer trust relationship may have a sign of a positive sign indicating trust and a sign of negative sign indicating distrust.

In yet another embodiment, the recommendation decision unit 230 may determine based on the non-connectivity of the trust relationship with the positive sign and the positive sign of the trust relationship.

Meanwhile, the approach of the present invention, together with the existing trust-recognition recommendation approach, will be described as follows.

Review existing collaborative filtering methods. In particular, the method based on matrix factorization is discussed in detail since the present approach is based on it. We then briefly describe the trust-aware recommendation approach that incorporates unsigned trust relationships. Finally, we introduce a trust - awareness recommendation approach that incorporates a signed trust relationship and point out the problem of that approach.

Simplicity, efficiency, and accuracy are important characteristics of collaborative filtering that are widely used in recommendation systems. There are two types of collaborative filtering: memory-based and model-based filtering. User-based collaborative filtering and item-based collaborative filtering are the most commonly used memory-based methods. User based (item based) collaborative filtering generates referrals for the target user using similarity relationships between users (items). A method based on matrix decomposition in model - based methods has gained popularity in recent years. The matrix decomposition based method is effective and linearly expands according to the number of evaluations.

The matrix R? ^D nxm of ^nxm is a user-item evaluation matrix, where each item r _ij represents an estimate given by user i to item j. The goal of matrix decomposition is to obtain two potential low-dimensional matrices U and V such that U? D ^nxk and V ^? D ^kxm can effectively recover the evaluation matrix R = UV.

In particular, the matrix decomposition attempts to obtain the potential user feature matrix U and the potential item feature matrix V by minimizing the following objective function.

&Quot; (2) "

Where ² _F represents the Frobenius norm, and the first term minimizes the difference between the user evaluation and the corresponding prediction estimate of the user obtained by U and V. This first term is called the factorization part. The two terms λ _U and λ _V are regularization parameters and are used to prevent over-fitting. These two terms can be called the regularization part. The evaluation r _{i, j} of user i for item j, which is not evaluated _, can be simply predicted by the dot product of the corresponding feature vectors U _i and V _j .

Let P _i and N _{i be the} set of trusted and unfriended individuals of user i in the previous step. Further, assume that all users included in P _i are similar to i. Thus, the preference of _i may be as close as possible to the user's preference contained in P _i .

In order to reflect the propagation of trust, the result of adding these terms to the normalization part of equation (2) is as follows.

&Quot; (3) "

는 사용자 i의 피신뢰자의 잠재적인(latent) 특징 벡터의 평균이다. 수학식 3에서 볼 수 있듯이, 사용자 i의 특징 벡터는 그 사용자의 직접적인 피신뢰자(direct trustees)의 특징 벡터에 종속적(dependent)이다. 재귀적으로, 그 피신뢰자의 특징 벡터 각각은 그 피신뢰자의 직접적인 피신뢰자의 특징벡터에 종속된다. 이러한 방법으로, 본 발명의 접근법에서 신뢰의 전파가 반영될 수 있다.Here, λ _s is a social regularization parameter,

Is the average of the latent feature vectors of the trustees of user i. As can be seen in Equation (3), the feature vector of user i is dependent on the feature vector of the direct trustees of the user. Recursively, each of the feature vectors of that trustee is dependent on the feature vector of the directly trusted trustee of that trustee. In this way, the propagation of trust can be reflected in the approach of the present invention.

In a similar manner, the user's preference may need to be kept as far from the user's unbeliever as possible. Since distrust is a feature of non-sense, unbeliever can be integrated into the decomposition part of (1) as follows.

&Quot; (4) "

는 사용자 i의 피불신자의 잠재적인(latent) 특징 벡터의 평균이다. 수학식 4는 N_i의 특징 벡터가 i의 특징 벡터 대신에 i의 평가에 영향을 미친다라는 점을 나타낸다. 이 점은 본 발명의 행렬 분해 모델이 불신의 전이성을 신뢰의 전이성과는 다르게 통합하고 있지 않아, 불신이 신뢰 네트워크를 통해 전파되는 것을 막고 있다는 것을 나타낸다.Here, alpha is a variable for tune the user's preference and the preference of the user's pseudocubic,

Is the average of the latent feature vectors of the user i's unbidden. Equation 4 shows that the feature vector of N _i affects the evaluation of i instead of the feature vector of i. This indicates that the matrix decomposition model of the present invention does not incorporate the metastability of distrust differently than the metastability of trust, thereby preventing distrust from propagating through the trust network.

If we obtain a set of trusted and unfriendly users for all users, we can model trust-aware recommendations that effectively integrate the implicit and explicit unsigned trust relationships using the evaluation information. Most existing trust - aware recommendations are based on matrix factorization techniques, taking efficiency and scalability into account. The present invention also uses a matrix decomposition technique and suggests an integrated model that utilizes trust and distrust relationships for effective recommendation. The proposed model solves the following optimization problem.

&Quot; (5) "

The local minimum value of the objective function used in Equation (6) can be found through a gradient descent for U _i and V _j .

&Quot; (6) "

Here, ψ _i and ψ _i are the unbeliever set and trust set of user i, respectively.

On the other hand, an experiment showing the effect of the proposed technique of the present invention on an existing excellent trust-recognition recommendation approach has been performed. These experiments were performed to answer the following key questions about the approach of the present invention.

(Q1) Is the proposed set of trust relationships really effective in improving the recommendation accuracy of the trust-awareness recommendation approach?

(Q2) Does the propagation of trust and non-propagation of distrust help to improve referral accuracy?

(Q3) Will the coldstart user get a more accurate recommendation when following the approach of the present invention?

(Q1), we use four types of signed trust relationships with SimSocial (our approach) and RecSSN (competitive approach). The four types of trust relationships are ExpRel, ClassExpRel, ImpRel, and (ClassExp + Imp) Rel. ExpRel is an exponential relationship between users in the original trust network. The ClassExpRel is an exponential relationship obtained by classifying ExpRel using the method of identifying the similarity and non-similarity of the above-mentioned extrinsically trusted and unbeliever. ImpRel is an implicit relationship created using a combination of the above-mentioned unsigned intrinsic and extrinsic trust relations. (ClassExp + Imp) Rel is simply a combination of ClassExpRel and ImpRel.

The results are shown in Fig. SimSocial is consistently superior to RecSSN under all types of trust. Both approaches also perform better on ClassExpRel than on ExpRel. This result supports the assertion that a pair of users may have different preferences despite having a trust relationship in the trust network. By classifying the extrinsic relationships based on the similarity scores between users, non-similar users that bias the user's feature vector may be excluded from the recommendation in the approach of the present invention. This method helps to learn the user's true preference, thereby making recommendations more accurate.

From the results of Fig. 4, we can observe the effects of handling the metonymity of trust metaphors and distrust. RecSSN works only in explicit, signed trust relationships, and it incorporates not only distrust but also the meta-trust. SimSocial also outperforms RecSSN in the case of explicit signed trusts.

On the other hand, both methods have somewhat less performance in ImpRel than ExpRel. The reason is that the continuity relationship is obtained from the rating data. Cold-start users have few evaluations, so there may not be a sufficient number of inherent relationships. Therefore, coldstart users may still not receive effective recommendations due to insufficient information. As a result, the overall accuracy of the recommendation approach can be reduced. Interestingly (ClassExp + Imp) Rel allows the trust-aware recommendation approach to achieve the best accuracy. This is because the combination of the two types of relationships increases the amount of information inferring the user's feature vector. This in turn contributes to understanding the user's true preferences and helping in a recommendation approach to creating effective referrals.

Figures 5 and 6 show the answers to (Q2) and (Q3) and show the calculated RMSE (Root Mean Square Error) and MAE (Mean Absolute Error) values for all users and cold start users. At this time, RMSE and MAE are the two most famous matrices for verifying rating prediction. All trust-based recommendation approaches perform better than MF, indicating that incorporating additional information on trust relationships is important in referral. Among the trust - awareness recommendation approaches, Impl shows the worst performance and SimSocial shows the best performance. Since Impl considers only the inter-user relationship between users, it can not provide a good recommendation item to the cold-start user as shown in Fig. Furthermore, it incorporates the transitivity of distrust. Thus, it can be seen that these two elements are less accurate.

It is interesting that PushTrust performs worse than RecSSN. The reason is that PushTrust learns user preferences from the user's most similar external and neutral trustees. Such extrinsic trustee is chosen most similar to a user whose similarity score is equal to or greater than the user's unbidden or neutral trustee. At this time, it may be possible that the neutral trusted user or unauthorized user of the user includes a more similar user than the trustee of the user. In this case, all outlier trusted are ignored, and the user's preferences can only be learned from the user's evaluation.

As a result, the overall accuracy of PushTrust is reduced. RecSSN, on the other hand, is surprisingly superior to PushTrust, but not better than SimSocial. This is because the RecSSN only considers the outer signed signatures, and therefore may not find an effective recommendation item for a user with little or no trust. Also, considering the metastability of distrust, user characteristics may not be properly modeled and accuracy may be degraded. On the other hand, SimSocial performs very well. The reasons are (1) a sufficient number of trust relationships, (2) similarity and dissimilarity identified by classification based on similarity scores, and (3) proper propagation of trust and distrust. All of these factors help the recommendation process to accurately learn the user's preferences and provide more satisfying recommendation items.

Thus, the present invention proposes a new trust-aware recommendation approach based on matrix decomposition using both coherent and explicit signed trust relationships. To more accurately grasp user preferences, we increase the number of trust relationships to include implicit trust relationships in addition to outer trust relationships. Based on the actual similarity score, we confirm the similarity and dissimilarity between intrinsic and extrinsic trust relationships. In addition, we propose a new model that learns the feature matrix of users and items by effectively handling the visibility of trust meta and distrust. Experiments on large real-life data sets show that the proposed trust-aware recommendation approach has consistently performed better than the other approaches.

The above-described embodiments of the present invention can be embodied in a general-purpose digital computer that can be embodied as a program that can be executed by a computer and operates the program using a computer-readable recording medium.

The computer readable recording medium includes a magnetic storage medium (e.g., ROM, floppy disk, hard disk, etc.), optical reading medium (e.g., CD ROM, DVD, etc.).

The present invention has been described with reference to the preferred embodiments. It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims. Therefore, the disclosed embodiments should be considered in an illustrative rather than a restrictive sense. The scope of the present invention is defined by the appended claims rather than by the foregoing description, and all differences within the scope of equivalents thereof should be construed as being included in the present invention.

Claims (10)

A method for recommending an item using an item recommending apparatus using an inner and outer trust relationship,
Based on evaluation information including a result of evaluating at least one of a plurality of items by each of a plurality of users, inferring an implicit trust relationship relating to a plurality of two-person combinations composed of two of the plurality of users ;
Verifying the implicit trust relationship for at least one combination in which an explicit trust relationship exists among the plurality of two-person combinations based on the evaluation information; And
Determining from among the plurality of items an item to be recommended to a recommending target person, which is one of the plurality of users, using the inferred ICT and the verified outer trust relationship
Lt; / RTI >
The inherent trust relationship and the implicit trust relationship
A positive sign indicating trust, and a negative sign indicating distrust,
The step of determining the item to be recommended
Wherein the determination is made based on the transitivity of the trust relationship having the positive sign and the intransitivity of the trust relationship having the negative sign. delete delete The method according to claim 1,
The step of inferring the inherent trust relationship
Calculating similarity between members of each of the plurality of two-person combinations using the evaluation information; And
Inferring the inherent trust relationship of each of the plurality of two-person combinations based on the predetermined threshold and the calculated similarity
And an item recommendation method using an inherent and extrinsic trust relationship. The method according to claim 1,
The step of verifying the implicit trust relationship
Calculating the degree of similarity between members for each of at least one combination in which the external trust relationship exists using the evaluation information; And
Verifying the implicit trust relationship of each of the at least one combination based on the predetermined threshold and the calculated similarity
And an item recommendation method using an inherent and extrinsic trust relationship. Based on evaluation information including a result of evaluating at least one of a plurality of items by each of a plurality of users, a reasoning unit for inferring an inner-strength trust relationship relating to a plurality of two- ;
A verification unit that verifies the external trust relationship with respect to at least one combination in which an outward trust relationship exists among the plurality of two-person combinations based on the evaluation information; And
A recommendation deciding unit for deciding among the plurality of items an item to be recommended to the recommending target person, which is one of the plurality of users, using the inferred inner trust relationship and the verified outer trust relationship,
Lt; / RTI >
The inherent trust relationship and the implicit trust relationship
A positive sign indicating trust, and a negative sign indicating distrust,
The recommendation decision section
Based on the non-transference of the trust relationship having the positive sign and the non-transference of the trust relationship having the negative sign. delete delete The method according to claim 6,
The reasoning unit
Calculates the degree of similarity among the members for each of the plurality of two-person combinations using the evaluation information,
Based on the predetermined threshold and the calculated degree of similarity, the inferior continuity relation of each of the plurality of two-person combinations is inferred. The method according to claim 6,
The verification unit
Calculating the degree of similarity among the members for each of at least one combination in which the external trust relationship exists using the evaluation information,
And the external trust relationship of each of the at least one combination is verified based on the predetermined threshold and the calculated similarity.

Images