Combining trust and reputation as user influence in cross domain group recommender system (CDGRS)

Abstract

Group recommender system provides suggestions for a group of users by exploring the choices of individual users of the group. Popularity of group recommender systems is increasing because many activities such as listening to music, watching movies, traveling, etc. are normally performed in groups rather individually. Group recommender systems like personal recommender systems also suffer from cold start and sparsity issues. The cold start and sparsity issues result into inaccurate recommendation computation which degrades the recommendation quality. To handle the cold start and sparsity issues in a Group Recommender System (GRS), this paper proposes to use cross domain approach and introduces Cross Domain Group Recommender System (CDGRS). The recommendations provided by trustworthy and reputed users in the group enhance the acceptance towards the presented recommendations as compared to the other individuals in the group. We have combined the social factors e.g. trust and reputation to get influential user in the group recommendation. A prototype of the system is developed for tourism domain that incorporates four sub-domains i.e. restaurants, hotels, tourist places and shopping places. The performance of CDGRS is compared with GRS. Spearman’s Correlation Coefficient, MAE, RMSE, Precision, Recall and F-measure are used to find the accuracy of the generated recommendations.

Keywords

Recommender system cross domain group recommender system multi-agent system user influence

1 Introduction

Due to ease of access, an exponential growth in the information has been recorded in the digital world resulting in information overload for the user. As the information overload problem increases, it becomes confusing for the end user to select the desired information. Recommender systems (RSs) helps to assist the user in the decision making process. RSs is an information filtering technique to deal with the information overload problem [1 –3]. They are broadly classified into three categories; collaborative filtering, content based and hybrid RSs [4, 5]. The basic idea behind the collaborative approach is that if two users have shared the similar choices in the past then they might share the same in future also. It is further categorized user-user collaborative filtering RSs and item-item collaborative filtering RSs. User-user collaborative filtering finds the similarity between the current (target) user and other users of the domain whereas Item-item collaborative filtering [6, 7] finds out the similarity among the items rather than users.

There are many activities which people would like to carry out in groups rather than individually e.g. consider an example where group of people wants recommendation for places to visit. To compute the group choices the system exploits the past individual preferences or any group preferences if available. So, recommendation for places to visit requires both the group choices as well as individual choices. Group recommender system provides suggestions for group of users based on collection of individual choices of users. Group recommendation is generated on the basis of the captured preferences using the profile of the users. Mostly the group choices are composed of the individual choices or preferences. As the similarity of likeness increases among the users in the group, the stronger group will be formed which results into effective recommendation generation. It is important to remove the problems associated with the ratings that reduce the effectiveness of presented recommendation e.g. the problem of cold start and sparsity. One of the solutions to the cold start and sparsity problem is cross domain approach.

In literature, cross domain approach is found as one of the potential solution to the sparsity and new user cold start problem [8, 9]. The goal of Cross Domain Recommender System (CDRS) is to exploit the ratings available in one domain, called source domain, and suggest the items in another domain, called target domain [10]. Lack of ratings in rating matrix may result into the inaccurate prediction computation. Mediation of user modeling data provided by other source domain enriches the available data for the target domain. This mediation process helps to increase accuracy in the prediction because in spite of less ratings of particular domain it is getting enriched data which helps to generate accurate prediction for the target user [11]. Sparsity is one of the major problems that a group recommender system faces. The problem occurs because of the unavailability of enough ratings in the system. Cold start problem in RSs is another problem due to the unavailability of the user or item details.

User Influence plays an important role in online social network and it has been effectively utilized in e-commerce business, communication and marketing also because it helps to explore the businesses recent trends [12, 13]. E-commerce and online social network effectively exploit the user influence to get the common interest or sometimes helps to get the breaking news. Adopting the concept of user influence in individual and group recommendation systems helps to get effective list of suggestions for the target user. Although, social influence has been extensively explored by various researchers in various areas, it will be interesting to obtain the measure in terms of trust and reputation that result into influence of a user. Trust in recommendation is always considered as one of the major factor that influences the decision making system. Trust of one user over another is considered as the positive encounter between the users. Positive encounter refers to the encounter between the two users in which a user is able to present accurate recommendation for another user [14]. Reputation of user is one of the deciding factors to determine the opinion about a user. Reputation in terms of recommendation can be stated as the opinion or view of community of user over one user [15]. Reputation is linked to subjectivity, it is considered as subjective opinion of a user. So, reputation of a user can be considered as the public belief or trust over one user. In other words it is a specific approach to discuss the trust between users.

In this paper, we combine the cross domain approach with the group recommendation which helps to solve the problems of sparsity and cold start. This paper utilizes the user influence using trust and reputation to compute the accurate recommendation. Since trust and reputation is considered as one of the major source to incorporate social notion in the recommendations, this paper incorporate the influence of a user characterized by trust and reputation.

Rest of the paper is organized as follows: in Section 2 literature survey is briefly presented describing the related topics. Proposed work and architecture of the system has been discussed in Section 3. Section 4 comprises of the algorithm. Finally experimental and evaluation are presented in Section 5 followed by the conclusion.

2 Literature survey

In the group recommendation each member of the group is important and to incorporate their opinion, a model is proposed by [16] that uses weighting members according to the degree of importance. Providing recommendation to a group rather than individual is complicated. This involves aggregating individual user’s preferences or recommendations. It needs various issues to be tackled while presenting recommendation to the group of users as mentioned in [17]. To generate quality in the group recommendation [18] has proposed Cascade TOPSIS. It selects the group recommendation from the group which consist only those members that provides good quality of recommendation.

Various domains are explored in the area of group recommendation where tourism is one of the most widely explored domains. A hybrid technique (combination of collaborative approach, content based approach and demographic) is proposed by [19] which recommends tourist attraction to the group of users. It unfolds the type of user’s relationship and suggests recommendation in the tourism domain to the group of users. The same domain is exploited by [20] to recommend travel destination to the group of users that uses hybrid recommendation approach. It includes various details of user i.e. user interest, the ratings provided by them to the items and other specific requirement. In some earlier work [21] various tourist activities are recommended to the user. It exploits the previous visited places along with the demographic details of the user. A conversational based collaborative group recommender CATS (Collaborative Advisory Travel System) is provided by [22] whose main focus is to provide planning skiing vacation to a group of users. Group recommendation can also be exploited in other domains e.g. suggesting movies to the group of users which incorporate the interest of the group as well as individual. A group recommendation is provided by [23] in the movie domain that uses collaborative filtering technique. They have combined the item-item and user-user collaborative filtering that provides homogeneous groups of user for the efficient group recommendation.

[24, 25] Provide an analysis of group personality composition. Role of the groups is very important in group recommendation. [26] made an effort for using different social choice strategy to form a group. Group recommendation is provided based on the prediction aggregation along with the group formation. In group recommendation every member’s contribution is important but even in group degree of importance of every user effect the group recommendation. PolyLens [27] provides recommendation for the group of users rather than individual using the collaborative filtering approach. They considered various issues of group recommender e.g. analysis of the primary design of group recommendation, nature of the group, social value functions, rights of the group members etc. There are various issues that are involved with a group recommender system. [28] raises some of the important aspects of group recommendation using a TRAVEL DECISION FORUM. After presenting recommendation they have suggested to visualize the effect of presented recommendation i.e. whether user is satisfied or least satisfied. [29] has presented various details about the group recommender system. They have presented various applications of the group recommender systems in various domains of music and TV programs, travel destination and events as well as various domain independent recommender systems. The evaluation of the group recommendation is similar to that of the evaluation of individual recommendation. Usually the user’s group is collected using the traditional approach of recommendation that means it is not possible to form group intrinsically. Another effort is made by [30] to find out the group by exploiting the individual recommendation and a classic clustering algorithm that identifies and form a group. The effectiveness of providing group recommendation by exploiting the personal preferences of the group members is examined by [31]. They accessed various recommendation algorithms in terms of accuracy, diversity, coverage and serendipity.

[13] has determined the influence of user using directed link and three measure of influence are used: one is in-degree, re-tweet and mentions. The reason for considering measures other than in-degree is that it is not necessary that the user that has higher in-degree may not be influential in terms of re-tweet and mentions. The accuracy for the provided recommendation is an important aspect while presenting a list of suggestions. In [32] trust information is combined to measure the influence of a user to generate future recommendations. Another effort is made by [33] to explore the benefits of user influence in marketing. A survey has been presented by [34] to measure the influence of a user on Online Social network e.g. Twitter. They have collected and classified different twitter influence measures. [24] provides an approach that uses social factors to generate the recommendation for group of user. They explored the similarity between the members of the group and then extract the user having higher influence that provides group’s opinion. [25] has provided a review that searches for various techniques in group recommendation. They have studied the effect of various social factors, e.g. trust, in group recommendations. A survey is provided by [35] based on three social computing services: recommender system, trust/reputation system and social networks along with two challenges: cold start problem and sparsity.

Trust and reputation has always played as two pillars of recommendation approach. [14] has argued that by adding additional factors such as trust improves the recommendation accuracy. A state of the art is presented by [36] for trust based recommendation along with the benefits using trust-enhanced recommendation. [37] has generated reputation score for the users by propagating trust. Trust metric helps to get the similarity assessment of users while reducing its computability. [38] explores the trust relationship that aggregate the opinions of other users to generate personalized recommendations. A fuzzy computational model is presented by [39] to compute both trust and reputation.

In this paper, we have combined the trust and reputation to generate the influential user that guide the system to generate the recommendation list for the other users in the group. The cross domain approach is considered when the system does not have enough rating for the target user. The combination of similar users and influential user leads to the final neighborhood for the target users.

3 Proposed trust and reputation based cross domain group recommender system (CDGRS)

The proposed work combines the group recommender system with cross domain approach. The cross domain approach helps to reduce the sparsity and cold start problem. The trust and reputation of a user manages to get the influence of user. The influential user helps to get the recommendation for the group of users. In group recommendation, one of the most important aspect is user profiling. Further, the subsection discusses the architecture and working of the proposed system.

3.1 User profiling of CDGRS

The proposed approach collects and stores the choices and interest of user as well as their demographic details and contextual preferences. The demographic details include the information about the user e.g. name, age, gender etc. The contextual preferences include the details about the user interest e.g. for the tourism domain which companion they prefer, budget preferences for the purpose of stay or to dine, any location or locality of preferences for the accommodation etc. The contextual information, demographic details and the ratings provided by the user is valuable which reflects varying user interest and also the likings of user towards the presented recommendation. So the system stores the related information and maintains the ratings along with the demographic details of the user in the repository.

3.2 Architecture of CDGRS

The architecture of the proposed system is shown in Fig. 1. The proposed system is a multi-agent based system where multiple agents help to get the desired information and various computations periodically. The group formation in the proposed approach is achieved using the similarity between users. In the group recommender system mostly the group choices are composed of the individual choices or preferences. While computing similarity between users, the ratings provided by the user plays an important role. Absence of enough rating results into the problem of sparse rating matrix and cold start problem. Cross domain approach uses mediation (i.e. import and aggregation) of data from various domains to solve the sparsity problem. In any e-commerce site exploiting user ratings for items in various domains will be helpful and suggest personalized recommendations of items that belongs to multiple domains e.g. suppose user have rated for movies then by suggesting books, music, video games, which are by some means related to the movies, offers as personalized multi-domain recommendation.

Fig. 1

Architecture of the proposed system.

In tourism domain, we have various sub-domains e.g. places to visit, places to dine at, places to stay for accommodation purpose and places to shop for shopaholic users. Since tourism has various domains so if there is less data in one domain then it can be enriched by other domains in the system. This will help to deal with sparse data and cold start problem. When a new user enters into the system then it may be possible that the user has shared some ratings with other domain which helps to get the neighborhood in the domain. Once the target domain doesn’t have enough rating, it requests the source domain to share the neighborhood for the target user. On the request of target domain the source domain computes the neighborhood for the target user (assuming the system is domain distributed where each domain shares the same structure locally and user identity is shared among the domains).

The remote domain compute the similar user for target user locally using the agents of the respected domain. After having the similar user the remote domain responds to the query by giving the neighborhood and the similarity score to the target domain. Upon receiving the response by the remote domains, target domain computes the overall similarity score between target user and other users. This results into the formation of neighborhood for the target user in the target domain. To compute the similarity between users similarity measure that helps to provide neighborhood for the target user Pearson’s correlation coefficient (PCC) is used. The set of similar user for the target user along with their similarity score is provided to the target domain. This provides the opportunity to form a group on the basis of similarity score of the users even if the particular domain doesn’t have enough ratings. As soon as the system has enough ratings then it forms group by collecting users having closest similarity score.

The aggregation of individual choices is done on the basis of user’s previous ratings followed by the user profiling. The aggregation of individual choices involves with the interest of the users. Multiple agents involve with the periodic update of the user’s interest that also helps to capture the variation in the choices. The similarity score of the users leads to the computation of confidence between users and reputation score of user. The final similarity score initiates the trust and reputation computation where trust computation is processed using confidence between users, similarity between users and reputation score. This provides the trust and reputation score of user that leads to categorizing influential user on the basis of trust and reputation score. The influential users are ranked and help to generate recommendation targeting the individual choices as well as the group choices.

4 Proposed approach

The processing of the system is divided into two phases, one is offline phase and another is online phase. These phases are described below:

4.1 Offline phase

The offline phase deals with the similarity computation between target user and other users in the system. This phase also compute the inter-domain correlation computation between domains. The computation steps are as follows:

a) Formation of input data

User’s preferences are stored within data repository in the form of user-item rating matrix for all recommendation items (hotel, restaurant, shopping places and places to visit). This input matrix consists of ratings in the discrete scale 1 to 5 and it contains a value 0 for unrated items. Each domain consists of two dimensional rating matrices which shares the same structure.

b) Group Similarity computation

Individual users are selected for the group formation whose similarity with the other user is above some threshold. To compute the similarity score between the individual users Pearson’s correlation coefficient [40] is used and is formulated as follows: $Sim (x, y) = \frac{\sum_{i = 1}^{n} (r_{xi} - {\bar{r}}_{x}) * (r_{yi} - {\bar{r}}_{y})}{\sqrt{\sum_{i = 1}^{n} {(r_{xi} - {\bar{r}}_{x})}^{2}} \sqrt{\sum_{i = 1}^{n} {(r_{yi} - {\bar{r}}_{y})}^{2}}}$ (1)

Where,

r_xi and r_yi denote the ratings of users x and y for i^th item respectively.

${\bar{r}}_{x}$ and ${\bar{r}}_{y}$ denote the average ratings of user x and y respectively.

c) Cross Domain Computation

Cross domain computation involves with the similarity computation (as in equation-1), overall similarity computation, inter-domain correlation coefficient and similarity computation between items. The steps are described in details as follows:

(a) Overall Similarity Calculation

In CDRS distributed neighborhood approach is used to find out similar users in the target domain.

Target domain computes the overall similarity score by aggregating the set of neighborhood from the remote domain. The overall similarity computation in target domain is performed by inter-domain correlation and averaging the similarity score of the neighborhood set from the remote domain. $Sim (x, y) = \frac{\sum_{t \in T} cor (t, s) {sim}_{t} (x, y)}{\sum_{t \in T} cor (t, s)}$ (2)

Where,

Sim(x, y) denote the local similarity value between user a and b in t^th domain.

cor(t, s) denote the correlation between the target domain t and remote domains.

(b) Inter-domain correlation computation

It is used as weight in the overall similarity computation. It computes how closely the two domain, target do-main and remote domains, are related. The overall computation between domains is computed as follows: ${cor}_{rating} (t_{1}, t_{2}) = AVG {sim (i, j) : i \neq j, i \in I_{t_{1}}, j \in I_{t_{2}}}$ (3)

Where,

Sim(i, j) is the similarity between two items.

I_t is the set of items in the domain t.

The correlation computation technique that we are using in this paper is rating based correlation. The rating based correlation is correlation between ratings given to the items in the corresponding domain with the assumption that the two domains share non empty set of common users.

Similarity computation between items is computed using item-item collaborative filtering. Item-item collaborative filtering compute the similarity between the two items and it is used for the inter-domain correlation computation and it is given by: $Sim (i, j) = \frac{\sum_{u \in U} (r_{u, i} - {\bar{r}}_{i}) * (r_{u, j} - {\bar{r}}_{j})}{{\sqrt{\sum_{u \in U} (r_{u, i} - \bar{r_{i}})}}^{2} {\sqrt{\sum_{u \in U} (r_{u, j} - {\bar{r}}_{j})}}^{2}}$ (4)

Where,

${\bar{r}}_{i}$ and ${\bar{r}}_{j}$ is the average rating of the i^th and j^th item.

r_u,i is the rating for an item i by user u.

This overall computation generates the set of k nearest neighbors which helps to computes the prediction score for the items to be recommended.

d) Trust and Reputation Computation

Trust is computed using three factors: similarity, confidence and users’ reputation measures to form User Trust matrix [41]. The i^th row and j^th column entry of this matrix represents the trust of user i on user j and this value lies in between 0 and 1. Trust is computed in this work as described below which utilizes the similarity between users, confidence between users and reputation score of user:

(a) Similarity Computation: We have used Pearson correlation coefficient for the similarity computation as shown in equation (1).

(b) Confidence between users: The confidence between two users (user x and user y) is computed as follows: $conf (y | x) \frac{no of common items rated by both x and y}{no of items rated by x}$ (5)

It is clear from the above equation that as the ratings behavior of two users matches then the confidence between them is high.

(c) Reputation of users: Initially at time t = 0 the reputation score the reputation score are set to zero to compute the user trust matrix using equation (7). Once the user trust matrix is intialized, the reputation score for each user is computed to form a user reputation vector. The reputation of jth user ROU_j in the UR vector is computed as: ${ROU}_{j} = \frac{\sum_{i = 1}^{n} τ_{i, j}}{n - 1} if i \neq j, 1 \leq j \leq n$ (6)

Where,

τ_i,j denotes the trust of user i on user j.

n denotes the total number of users.

The dynamic trust of user x on user yτ_x,y (∀ x ≠ y) in UT matrix is computed by combining similarity between users x and y, confidence of user x on user y and reputation of user y as follows: $τ_{x, y} {\begin{matrix} H (x_{1}, x_{2}, x_{3}) & if x_{1} \neq 0, x_{2} \neq 0, x_{3} \neq 0 \\ H (x_{1}, x_{2}, k_{1}) & if x_{1} \neq 0, x_{2} \neq 0, x_{3} = 0 \\ H (k_{1}, x_{2}, x_{3}) & if x_{1} = 0, x_{2} \neq 0, x_{3} \neq 0 \\ k_{2} \times x_{2} & if x_{1} = 0, x_{2} \neq 0, x_{3} = 0 \\ k_{2} \times x_{3} & if x_{1} = 0, x_{2} = 0, x_{3} \neq 0 \\ 0 & if x_{1} = 0, x_{2} = 0, x_{3} = 0 \end{matrix}$ (7)

Where, $H (x_{1} x_{2} \dots x_{n}) = {(\frac{1}{n} \times \sum_{i = 1}^{n} x_{i}^{- 1})}^{- 1}$

x₁: Sim (x, y) represents similarity between users x and y.

x₂: conf (y|x) represents confidence of user x on user y

x₃: ROU_j represents reputation of user y

k₁ and k₂ are very small positive constants

Diagonal values of the user trust represent the trust value on themselves, so it is set to 1. Initially at time t = 0, τ_x,y may be negligible but as the time passes, user y will become trustworthy of user x. As the similarity between users increases, correspondingly the reputation of user y increases. The recommendations are continuously generated by the user y for user x and user x gives a positive feedback against the generated recommendations.

The data repository consists of the reputation of each item, trust between users and items rated by trustworthy users. The information is stored for the next phase computation. The information in repository is stored once the items reputation vectors, the normalized user-item rating matrix and user trust matrix are created.

e) Generating Influential User by using Trust and Reputation

The reputation score is directly proportional to the influence of user. As the reputation and trust score is generated, rank the users in decreasing reputation score. The rank of the user decides the influence of user. The assumption behind is that the high reputed user has high influence in the group. The ranking of the users changes over time as the reputation score changes depending on the three factors (similarity between users, confidence between users and reputation score).

4.2 Online phase

The online computation initiate when a query is received by UA from either the individual user or group of users. After getting the similar user the prediction score for the items is computed using following steps:

a) Selection of group of similar users – On the basis of overall similarity score neighborhood of the target user is computed. This overall similarity computation helps to get the group of similar users on the basis of similarity score. This group of similar users and the influential users is used to generate prediction for the items.

b) Prediction Computation– The prediction computation of an item for the user gives the prediction score for the item. This score decides that whether an item will be recommended to the group of user or not. The formula for prediction computation is given below: $Pred (x, i) = {\bar{r}}_{x} + \frac{\sum_{y \in U} sim (x, y) * (r_{y, i} - {\bar{r}}_{y})}{\sum_{y \in U} sim (x, y)}$ (8)

Where,

sim (x, y) denote the similarity between the user x and y.

${\bar{r}}_{x}$ and ${\bar{r}}_{y}$ denote the average rating of user x and y.

r_y,i denote the rating of user x for an item i.

c) Select top-n items for generating group recommendation– Using prediction formula top-n item is selected according to decreasing order of prediction score and the list is send to UA responding the query. This list of recommendation further saved for future recommendation.

5 Experimental evaluation

The evaluation of the proposed method includes the standard evaluation measures: MAE, RMSE, Precision, Recall and F-measure. A prototype of the system is developed using various Java technologies. JADE (Java Agent Development Environment) is used to develop the Multi-Agent system and MySql 5.0.21 provides the data repository.

5.1 Dataset

We have used the tourism dataset which includes four sub domain of available restaurants, tourist places, shopping places and places to stay of Delhi (India) is collected. The information about restaurants includes restaurants name, address, their opening and closing time, average cost per person etc. For hotels, this information includes hotel name, their location, charges etc. For shopping places and travel places, it includes their name, location, opening and closing time etc.

The detail of restaurants, hotels, places and shopping location is collected using the website Zomato, MakeMyTrip, TripAdvisor, Delhi Tourism, ShopKhojc. This information is stored in the database and further processed to get longitudes and latitudes of each entry of the above mentioned. http://www.distancesfrom.com/latitude-longitude.aspxis used to collect the longitude and latitude by using available reverse geo-coding tools. The dataset contains 8857 restaurants, 1023 hotels, 1139 places to shop and 115 places to visit for entertainment or tourist spot.

5.2 Evaluation metric

Table 1
Inter-domain similarity among domains

Restaurant Hotel Travel Places Shopping places

Restaurant 0.55858 0.54377 0.53797 0.54571

Hotel 0.54377 0.65080 0.60141 0.62695

Travel Place 0.53797 0.60141 0.86021 0.77301

Shopping Places 0.54571 0.62695 0.77301 1.01146

	Restaurant	Hotel	Travel Places	Shopping places
Restaurant	0.55858	0.54377	0.53797	0.54571
Hotel	0.54377	0.65080	0.60141	0.62695
Travel Place	0.53797	0.60141	0.86021	0.77301
Shopping Places	0.54571	0.62695	0.77301	1.01146

The performance of the proposed system is evaluated in terms of processing time and accuracy for the system. To compute the inter-domain correlation rating based approach is used and the correlation is shown in Table 1. The inter-domain correlation conveys the similarity among the domain which is further used to compute the overall similarity computation for the target user.

MAE is the mean absolute error which is deviation from the actual rating and the predicted rating are commonly used in the evaluation of group recommender system. MAE is formulated as: $MAE = \frac{\sum_{i = 1}^{n} | p_{i} - a_{i} |}{n}$ (9)

The comparison of MAE is between the traditional approach used in group recommendation and cross domain based group recommendation. RMSE is the Root Mean Square Error which is the measure of the error between two sets. Basically it compares the predicted value and actual value.

The formula for RMSE is given as: $RMSE = \sqrt{\frac{\sum_{i = 1}^{n} {(p_{i} - a_{i})}^{2}}{n}}$ (10)

Precision is the ratio of relevant and retrieved from the number of items retrieved by the system. Recall is the ratio of relevant and retrieved from the items actually relevant. Precision and recall is formulated as: $Precision = \frac{| {RelevantItem} \cap {RetrievedItem} |}{| {RetrievedItem} |}$ (11) $Recall = \frac{| {RelevantItem} \cap {RetrievedItem} |}{| {RelevantItem} |}$ (12)

Precision and recall are conflicting in nature. Both precision and recall are the important factors that evaluate the system performance to generate the Top-n recommendations. So both precision and recall are combined to get the metric F-measure. The formula of F-measure is: $F - measure = \frac{2 * Precision * Recall}{Precision + Recall}$ (13)

The comparison between Precision, Recall, F-measure and MAE is shown below in tabular format in Table 2(a) as well as in Fig. 3.

Table 2

Comparison of Precision, Recall, F-measure and MAE between Traditional Group Recommender system and Cross Domain based Group Recommender System (a) In Restaurant Domain

	Precision	Recall	F-Measure	MAE
GRS	0.534	0.039	0.0727	0.41035
CDGRS	0.59	0.045	0.0836	0.3087
(b) In Hotel Domain
GRS	0.588	0.0492	0.05357	0.266
CDGRS	0.515	0.113	0.1914	0.235
(c) In Travel Places
GRS	0.545	0.5084	0.526	0.2675
CDGRS	0.75	0.372	0.4886	0.0558
(d) Shopping Places
GRS	0.54	0.3539	0.4247	0.2511
CDGRS	0.772	0.539	0.688	0.2198

Fig. 2

Comparison of precision, recall, F-measure and MAE between traditional group recommender system and cross domain based group recommender system in restaurant domain.

Fig. 3

Comparison of precision, recall, F-measure and MAE between traditional group recommender system and cross domain based group recommender system in hotel, travel & shopping domain.

The result shows that the proposed approach CDGRS have better accuracy in terms of MAE than GRS approach. Also precision, recall and F-measure is compared for GRS and CDGRS where it was found that the proposed approach outperforms in restaurant domain. The experiment is continued for the other domains (as shown in Table 2(b), (c) and (d) and in Fig. 4) to observe the variation in precision, recall and F-measure.

Fig. 4

The comparison of MAE and RMSE when the number of users varies in the group.

Again the system observes the better accuracy in terms of Precision, Recall and F-measure for the hotel domain as well as travel places and shopping places. The effect of increase in number of users in the group is considered for MAE and RMSE are computed to compare both GRS and CDGRS. To check the effect of varying number of users in the group, MAE and RMSE is computed for the same. The result (Fig. 4) shows that even the number of users in the group varies for 3, 11, 16 and 24, both MAE and RMSE does not get affect much.

The Precision, Recall, F-measure along with the comparison of MAE and RMSE on the varying number of group of users suggest that the proposed approach outperforms as compared to the traditional approach of group recommendation. To evaluate the social influence of user we compare the rating list given by different methods to examine their correlation. To quantify the metric Spearman’s correlation coefficient [42] is widely used and is formulated as: $ρ = 1 - \frac{6 \sum {(x_{i} - y_{i})}^{2}}{n (n^{2} - 1)}$ (14)

Where,

x_i and y_i are the rankings of the users i in the lists x and y respectively.

n is the number of users.

We have compared the change of social influence variation (Fig. 5) as the number of top-N user increases. The top-k most influential users are used to access the social influence. We have compared the ranking provided by the traditional group recommender approach and the proposed approach. The top-N varies with the number 5, 11, 16, and 24 users and as shown in Figure it does not get much affected as Top-N increases.

Fig. 5

Change of social influence over top-N users.

6 Conclusion

A novel approach for group recommendation using cross domain data is proposed and presented in this paper. The approach tries to handle the sparsity and cold start problems in target domain by utilizing data from the source domain. To boost the grouping of users based on the similarity between them, cross domain approach is used. Once the target domain does not have enough rating to perform the grouping of users, it requests the source domain to begin the similarity computation for the target user. The remote domain respond back by sending the similar user’s along with their similarity score for the target user. As soon as the target domain gets the response from the remote domains; it computes overall similarity score for the target user. This overall similarity computation helps to get the neighborhood for the target even when the target domain does not have enough ratings to compute similarity score. The trust and reputation score is computed using similarity, confidence and reputation score. This leads to generate the influential user in the group that helps to generate the recommendations for the group having better acceptance towards presented list. The presented approach is a multi-agent based approach where multiple agents work in a cooperative way to provide the solution to an assigned problem. A prototype of the system is developed using Java and JADE technologies. For the experimental evaluation tourism domain is used which is divided into four sub-domains i.e. restaurant, travel places, hotel and shopping places. The experimental results show that the proposed system outperforms as compared to the traditional approach of group recommender.

References

Ricci

, Rokach

, Shapira

and Kantor

, Recommender Systems Handbook. New York, USA: Springer, 2011.

Das

Abhinandan S.

, Datar

Mayur

, Ashutosh

Garg

and Rajaram

Shyam

, Google news personalization: scalable online collaborative filtering,” in Proceedings of the 16th international conference on World Wide Web, Banff, Alberta, Canada, 2007, pp. 271–280.

Bedi

, Agarwal

S.K.

, Jindal

and Richa , MARST: Multi-Agent Recommender System for e-Tourism Using Reputation Based Collaborative Filtering,” in International Workshop on Databases in Networked Information Systems, Aizu-Wakamatsu City, Japan, 2014, 189–201.

Pazzani

Michael J.

and Billsus

Daniel

, Content-based recommendation systems,” in The adaptive web, 2007, pp. 325–341.

Burke

, Hybrid recommender system: survey and experiments,”User Modeling and User-Adapted Interaction Journal12(4) (2002), 331–370.

Badrul

Sarwar

, Karypis

George

, Konstan

Joseph

and Riedl

John

, Item-Based Collaborative Filtering Recommendation Algorithms,” in Proceedings of the 10th international conference on World Wide Web, Hong Kong, 2001, pp. 285–295.

Deshpande

Mukund

and Karypis

George

, Item-based top-n recommendation algorithms, ACM Transactions on Information Systems (TOIS)22(1) (2004), 143–177.

Cremonesi

Paolo

, Tripodi

Antonio

and Turrin

Roberto

, Cross-domain recommender systems,” in Data Mining Workshops (ICDMW), Vancouver, Canada, 2011, pp. 496–503.

Murad Khan

Muhammad

and Ibrahim

Roliana

, Cross domain recommender systems: A systematic literature review, ACM Computing Surveys (CSUR)50(3) (2017), 1–34.

10.

Berkovsky

Shlomo

, Kuflik

Tsvi

and Ricci

Francesco

, Mediation of user models for enhanced personalization in recommender systems,”User Modeling and User-Adapted Interaction18(3) (2008), 245–286.

11.

Richa and Bedi

Punam

, Parallel proactive cross domain context aware recommender system, Journal of Intelligent & Fuzzy Systems34(3) (2018), 1521–1533.

12.

Alina

Ingrid Christensen

and Schiaffino

Silvia

, Social influence in group recommender systems,”Online Information Review38(4) (2014), 524–542.

13.

Cha

Meeyoung

, Haddadi

Hamed

, Benevenuto

Fabricio

and Gummadi

Krishna P.

, Measuring user influence in twitter: The million follower fallacy,” in fourth international AAAI conference on weblogs and social media, 2010.

14.

O’Donovan

John

and Smytha

Barry

, Trust in recommender systems,” in 10th international conference on Intelligent user interfaces, 2005, pp. 167–174.

15.

Yan

Zheng

, Zhang

Peng

and Deng

Robert H.

, TruBeRepec: a trust-behavior-based reputation and recommender system for mobile applications,”Personal and Ubiquitous Computing16(5) (2012), 485–506.

16.

Wang

Wei

, Zhang

Guangquan

and Lu

Jie

, Member contribution-based group recommender system, Decision Support Systems87(1) (2016), 80–93.

17.

Masthoff

Judith

, Group recommender systems: aggregation, satisfaction and group attributes, in Recommender Systems Handbook.: SpringerBoston MA, 2015, pp. 743–776.

18.

Dixit

Veer Sain

, Mehta

Harita

and Bedi

Punam

, A proposed framework for group-based multi-criteria recommendations, Applied Artificial Intelligence28(10) (2014), 917–956.

19.

Christensen

Ingrid

, Schiaffino

Silvia

and Armentano

Marcelo

, Social group recommendation in the tourism domain,”Journal of Intelligent Information Systems47(2) (2016), 209–231.

20.

De Pessemier

Toon

, Dhondt

Jeroen

, Vanhecke

Kris

and Martens

Luc

, TravelWithFriends: a hybrid group recommender system for travel destinations,” in In Workshop on tourism recommender systems (touRS15) in conjunction with the 9th ACM conference on recommender systems (recsys 2015), 2015, pp. 51–60.

21.

Garcia

Inma

, Sebastia

Laura

, Onaindia

Eva

and Guzman

Cesar

, A group recommender system for tourist activities,” in International conference on electronic commerce and web technologies, 2009, pp. 26–37.

22.

McCarthy

Kevin

, et al.Group recommender systems: A critiquing based approach,” in 11th international conference on Intelligent user interfaces, 2006, pp. 267–269.

23.

Pujahari

Abinash

and Padmanabhan

Vineet

, Group Recommender Systems: Combining user-user and item-item Collaborative filtering techniques,” in International Conference on Information Technology (ICIT), 2015, pp. 148–152.

24.

Quijano-Sanchez

Lara

, Recio-Garcia

Juan A.

, Diaz-Agudo

Belen

and Jimenez-Diaz

Guillermo

, Social factors in group recommender systems, ACM Transactions on Intelligent Systems and Technology (TIST)4(1) (2013), 1–8.

25.

Quijano-Sanchez

Lara

, Recio-Garcia

Juan A.

, Diaz-Agudo

Belen

and Jimenez-Diaz

Guillermo

, Social factors in group recommender systems,”ACM Transactions on Intelligent Systems and Technology (TIST)4(1) (2013), 1–8.

26.

Cantador

Iván

and Castells

Pablo

, Group recommender systems: new perspectives in the social web,” in Recommender systems for the social web, 2012, pp. 139–157.

27.

Riedl

John

, O’connor

Mark

, Cosley

Dan

and Konstan

Joseph A.

, PolyLens: a recommender systemfor groups of users,” in ECSCW, 2001, pp. 199–218.

28.

Jameson

Anthony

, More than the sum of its members: challenges for group recommender systems, in working conference on Advanced visual interfaces, 2004, pp. 48–54.

29.

Amer-Yahia

Sihem

, Basu Roy

Senjuti

, Chawla

Ashish

, Das

Gautam

and Yu

Cong

, Group recommendation: Semantics and efficiency,”VLDB Endowment2(1) (2009), 754–765.

30.

Boratto

Ludovico

, Carta

Salvatore

and Sa

Michele

, Groups Identification and Individual Recommendations in Group Recommendation Algorithms,” in PRSAT@ recsys, 2010, pp. 27–34.

31.

De Pessemier

Toon

, Dooms

Simon

and Martens

Luc

, Comparison of group recommendation algorithms,”Multimedia Tools and Applications72(3) (2014), 2497–2541.

32.

Oshodin

Eseosa

and Chiclana

Francisco

, Effect of influential users on recommendation, in SAI Intelligent Systems Conference (IntelliSys), 2015, pp. 731–732.

33.

Yerasani

Sinjana

, Appam

Deepthi

, Sarma

Monalisa

and Tiwari

Manoj Kumar

, Estimation and maximization of user influence in social networks, International Journal of Information Management47 (2019), 44–51.

34.

Riquelme

Fabián

and González-Cantergiani

Pablo

, Measuring user influence on twitter: A survey,”Information Processing & Management52(5) (2016), 949–975.

35.

Tavakolifard

Mozhgan

and Almeroth

Kevin C.

, Social computing: an intersection of recommender systems, trust/reputation systems, and social networks,”IEEE Network26(4) (2012), 53–58.

36.

Victor

Patricia

, De Cock

Martine

and Corne

Chris

, Trust and recommendations,” in Recommender systems handbook.: Springer, Boston, MA, 2011, pp. 645–675.

37.

Massa

Paolo

and Avesani

Paolo

, Trust-aware collaborative filtering for recommender systems, in OTM Confederated International Conferences On the Move to Meaningful Internet Systems, 2004, pp. 492–508.

38.

Andersen

Reid

, et al., Trust-based recommendation systems: an axiomatic approach,” in 17th international conference on World Wide Web, 2008, pp. 199–208.

39.

Bharadwaj

Kamal K.

and Al-Shamri

Mohammad Yahya H.

, Fuzzy computational models for trust and reputation systems, Electronic Commerce Research and Applications8(1) (2009), 37–47.

40.

Benesty

, Huan

and Chen

, Pearson Correlation Coefficient,” in Springer Berlin Heidelberg: Springer Topics in Signal Processing, 2009, pp. 1–4.

41.

Bedi

Punam

, Kumar Agarwal

Sumit

and Richa , Trust and reputation-based multi-agent recommender system, International Journal of Computational Science and Engineering16(4) (2018), 350–362.

42.

Shaozhi

and Felix Wu

, Measuring message propagation and social influence on Twitter. com, in International conference on social informatics, 2010, pp. 216–231.

Combining trust and reputation as user influence in cross domain group recommender system (CDGRS)

Abstract

Keywords

1 Introduction

2 Literature survey

3 Proposed trust and reputation based cross domain group recommender system (CDGRS)

3.1 User profiling of CDGRS

3.2 Architecture of CDGRS

4.1 Offline phase

5.1 Dataset

5.2 Evaluation metric

Table 1 Inter-domain similarity among domains Restaurant Hotel Travel Places Shopping places Restaurant 0.55858 0.54377 0.53797 0.54571 Hotel 0.54377 0.65080 0.60141 0.62695 Travel Place 0.53797 0.60141 0.86021 0.77301 Shopping Places 0.54571 0.62695 0.77301 1.01146

References

Table 1
Inter-domain similarity among domains

Restaurant Hotel Travel Places Shopping places

Restaurant 0.55858 0.54377 0.53797 0.54571

Hotel 0.54377 0.65080 0.60141 0.62695

Travel Place 0.53797 0.60141 0.86021 0.77301

Shopping Places 0.54571 0.62695 0.77301 1.01146