Merging trust in collaborative filtering to alleviate data sparsity and cold start Guibing Guo ⇑ , Jie Zhang, Daniel Thalmann School of Computer Engineering, Nanyang Technological University, Singapore article info Article history: Received 29 May 2013 Received in revised form 17 November 2013 Accepted 7 December 2013 Available online 12 December 2013 Keywords: Recommender systems Collaborative filtering Cold start Data sparsity Trusted neighbors abstract Providing high quality recommendations is important for e-commerce systems to assist users in making effective selection decisions from a plethora of choices. Collaborative filtering is a widely accepted tech- nique to generate recommendations based on the ratings of like-minded users. However, it suffers from several inherent issues such as data sparsity and cold start. To address these problems, we propose a novel method called ‘‘Merge’’ to incorporate social trust information (i.e., trusted neighbors explicitly specified by users) in providing recommendations. Specifically, ratings of a user’s trusted neighbors are merged to complement and represent the preferences of the user and to find other users with similar preferences (i.e., similar users). In addition, the quality of merged ratings is measured by the confidence considering the number of ratings and the ratio of conflicts between positive and negative opinions. Further, the rat- ing confidence is incorporated into the computation of user similarity. The prediction for a given item is generated by aggregating the ratings of similar users. Experimental results based on three real-world data sets demonstrate that our method outperforms other counterparts both in terms of accuracy and coverage. Ó 2013 Elsevier B.V. All rights reserved. 1. Introduction The emergence of Web 2.0 applications has greatly changed users’ styles of online activities from searching and browsing to interacting and sharing [6,40]. The available choices grow up exponentially, and make it challenge for users to find useful information which is well-known as the information overload problem. Recommender systems are designed and heavily used in modern e-commerce applications to cope with this problem, i.e., to provide users with high quality, personalized recommenda- tions, and to help them find items (e.g., books, movies, news, music, etc.) of interest from a plethora of available choices. Collaborative filtering (CF) is one of the most well-known and commonly used techniques to generate recommendations [1,17]. The heuristic is that the items appreciated by those who have sim- ilar taste will also be in favor of by the active users (who desire rec- ommendations). However, CF suffers from several inherent issues such as data sparsity and cold start. The former issue refers to the difficulty in finding sufficient and reliable similar users due to the fact that users in general only rate a small portion of items, while the latter refers to the dilemma that accurate recommenda- tions are expected for the cold users who rate only a few items and thus whose preferences are hard to be inferred. To resolve these issues and model user preferences more accu- rately, additional information from other sources is studied and incorporated into CF including friendship [19], membership [38,12] and social trust [41,2], where trust is believed less ambig- uously and more reliable than friendship and membership. In this paper, trust is defined as one’s belief toward others in providing accurate ratings relative to the preferences of the active user. Both implicit trust (e.g., [26,28]) and explicit trust (e.g., [4,8,25,27]) have been investigated in the literature. The former trust is inferred from user behaviors such as ratings whereas the latter is directly specified by users. By definition, the explicit trust tends to be more accurate and reliable than the implicit one. We focus on the expli- cit trust in this paper. Although many trust-based approaches have been proposed and the improvements to some extent have been achieved, there is still much room left for a better trust-based ap- proach as stressed by [32]. In this paper, we propose a novel trust-based approach called ‘‘Merge’’ by incorporating the trusted neighbors explicitly specified by the active users in the systems, aiming to improve the overall performance of recommendations and to ameliorate the data spar- sity and cold-start problems of CF. Specifically, we merge the rat- ings of trusted neighbors of an active user by averaging the ratings on the commonly rated items according to the extent to which the trusted neighbors are similar to the active user. The 0950-7051/$ - see front matter Ó 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.knosys.2013.12.007 ⇑ Corresponding author. Tel.: +65 84005322. E-mail addresses: gguo1@ntu.edu.sg (G. Guo), zhangj@ntu.edu.sg (J. Zhang), danielthalmann@ntu.edu.sg (D. Thalmann). Knowledge-Based Systems 57 (2014) 57–68 Contents lists available at ScienceDirect Knowledge-Based Systems journal homepage: www.elsevier.com/locate/knosys