User recommendation algorithm in social tagging system based on user trust method

Collaborative Tagging Systems such as Flickr, del.icio.us, and BibSonomy are examples of Web 2.0 applications that have recently gained widespread popularity, where users label digital resources by means of personalized tags. The simplistic and user-centered design of those systems have encouraged many Web users to annotate their data using tags to provide easy search and retrieval of non-textual Web sources such as photos or videos, hence resulting in huge amount of data and metadata becoming available over the Web. This causes the task of searching to be out of reach especially among the common Internet users. This is where recommendation systems or tools come in handy. A lot of methods can be used for the purpose of recommendation. Collaborative filtering is the most popular technique among recommendation system that makes use only past Muser activities such as transaction history or user satisfaction expressed in ratings. Collaborative filtering has been a substantial success; however they do not rely on the actual content of the items. To improve recommendation quality, metadata such as content information in items and tags have been typically used as additional knowledge. Nonetheless, this type of recommendation is not entirely reliable since the knowledge are sourced from people whom we do not know or trust. The accuracy of recommendation system will generally be improved through incorporation of user trust information into the systems due to the fact that acquaintances might share professional interest while social friends might share hobbies. Unfortunately, the level of existing recommendation accuracy to date is still at unsatisfactory level among the users. In effort to improve recommendation in terms of accuracy and coverage, we propose a hybrid method for user recommendation approach based on User Trust method to allow users to easily find other users with similar interest in social tagging system. This method is a combination of developing trust network based on user interest similarity and trust network from social network analysis. The user interest similarity is derived from personalized user tagging information. The User Trust method is able to find the similar users and selected them as neighbors to make automated recommendations. The proposed method is tested using the Del.icio.us dataset. The experiment results showed that the proposed User Trust method outperforms the user-based collaborative filtering in making recommendations with the Pearson Correlation Coefficient (PCC) (Resnick et al., 1994) around 49%, Tidal Trust (TT) (Golbeck, 2006) around 32%, UserRec (Zhou et al., 2010) around 39%, tag-based Similarity Trust (ST) (Bhuiyan et al., 2010) around 45%, as well as incorporation of social network information in collaborative filtering (PCC-SN) (Liu and Lee, 2010) around 29%.