Collaborative Item Recommendations Based on Friendship Strength in Social Network
Normaslina Jamil, Shahrul Azman Mohd Noah, Masnizah Mohd · International Journal of Machine Learning and Computing · 2020
Recommender Systems (RSs) are among the solutions in addressing the information overload problems.One of the RS main problem is cold start users where RSs do not have enough information to identify the user's preferences and thus unable to recommend relevant items.One of the approaches to overcome such problem is through social relationship which can be extracted from social networks.Existing studies used friendship relation to find nearest neighbors combined with implicit data such as tweet content, posts, like or tags to identify user's preference.This study proposed friendship strength through user's interaction and item rating values to identify users' preferences.User-based collaborative filtering methods are used in developing the system.Two datasets are synthetically obtained from MovieLens database for user's rating information.Meanwhile Twitter users' interactions data, was retrieved from Higgs Boson topic available from the Standford University.There are four phases of development namely user profile phase, friendship strength phase, user's preference phase and items recommendation phase.The findings of the study show that there is an improvement in the proposed method compared to the baseline based on the recall (R), precision (P) and F score measures.