Directional user similarity model for personalized recommendation in online social networks

With the huge amount of information available in online social networks and the increasing spread of user generated data in different forms, personal recommendation systems represent ideal solutions for optimising users' efforts and advocating the right information that suits the user's context. This research proposes a user-based collaborative filtering recommendation technique for social networks that exploits the context and characteristics of social networks in order to recommend personalised content to users. The originality of the approach is to streamline contextual features and social network data to reduce data dimensionality and use learning techniques to uncover contextual information that is not normally included as a typical interest to rank tweets for users. The proposed approach uses a hybrid directional relation user similarity model (DRUSM) that exploits subtle users' actions and directional interactions' diversity to match users' contexts. DRUSM includes two similarity algorithms - author directional similarity (AUDS) and action directional similarity (ACDS) - to account for the abovementioned actions and interactions. It employs a learning algorithm to rank tweets according to a user's interests. DRUSM was tested and compared with AUDS, ACDS and other algorithms, such as the new heuristic similarity model, content relevance and publisher authority. Experiments show that DRUSM outperformed all the other algorithms. It achieves higher accuracy in recommending tweets due to its suitability to appropriately handle fine details of the user's context.

Automated summary

This research proposes a user-based collaborative filtering recommendation technique for social networks that utilizes the context and characteristics of social networks to recommend personalized content to users. The approach streamlines contextual features and social network data to reduce data dimensionality and uses learning techniques to uncover contextual information for ranking tweets. The proposed approach outperformed other algorithms in recommending tweets, achieving higher accuracy by appropriately handling the fine details of the user's context.