Improving the Recommendation Accuracy for Cold Start Users in Trust-Based Recommender Systems

Recommender systems have become extremely popular in recent years due to their ability to predict a user's preference or rating of a certain item by analyzing similar users in the network. Trust-based recommender systems generate these predictions by using an explicitly issued trust between the users. In this paper we propose a recommendation algorithm called Averaged Localized Trust-Based Ant Recommender (ALT-BAR) that follows the methodology applied by Ant Colony Optimization algorithms to increase the accuracy of predictions in recommender systems, especially for cold start users. Cold start users are considered challenging to deal with in any recommender system because of the few ratings they have in their profiles. ALT-BAR reinforces the significance of trust between users, to overcome the lack of ratings, by modifying the way the initial pheromone levels of edges are calculated to reflect each edge's associated trust level. An appropriate initialization of pheromone in ant algorithms in general can guarantee a proper convergence of the system to the optimal solution. ALT-BAR's approach allows the ants to expand their search scope in the solution space to find ratings for cold start users while exploiting discovered good solutions for the sake of heavy raters. When compared to other algorithms in the literature, ALT-BAR proved to be extremely successful in enhancing the prediction accuracy and coverage for cold start users while still maintaining fairly good results for heavy raters.