Trust-based Ant Recommender (T-BAR)

Recommender Systems suggest to users items that may be of interest to them. Collaborative filtering recommender systems suggest the items based on the item ratings provided by similar users in the network. Trust-based recommender systems utilize an explicitly issued trust between users to increase the accuracy of the recommendations. In this paper, we propose a bio-inspired algorithm, called Trust-based Ant Recommender (T-BAR), to further increase the accuracy and the coverage of the recommendations in trust-based networks. T-BAR uses the Ant Colony System computational model to imitate the behavior of ants during their search for a good food source. T-BAR´s advantage over other known algorithms is that it considers all the target item ratings along the paths rather than just using the ratings found at the end of each path. The Epinions.com dataset was used for the empirical evaluation of Trust-based Ant Recommender and proved its success by drastically improving the coverage of the recommendations while maintaining a reasonable level of accuracy of the results. T-BAR outperforms the basic CF algorithm that uses the Pearson Similarity and Massa´s MoleTrust (MT) by achieving a balanced trade-off between accuracy and coverage.