Dt-bar: a Dynamic Ant Recommender to Balance the Overall Prediction Accuracy for All Users

Ant colony algorithms have become recently popular in solving many optimization problems because of their collaborative decentralized behavior that mimics the behavior of real ants when foraging for food. Recommender systems present an optimization problem in which they aim to accurately predict a user’s rating for an unseen item by trying to find similar users in the network. Trust-based recommender systems make use of trust between users. T-BAR was the first successful application of an ant colony algorithm to trust-based recommender systems but it lacked the ability to deal with cold start users. In this paper we propose a dynamic trust-based ant recommender (DT-BAR) that solves the problem of cold start users by locally initializing the pheromone level on edges using the dynamically changing information within each neighborhood as ants pass by. DT-BAR increases the communication between ants and emphasizes the importance of trust in the pheromone initialization process.