Solving cold start problem in tag-based recommender systems using discrete imperialist competitive algorithm

Recommender systems detect users´ favorites based on their past behavior and provide them with proper suggestions; however, these systems would encounter problems while dealing with users with low or empty usage data. This issue leads to the most prominent challenge of such systems called cold start. In thispaper, we proposea system based on which a modified discrete imperialist competitive algorithm where tags are clustered using K-medoids algorithm. When a new user logs in and enters his/her tags then the system will suggest just a few sources with the largest weight. Experimental results demonstrate improvement of evaluation criteria for recommender system in comparison with other methods.