Solving cold start problem in tag-based recommender systems using discrete Imperialist Competitive Algorithm

Recommender systems detect users' favorites basedon their past behavior and provide them with propersuggestions; however, these systems would encounter problemswhile dealing with users with low or empty usage data. Thisissue leads to the most prominent challenge of such systemscalled cold start. In this paper, we propose a system based onwhich a modified discrete imperialist competitive algorithmwhere tags are clustered using K-medoids algorithm. When anew user logs in and enters his/her tags then the system willsuggest just a few sources with the largest weight. Experimentalresults demonstrate improvement of evaluation criteria forrecommender system in comparison with other methods.