Refinement of Group Recommendations Using User Preferences and Item Attributes

Abstract

Providing  Group  Recommendations is  an  open  research  area. In the  proposed scheme, the  combined  entropy  based similarities using positive and negative preference ratings  among  training  users are used to extract Similar Taste Users  (STUs). Such STUs build a group for the target user from which  group recommendations are generated. Using information gain, it further computes top  N individual recommendations  from these  group  recommendations based  on opposite user  preferences. In this paper, a method is proposed to overcome the sparsity  among  preferences  of group members. Genre based similarity  (based on implicit multi criteria information) among target user  and each group member generates genre based profile of target user which in turn increases the density of preferences among  group members. Movie Lens dataset is used for experiments. It shows significant improvements in overcoming sparsity problem in group recommender systems and performance measures used  shows improvement in recommendation quality