Entropy for item inclination in sub-community based recommender system

To overcome the new user cold-start problems in collaborative filtering, an innovative framework has been propsed that used entropy for item inclination in sub-community-based recommender system (EISR). It administered demographic filtering on user and item attributes for finding similar users and applied collaborative filtering on rating preferences. The proposed framework leveraged the advantages of traditional group aggregation strategies for delivering good quality recommendations using item preferences of the members of a refined group detected using two-tier approach. At Tier-I, user communities were detected using demographic attributes, which were decomposed into discernible sub-communities by exploiting the item preferences of users. A novel entropy-based hybrid group aggregation method called pragmatic propensity was used to combine the item preferences of members of these sub-communities. Also, experiments conducted using the MovieLens and Book-crossing datasets revealed the better quality of recommendations and the comparison with other algorithms confirmed the effectiveness of the proposed framework.