Collaborative Preference Elicitation based on Dynamic Peer Recommendations

Abstract

Recommender Systems, in order to recommend correctly, demand huge information related to the past transactions and behavior of the user. In the events, where the data is inconsistent or sparse, the systems show a decline in its predictions or recommendations. Here we propose a new preference elicitation system that is based on preference from closed user group. The implicit behavior of the user is tracked when the user picks up an item. The explicit behavior is tracked by the user-ratings for the given item. The userpreference is computed on a memory-based model taking in account the implicit behavior. The peers are identified based on user-similarity on the explicit-preference indicator. The peer preferences are used on the test-dataset to find the percentage of preference that could be matched. The algorithm has been tested on MovieLens dataset and has given competitive results over the comparable techniques like sliding window method or collaborative filtering methods in isolation.