The traditional method of recommender systems suffers from the Sparsity problem whereby incomplete dataset results in poor recommendations. Another issue is the drifting preference, i.e. the change of the user's preference with time. In this paper, we propose an algorithm that takes minimal inputs to do away with the Sparsity problem and takes the drift into consideration giving more priority to latest data. The streams of elements are decomposed into the corresponding attributes and are classified in a preferential list with tags as "Sporadic", "New", "Regular", "Old" and "Past" - each category signifying a changing preference over the previous respectively. A repeated occurrence of attribute set of interest implies the user's preference for such attribute(s). The proposed algorithm is based on drifting preference and has been tested with the Yahoo Webscope R4 dataset. Results have shown that our algorithm have shown significant improvements over the comparable "Sliding Window" algorithm.
Mendeley saves you time finding and organizing research
Choose a citation style from the tabs below