Recommendation system based on items selection records

Abstract

This paper proposes a system which recommends items based on sales records. The system is designed under the premise that (1) user preference is more apparent in the recent item than the previous ones, and (2) it is useful for user to recommend items derived from the preference structure of many people. Then, the proposed system establishes the preference structures as a directed graph. Constructing a prototype system based on real sales records, the effectiveness and limitation are discussed.