Extending Content-Based Recommendation by Order-Matching and Cross-Matching Methods

We propose TwinFinder: a recommender system for an on-line bookstore. TwinFinder provides two recommendation methods, the Order-Matching Method (OMM) and the Cross-Matching Method (CMM). TwinFinder profiles a customer's interest based on his/her purchase history. Thus, it generates a vector of keywords from titles, authors, synopses, and categories of books purchased. OMM keeps this vector to each category the books belong to. Thus, OMM avoids recommending books that share only one or two keywords but belong to the categories in which the customer has no interest. When a customer has purchased several books that range over two or more categories, TwinFinder generates recommendations based on CMM. CMM looks for books in a category based on the keywords generated from the purchased books in other categories. Thus, TwinFinder can generate rather useful and surprising recommendations by OMM and CMM. We have implemented and validated TwinFinder in the e-business system of a bookstore in Japan.