Outline • Recommendation problem • Content-based approach • Collaborative approach • Profile association rules discovery • Rule discovery • Evaluation
Recommendation problem • The effectiveness of targeting a small, selected portion of customers for advertising has long been recognized by businesses. • Avoid information overload • Reduce cost • Ex. Amazon, UDN, Citeseer, etc.
Recommendation problem • Content-based approach • Recommendable products are characterized by a set of content features. • Customer’s interest is represented by the same feature set. • Target customers whose interests have a high degree of similarity with the product are selected.
Recommendation problem • Collaborative approach • Looks for relevance among users by observing the ratings • The nearest-neighbor users are those that exhibit the strongest relevance to the target user. • Products that appear in the profiles of partners are recommended to the target user.
Recommendation problem • Disadvantages of content-based approach • Inability to advertise products to a new user. • Inability to provide novel advertisements. • Disadvantages of collaborative approach • Inability to advertise products to a new user. • Inability to advertise new products. • Poor prediction performance when data are sparse.
Profile association rule discovery • The new book recommendation problem in NSYSU library: • NSYSU has approximately 7000 enrolled students • Its library approximately buys 1000 new books per month • One of the library’s e-services is promoting new books to NSYSU’s students (via email).
Profile association rule discovery • The new-book feature makes the collaborative approach inapplicable. • The new-user feature makes the content-based approach inapplicable. • The library uses an online public access catalog (OPAC) system. Huge amount of data has been collected in their database.
Profile association rule discovery • Patron hierarchy • Book hierarchy • ISBN standard
Profile association rule discovery • Transaction database (OPAC log) • Translated transactions
Profile association rule discovery • The association rule algorithm is applied on the transaction data. • Rules whose right hand side are not book types are removed. • For a new book, some rules are triggered. Target users are accordingly selected to promote the new book.
Profile association rule discovery • Evaluation