合作式過濾推薦之實作與比較

Abstract

市場動向的變化相當激烈，顧客的需求也變得多樣化，現在的企業必須配合顧客的需求狀況來設定經營策略，在這樣的需求下，推薦系統應運而生。推薦系統的目的是在特定領域上，幫助使用者從龐大的資料量中，過濾出可能會喜愛的項目並推薦給使用者，以縮短資料搜尋的時間，然而在許多情形下推薦結果的精確度很低，顯示無法對使用者做出適切的推薦。 本研究實作合作式過濾推薦系統，嘗試依據使用者的興趣、喜好，找出偏好相近的夥伴，為使用者提供產品項目推薦，希望藉著有效的推薦服務提高預測精確度，達成個人化推薦之目的。本研究也實作整合內容為基礎方法與結合類別權重調整方法之合作式過濾推薦系統，並進行實驗評估，以驗證與比較各種過濾推薦方法之推薦精確度。

The marketplace is dramatically changed nowadays so as customers’ various demands. Today, organizations must design business strategies in accordance with customers’ needs. Due to this circumstance, recommender system has formed and deployed. The feature and purpose of the recommender system is to filter probable preferred items among piles of information in a specific field and in turn to recommend to users. In that sense users are able to reduce time for searching information. However, in certain conditions the preciseness of results from the system is quite low that reveals the system’s incapability of making appropriate recommendations to users. This research implements a collaborative filtering recommender system, and attempts to provide recommendations of filtered product items according to users’ interest and preference. With effective recommendation services, the recommender system enables to improve its preciseness of the estimation and hence reaches the goal of personalized recommendation. Moreover, this research integrates content-based approach and category weighting adjustment approach with collaborative filtering recommender system. Experiment evaluations are conducted to compare various recommendation methods.