以模糊分群技術為基礎建構向量式輪廓混合型推薦系統

A Hybrid Recommendation System with Vector Profile Based on Fuzzy Clustering

陳羿逞(Chuck Chen)

指導教授 ： 賴國華

元智大學/資訊學院/資訊工程學系/碩士(2002年)

摘要

蘊藏於網路中的資料隨著Internet的發展以驚人的速度成長，如何從中找到需要的資料相對越顯困難。近幾年，資訊過濾器(Information filtering)技術普遍被使用來解決此問題，除了被動式強化搜尋引擎能力外，還可以進一步使網站主動推薦個人化的資訊。相關應用技術大致分為內容導向式過濾(content-based filtering)、協調式過濾(collaborative filtering)以及混合式過濾三種類型。
本研究企圖在e-Learning環境下以輪廓整合的方式建構混合式推薦系統，兩種過濾方式併行，避免單一方式所衍生的問題，系統主要知識來自網頁內容輪廓(content profile)以及使用者行為輪廓(user profile)，前者主要是針對網站內教材特徵萃取而成的向量；而後者的設計是依據前者所延伸，除包含分析網站日誌所得的瀏覽行為特徵外，並納入使用者其他資訊。針對使用者與網頁分群，本論文提出新的模糊分群技術，不僅加強雜訊的偵測力更可接受不理想的初始值。系統藉由分類後形成的群集，進行適當的推薦，其中文章的選擇因輪廓整合而加速，此外為因應使用者興趣的快速變化，一適當的回饋機制也包含於此論文中。為評估系統的可行性與準確性，我們實作了一教學網站On-Line Learner (OLLer)，並且設計一些相關實驗以作效能驗證。

關鍵字

遠距學習 ； 資訊過濾 ； 模糊分群 ； 推薦系統

並列摘要

Rapid development of the Internet makes users differentiating what information is really needed more difficult. To cope with the problems of information overloading and low signal to noise ratio, a number of solutions based on information filtering, mainly content-based filtering, collaborative filtering and hybrid filtering, have been applied to develop an adaptive web site.
In this thesis, we present a hybrid filtering mechanism into the development of a recommendation system for e-Learning. The web pages and users’ logs are represented into vector profiles first. The resulting profiles, then, are classified into peer-groups based on a novel fuzzy clustering algorithm free from the effects of noise and poor initial values immunity. The similarity among profiles is characterized as a membership function allowing system to provide recommendation flexibly. Case-based reasoning is also incorporated into the system to support the controlling of browsing order in e-Learning. Besides, in order to catch the rapid changes of users’ interests, a corresponding feedback mechanism is also included. Finally, we have also implemented a web site, On-Line Learner (OLLer), and set up several experiments to evaluate the effectiveness of our system.

並列關鍵字

e-Learning ； Information Filtering ； Fuzzy Clustering ； Recommendation System

參考文獻

[1] Barni, M., Cappellini, V. and Mecocci, A., 1996. “Comments on A possibilistic approach to clustering,” IEEE Transactions on Fuzzy Systems, vol 4, issue 3, pp. 393-396.

[3] Beni, G. and Xiaomin Liu, 1994. “A least biased fuzzy clustering method,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol 16, issue 9, pp. 954-960.

[5] Borchers, A., et al., 1998, “Ganging up on information overload,” Computer ,

[6] Bradly, Keith, Rachael Rafter and Barry Smyth, 2000, “Case-Based User Profiling for Content Personalization,” AH, pp. 62-72.

[7] Chen, Ming-Syan, Jong Soo Park and Yu, P.S., 1998, “Efficient data mining for path traversal patterns,” in IEEE Trans. on Knowledge and Data Engineering,