以移動物件之密度分群實現專家推薦系統

Enabling Expert-driven Recommendation by Density-based Clustering on Moving Objects

沈志鴻(Jhih-Hong Shen)

指導教授 ： 吳宜鴻

中原大學/工學院/資訊工程研究所/碩士(2008年)

https://doi.org/10.6840/cycu200900530

摘要

在電子商務應用中，使用推薦系統可以針對個別消費者推薦適合的商品，而常見的推薦系統會先分類欲推薦的項目，並依照這些項目上的消費行為將使用者分群，再以分群結果作適當的推薦。分群計算往往需要在使用者上線的當下執行，因此推薦系統的反應時間就顯得相對重要；有鑑於此，本論文以找出可呈現實際行為分佈的分群為目標，提出一個有效率的方法來快速計算分群，以兼顧推薦結果的準確度與系統執行的效率。在準確度方面，由於以距離為基礎分群法不易找出不規則形狀的分群，故改用以密度為基礎的分群方法，並以專家概念選出各分群中較具代表性的使用者，作為推薦的決策依據；在效率方面，我們建立多維度空間索引的機制，並盡量減少更新分群所需的計算量，進而減輕傳統以密度為基礎分群法在運算速度上過慢的問題。我們建立了一個音樂推薦系統以驗證上述方法，該系統會根據每個人聆聽不同類型音樂的經驗，為個人決定最值得推薦的音樂清單。

關鍵字

分群 ； 推薦系統 ； 移動物件 ； 密度

並列摘要

In electronic commerce, recommendation systems can be used to suggest suitable products for individual customers. Usually, such systems first classify the items, then divide users into clusters according to their behaviors on the items, and finally make the recommendation based on the clustered results. The clusters are often computed when the user is online and the response time is thus critical. Therefore, this paper aims at clustering the users to fit the distribution of their actual behaviors and proposes an efficient method for it. For accuracy, we adopt the density-based instead of distance-based clustering because the latter is not suitable for clusters with irregular shapes. Moreover, we use the concept of experts to select the representative users in each cluster. For efficiency, we integrate the concept of density into a grid-based index to reduce the computation for cluster update. With this mechanism, the difficulty of costly computation in density-based clustering methods is thus alleviated. A music recommendation system has been built for evaluation. Our system can recommend a personalized music list based on user experience in listening to different kinds of music.

並列關鍵字

moving objects ； clustering ； density ； recommendation system

參考文獻

1. Cai-Nicolas Ziegler, Georg Lausen and Lars Schmidt-Thieme, “Taxonomy-driven computation of product recommendations,” CIKM, pp. 406-415, 2004.

2. Charu C. Aggarwal, Jiawei Han, Jianyong Wang and Philip S. Yu, “A Framework for Clustering Evolving Data Streams,” VLDB, pp. 81-92, 2003.

4. George Karypis, “Evaluation of Item-Based Top-N Recommendation Algorithms,” CIKM, pp. 247-254, 2001.

5. Huanliang Sun, Ge Yu, Yubin Bao, Faxin Zhao and Daling Wang, “CDS-Tree: An Effective Index for Clustering Arbitrary Shapes in Data Streams,” RIDE-SDMA, pp. 81-88, 2005.

6. Hung-Chen Chen and Arbee L.P. Chen, “A Music Recommendation System Based on Music Data Grouping and User Interests,” CIKM, pp. 231-238, 2001.