整合內容導向式方法與混合式協同過濾之電影推薦

Integration of Content-based approach and Hybrid Collaborative Filtering for Movie Recommendation

李佳馨(Chia-Hsing Lee)

指導教授 ： 翁頌舜

國立臺北科技大學/管理學院/資訊與運籌管理研究所/碩士(2013年)

摘要

隨著電子商務規模不斷擴大，企業為了節省消費者的搜尋時間與成本，個人化推薦系統應運而生。個人化推薦系統的核心技術中，應用最為廣泛的推薦方法之一的協同過濾，仍然存在著幾個問題。第一為評分矩陣稀疏問題(Sparse)，難以找到相似的使用者，影響預測的準確度，第二為冷啟動(Cold Start)，包含新使用者與新項目的問題，缺乏評分依據，導致無法預測使用者的喜好程度進行推薦。
本研究模擬推薦系統所面臨的真實環境，在評分矩陣稀疏的情況下，考量新使用者與新電影的因素，以電影屬性為基礎的內容導向式方法(Content-based approach)預測一般使用者對新電影的評比，並且修正協同過濾(Collaborative Filtering)傳統的相似度計算結合矩陣分解(Matrix Factorization)方法預測新使用者與一般使用者尚未評分之電影評比。在不同稀疏程度的資料集下，以本研究方法對整體評分預測有較低的MAE誤差值，代表預測的分數越接近實際的評分也較符合使用者的喜好，實驗證實本研究方法在稀疏評分矩陣的預測準確率較高且優於傳統的協同過濾方法。

關鍵字

協同過濾 ； 內容導向式方法 ； 矩陣分解 ； 冷啟動

並列摘要

As the scale of e-commerce continues to expand, personalized recommendation systems have been developed for general users in the hope of saving their search cost and time. In the core methods of personalized recommendation systems, collaborative filtering, one of the most widely-used recommended methods, still leaves two major problems. One is sparsity problem, the difficulty of finding similar users results in poor accuracy. The other is cold start, new users and new items make it hardly possible to estimate the preferences because of the lack of past ratings.
This work simulates a real environment for movie recommendation. In the case of considering the factors of the new users and new movies in the sparse rating matrix, we conduct a content-based approach based on movie genre to predict user ratings on new movies. Furthermore, we integrate the modification of similar measures in memory-based collaborative filtering with matrix factorization(model-based collaborative filtering). In experiments, we observe our methodology brought out a lower MAE in overall rating prediction. Finally, our approach has been shown to have better recommendation quality than basic collaborative filtering in different sparsity level dataset.

並列關鍵字

Collaborative Filtering ； Content-based Approach ； Matrix Factorization ； Cold Start

參考文獻

1. 郭秉仁，基於個人本體論與MapReduce技術之圖書推薦系統，碩士論文，國立中興大學資訊科學與工程學系所，2012年。

3. Adomavicius, G. & Tuzhilin, E., “Toward the next generation of recommender systems: A survey of the state-of-the-art and possible extensions,” IEEE Transactions on Knowledge and Data Engineering, vol. 17, no. 6, 2005, pp. 734-749.

4. Alspector, J. A., Kolcz, A. & Karunanithi, N., “Comparing feature-based and clique-based user models for movie selection,” Proceedings of the 3rd ACM Conference on Digital Libraries, Pittsburgh, Pennsylvania, 1998, pp. 11-18.

5. Armstrong, R., Freitag, D., Joachims, T. & Mitchell, T., “WebWatcher: A Learning Apprentice for the World Wide Web,” Proceedings of Fourteenth International Joint Conference on Artificial Intelligence, Stanford, CA, 1995.

6. Balabanovic, M. & Shohan, Y., “Fab: content-based, collaborative recommendation,” Communications of ACM, vol. 40, no. 3, 1997, pp. 66-72.