An Effective Two-Phase Collaborative Filtering Algorithm for Recommender Systems

二階段協同式資訊過濾演算法的推薦系統

一般而言，推薦系統主要是可以藉由使用者的行為模式，利用機器學習的方法來處理資訊過量的問題。雖然傳統的協同式資訊過濾技術可以有效地預測使用者的喜好樣式，但還是有資料稀疏性的問題要去克服。為了要降低資料稀疏性的影響，本研究提出一種創新的協同式資訊過濾技術的推薦方法，這個推薦方法將預測的程序分解成二個階段。在第一個階段，會將使用者沒有評分過的影片來做預估值的填補，然後再利用這些填補值在第二階段做預測的計算。實驗的結果顯示本方法有很好的效率，尤其是當資料庫的資料為非常稀疏的情況之下，本方法之表現顯著優於傳統協同式資訊過濾技術的方法。

A recommender system is a set of machine learning methods that adapt to a user’s behavior to deal with the problem of information overload. Although traditional collaborative filtering has been shown to be effective in predicting a user’s preferences, it suffers from a data sparsity problem. To alleviate this sparsity problem, in the present study, an innovative collaborative filtering recommender that decomposes the prediction procedure into two phases is proposed. In the first phase, the user’s unknown ratings are estimated as the initial ratings to provide information for the second prediction phase. Experimental evaluation results show the effectiveness of the proposed method, especially for situations with very sparse data.