以信任為基之過濾建構多準則推薦環境.pdfVIP

以信任為基以信任為基之過濾建構之過濾建構多準則推薦環境多準則推薦環境 以以信任為基信任為基之過濾建構之過濾建構多準則推薦環境多準則推薦環境 1 2 2 3 翁頌舜 楊銘賢 葉建宏 吳宗憲 (1 台北科技大學資訊與運籌管理研究所，2 輔仁大學資訊管理研究所，3 輔仁大學商學研究所) 摘 摘 要要：：為了建立多構面的使用者輪廓，本研究採用多準則評分(Multi-criteria Ratings)作為電影喜好蒐集，並且以統 摘摘 要要：： 計方法的複迴歸分析將迴歸係數作為評估使用者各準則偏好程度，為了增加各準則下評分預測品質以降低傳統協同 過濾稀疏性問題的影響，首先將各個準則下成員偏好程度以及評分可靠程度作為信任權重計算，並且以信任權重門 檻值作為推薦群體篩選標準，並利用信任為基之過濾(Trust-Based Filtering)方法於各個準則評分預測，在資料相對稀 疏情況下可減少預測誤差。在最終預測聚合實驗中發現以信任為基之過濾對整體評分預測有較低的 MAE 值，在各資 料密度下對於推薦評估指標 F1 顯示也有較好的推薦效能。由實驗結果發現本研究所提出方法有助於提升多準則推薦 環境下之推薦品質。 關鍵詞關鍵詞：：信任為基之過濾、多準則評分、推薦系統、稀疏性、複迴歸分析 關鍵關鍵詞詞：： Using Trust-Based Filtering to Develop a Multi-criteria Recommendation Environment Abstract: In this research, we applied multi-criteria ratings in movie preference collection for building users’ profiles of different aspects profile and we used regression coefficient, which was derived from multiple regression analysis, for each criterion preference level. In order to improve each criterion prediction quality and alleviate the sparse problem of CF, we firstly combine users’ preference levels and trust-values as trust-weight in different criterions and then set trust-weight threshold to filter users and to find recommendation members. In a sparser data set, trust-based filtering method can reduce the predictable errors and its performance is better than the traditional CF technique. In the final aggregation experiments, we observed that trust-based filtering method also brought out a lower MAE in overall rating prediction and the F1 value has a better recommendation performance in different sparsity level dataset. Finally, the approach we proposed can improve the recommendation quality in the multi-criteria recommend