一個基於新聞流與相關回饋為基礎之個人資料檔強化的混合式新聞推薦機制

Abstract

隨著電腦科技與網際網路的發展，資訊傳遞與交換的質與量都隨之俱增，為避免資訊過載造成決策時的錯誤判斷，推薦系統的發明與導入成為重要關鍵。在以內容為基礎的推薦系統，其推薦結果優劣除了受到所使用的演算法外，也受到個人資料檔品質的影響。此外個人資料檔雖能代表個人長期的喜好，卻無法反映出短期的興趣所在。 為此本論文提出一混合式推薦演算法，結合新聞流與相關回饋兩技術，藉此反應使用者短期與長期的喜好，以獲得較佳的推薦結果。本文並實作一新聞推薦網站，藉以驗證與比較不同演算法。經實作一新聞推薦網站進行實驗證實，本文所提出之混合式推薦演算法能較單一使用語意式內容基礎推薦方法將推薦精準度從0.311提升至0.744，達到更佳的推薦結果供使用者使用。

With the development of computer science and Internet, the exchange and deli-very of information increases rapidly. To avoid decision mistake caused by informa-tion overload, recommendation system has been invented and introduced. For con-tent-based recommendation system, the quality of recommendation result is affected not only by algorithm itself, but also the quality of user profile. Besides, a con-tent-based recommendation system is unable to reflect short-term interest of users. In this thesis we proposed a hybrid recommendation algorithm combined news flow and relevance feedback. With these two techniques, the system can reflect short-term and long-term user interests. We also implemented the algorithm in a news recommendation website, which helped us to validate our algorithm and compare to other algorithm. Through the experiment based on a news recommendation website, our algorithm has been proven that the hybrid algorithm performs better than semantic content-based recommendation algorithm which enhanced precision from 0.311 to 0.744, and provides better recommendation result to users.