- 學位論文
應用個人與群體信任之文件推薦
Applying personal and group trust to document recommendation
摘要
協同過濾(CF)推薦方法已被普遍應用在各領域中並用以解決資訊過載的問題，在知識密集的組織環境中，很多工作都是以團隊合作的方式進行，使用者會參考群體成員的喜好或意見來做決策，使得傳統的個人化推薦系統將無法滿足群體使用者的需求，因此本研究提出以群體信任為基礎之推薦方法，將分析使用者在群體中受信任之程度，並尋找值得信賴的工作者。群體信任包含兩個部分: 混合式個人信任模型與個人在群體中的重要性。混合式個人信任模型，結合文章層級個人信任與主題層級個人信任來計算個人信任值，並利用主題層級個人信任來彌補文章層級個人信任之不足。評估使用者在群體的重要性分成三種方法，分別為活躍度、相似度、聲譽。本研究將群體信任與協同過濾方法整合，透過群體信任的分析，得到較可靠的信任值，期望提升推薦品質。經實驗結果可知，本研究所提出的以群體信任為基礎之推薦，比其他傳統以信任為基礎的推薦方法，更能得到較好的推薦結果，提高評分預測的準確度，並提升推薦品質。
並列摘要
Collaborative Filtering(CF) recommender systems have been applied on various of field to solve the problems of information overloading. In the environment of information concentrated, most of the works are processing with group. The preference and the suggestion will provide reference for the users to make decisions. This kind of situation makes the previous personalized recommender systems no longer to meet the demand of the users. Therefore, this research proposes a new recommender system based on the group-trusted. Analyze the extent of trusted users in the group, then find trustworthy users. There are two parts included in the group-trusted: hybrid personal trust model and importance of a person in the group. Hybrid personal trust model, combines the document-level trust and topic-level trust metric to resolve the co-rated of articles too little shortcomings. To define the importance of the users has three methods: namely activity, similarity, and reputation. This research integrate the group-trusted and trustworthiness of users, through the analyzed of group-trusted to get the reliable trust value and expect it to promote the quality of recommendation. Form the result of this research, the recommender system which based on the group-trusted could get the better recommendation than the previous recommender which just based on the trust. The result of this research also could help to improve the accuracy of the prediction and promote the quality of the recommendation.