- 學位論文
目標導向屬性評估於推薦系統之研究
A Goal-Driven Attribute Evaluation Method for Recommendation Systems
摘要
推薦系統能夠將可能喜好的資訊或物品推薦給使用者，而過去學者們不斷提出新方法或是架構，希望能提高推薦精準度。在進行推薦時，群組中若有大量使用者，會導致系統在進行推薦的準確率下降，學者們提出了許多分類方法，試圖解決這個問題，而本研究以使用者之間的關係進行相似度比較，所以採用了協同式過濾。但採用協同式過濾技術會伴隨著冷啟始問題以及稀疏問題，為了解決這些問題，利用加入使用者的屬性來輔助解決。本論文為了使用者屬性的選擇，能夠以更客觀的方式做擷取，而採用了目標導向使用案例的方式對屬性做篩選，選出最適當且符合使用者要求的屬性，加入系統讓推薦時有更多的依據，同時也幫助資料量較少系統作為推薦參考依據。本研究利用MovieLens作為實驗資料，因為此資料庫的資料完整度高，推薦參考資料也較多，所以推薦精準度會較高，此外在推薦系統中加入了最近鄰居法，驗證對相似使用者進行分群會影響推薦的效果；而在本研究與另一個資料較少的電影資料庫做比較，發現推薦精準度變差，但當加入使用者額外訊息時，對於推薦效果是有正向影響的。
並列摘要
User classification in a recommendation system is a very important research direction, because it shortens the process time of the recommendation system. However, some negative information will be accompanied at the same time when the number of users grows. Without proper processing of information, the cold-start problem or the sparsity problem may arise. When the system has a new user, recommendation may be unsuccessful, because the new user does not have his/her own usage. In order to avoid these things to happen in a recommendation system, this research has added user data and status to aid the recommendation system. In this study, we use the goal-oriented way to find an appropriate user characteristics based on user’s status attributes to be include in the recommendation system. The recommendation system will be more flexible by adding status information, because the recommendation system has more reference information. In this study, MovieLens is used as the experimental data. Because data integrity is high and usable reference materials are also more, its recommended accuracy will be higher. By adding the nearest neighbor method to the recommendation system, we obtain a better result. When the same method is applied to a small database the recommendation accuracy deteriorates. However, when adding the user information, we have a positive effect for the recommendation results.