許永真臺灣大學：資訊工程學研究所林友宣Lin, Yu-HsuanYu-HsuanLin2007-11-262018-07-052007-11-262018-07-052006http://ntur.lib.ntu.edu.tw//handle/246246/53864Nowadays the amount of information in the world is increasing far more quickly than our ability to process them. How people can use their limited time to get interesting information has become an important issue in our daily life. Collabora-tive ‾ltering recommender system is one of the prevailing approaches that can help users to ‾lter unsuitable information. However, traditional collaborative ‾ltering recommender systems do not take the changing behavior of each user's interests into account. This research proposes a new time-weighted collaborative ‾ltering recommender system to capture each user's current interests precisely. The ex-perimental results show that the time-weighted collaborative ‾ltering recommender system outperforms the traditional collaborative ‾ltering recommender system with 11.2% in recommendation accuracy.Acknowledgments i Abstract iii List of Figures viii List of Tables ix Chapter 1 Introduction 1 1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.3 Problem De‾nition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.4 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Chapter 2 Literature Survey 7 2.1 Recommender System . . . . . . . . . . . . . . . . . . . . . . . . . . 7 2.2 Demographic Filtering Approach . . . . . . . . . . . . . . . . . . . . 9 2.3 Content-Based Filtering Approach . . . . . . . . . . . . . . . . . . . . 10 2.3.1 Standard Keyword Matching . . . . . . . . . . . . . . . . . . . 11 2.3.2 Cosine Similarity . . . . . . . . . . . . . . . . . . . . . . . . . 11 2.3.3 Classi‾cation . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 2.4 Collaborative Filtering Approach . . . . . . . . . . . . . . . . . . . . 12 2.4.1 Nearest Neighbors . . . . . . . . . . . . . . . . . . . . . . . . 15 2.4.2 Item-Based Filtering . . . . . . . . . . . . . . . . . . . . . . . 15 2.4.3 Clustering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 2.5 Hybrid Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 Chapter 3 Time-Weighted Collaborative Filtering Recommender Sys- tem 19 3.1 Recommendation Process . . . . . . . . . . . . . . . . . . . . . . . . . 19 3.2 Correlation Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 3.2.1 Traditional Correlation Process . . . . . . . . . . . . . . . . . 23 3.2.2 Time-Weighted Correlation Process . . . . . . . . . . . . . . . 26 3.3 Aggregation Process . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 Chapter 4 Experiments and Analysis 35 4.1 Experiment Data Set . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 4.2 Experiment Evaluation and Results . . . . . . . . . . . . . . . . . . . 36 4.2.1 Experiment Evaluation . . . . . . . . . . . . . . . . . . . . . . 37 4.2.2 Sensitivity of Half-Life Scaling Parameter . . . . . . . . . . . . 37 4.2.3 Comparison with Traditional Collaborative Filtering Recom- mender System . . . . . . . . . . . . . . . . . . . . . . . . . . 39 Chapter 5 Conclusion 41 5.1 Summary of Contributions . . . . . . . . . . . . . . . . . . . . . . . . 42 5.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 Bibliography 43506751 bytesapplication/pdfen-US合作過濾推薦系統時間加權collaborative filteringtime-weightedrecommender systemthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/53864/1/ntu-95-R93922122-1.pdf