使用擴充資料進行共分群的協同式推薦系統; Collaborative Filtering based on Co-clustering with CCAM

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/48484

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/48484

Title:	使用擴充資料進行共分群的協同式推薦系統;Collaborative Filtering based on Co-clustering with CCAM
Authors:	劉睿哲;Rui-Zhe Liu
Contributors:	資訊工程研究所
Keywords:	分群;協同式推薦;資訊理論;額外資料;共分群;推薦系統;文本式推薦;廣告;information theory;clustering;recommender system;content-based filtering;collaborative filtering;ad;co-clustering;coclustering;extensive data;augmented
Date:	2011-08-03
Issue Date:	2012-01-05 14:56:03 (UTC+8)
Abstract:	近二十年來，推薦系統已成為一門廣受歡迎的研究領域，無論是學術界或者是業界都投以高度的興趣。身為推薦系統的其中一個分支，協同式過濾推薦系統尤其以藉由分享同儕的意見而能夠高度精確的預測使用者喜好而經常受到學者專家的討論。然而，協同式過濾推薦系統仍有它所需要面對的問題，最常見的就是缺乏已知資料的議題。相對於大量的未知資料，通常資料集會嚴重缺乏可加以利用的已知的資料，因此預測使用者的喜好所需的資訊將會非常稀少。在這篇論文中，我們引用一種新奇的協同式過濾推薦系統，同時結合基於使用者的喜好進行推薦的演算法、基於廣告的受歡迎程度進行推薦的演算法以及考慮擴充矩陣的共分群演算法 - 一種在透過最佳化共分群以減少因共分群而損失的資訊時又能夠從多個相關矩陣中考量有用的資訊而能夠運算出更佳的分群效果的演算法以產生最終的推薦。在我們提出的混合模型實驗結果中，它在預測使用者喜好的誤差相對於相關的演算法例如k-Means、k-NN以及ITCC表現的更為出色，顯示出在缺乏已知資料的議題上能藉由同時應用多重矩陣之間的資訊而能夠產生更佳的處理結果。 Recommender system has become an important research area since the high interest of academia and industry. As a branch of recommender systems, collaborative filtering systems take its roots from sharing opinions with others and have been shown to be very effective for generating high quality recommendations. However, CF often confronts a problem of sparsity which is caused by relevantly less number of ratings against the unknowns that need to be predicted. In this paper, we consider a hybrid approach which combines the content-based approach with collaborative filtering under a unified model called co-clustering with augmented data matrix (CCAM) that overrides information-theoretic co-clustering (ITCC) in order to further consider augmented data like user profile and item description. By presenting results on a better error of prediction, we show that when model-based CF method was integrated with memory-based method, our algorithm is the more effective than state-of-the-art algorithms k-NN, k-Means and ITCC through optimizing the co-cluster in mutual information loss between multiple tabular data in addressing sparsity.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	444	View/Open

社群 sharing

Loading...