一般圖書館查詢系統只能依靠關鍵字詞逐一尋找所需圖書，若遇到書名用詞或語言不同情況，就會增加找到符合需求圖書的困難度。本文推薦系統有別於一般關鍵字搜尋，希望能利用同儕借閱紀錄做協同推薦且使用圖書分類號輔助推薦排序。另外，由於個資問題，可能無法取得當下借閱者過去借閱紀錄，因此在不能直接使用人推薦物方法前提下，本文提出一套方法可同時適用於物推薦物及人推薦物之場合進行圖書推薦。
為輔助評估，本文使用機器學習軟體-Mahout的三種人推薦物方法，分別為斜率1、用戶為本和物品為本。此外本文也針對四種物推薦物方法進行比較，分別是本文提出的兩層關聯式物推薦物方法、Mahout物推薦物方法、兩層關聯式物推薦物結合分類號方法、不含本身兩層關聯式物推薦物結合分類號方法。本文以精確率為評估指標，結果發現本文提出方法優於Mahout三種方法，結合分類號比未結合佳，推薦時含給定圖書本身比不含佳。

Most library search systems only rely on keywords input by the user to find the desired books. When there are alternative synonyms or translations for the input keywords, one would often find it difficult to locate other related books. Instead of using the keywords approach, this work considers the use of collaborative filtering on the library circulation data for book borrowing recommendation. Under the premise that a user might not be willing to disclose the past borrowing records for privacy reasons, an item-based recommendation framework is proposed which will work for both use cases of item-based and user-based book recommendation. Given a book item, our two-layer relational item-based recommendation method will consider those books borrowed by the common users and use sorting keys first based on the past borrowing count and then on the call-number-based distance.
For benchmarking, the Mahout open source machine learning package is adopted where the slope-one, user-based, and item-based recommendation methods are evaluated. In addition, four item-based recommendation methods are evaluated which include our two-layer relational item-based method, Mahout item-based method, our two-layer relational item-based method combined with the call number, and our two-layer relational item-based method excluding self recommendation. This work uses the precision as the performance index. The experimental results show that our method is better than the three Mahout methods. Our method combined with the call number is better than that without the call number. Also, our method with self recommendation is better than that without self recommendation too.

目錄
一、緒論...1
二、文獻探討...2
2.1推薦系統...2
2.2 Amazon購物網站...6
2.3圖書分類法...6
2.4 Mahout機器學習軟體...8
三、方法介紹...11
3.1問題定義...11
3.2兩層關聯式物推薦物方法...12
3.3結合分類號的兩層關聯式物推薦物方法...13
3.4人推薦物方法...14
四、實驗...15
4.1資料來源及前處理...15
4.2實驗步驟和結果...17
4.2.1 Mahout推薦法評估...18
4.2.2兩層關聯式物推薦物方法評估...23
4.3 網路服務...30
五、結論...33
六、參考文獻...35

圖目錄
圖2. 1、中國圖書分類法...8
圖2. 2、User-Based演算法示意圖...9
圖2. 3、Item-Based演算法示意圖...10
圖2. 4、Slope-One演算法示意圖...10
圖3. 1、研究架構...12
圖3. 2、兩層關聯式推薦法...13
圖3. 3、分類號距離計算示意圖...14
圖3. 4、嵌入物推薦物後的人推薦物方法...15
圖4. 1、各種借閱者百分位數下的最高借書數分布(log)...16
圖4. 2、全年各種借書總數的人數分布(log)...16
圖4. 3、Mahout三種推薦方法精確率比較圖...22
圖4. 4、本文幾種推薦方法精確率比較圖...28
圖4. 5、七種推薦法精確率比較圖...29
圖4. 6、提供圖書推薦網路服務示意圖...30
圖4. 7、網路服務製作概要...31
圖4. 8、推薦畫面(1)...32
圖4. 9、推薦畫面(2)...32

表目錄
表4. 1、圖書館流通紀錄格式...15
表4. 2、訓練測試集摘要...17
表4. 3、MahoutSlopeOne平均精確率...19
表4. 4、MahoutUserBase平均精確率...20
表4. 5、MahoutItemBase平均精確率...21
表4. 6、使用兩層關聯式物推薦物方法在未結合分類號下進行人推薦物結果...23
表4. 7、使用Mahout物推薦物方法進行人推薦物結果...24
表4. 8、使用兩層關聯式物推薦物結合分類號方法下進行人推薦物結果...25
表4. 9、使用不含本身兩層關聯式推薦物結合分類號方法下進行人推薦物結果..27

[1]卜小蝶，「以圖書借閱紀錄探勘加強圖書資源利用之探討」，中國圖書館學會會報，第66期，頁59-72，2001。
[2]林慶隆，陳建民，邱重毅，圖書分類法1－杜威十進制圖書分類法，國家研究教育院電子報，第46期，http://epaper.naer.edu.tw/index.php?edm_no=46&content_no=1277，2012。
[3]洪志淵，「圖書流通紀錄之一般化相關規則找尋之研究」，國立中山大學資訊管理研究所碩士論文，2001。
[4]康勝修，「應用資料探勘技術於圖書館借閱紀錄分析與館藏查詢服務」，國立雲林科技大學資訊管理研究所碩士論文，2006。
[5]郭俊桔，張育蓉，張瑞珊，「導入矩陣分群與主題地圖之圖書推薦系統」，Workshop of Current Research in Information and Library Sciences (CRILS)，2011。
[6]陳麥麟屏，林國強，美國國會圖書館主題編目，三民書局，2001。
[7]黃淑霞，「資料探勘技術於圖書館借閱歷程分析及推薦系統」，國立東華大學資訊工程系碩士在職班論文，2008。
[8]賴永祥，中國圖書分類法，文華圖書管理資訊股份有限公司，2001。
[9]戴玉旻，「圖書館借閱紀錄探勘系統」，國立交通大學資訊科學研究所碩士論文，2002。
[10]A. Alsalama, “A Hybrid Recommendation System Based on Association Rules, ” Western Kentucky University, 2013.
[11] M. Balabanovi’c and Y. Shoham, “Fab: content-based, collaborative recommendation,” Communications of the ACM, vol. 40,  no. 3, 66–72, 1997.
[12] C. Basu, H. Hirsh, and W. Cohen, “Recommendation as classification: using social and content-based information in recommendation,” in Proceedings of the 15th National Conference on Artificial Intelligence (AAAI’98) (Madison, WI), American Association for Artificial Intelligence, 714–720, 1998. 
[13] D. Billsus and M. J. Pazzani, “Learning collaborative information filters,” Proceedings of the 15th International Conference on Machine Learning (ICML’98), Morgan Kaufmann, San Francisco, 46–54,1998.
[14] R. Burke, “Knowledge-based recommender systems,” Encyclopedia of Library and Information Science, vol. 69, no. 32, 180–200, 2000. 
[15] M. Claypool, A. Gokhale, T. Miranda, P. Murnikov, D. Netes, and M. Sartin, “Combining content-based and collaborative filters in an online newspaper,” Proceedings of the ACM SIGIR Workshop on Recommender Systems: Algorithms and Evaluation (Berkeley, CA), 1999.
[16] J. Han, M. Kamber, “Data Mining: Concepts and Techniques”, Morgan Kaufmann, 2006.
[17] J. Han, J. Pei, and Y. Yin, “Mining Frequent Patterns without Candidate Generation,” ACM SIGMOD International Conference on Management of Data, pp.241-250,2000.
[18] D. Jannach, M. Zanker, A. Felfernig, G. Friedrich, “Recommender Systems: An Introduction,” Cambridge, 2011.
[19] G. Linden, B. Smith, J. York, “Amazon.com Recommendations: Item-to-item Collaborative Filtering,” IEEE Internet Computing, pp.76-80, Jan/Feb, 2003.
[20] S. Liao, K. Kao, I. Liao, H. Chen, S. Huang, “PORE: a personal ontology recommender system for digital libraries,” The Electronic Library, vol. 27, no. 3, pp. 496-508, 2009. 
[21]R. Nakamoto, S. Nakajima, J. Miyazaki, S. Uemura, “Tag-Based Contextual Collaborative Filtering,” IAENG International Journal of Computer Science, 34:2, IJCS_34_2_08, 2007.
[22] S. Owen, R. Anil, T. Dinning, E. Friedman, Mahout in Action, Manning Publications, 2012.
[23] M. J. Pazzani, D. Billsus, Content-Based Recommendation Systems, 2007.