近年來，由於資訊發展的快速，依賴傳統CD音樂市場，轉變為數位音樂系統，為了增加聆聽音樂的方便性，線上數位音樂逐漸成為當代的主流市場，由於網路使用數量的上升，使線上服務可以藉由聽眾使用習慣收集大量的資訊訊息。本研究透過知名音樂平台KKBOX聽眾的歷史聆聽紀錄找出聽眾與歌手之間的關聯，過去研究都是關於大型銷售市場的分析，卻忽略了有忠實粉絲所喜愛的歌手。雖然表面上並沒有達到可觀的利益，但是將這些歌手進行分析，將可以達到不錯的經濟效益。所以本研究將會從聽眾的聆聽紀錄中，挖掘線上音樂市場中的小眾市場，藉由小眾市場進行市場分析與線上音樂排行榜的預測。
    因此，本研究將會藉由聽眾對於歌手的聆聽習慣和歌手被聆聽的歷史紀錄，使用雙分圖(Bipartite graph)，G(U, S, W)的方式實現雙聚類(Co-clustering)交互參考的概念，U表示我們的聽眾、S表示歌手，W則表示聽眾跟歌手之間的聆聽次數，首先藉由聆聽區間算出聽眾歷史聆聽紀錄忠誠度找出忠實聽眾，接著藉由忠實聽眾找出小眾歌手及透過小眾歌手找出小眾聽眾，藉由聆聽區間與資訊熵(Information Entropy)計算小眾歌手忠誠分數當作權重值找出每位歌手的小眾分數並進行排序，最後預測小眾市場後起之秀。我們藉由驗證歌手聆聽頻率呈現冪次定律判斷是否符合帕累托法則(Pareto principle)，根據帕累托法則選擇小眾歌手候選人，最後透過忠誠度的計算給予小眾歌手小眾分數指標將小眾歌手進行排序。根據我們提出的框架可以有效產生小眾歌手列表，經由小眾歌手銷售的狀況進行預測模型的建立，預測部份主要分成三個項目預測銷售量位置、預測排名區間及預測排名動態，將透過決策樹、隨機森林、K最近鄰、樸素貝葉斯與支持向量機建立預測模型。
    實驗結果，決策樹在三個實驗中表現優異，實驗最後選擇了精確率、召回率與F值作為評估模型的衡量標準，我們藉由決策樹針對三個實驗所得到的最優結果分別為72%、97%與95%，因此可以發現在預測歌手排名區間與動態實驗中，決策樹展現樹狀結構優勢。

Recently, the rapid development of information dissemination approaches has facilitated the transformation of the conventional music market based on compact discs into a digital music system. Online digital music has gradually become the mainstream market because of the convenience it affords for music listening. In addition, the growing number of Internet users has enabled online music services to collect a considerable amount of information according to users’ listening patterns.The objective of this study was to reveal the connections among users and artists by examining users’ listening histories on the renowned music-streaming platform KKBOX. Past research has focused on the analysis of mass markets while neglecting the analysis of artists with loyal fans. Although such artists mostly do not gain considerate profits, analyzing the artists can contribute to the economic benefits the music market. Therefore, this study explored niche markets within the online music market by analyzing users’ listening histories; in addition, market analyses were conducted to predict online music charts in the niche markets. 
   This study applied a bipartite graph G = (U, S, W) to execute co-clustering, a cross-reference concept, according to the listening patterns of users and listening histories of artists. Let U represent users, S represent artists, and W represent users’ listening frequency for artists. First, on the basis of listening histories, loyal users were determined by calculating their degree of loyalty to artists, which was derived through listening intervals. The loyal users could thus serve as indicators of niche artists, which could then be used to determine niche users. Listening intervals and information entropy were used to calculate the loyalty score for each artist. The loyalty scores served as the weights for calculating the niche’s score for each artist, and a ranking list of the artists could be produced accordingly. Finally, emerging talents in the niche market could be predicted. Listening frequencies exhibit a power-law distribution; this study thus investigated the conformity of such a distribution with the Pareto principle. The study selected niche artist candidates according to the Pareto principle. Finally, the study created a ranking list of niche artists according to the niche’s score index, which was calculated using the derived degree of loyalty. The proposed study framework can enable the effective establishment of a ranking list of niche artists. A predictive model was established according to the number of sales, and three prediction targets were set: the sales distribution, ranking range, and ranking dynamic. Decision trees, random forests, k-nearest neighbors, Naive Bayes, and support vector machines were used to establish the estimation model.
    The results revealed that the decision trees exhibited favorable performance levels in the three experiments. Precision ratio, recall ratio, and F distribution were used for model assessment after the experiments. The optimal performance levels of the decision trees in the three experiments were 72%, 97%, and 95%. Therefore, the tree structures of decision trees were advantageous for predicting the range and dynamic of ranking in the experiments.

目錄
第一章	緒論	1
第一節	研究背景與動機	1
第二節	研究目的	2
第三節	論文組織與架構	3
第二章	文獻探討	4
第一節	音樂分類	4
第二節	協同過濾推薦	5
第三節	客戶忠誠度	7
第四節	小眾市場	8
第三章	研究方法	9
第一節	雙分圖網路建構之忠誠度計算	11
第二節	帕累托法則	13
第三節	資訊熵	16
第四節	分類模型建立	17
第四章	實驗結果	26
第一節	實驗說明	26
第二節	實驗結果	29
第三節	實驗討論	48
第五章	結論與未來展望	50
第一節	結論	50
第二節	問題討論與未來展望	51
參考文獻	52
附錄-英文論文	56

圖目錄
圖3-1系統架構圖	10
圖3-2聽眾與歌手雙分圖網路建立	11
圖3-3帕累托法則驗證圖	14
圖3-4歌手被聆聽總人數統計圖	15
圖3-5帕累托法則第二次驗證	15
圖3-6歌手被聆聽總次數統計圖	16
圖3-7歌手資訊熵	17
圖3-8 SVM線性可分割	19
圖3-9 SVM線性不可分割	19
圖3-10排名特徵轉換	23
圖4-1聽眾與歌手交叉參考	26
圖4-2原始資料	27
圖4-3分類問題與預測問題比較	28
圖4-4小眾歌手清單比較圖	30
圖4-5帕累托法則80/20(左)與 90/10(右)比較圖	31
圖4-6歌手排名漲幅	32
圖4-7 預測銷售量位置分類器精確率比較	33
圖4-8預測銷售量位置分類器召回率比較	35
圖4-9預測銷售量位置分類器F值比較	36
圖4-10預測排名區間分類器精確率比較	37
圖4-11預測排名區間分類器召回率比較	39
圖4-12預測排名區間分類器F值比較	40
圖4-13預測排名動態分類器精確率比較	41
圖4-14預測排名動態分類器召回率比較	42
圖4-15預測排名動態分類器F值比較	44
圖4-16特徵值重要性	45
圖4-17預測銷售位置決策樹規則	46
圖4-18預測排名區間決策樹規則	47
圖4-19預測排名動態決策樹規則	48

表目錄
表3-1特徵說明	12
表3-2決策樹分類器特徵說明	20
表4-1預測銷售量位置精確率比較	33
表4-2預測銷售量位置召回率比較	34
表4-3預測銷售量位置F值比較	35
表4-4預測排名區間精確率比較	37
表4-5預測排名區間召回率比較	38
表4-6預測排名區間F值比較	39
表4-7預測排名動態精確率比較	41
表4-8預測排名動態召回率比較	42
表4-9預測排名動態F值比較	43

[1]	Ruth Dhanaraj and Beth Logan, “Automatic Prediction of Hit Songs,” in Proc. 2005 International Society for Music Information Retrieval (ISMIR),” London, UK, Jan, 2005.
[2]	Stephen Mussmann, John Moore, and Brandon Coventry, “Using Machine Learning Principles to Understand Song Popularity,” [Online]. Available: https://ecitydoc.com/
download/using-machine-learning-principles-to-understand-song-popularity-1_pdf, Jan. 19, 2018.
[3]	Dorien Herremans, David Martens, and Kenneth Sörensen, “Dance Hit Song Prediction,” Journal of New Music Research, 43(3):291–302, July, 2014.
[4]	Chia-Hao Chung, Yian Chen, and Homer Chen, “Exploiting Playlists for Representation of Songs and Words for Text-Based Music Retrieval,” in Proc. 2017 18th International Society for Music Information Retrieval Conference, Oct. 23-27, 2017.
[5]	Nicanor Marco P. Valdez and Rowena Cristina L. Guevara, “Feature Set for Philippine Gong Music Classification by Indigenous Group,” in Proc. 2011 IEEE Region 10 Conference, pp. 339–343, Nov. 21-24, 2011.
[6]	Jen-Yu Liu and Yi-Hsuan Yang, “Inferring Personal Traits From Music Listening History,” in Proc. 2012 ACM Music Information Retrieval with User-Centered and Multimodal Strategies (MIRUM), pp. 31-36, Nara, Japan, Nov. 2-2, 2012.
[7]	Tianqi Zhou, Lina Chen, and Jian Shen, “Movie Recommendation System Employing the User-Based CF in Cloud Computing,” in Proc. 2017 IEEE International Conference on Computational Science and Engineering (CSE) and IEEE International Conference on Embedded and Ubiquitous Computing (EUC), Guangzhou, China, July. 21-24, 2017.
[8]	Shohei Kinoshita, Takahiro Ogawa, and Miki Haseyama, “LDA-Based Music Recommendation with CF-Based Similar User Selection,” in Proc. IEEE 4th Global Conference on Consumer Electronics(GCCE), pp.215-216, Feb. 14-16, 2016.
[9]	Mubbashir Ayub, Mustansar Ali Ghazanfar, Muazzam Maqsood, and Asjad Saleem, ”A Jaccard Base Similarity Measure to Improve Performance of CF Based Recommender Systems,” in Proc. 2018 International Conference on Information Networking (ICOIN), Jan. 10-12, 2018.
[10]	Arezou Koohi, Houman Homayoun, and Jie Xu, ”Co-Clustering of Diseases, Genes, and Drugs for Identification of Their Related Gene Modules,” in Proc. 2016 8th International Conference on Advanced Computational Intelligence Chiang Mai, Thailand, Feb. 14-16, 2016.
[11]	Hiroyuki Oikawa, Kohei Otake, and Takashi Namatame, ”Loyal Customer Discriminant Model Using Customer Features on EC Site,” in Proc. 2016 Future Technologies Conference (FTC), San Francisco, United States, Dec. 6-7, 2016.
[12]	Lu-An Tang, Xiao Yu, and Sangkyum Kim, “Tru-Alarm: Trustworthiness Analysis of Sensor Networks in Cyber-Physical Systems,” in Proc. 2010 IEEE International Conference on Data Mining series (ICDM), pp. 1079–1084, Dec. 13-17, 2010.
[13]	Jun-Lin Zhou, Yan Fu, Hua Lu, and Chong-Jing Sun, “From Popularity to Personality A Heuristic Music Recommendation Method for Niche Market,” Journal of Computer Science and Technology, 26(5):816–822,2011, Sept. 23, 2011.
[14]	Kuei-Kuei Lai, Ta-Sheng Hung, and Mu-Yen Hsu, “An Innovative Model of Blue Ocean Strategy and Niche Marketing in Green Industry: A Case Study of The Smart LED Industry,” in Proc. 2011 Portland International Conference on Management of Engineering and Technology (PICMET), July. 31-Aug. 4, 2011.
[15]	Sarah Lubik, Elizabeth Garnsey, and Tim Minshall, “Beyond Niche Thinking: Market Selection in Science-Based Ventures,” in Proc. 2012 Technology Management in the Energy Smart World (PICMET), July. 29-Aug. 2, 2012.
[16]	Tao Zhou, Zoltán Kuscsik, Jian-Guo Liu, Matúš Medo, Joseph Rushton Wakeling, and Yi-Cheng Zhang, “Solving the Apparent Diversity-Accuracy Dilemma of Recommender Systems,” in Proc. 2010 National Academy of Sciences(PNAS), Mar. 9, 2010.
[17]	Jigar Patel, SahilShah, Priyank Thakkar, and K Kotecha, “Predicting Stock and Stock Price Index Movement Using Trend Deterministic Data Preparation and Machine Learning Techniques,” Journal of Expert Systems with Applications, 42(1), 259–268.1, Jan, 2015.
[18]	Dorin Militaru and Costin Zaharia, “A Survey of Collaborative Filtering-Based Systems for Online Recommendation,” in Proc. 2010 12th International Conference on Electronic Commerce: Roadmap for the Future of Electronic Business, ser. ICEC '10, pp. 43-47, New York, USA, Aug. 2-4, 2010.
[19]	Chee Seng Chong et al., “Collaborative Analytics for Predicting Expressway-Traffic Congestion,” in Proc. 2012 14th Annual International Conference on Electronic Commerce (ICEC'12), pp. 35-38, Aug, 2012.
[20]	Chen Yiling, Kash IA, and Ruberry Michael, “Eliciting Predictions and Recommendations for Decision Making,” Journal of ACM Transactions on Economics and Computation, DOI:10.1145/2556271, 2(2):6:1–6:27, June, 2014
[21]	Efthimios Bothos, Konstantinos Christidis, Dimitris Apostolou, and Gregoris Mentzas, “Information Market Based Recommender Systems Fusion,” in Proc. 2011 2nd International Workshop on Information Heterogeneity and Fusion in Recommender Systems, Oct. 27 -27, 2011.
[22]	I-Hsien Ting and Pei-Lun Yu, “Measuring the Effect of Social Network Data on Music Recommendation,” in Proc. 2016 11th International Knowledge Management in Organizations(KMO) Conference on The changing face of Knowledge Management Impacting Society, Hagen, Germany, July. 25 - 28, 2016.
[23]	Stephen Guo, Mengqiu Wang, and Jure Leskovec, “The Role of Social Networks in Online Shopping: Information Passing, Price of Trust, and Consumer Choice,” in Proc. 12th ACM conference on Electronic Commerce, pp. 157-166, June. 5 - 9, 2011.
[24]	Chen De-Fu, Wei Yong-Zhi, and Li Shi-Ming, ”The Evolutionary Path of Technological Capability and The Choice of Niche Market: A Longitudinal Research From QOK Corporation,” in Proc. 2011 International Conference on E-Business and E-Government (ICEE), Shanghai, China, May. 6-8,2011.
[25]	Sarah Lubik, Elizabeth Garnsey, and Tim Minshall, “Beyond Niche Thinking: Market Selection in Science-Based Ventures,” in Proc. 2012 PICMET '12: Technology Management for Emerging Technologies, Vancouver, BC, Canada, July. 29-Aug. 2, 2012.
[26]	Lili Tong, Yiting Wang, Fan Wen, and Xiaowen Li, “The Research of Customer Loyalty Improvement in Telecom Industry Based on NPS Data Mining,” Journal of China Communications, vol. 14, issue. 11, pp. 260-268, DOI: 10.1109/CC.2017.8233665, Dec. 22, 2017.
[27]	Yi-Lin Yu, “A Research of The Relationship Between Customers' Loyalty and Degree of Involvement in Virtual Game Props Trading Site,” in Proc. 2017 International Conference on Information and Communication Technology Convergence (ICTC), Jeju, South Korea, Oct. 18-20, 2017.
[28]	Hilal Alhulail, Martin Dick, and Ahmad Abareshi, “An Investigation of Customers' Loyalty to Social Commerce Websites,” in Proc. 2015 12th International Joint Conference on e-Business and Telecommunications (ICETE), Colmar, France, July. 20-22, 2015.
[29]	Lama Soboh, Islam Elkabani, and Ziad Osman, “Arabic Cultural Style Based Music Classification,” in Proc. 2017 International Conference on New Trends in Computing Sciences (ICTCS), Amman, Jordan, Oct. 11-13, 2017.
[30]	Keunwoo Choi, György Fazekas, Mark Sandler, and Kyunghyun Cho, “Convolutional Recurrent Neural Networks for Music Classification,” in Proc. 2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), New Orleans, LA, USA, Mar. 5-9, 2017.
[31]	Nirbhay Kashyap, Tanupriya Choudhury, Dev Kumar Chaudhary, and Roshan Lal, “Mood Based Classification of Music by Analyzing Lyrical Data Using Text Mining,” in Proc. 2016 International Conference on Micro-Electronics and Telecommunication Engineering (ICMETE), Ghaziabad, India, Sept. 22-23, 2016.
[32]	Peter Foster, Simon Dixon, and Anssi Klapuri, ”Identifying Cover Songs Using Information-Theoretic Measures of Similarity,” Journal of IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 23, issue. 6, pp. 993-1005, DOI: 10.1109/TASLP.2015.2416655, Mar. 25, 2015.