對於一些以內容為主的溝通工具例如部落格(Blogs)，情緒的辨識顯(Recognizing Emotion)的格外重要。在網路空間的企業部落格中，部落客所給予產品的負面內容或評價將會散播的非常迅速。而且，在部落格圈(Blogosphere)中揭露一些不為人知的企業內幕，將可能會影響到公司的商譽。而這些負面的內容通常會傷害到企業，甚至可能造成更大的損害。近幾年，有許多研究學者專注於情感分類(Sentiment Classification)的研究上。有效地辨識客戶的負面內容，將可協助企業小心地回應客戶的內容。本文目的主要有(1)發展一個新的語意導向指標，即“語意導向-結合PMI與LSA”(Semantic Orientation from Combining PMI with LSA；SO-CPL)，有效地改善傳統語意指標的分類效率。(2)結合機器學習(包含倒傳遞類神經網路、支援向量機與貝氏分類器)及資訊檢索技術的優點，以提出三個新穎的、以機器學習為主的語意導向指標。以協助企業迅速地、有效率地偵測具有傷害的負面部落格內容。實驗結果顯示，我們所提出的方法皆優於先前傳統的方法，包含倒傳遞類神經網路(Back-Propagation Neural Network；BPN)、支援向量機(support vector machines；SVM)、貝氏分類器(Navie-Bayes Classifier)及數個傳統語意導向指標。

Recognizing emotion is extremely important for some text-based communication tools such as blogs. On commercial blogs, bloggers’ negative comments or evaluations of products spread quickly in the cyber space. Moreover, an exposure of an inside story in the blogosphere may influence a company’s reputation. These negative comments are often harmful to enterprises and might result in great damage. Recently, researchers have paid much attention on sentiment classification to efficiently identify customers’ negative emotions for helping companies to carefully response customers’ comments. Following this trend, the main purposes of this study are (1) to develop new semantic orientation indexes, namely “Semantic Orientation from Combining PMI with LSA, (SO-CPL)” to improve the effectiveness of traditional indexes; (2) to propose three novel machine learning based semantic orientation indexes, which combines the advantages of machine learning techniques (including back-propagation neural network, support vector machines and Navie-Bayes classifier) and information retrieval (semantic orientation indexes), to help companies detecting harmfully negative bloggers’ comments quickly and effectively. Experimental results indicated that our proposed methodologies outperform traditional approaches, including back-propagation neural network, support vector machines, Navie-Bayes classifier and several traditional semantic orientation indexes.

中文摘要..............................................I
英文摘要.............................................II
目錄................................................III
圖目錄...............................................VI
表目錄................................................X
第一章 緒論...........................................1
1.1研究背景...........................................1
1.2研究動機...........................................4
1.3研究目的...........................................7
1.4研究架構...........................................8
第二章 文獻探討......................................10
2.1語意導向相關研究..................................10
2.2非監督學習相關方法................................13
2.2.1關連式語意導向..................................13
2.2.2語意導向-PMI....................................14
2.2.3語意導向-潛在語意分析...........................15
2.3機器學習相關方法..................................19
2.3.1倒傳遞類神經網路................................19
2.3.2支援向量機......................................22
2.3.3貝氏分類器......................................26
第三章 研究方法......................................28
3.1實驗架構與實驗模組................................28
3.2 SO-CPL語意導向指標...............................30
3.3基於機器學習為主之語意導向指標....................33
3.3.1基於倒傳遞類神經網路之語意導向指標..............36
3.3.2基於支援向量機之語意導向指標....................39
3.3.3基於貝氏分類器之語意導向指標....................41
第四章 實驗結果與分析................................44
4.1實驗文集..........................................44
4.2資料預先處理......................................46
4.3實驗環境..........................................49
4.4實驗結果..........................................50
4.4.1非監督語意導向指標..............................50
4.4.2基於類神經網路之語意導向指標....................61
4.4.3基於支援向量機之語意導向指標....................75
4.4.4基於貝氏分類器之語意導向指標....................87
第五章 結論與未來研究方向...........................100
5.1討論.............................................100
5.2結論.............................................104
參考文獻............................................106
附錄................................................109

表目錄
表 2- 1語意導向與情緒辨識相關研究整理...................11
表 2- 2前饋式與回饋式類神經網路相關種類.................19
表2 - 3常見核函數之種類.................................26
表 3 - 1 SO-CPL MATRIX..................................32
表 3 - 2基於機器學習為主之語意導向指標使用資料集整理表..35
表 4 - 1實驗文集與模組..................................41
表 4 - 2詞彙-關鍵字矩陣之範例...........................46
表 4 - 3 PART-OF-SPEECH 之規............................49
表 4 - 4 非監督語意導向指標分類成效整理(一).............50
表 4 - 5 非監督語意導向指標分類成效整理(二).............51
表 4 - 6 實驗一：傳統SO與SO-CPL實驗結果比較.............53
表 4 - 7實驗二：傳統SO與SO-CPL實驗結果比較..............55
表 4 - 8實驗三：傳統SO與SO-CPL實驗結果之比較............56
表 4 - 9 實驗四：傳統SO與SO-CPL實驗結果之比較...........58
表 4 - 10 實驗五：傳統SO與SO-CPL實驗結果之比較..........59
表 4 - 11 實驗六：傳統SO與SO-CPL實驗之結果..............61
表 4 - 12 基於類神經網路之語意導向指標分類成效整理......62
表 4 - 13 傳統TDM與基於NN為主之語意導向指標之平均處理時間......................................................63
表 4 - 14 實驗一：傳統TDM與基於NN為主之語意導向指標實驗結果......................................................65
表 4 - 15 實驗二：傳統TDM與基於NN為主之語意導向指標實驗結果......................................................67
表 4 - 16 實驗三：傳統TDM與基於NN為主之語意導向指標實驗結果......................................................69
表 4 - 17 實驗四：傳統TDM與基於NN為主之語意導向指標實驗結果......................................................71
表 4 - 18 實驗五：傳統TDM與基於NN為主之語意導向指標實驗結果......................................................72
表 4 - 19 實驗六：傳統TDM與基於NN為主之語意導向指標實驗結果......................................................74
表 4 - 20 基於支援向量機之語意導向指標分類成效整理......................................................75
表 4 - 21 傳統TDM與基於SVM為主之語意導向指標之平均處理時間......................................................76
表 4 - 22 實驗一：傳統TDM與基於SVM為主之語意導向指標實驗結果......................................................78
表 4 - 23 實驗二：傳統TDM與基於SVM為主之語意導向指標實驗結果......................................................80
表 4 - 24 實驗三：傳統TDM與基於SVM為主之語意導向指標實驗結果......................................................81
表 4 - 25 實驗四：傳統TDM與基於SVM為主之語意導向指標實驗結果......................................................83
表 4 - 26 實驗五：傳統TDM與基於SVM為主之語意導向指標實驗結果......................................................85
表 4 - 27 實驗六：傳統TDM與基於SVM為主之語意導向指標實驗結果......................................................86
表 4 - 28 基於貝氏分類器之語意導向指標分類成效整理......................................................87
表 4 - 29 傳統TDM與基於SVM為主之語意導向指標之平均處理時間......................................................88
表 4 - 30 實驗一：傳統TDM與基於NB為主之語意導向指標實驗結果......................................................90
表 4 - 31實驗二：傳統TDM與基於NB為主之語意導向指標實驗結果......................................................92
表 4 - 32 實驗三：傳統TDM與基於NB為主之語意導向指標實驗結果......................................................93
表 4 - 33 實驗四：傳統TDM與基於NB為主之語意導向指標實驗結果......................................................95
表 4 - 34 實驗五：傳統TDM與基於NB為主之語意導向指標實驗結果......................................................97
表 4 - 35 實驗六：傳統TDM與基於NB為主之語意導向指標實驗結果......................................................98
表 5 - 1 最重要節點之估算結果.....................................................103

圖目錄
圖 1- 1研究架構..........................................9
圖 2 - 1奇異值分解（DEERWESTER ET AL., 1990）...........18
圖 2 - 2倒傳遞類神經網路架構圖..........................20
圖 2 - 3支援向量機區分超平面示意圖......................23
圖 2 - 4原始資料映射到特徵空間示意圖....................25
圖 3 - 1實驗流程........................................29
圖 3 - 2 SO-LSA與SO-CPL步驟流程.........................31
圖 3 - 3 字彙頻率矩陣圖.................................32
圖 3 - 4 字彙連續相乘頻率矩陣...........................32
圖 3 - 5 PMI值矩陣......................................32
圖 3 - 6傳統NN分類與基於NN為主之語意導向指標差異........36
圖 3 - 7傳統NN分類與基於NN為主之語意導向指標差異........37
圖 3 - 8基於NN為主之語意導向指標流程....................38
圖 3 - 9基於支援向量機之語意導向指標....................40
圖 3 - 10基於貝氏分類器之語意導向指標...................42
圖 4 - 1電影評論內容範例................................45
圖 4 - 2部落客撰寫之內容範例............................46
圖 4 - 3 EC及MP3之評論撰寫範例..........................46
圖 4 - 4 RUBRYX2.0 軟體計算詞頻之功能...................48
圖 4 -5 實驗一：傳統SO與SO-CPL實驗結果之比較............52
圖 4 -6 實驗二：傳統SO與SO-CPL實驗結果之比較............54
圖 4 -7 實驗三：傳統SO與SO-CPL實驗結果之比較............56
圖 4 -8 實驗四：傳統SO與SO-CPL實驗結果之比較............57
圖 4 - 9實驗五：傳統SO與SO-CPL實驗結果之比較............59
圖 4 - 10 實驗六：傳統SO與SO-CPL實驗結果之比較..........60
圖 4 - 11 實驗一：傳統TDM與基於NN為主之語意導向指標實驗結果......................................................64
圖 4 - 12 實驗二：傳統TDM與基於NN為主之語意導向指標實驗結果......................................................66
圖 4 - 13 實驗三：傳統TDM與基於NN為主之語意導向指標實驗結果......................................................68
圖 4 - 14 實驗四：傳統TDM與基於NN為主之語意導向指標實驗結果......................................................70
圖 4 - 15 實驗五：傳統TDM與基於NN為主之語意導向指標實驗結果......................................................72
圖 4 - 16 實驗六：傳統TDM與基於NN為主之語意導向指標實驗結果......................................................73
圖 4 - 17 實驗一：傳統TDM與基於SVM為主之語意導向指標實驗結果......................................................77
圖 4 - 18 實驗二：傳統TDM與基於SVM為主之語意導向指標實驗結果......................................................79
圖 4 - 19 實驗三：傳統TDM與基於SVM為主之語意導向指標實驗結果......................................................81
圖 4 - 20 實驗四：傳統TDM與基於SVM為主之語意導向指標實驗結果......................................................82
圖 4 - 21 實驗五：傳統TDM與基於SVM為主之語意導向指標實驗結果......................................................84
圖 4 - 22 實驗六：傳統TDM與基於SVM為主之語意導向指標實驗結果......................................................85
圖 4 - 23 實驗一：傳統TDM與基於NB為主之語意導向指標實驗結果......................................................89
圖 4 - 24 實驗二：傳統TDM與基於NB為主之語意導向指標實驗結果......................................................91
圖 4 - 25 實驗三：傳統TDM與基於NB為主之語意導向指標實驗結果......................................................93
圖 4 - 26 實驗四：傳統TDM與基於NB為主之語意導向指標實驗結果......................................................94
圖 4 - 27 實驗五：傳統TDM與基於NB為主之語意導向指標實驗結果......................................................96
圖 4 - 28 實驗六：傳統TDM與基於NB為主之語意導向指標實驗結果......................................................97

[1]A. Abbasi, H. Chen, S. Thoms, T. Fu (2008), “Affect analysis of web forums and blogs using correlation ensembles,” IEEE Transactions on Knowledge and Data Engineering, vol. 20, issue. 9, pp.1168-1180.
[2]A. J. Smola(1998), “Learning with kernels” PhD Thesis, University of Birlinghoven Birlinghoven.
[3]B. Pang, L. Lee and S. Vaithyanathan (2002), “Thumbs up?: Sentiment classification using machine learning techniques,” Annual Meeting of the ACL Proceedings of the ACL-02 conference on Empirical methods in natural language processing, vol. 10, pp.79-86 .
[4]B. Pang, L. Lee, and S. Vaithyanathan (2002), “Thumbs up? Sentiment classification using machine learning techniques, ” presented at the 2002 Conference on Empirical Methods in Natural Language Processing (EMNLP''2002), pp.79-86.
[5]C. C. Chang and C. J. Lin (2001), LIBSVM A Library for Support Vector Machines, Software available at http://www.csie.ntu.edu.tw~cjlinlibsvm.
[6]C. Cortes, and V. Vapnik (1995), “Support-vector networks,” Machine Learning, vol. 20, pp. 273-297.
[7]C. H. Wu, Z. J. Chung and Y. C. LIN (2006), “Emotion recognition from text using semantic labels and separable mixture models.” TTACM Transactions on Asian Language Information Processing (TALIP), vol. 5 , issue. 2, pp. T165-183.
[8]C. Wang, J. Lu and G. Zhang (2005), “A semantic classification approach for online product reviews, ” IEEE/WIC/ACM International Conference on Web Intelligence (WI''05), pp.276-279.
[9]C. Yang, K. H. Y. Lin and H. H. Chen (2007), “Emotion classification using web blog corpora,” Proceedings of the IEEE/WIC/ACM International Conference on Web Intelligence, pp. T275-278.
[10]C.W. Hsu, C.C. Chang and C.J. Lin (2003), “A practical guide to support vector classification,” Technical Report, National Taiwan University.
[11]D. Winer(2002), The history of weblogs. http://newhome.weblogs.comhistory of Weblogs.
[12]G. Leshed and J. ‘Jofish’ Kaye (2006), “Understanding how bloggers feel: Recognizing affect in blog posts,” Conference on Human Factors in Computing Systems CHI ''06 extended abstracts on Human factors in computing systems, pp. 1019-1024.
[13]G. Mishne(2005), “Experiments with Mood Classification in Blog Posts,” In Style2005-1st Workshop on Stylistic Analysis of Text for Information Access, at SIGIR.
[14]J. C. Na, C. Khoo and P. H. J. Wu(2005), “Use of negation phrases in automatic sentiment classification of product reviews,” Library Collections, Acquisitions, & Technical Services, Vol. 29, pp. 180-191.
[15]J. Read (2004), “Recognizing affect in text using point-wise mutual information,” Masters Thesis, University of Sussex.
[16]J. S. Chen, R. Ching, H. T. Tsai and Y. J.Kuo1 (2008), “Blog effects on brand attitude and purchase intention,” Service Systems and Service Management, 2008 International Conference on, pp. 1-6.
[17]L. Zhuang, F. Jing and X. Y. Zhu (2006), “Movie review mining and summarization,” Conference on Information and Knowledge Management Proceedings of the 15th ACM international conference on Information and knowledge management, pp. 43-50.
[18]P. D. Turney and M. L. Littman (2003), "Measuring praise and criticism: inference of semantic orientation from association," ACM Transactions on Information Systems, vol. 21, pp. 315-346.
[19]P. Subasic, and A. Huettner (2001), “Affect analysis of text using fuzzy semantic typing,” IEEE Transactions on Fuzzy Systems, volume: 9, issue. 4, pp. 483-496.
[20]P.D. Turney (2002), “Thumbs up or thumbs down? Semantic orientation applied to unsupervised classification of reviews,” Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics (ACL''02). Philadelphia, Pennsylvania, USA, pp.417-424.
[21]Q. Ye, W. Shi and Y. Li(2006), “Sentiment Classification for Movie Reviews in Chinese by Improved Semantic Oriented Approach,” Proceedings of the 39th Hawaii International Conference on System Sciences, Vol. 3, pp. 53b-53b.
[22]R.M. Tong (2001), “An operational system for detecting and tracking opinions in on-line discussions,” Working Notes of the ACM SIGIR 2001 Workshop on Operational Text Classification, pp. 1-6.
[23]S. C. Herring, L. A. Scheidt, S. Bonus and E. Wright (2004), “Bridging the gap: A genre analysis of weblogs,” System Sciences, Proceedings of the 37th Annual Hawaii International Conference on.
[24]S. Deerwester, S. T. Dumais, T. K. Landauer, G. W. Furnas and R. A. Harshman(1990), “Indexing by latent semantic analysis,” Journal of the Society for Information Science, vol. 41, pp. 391-407.
[25]S. R. Gumn (1998), “Support vector machines for classification and regression,” Technical Report, University of Southampton.
[26]T. Singh, L.Veron-Jackson, J. Cullinane (2008), “Blogging: A new play in your marketing game plan,” Business Horizons 51, pp.281-292.
[27]T.K. Landaure and S.T. Dumais (1997), “A solution to plato’s problem: The latent semantic analysis theory of the acquisition, induction, and representation of knowledge,” Psychological Review, vol. 104, pp. 211-240.
[28]V. Hatzivassiloglou and K. R. McKeown (1997), "Predicting the semantic orientation of adjectives," presented at Proceedings of the 35th Annual Meeting of the Association for Computational Linguistics and the 8th Conference of the European Chapter of the ACL, pp.174-181.
[29]Y. Chen, F.S. Tsai and K.L. Chan (2008), “Machine learning techniques for business blog search and mining,” Expert Systems with Applications: An International Journal, Vol. 35, pp. 581-590.
[30]Z. Fei, J. Liu and G. Wu (2004), “Sentiment classification using phrase patterns,” Computer and Information Technology The Fourth International Conference on, pp. 1147-1152.
[31]Z. Teng, F. Ren and S. Kuroiwa (2007), “Emotion recognition from text based on the rough set theory and the support vector machines,” Natural Language Processing and Knowledge Engineering International Conference on, pp. 36-41.
[32]D. Sifry (2007), The state of the live web, from http://www.sifry.com/alerts/archives/000493.html.
[33]張儀興、黃秀怡(2008)，“基於貝氏理論及本體論之自動化之文件分類系統”， 第三屆數位教學暨資訊實務研討會論文集，台南。
[34]王景南(2003)，多 類 支 向 機 之 研 究 ，碩士論文，元 智 大 學資訊管理學系碩士班，桃園。
[35]洪淑芬(2006)，潛在語意索引在生醫文件分類之應用，碩士論文，樹德科技大學資訊管理學系碩士班，高雄。
[36]張俊雄(2008)，以倒傳遞網路預測環孢靈在腎臟移植後患者之用量適當性，碩士論文，大同大學資訊工程研究所，台北。
[37]彭華瑞(2002)，應用潛在式語意分析於語言模型之研究，碩士論文，成功大學資訊工程學系，台南。
[38]黃啟信(2003)，應用資料修補技術於醫療環境之研究，碩士論文，東海大學工業工程與經營資訊所，台中。
[39]洪淑芬(2006)，潛在語意索引在生醫文件分類之應用，碩士論文，樹德科技大學資訊管理系碩士班。
[40]顏彰慶(2007)，利用空間分析研究纖毛蟲在不同餵食條件下的行為特徵，碩士論文，中山大學海洋環境及工程研究碩士班。