臺灣博碩士論文加值系統

情緒分析旨在分析一段自然語言文字中所隱含的情緒。為了找出文字中的情緒，許多情緒分析的研究仰賴情緒辭典查詢文字片段中隱含的情緒值，並據此總結出整段文字的情緒。在先前的研究中，我們提出了一個以 ConceptNet 為基礎的半監督式方法，將情緒值從已知情緒值的概念（concept）傳遞到其它的概念上，並建立一個具有情緒數值的概念層級情緒辭典。然而，由於已知情緒值的概念仍不足以在這個巨大的圖中傳遞情緒值。並且，收集大量情緒標註也相當耗費成本。在這個研究，我們藉由加入一個主動式學習元件來改良我們先前的方法。原有的數值傳遞方法被略微修改，以估計每個概念情緒值的不確定性分數。基於這些不確定性分數，我們提出「最不確定」與「最大影響」這兩個查詢策略（query strategy）以選擇需要情緒標註的概念。實驗結果顯示，我們提出的不確定性估算方式能夠合理地區分確定的概念與不確定的概念。並且，「最不確定」與「最大影響」兩者皆優於「隨機選取」策略。此外，「最大影響」能夠降低比「最不確定」策略更多的錯誤，原因在於其同時考慮了概念的不確定性與影響力。我們證實了我們提出的主動式學習元件確實能夠改進現有情緒辭典的品質。

Sentiment analysis aims to analyze the sentiments behind natural language text. Most sentiment analysis methods rely on sentiment dictionaries to identify sentiments in text. Our previous work proposed a ConceptNet-based semi-supervised approach, which propagated sentiment values from seed concepts to other concepts in ConceptNet. However, the seed concepts are insufficient to propagate sentiment values in a larger graph, and collecting large numbers of annotated seed concepts can be expensive. In this work, we refine our previous method by adding an active learning component. We also modify our previous value propagation method to estimate certainty score for each concept''s sentiment value. Based on these certainty scores, two query strategies, maximal uncertainty (MU) and maximal impact (MI), are proposed for choosing which concepts to send for sentiment annotation. Our experiment shows that our proposed certainty estimation methods can discriminate certain concepts from uncertain ones. Also, we show that both MU and MI strategies outperform the ``random'' strategy. Furthermore, MI corrects more concepts than MU, since it considers both uncertainty and influence of concepts. We conclude that our proposed active learning component can improve the quality of existing sentiment dictionaries.

致謝 iii
摘要 v
Abstract vii
1 Introduction 1
1.1 Background ................................. 1
1.2 Problem Definition ............................. 2
1.3 Thesis Structure ............................... 2
2 Literature Survey 5
2.1 Sentiment Dictionary Construction ..................... 5
2.2 Sentiment Annotation Collection via Crowdsourcing ........... 7
3 Methodology 11
3.1 RelationSelection .............................. 11
3.2 SentimentValuePropagation ........................ 14
3.3 Bias Correction ............................... 15
3.4 Certainty Estimation ............................ 16
3.5 Instance Selection .............................. 16
3.5.1 Maximal Uncertainty (MU) .................... 16
3.5.2 Maximal Impact (MI) ....................... 17
3.6 Label Correction .............................. 18
4 Evaluation 21
4.1 Experiment Settings ............................. 21
4.2 Reliability of Certainty Estimation ..................... 21
4.2.1 Dataset ............................... 22
4.2.2 Evaluation Metrics ......................... 23
4.2.3 Experiment Design and Results .................. 24
4.3 Effectiveness of Active Learning...................... 24
4.3.1 Data Collection and Analysis.................... 24
4.3.2 Evaluation Metrics ......................... 26
4.3.3 Experiment Design and Results .................. 27
5 Conclusion and Future Work 29
5.1 Conclusion ................................. 29
5.2 FutureWork................................. 30
Bibliography 31

[1] S. Baccianella, A. Esuli, and F. Sebastiani. SentiWordNet 3.0: An enhanced lexical resource for sentiment analysis and opinion mining. In Proceedings of the 7th Inter- national Conference on Language Resources and Evaluation (LREC ’10), Valletta, Malta, May 2010. European Language Resources Association (ELRA).
[2] J.Bernard,editor.TheMacquarieThesaurus.MacquarieLibrary,Sydney,Australia, 1986.
[3] M. M. Bradley and P. J. Lang. Affective norms for English words (ANEW) instruc- tion manual and affective ratings. Technical Report C-1, The Center for Research in Psychophysiology, 1999.
[4] E. Cambria, C. Havasi, and A. Hussain. SenticNet 2: A semantic and affective re- source for opinion mining and sentiment analysis. In Proceedings of the 25th Inter- national Florida Artificial Intelligence Research Society Conference (FLAIRS ’12). AAAI Press, 2012.
[5] E. Cambria, A. Hussain, C. Havasi, and C. Eckl. AffectiveSpace: Blending common sense and affective knowledge to perform emotive reasoning. In Proceedings of the 1st Workshop on Opinion Mining and Sentiment Analysis (WOMSA ’09), pages 32– 41, 2009.
[6] E.Cambria,A.Livingstone,andA.Hussain.Thehourglassofemotions.InCognitive Behavioural Systems, volume 7403 of Lecture Notes in Computer Science, pages 144–157. Springer Berlin Heidelberg, 2012.
[7] E. Cambria, D. Olsher, and D. Rajagopal. SenticNet 3: A common and common- sense knowledge base for cognition-driven sentiment analysis. In Proceedings of the 28th AAAI Conference on Artificial Intelligence (AAAI ’14), pages 1515–1521, 2014.
[8] E. Cambria, R. Speer, C. Havasi, and A. Hussain. SenticNet: A publicly available semantic resource for opinion mining. In Proceedings of the 2010 AAAI Fall Sym- posium Series on Commonsense Knowledge, 2010.
[9] J. L. Fleiss. Measuring nominal scale agreement among many raters. Psychological Bulletin, 76(5):378–382, 1971.
[10] C.Havasi,R.Speer,andJ.B.Alonso.ConceptNet3:Aflexible,multilingualseman- tic network for common sense knowledge. In Proceedings of the Recent Advances in Natural Language Processing, Borovets, Bulgaria, Sept. 2007.
[11] Y. Hong, H. Kwak, Y. Baek, and S. Moon. Tower of babel: A crowdsourcing game building sentiment lexicons for resource-scarce languages. In Proceedings of the 22nd International Conference on World Wide Web Companion (WWW ’13), pages 549–556, Republic and Canton of Geneva, Switzerland, 2013. International World Wide Web Conferences Steering Committee.
[12] J. Howe. Crowdsourcing: A definition. http://crowdsourcing.typepad.com/cs/ 2006/06/crowdsourcing_a.html, June 2006. Online; Accessed 2014-03-26.
[13] F. Keshtkar and D. Inkpen. Using sentiment orientation features for mood classifi- cation in blogs. In International Conference on Natural Language Processing and Knowledge Engineering (NLP-KE ’09), pages 1–6, 2009.
[14] R.J.LandisandG.G.Koch.Themeasurementofobserveragreementforcategorical data. Biometrics, 33(1):159–174, Mar. 1977.
[15] D. D. Lewis and W. A. Gale. A sequential algorithm for training text classifiers. In Proceedings of the 17th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR ’94), pages 3–12. Springer-Verlag, 1994.
[16] Y. Liu and J. Kender. Video feature selection using fast-converging sort-merge tree. In Proceedings of the 2004 IEEE International Conference on Multimedia and Expo (ICME ’04), volume 3, pages 2083–2086, June 2004.
[17] G. A. Miller. WordNet: A lexical database for English. Communications of the ACM, 38(11):39–41, Nov. 1995.
[18] S. M. Mohammad and P. D. Turney. Crowdsourcing a word–emotion association lexicon. Computational Intelligence, 29(3):436–465, 2013.
[19] A. Mudinas, D. Zhang, and M. Levene. Combining lexicon and learning based ap- proaches for concept-level sentiment analysis. In Proceedings of the 1st Interna- tional Workshop on Issues of Sentiment Discovery and Opinion Mining (WISDOM ’12), pages 5:1–5:8, New York, NY, USA, 2012. ACM.
[20] C.-C.MusatThisone,A.Ghasemi,andB.Faltings.Sentimentanalysisusinganovel human computation game. In Proceedings of the 3rd Workshop on the People’s Web Meets NLP: Collaboratively Constructed Semantic Resources and Their Applica- tions to NLP, pages 1–9, Stroudsburg, PA, USA, 2012. Association for Computa- tional Linguistics.
[21] K. Pearson. The problem of the random walk. Nature, 72(1865):294, 1905.
[22] R. Plutchik. A general psychoevolutionary theory of emotion. 1(3):3–33, 1980.
[23] K. R. Scherer. Psychological Models of Emotion. Oxford University Press, 2000.
[24] B. Settles. Active learning literature survey. Computer Sciences Technical Report 1648, University of Wisconsin–Madison, 2009.
[25] P. Singh, T. Lin, E. T. Mueller, G. Lim, T. Perkins, and W. L. Zhu. Open mind common sense: Knowledge acquisition from the general public. In On the Move to Meaningful Internet Systems 2002: CoopIS, DOA, and ODBASE, volume 2519 of Lecture Notes in Computer Science, pages 1223–1237. Springer Berlin Heidelberg, 2002.
[26] R. Snow, B. O’Connor, D. Jurafsky, and A. Y. Ng. Cheap and fast—but is it good? evaluating non-expert annotations for natural language tasks. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP ’08), pages 254–263, Stroudsburg, PA, USA, 2008. Association for Computational Lin- guistics (ACL).
[27] R. Speer and C. Havasi. Representing general relational knowledge in ConceptNet 5. In Proceedings of the 8th International Conference on Language Resources and Evaluation (LREC ’12), Istanbul, Turkey, May 2012. European Language Resources Association (ELRA).
[28] C. Strapparava and A. Valitutti. WordNet-Affect: An affective extension of Word- Net. In Proceedings of the 4th International Conference on Language Resources and Evaluation (LREC ’04), pages 1083–1086, Lisbon, Portugal, May 2004. European Language Resources Association (ELRA).
[29] M. Taboada, J. Brooke, M. Tofiloski, K. Voll, and M. Stede. Lexicon-based methods for sentiment analysis. Computational Linguistics, 37(2):267–307, June 2011.
[30] L. Talavera. An evaluation of filter and wrapper methods for feature selection in categorical clustering. In Advances in Intelligent Data Analysis VI, volume 3646 of Lecture Notes in Computer Science, pages 440–451. Springer Berlin Heidelberg, 2005.
[31] A. C.-R. Tsai, C.-E. Wu, R. T.-H. Tsai, and J. Y.-j. Hsu. Building a concept-level sentiment dictionary based on commonsense knowledge. IEEE Intelligent Systems, 28(2):22–30, Mar. 2013.
[32] A. Weichselbraun, S. Gindl, and A. Scharl. Using games with a purpose and boot- strapping to create domain-specific sentiment lexicons. In Proceedings of the 20th ACM International Conference on Information and Knowledge Management (CIKM ’11), pages 1053–1060, New York, NY, USA, 2011. ACM.
[33] C.-E.WuandR.T.-H.Tsai.Usingrelationselectiontoimprovevaluepropagationin a conceptnet-based sentiment dictionary. To Appear in Knowledge-Based Systems, 2014.
[34] L. Yu and H. Liu. Feature selection for high-dimensional data: a fast correlation- based filter solution. In Proceedings of the 20th International Conference on Ma- chine Learning (ICML ’03), volume 20, pages 856–863, 2003.