臺灣博碩士論文加值系統

在推薦系統中，研究顯示機率矩陣分解模型能產生不錯的推薦結果，但其缺點為在模型中只考慮了數值的評分，且在其他研究中，使用機率矩陣分解產生的推薦結果無法很好的被解釋。另一方面，因為我們能從評論中發現使用者的真正喜好商品特徵，評論應被視為推薦系統中的一大重要元素。在於釐清使用者喜好特性方面，文字情感分析相關的方法能有良好的表現，能從評論中萃取出商品特性與情感文字。然而，大多數的文字情感分析方法無法針對使用者喜好的商品特性給予先後或是優劣順序。因此，本論文結合了文字情感分析和機率矩陣分解，找出使用者真正的喜好特性。除此之外，本論文假設使用者在給予評分以及評論時，會有相似的行為，因為人們傾向表現出一定程度的一致性。在這樣的假設下，我們專注於處理評分與評論對於推薦系統的影響。在本論文中，透過結合機率矩陣分解與文字情感分析，學習使用者喜好的商品特性，實驗顯示本論文提出的模型，能產生高準確率的推薦結果，並且更進一步的了解機率矩陣分解模型的缺點，提供了一種解釋推薦結果的方式。

In recommender systems, probabilistic matrix factorization model has been examined and showed to give promising recommendation, but it only takes ratings into consideration and has long been criticized for its inability to provide explainable recommendation. On the other hand, because we can unearth users' preferred features from review text, review text should be viewed as an essential element in recommender systems. And when it comes to understanding users' preferred features, sentiment analysis methods report great performance on extracting product features and sentiment toward product attributes. However, most sentiment analysis methods are incapable of ranking user preferred features. Therefore, we incorporate sentiment analysis and probabilistic matrix factorization model to find user preferred features. In addition, we assume that when users give products or services ratings and review, they behave in similar fashions because human behavior tend to exhibit a certain level of consistency. Under such assumption, we focus on both ratings and reviews. In this study, we learn user preferred features by fusing probabilistic matrix factorization with sentiment analysis, and experiments show that our model is able attain high accuracy recommendation, and also take a step closer to solving the problem of matrix factorization technique, which is its inability to provide interpretable recommendation.

Chinese Abstract i
Abstract ii
Table of Contents iii
List of Tables iv
List of Figures v
List of Algorithms vi
Chapter 1 Introduction 1
1.1 Background and Motivation 1
1.2 Contributions 3
1.3 Research Limitations 3
1.4 Research Architecture 4
Chapter 2 Related Work 5
2.1 Matrix Factorization 5
2.2 Sentiment Analysis Approach 8
2.2.1 Feature and Opinion Extraction 8
2.2.2 Polarity Classification 13
2.3 Principal Component Analysis 13
2.4 Hybrid Methodology 14
Chapter 3 Research Methodology 15
3.1 Feature-Opinion Set Extraction and Assigning Opinion Polarity 18
3.2 Combining Probabilistic Matrix Factorization and Sentiment Analysis 18
Chapter 4 Experiments 22
4.1 Data Sets 22
4.2 Evaluation Metrics 22
4.3 Experiment Results 23
Chapter 5 Conclusions and Future Work 29
References 30

Baccianella, S., ＆ Sebastiani, A. E. F. (2010, May). SentiWordNet 3.0: An enhanced lexical resource for sentiment analysis and opinion mining. In LREC (Vol. 10, pp. 2200–2204).
Bell, R., ＆ Volinsky, C. (2009, August). Matrix factorization techniques for recommender systems. Computer(8), 30–37.
Diao, Q., Qui, M., Wu, C.-Y., Smola, A. J., Jiang, J., ＆ Wang, C. (2014, August). Jointly modeling aspects, ratings and sentiments for movie recommendation (JMARS). In Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining (pp. 193–202). ACM.
Hibbing, J. R., ＆ Alford, J. R. (2004, January). Accepting authoritative decisions: Humans as wary cooperators. American Journal of Political Science, 48(1), 62–76.
Liu, R. S., ＆ Yang, T. C. (2015). A recommender system based on collective matrix factorization with social trust and ewom ensemble (Master Thesis). NCKU.
McAuley, J., ＆ Leskovec, J. (2013, October). Hidden factors and hidden topics: understanding rating dimensions with review text. In Proceedings of the 7th ACM conference on Recommender systems (pp. 165–172). ACM.
Medhat, W., Hassan, A., ＆ Korashy, H. (2014). Sentiment analysis algorithms and applications: A survey. Ain Shams Engineering Journal, 5(4), 1093–1113.
Park, D. H., Kim, H. K., Choi, I. Y., ＆ Kim, J. K. (2012, February). A literature review and classification of recommender systems research. Expert Systems with Application, 39(11), 10059–10072.
Qiu, G., Liu, B., Bu, J., ＆ Chen, C. (2011). Opinion word expansion and target extraction through double propagation. Computational linguistics, 37(1), 9–27.
Qiu, M., Yang, L., ＆ Jiang, J. (2013, June). Mining user relations from online discussions using sentiment analysis and probabilistic matrix factorization. In Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies(NAACL’13) (pp. 401–410). Research Collection School Of Information Systems.
Rennie, J. D. M., ＆ Srebro, N. (2005, August). Fast maximum margin matrix factorization for collaborative prediction. In Proceedings of the 22nd international conference on Machine learning (pp. 713–719). ACM.
Salakhutdinov, R., ＆ Mnih, A. (2007, January). Probabilistic matrix factorization. In Advances in neural information processing systems (pp. 1257–1264).
Sarwar, B. M., Karypis, G., Konstan, J. A., ＆ Riedl, J. T. (2000a, October). Analysis of recommendation algorithms for e-commerce. In Proceedings of the 2nd ACM conference on Electronic commerce (pp. 158–167). ACM.
Sarwar, B. M., Karypis, G., Konstan, J. A., ＆ Riedl, J. T. (2000b). Application of demensionality reduction in recommender system – a case study (Tech. Rep. No. TR-00-043). Minneapolis, MN 55455: Minnesota Univ Minneapolis Dept of Computer Science.
Smith, L. I. (2002). A tutorial on principal components analysis. Retrieved from: http://www.cs.otago.ac.nz/cosc453/student_tutorials/principal_components.pdf.
Xu, Y., Chen, Z., Yin, J., Wu, Z., ＆ Yao, T. (2015, July). Learning to recommend with user generated content. In Web-Age Information Management (pp. 221–232). Springer International Publishing.
Zhang, Y., Lai, G., Zhang, M., Zhang, Y., Liu, Y., ＆ Ma, S. (2014, July). Explicit factor models for explainable recommendation based on phrase-level sentiment analysis. In Proceedings of the 37th international ACM SIGIR conference on Research ＆ development in information retrieval (pp. 83–92). ACM.