本研究提出一個創新的方法，將所有產品皆表達成為元素組合的服務套裝形式，並根據當時的社群網站評論意見，進行市場喜好分析，預測次季熱門的服務套裝，推薦給企業行銷部門做為調整產品內容的參考方向。透過比較上市產品的實際銷售紀錄以及模型預測的銷售結果，兩年間共八季的實驗數據顯示，本研究的分析方法產生的推薦排序相關性超越傳統銷售預測的效能，因而能夠做為服務套裝產品連結輿情分析的解決方案。

This study proposes an innovative method, which presents all the products in packages of multiple elements and conducts market preference analysis based on the opinions in social media at the time collected by Web Crawler, to forecast popular packages for advising marketing departments on the adjustment of product content. An experiment is made to compare the actual sales records by a large enterprise during two years with the sales data predicted by the trained model that can forecast the next quarter according to the relevant online opinions about the previous quarter. With the comparison results of eight quarters during the two years, this analysis method proves to surpass the efficiency of traditional sales forecasting approaches, and therefore can be the solutions of public opinion analysis for service packages.

誌謝 i
中文摘要 ii
Abstract iii
Table of Contents iv
List of Figures vi
List of Tables vii

Chapter 1 Introduction 1
1.1 Motivation 1
1.2 Explanation of Terms 2
1.2.1 Market Preference 2
1.2.2 Service Packages 2
1.3 Problem Statement 3
1.4 Contributions 4
1.5 Thesis Organization 4

Chapter 2 Related Work 6
2.1 Recommendation 6
2.2 Chinese Sentiment Analysis 8

Chapter 3 Datasets 10
3.1 Sales Data of Service Packages 10
3.1.1 Data Exploration 10
3.1.2 Data Preprocessing 14
3.2 The Public Opinion Corpus 14
3.2.1 Data Exploration 14
3.2.2 Data Preprocessing 16
3.2.2.1 Objectivity Detection 17
3.2.2.2 Abnormal User Detection 20
3.2.2.3 Vocabulary Construction 22
3.2.2.4 Coreference Resolution 24

Chapter 4 Feature Extraction for the Proposed Model 27
4.1 Document Level 27
4.2 Sentence Level 28
4.3 Entity Level 29
4.4 Structure of the Proposed Model 31

Chapter 5 Experiments 33
5.1 Experimental Setup 33
5.2 Evaluation Metrics 33
5.3 Baseline Models for Comparison 35
5.3.1 Trained by Sales Data 35
5.3.2 Trained by Sentiment Scores from SnowNLP Toolkit 36
5.3.3 Trained by Sales Data and Sentiment Scores 38
5.4 Results and Analysis 39
5.4.1 NDCG Evaluation Results 39
5.4.2 Performance Comparison of Baseline Models 40
5.4.3 Performance Comparison of the Proposed Model 42
5.4.4 Evaluation Comparison of NDCG@50 and NDCG@100 44

Chapter 6 Conclusion and Future Work 46
6.1 Conclusion 46
6.2 Future Work 48
References 49

[1] Asghar, M. Z., Khan, A., Ahmad, S., and Kundi, F. M., “A Review of Feature Extraction in Sentiment Analysis,” Journal of Basic and Applied Scientific Research, vol. 4, no. 3, pp. 181-186, 2014.
[2] Bai, T., Nie, J. Y., Zhao, W. X., Zhu, Y. T., Du, P., and Wen, J. R., “An Attribute-aware Neural Attentive Model for Next Basket Recommendation” in Proceedings of the 41st International ACM SIGIR Conference on Research and Development in Information Retrieval. SIGIR ‘18, 2018, pp. 1201-1204.
[3] Bhattacharjee, S., Das, A., Bhattacharjee, U., Parui, S. K., and Roy, S., “Sentiment Analysis using Cosine Similarity Measure” in Proceedings of the 2nd IEEE International Conference on Recent Trends in Information Systems. ReTIS ’15, 2015, pp. 27-32.
[4] Chen, H. C., Li, F., Zhu, X. H., and Ma, R. C., “A Path and Depth-Based Approach to Word Semantic Similarity Calcalation in Cilin,” Journal of Chinese Information Processing, vol. 30, no. 5, pp. 80-88, 2016.
[5] Ku, L. W., Liang, Y. T., and Chen, H. H., “Opinion Extraction, Summarization and Tracking in News and Blog Corpora” in Proceedings of the AAAI 2006 Spring Symposium on Computational Approaches to Analyzing Weblogs. AAAI-CAAW ’06, 2006, pp. 100-107.
[6] Lu, J., Venugopal, D., Gogate, V., and Ng, V., “Joint Inference for Event Coreference Resolution” in Proceedings of the 26th International Conference on Computational Linguistics. COLING ’16, 2016, pp. 3264-3275.
[7] Mukherjee, A., Kumar, A., Liu, B., Wang, J., Hsu, M., Castellanos, M., and Ghosh, R., “Spotting Opinion Spammers using Behavioral Footprints” in Proceedings of the 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. KDD ’13, 2013, pp. 632-640.
[8] Oliveira, N., Paulo, C., and Areal, N., “Stock Market Sentiment Lexicon Acquisition using Microblogging Data and Statistical Measures,” Decision Support Systems, vol. 85, pp. 62-73, 2016.
[9] Peng, H. Y., Cambria, E., and Hussain, A., ‘‘A Review of Sentiment Analysis Research in Chinese Language,’’ Cognitive Computation, vol. 9, no. 4, pp. 423-435, 2017.
[10] Peng, H. Y., Ma, Y. K., Li, Y., and Cambria, E., “Learning Multi-grained Aspect Target Sequence for Chinese Sentiment Analysis,” Knowledge-Based Systems, vol. 148, pp. 167-176, 2018.
[11] Peng, Q., Zhu, X. H., Chen, Y. S., Sun, L., and Li, F., “IC-based Approach Calculating Word Semantic Similarity in Cilin,” Application Research of Computers, vol. 35, no. 2, pp. 400-404, 2018.
[12] Wang, Y. N., Wang, L. W., Li, Y. Z., He, D., Liu, T. Y., and Chen, W., “A Theoretical Analysis of NDCG Ranking Measures” in Proceedings of the 26th Annual Conference on Learning Theory. COLT ‘13, 2013, pp. 1-30.
[13] Wang, Y. X., Zhiguo, H., and Shi, M. Y., "Research on LDA Model Algorithm of News-oriented Web Crawler" in Proceedings of the 17th IEEE International Conference on Computer and Information Science. ICIS ’18, 2018, pp. 748-753.
[14] Zhang, D., Xu, H., Su, Z., and Xu, Y. F., “Chinese Comments Sentiment Classification Based on Word2vec and SVMperf,” Expert Systems with Applications, vol. 42, no. 4, pp. 1857-1863, 2015.
[15] Zhao, M., Zhang, T. Z., and Chai, J. P., “Based on SO-PMI Algorithm to Discriminate Sentimental Words'' Polarity in TV Programs’ Subjective Evaluation” in Proceedings of the 8th IEEE International Symposium on Computational Intelligence and Design. ISCID ’15, 2015. pp. 38-40.
[16] Zheng, L. J., Wang, H. W., and Gao, S., “Sentimental Feature Selection for Sentiment Analysis of Chinese Online Reviews,” International Journal of Machine Learning and Cybernetics, vol.9, no.1, pp. 75-84, 2018.
[17] Zhu, X. H., Ma, R. C., Sun, L., and Chen, H. C., “Word Semantic Similarity Computation Based on Hownet and Cilin,” Journal of Chinese Information Processing, vol. 30, no. 4, pp. 29-36, 2016.