臺灣博碩士論文加值系統

因為社群媒體發展迅速，各式各樣的資訊如電影評論、商品評價皆在社群媒體中進行分享，因此在進行消費行為前查詢相關評論已經變成了一種習慣，評論內顯示的情緒極性更成為了影響消費者是否進行消費行為的重要因素。
因此本論文提出一套針對英文電影評論擷取出關鍵字詞的方法，採用NLTK和IMDB的英文電影評論作為數據集，由於NLTK英文電影評論已經標註好情緒極性能快速驗證實驗方法，而IMDB則能實驗在大量數據下的訓練結果本論文的方法是否有效，將電影評論利用縮寫還原去除掉會造成斷詞錯誤的縮寫字詞，斷詞後進行篩選評論內的字詞詞性保留與情緒相關的關鍵字詞，並透過WordNet將還原字詞至原始型態，並實驗了其餘兩種不同前處理方法分析訓練結果。
將關鍵字詞使用三種不同的模型分別是SUPPORT VECTOR MACHINE、Naive Bayes classifier、Logistic regression進行情感分析訓練，以及文本特徵提取，再透過網格搜尋調整參數顯示出訓練結果，NLTK數據集經由網格搜尋最佳訓練結果為Logistic regression模型，準確率為87.3%，而IMDB數據集則為SUPPORT VECTOR MACHINE模型有最佳訓練結果，準確率為91.9%。
使用機率性質預測來驗證在評論內，模型預測正負向標籤的電影評論結果，並以特徵遞迴消除利用準確率判斷出哪些字詞預測結果影響較大，顯示出模型內關鍵字詞以及字詞重要性程度，生成正負向關鍵字詞圖，可以讓使用者快速得知評論內的正負向前十個關鍵詞。

Because of the rapid development of social media, all kinds of information, such as movie reviews and product reviews, are shared. Therefore, checking relevant reviews before making consumption behaviors has become a habit. The emotional polarity displayed in the studies has become an essential factor affecting whether consumers conduct consumption behavior.
Therefore, this paper proposes a method for retrieving critical words for English movie reviews, using the English movie reviews of NLTK and IMDB as the data set. Since the NLTK English movie reviews have been marked with emotional poles, they can quickly verify the experimental method, while IMDB can soon verify it. The training results of the experiment under a large amount of data are the methods of this paper effective? Use abbreviations to restore movie reviews to remove abbreviated words that will cause hyphenation errors. After hyphenation, screen the comments in the reviews to retain keywords related to emotions. Words restore the words to the original form through WordNet and experiment with the other two preprocessing methods to analyze the training results.
Three different models are used for keywords, namely SUPPORT VECTOR MACHINE, Naive Bayes classifier, and Logistic regression for sentiment analysis training, as well as text feature extraction, and then adjust the parameters through grid search to display the training results. The NLTK dataset is searched through the grid. The best training result is the Logistic regression model with an accuracy rate of 87.3%. The IMDB dataset has the best training result for the SUPPORT VECTOR MACHINE model, with an accuracy rate of 91.9%.
Using recursive elimination of features, you can verify that the model predicts movie review results with positive and negative labels. Using this, you can determine which words significantly impact the prediction results, resulting in a list of key terms and phrases. Importance, generate a positive and negative keyword map, allowing users to quickly know the positive and negative top ten keywords in the comment.

中文摘要 i
英文摘要 ii
致謝 iv
目錄 v
表目錄 vii
圖目錄 viii
壹、緒論 1
一、研究背景 1
二、研究動機 1
三、研究目的 2
四、論文架構 3
貳、文獻探討 4
一、語意網路 4
二、情緒分析 5
三、機器學習 6
四、網格搜尋 7
參、研究方法 9
一、研究架構 9
二、數據處理流程 10
(一)縮寫字詞還原 10
(二)斷詞與詞性篩選 12
(三)詞形還原 14
三、實驗方法與工具 15
(一)機器學習環境和模型使用 15
(二)文本特徵提取 16
肆、模型訓練結果 17
一、原模型訓練結果 17
二、網格搜尋調整模型參數訓練結果 19
三、與僅斷詞處理模型訓練比較 20
四、機率性質預測 22
五、特徵遞迴消除 23
六、其餘實驗方法 25
(一)使用NRC情緒詞典代替詞性篩選 25
(二)使用WordNet尋找替換反義詞 27
七、IMDB英文電影評論數據集訓練結果 29
八、相關文獻比較 34
伍、結論與未來展望 35
參考文獻 36

[1] DIGITAL 2022: GLOBAL OVERVIEW REPORT (https://datareportal.com/reports/digital-2022-global-overview-report)
[2] 消費者相信誰？線上評論如何影響消費者線上/線下購買決策(https://www.smartm.com.tw/article/35343537cea3)
[3] Sowa, John F. "Semantic networks." Encyclopedia of Cognitive Science (2012).
[4] 語意網路(https://terms.naer.edu.tw/detail/1679016/)
[5] Radicioni, Tommaso, et al. "Analysing Twitter semantic networks: the case of 2018 Italian elections." Scientific Reports 11.1 (2021): 1-22.
[6] Huang, Chong, et al. "Keyphrase extraction using semantic networks structure analysis." Sixth International Conference on Data Mining (ICDM'06). IEEE, 2006.
[7] Mejova, Yelena. "Sentiment analysis: An overview." University of Iowa, Computer Science Department (2009).
[8] Tyagi, Priyanka, and R. C. Tripathi. "A review towards the sentiment analysis techniques for the analysis of twitter data." Proceedings of 2nd international conference on advanced computing and software engineering (ICACSE). 2019.
[9] Schweidel, David A., Wendy W. Moe, and Chris Boudreaux. "Social media intelligence: measuring brand sentiment from online conversations." Available at SSRN 1874892 (2011).
[10] Pastor, Cherish Kay. "Sentiment analysis of Filipinos and effects of extreme community quarantine due to coronavirus (COVID-19) Pandemic." Available at SSRN 3574385 (2020).
[11] Dubey, Akash Dutt. "Twitter sentiment analysis during COVID-19 outbreak." Available at SSRN 3572023 (2020).
[12] Medhat, Walaa, Ahmed Hassan, and Hoda Korashy. "Sentiment analysis algorithms and applications: A survey." Ain Shams engineering journal 5.4 (2014): 1093-1113.
[13] Gonçalves, Pollyanna, et al. "Comparing and combining sentiment analysis methods." Proceedings of the first ACM conference on Online social networks. 2013.
[14] Onovo, Amobi Andrew, et al. "Using supervised machine learning and empirical Bayesian kriging to reveal correlates and patterns of COVID-19 disease outbreak in sub-Saharan Africa: exploratory data analysis." medRxiv (2020).
[15] Denny, Matthew J., and Arthur Spirling. "Text preprocessing for unsupervised learning: Why it matters, when it misleads, and what to do about it." Political Analysis 26.2 (2018): 168-189.
[16] Palacio, Sebastián M. "Detecting Outliers with Semi-Supervised Machine Learning: A Fraud Prediction Application." XREAP WP 2 (2018).
[17] Jiang, Lanxin and Vasarhelyi, Miklos and Zhang, Chanyuan (Abigail), The Enigma of Unlabeled Positives - Using Semi-Supervised Learning to Detect and Predict Unlabeled Restatements (January 7, 2022). Available at SSRN: https://ssrn.com/abstract=4003621 or http://dx.doi.org/10.2139/ssrn.4003621
[18] Tyagi, Amit Kumar and G, Rekha, Machine Learning with Big Data (March 20, 2019). Proceedings of International Conference on Sustainable Computing in Science, Technology and Management (SUSCOM), Amity University Rajasthan, Jaipur - India, February 26-28, 2019, Available at SSRN: https://ssrn.com/abstract=3356269 or http://dx.doi.org/10.2139/ssrn.3356269
[19] Novaković, Jasmina Dj, et al. "Evaluation of classification models in machine learning." Theory and Applications of Mathematics & Computer Science 7.1 (2017): 39-46.
[20] Binary and Multiclass Classification in Machine Learning(https://www.analyticssteps.com/blogs/binary-and-multiclass-classification-machine-learning)
[21] Dreiseitl, Stephan, and Lucila Ohno-Machado. "Logistic regression and artificial neural network classification models: a methodology review." Journal of biomedical informatics 35.5-6 (2002): 352-359.
[22] Ahmad, Ghulab Nabi, Hira Fatima, and Abdelaziz Salah Saidi. "Efficient Medical Diagnosis of Human Heart Diseases using Machine Learning Techniques with and without GridSearchCV." IEEE Access (2022).
[23] Kartini, Dwi, Dodon Turianto Nugrahadi, and Andi Farmadi. "Hyperparameter tuning using GRIDSEARCHCV on the comparison of the activation function of the ELM method to the classification of pneumonia in toddlers." 2021 4th International Conference of Computer and Informatics Engineering (IC2IE). IEEE, 2021.
[24] Chauhan, Anamika. "Detection of lung cancer using machine learning techniques based on routine blood indices." 2020 IEEE international conference for innovation in technology (INOCON). IEEE, 2020.
[25] Tripathi, Sandesh, et al. "Analyzing sentiment using IMDb dataset." 2020 12th International Conference on Computational Intelligence and Communication Networks (CICN). IEEE, 2020.
[26] Mahyarani, Meta, et al. "Implementation of sentiment analysis movie review based on imdb with naive bayes using information gain on feature selection." 2021 3rd International Conference on Electronics Representation and Algorithm (ICERA). IEEE, 2021.