臺灣博碩士論文加值系統

COVID-19疫情對全球人民的生活產生了深遠影響，改變了我們的工作方式、學習方式，甚至是我們與他人的互動方式。在這種情況下，理解公眾的情感反應變得非常重要。本研究主要是使用深度學習技術來分析COVID-19期間公眾的情感變化。
本研究使用了基於轉換器的雙向編碼器表示技術(Bidirectional Encoder Representations from Transformer, BERT)和雙向長短期記憶(Bidirectional Long Short-Term Memory, BiLSTM)兩種深度學習模型來進行情感分析。BERT是一種預訓練的深度學習模型，專門用於自然語言處理任務，而BiLSTM則是一種特殊的遞迴神經網路，專門用於處理序列資料，如文章內容或社群媒體的貼文內容。
本研究的目標是使用這些模型來評估COVID-19各階段的推文資料，並進行情感分析。希望通過這種方式，能夠更好的理解公眾在疫情期間的情感變化。本研究也在BERT模型的字典中增加了與COVID-19相關的單字，以提高模型的準確率和泛化能力。透過這種方式可以更準確的捕捉到公眾對COVID-19的情感反應，且提供更準確的情感分析結果，並使用VADER技術的情感分析結果進行標籤並與本研究的模型分類結果進行比較，透過這種比較，我們可以更深入的理解我們的模型在情感分析上的性能。
從三種不同階段資料集的實驗結果來看，BERT模型在情感分析上已經有非常出色的表現。但是如果BiLSTM模型沒有配合適當的向量轉換工具或模型，其表現會變得非常不準確。實驗結果也證實了這一點，BiLSTM在第一個階段資料集配合舊的向量轉換工具的各項準確率、精確率、召回率與F1 Score的指標分別為76.71%、62.19%、64.58%與63.36%，如果配上新的向量轉換工具BERT，其各項指標分別提升至90.61%、84.66%、84.72%與84.69%，在每個階段的資料集組合模型都比單一模型表現還要優秀，表示組合模型是能夠提升其各項指標的。

COVID-19 has had a profound impact on people’s lives worldwide, changing the way we work, learn, and interact with others. In this situation, understanding the public’s sentiment response has become very important. This research mainly uses deep learning technology to analyze the public’s sentiment changes during COVID-19.
This research uses two deep learning models, BERT and BiLSTM, for sentiment analysis. BERT is a pre-trained deep learning model specifically designed for natural language processing tasks, while BiLSTM is a special type of recurrent neural network designed to handle sequence data such as article content or social media posts.
The goal of this research is to use these models to evaluate tweet data from different stages of COVID-19 and perform sentiment analysis. It is hoped that this approach can better understand the public’s sentiment changes during the pandemic. This research also added COVID-19-related words to the dictionary of the BERT model to improve its accuracy and generalization ability. Through this approach, the public’s sentiment reactions to COVID-19 can be more accurately captured, and more accurate sentiment analysis results can be provided. The sentiment analysis results using VADER technology are labeled and compared with the model classification results of this research. Through this comparison, we can gain a deeper understanding of the performance of our model in sentiment analysis.
From the experimental results of the three different stage datasets, the BERT model has performed very well in sentiment analysis. However, if the BiLSTM model is not paired with an appropriate vector conversion tool or model, its performance will become very inaccurate. The experimental results also confirmed this point. The indicators of accuracy, precision, recall, and F1 score of BiLSTM in the first stage dataset with the old vector conversion tool were 76.71%, 62.19%, 64.58%, and 63.36%, respectively. If paired with the new vector conversion tool BERT, the indicators improved to 90.61%, 84.66%, 84.72%, and 84.69%, respectively. The combined model of each stage dataset performed better than a single model, indicating that the combined model can improve its various indicators.

摘要......i
Abstract......ii
誌謝......iii
目錄......iv
表目錄......vi
圖目錄......vii
第一章 緒論......1
1.1 研究背景與動機......1
1.2 研究目的......1
1.3 論文架構......2
第二章 文獻回顧......3
2.1 情感分析......3
2.2 資料前處理......4
2.3 深度學習方法......5
2.3.1 循環神經網路(RNN)......5
2.3.2 長短期記憶網路(LSTM)......6
2.3.3 雙向長短期記憶網路(BiLSTM)......10
2.4 基於轉換器的雙向編碼器表示技術(BERT)......10
2.4.1 預訓練(Pre-Training) BERT......12
2.4.2 微調(Fine-Tuning) BERT......12
2.5 混淆矩陣(Confusion Matrix)......14
第三章 研究設計與方法......17
3.1 研究架構......17
3.2 模型實作流程圖......18
3.3 資料收集與前處理......19
3.3.1 資料收集......19
3.3.2 資料前處理......20
3.4 建構模型......22
3.4.1 BERT模型......22
3.4.2 BERT’模型......22
3.4.3 BiLSTM模型......24
3.4.4 BERT-BiLSTM組合模型......25
3.4.5 BERT’-BiLSTM組合模型......26
3.5 情感預測與結果比較......26
3.5.1 情感詞典與推理的價感知詞典(VADER)......26
3.5.2 模型結果比較......28
第四章 實驗結果與分析......29
4.1 實驗環境......29
4.2 實驗結果......29
4.3 模型評估......34
第五章 結論與建議......37
5.1 結論......37
5.2 建議與未來方向......37
參考文獻......38
Extended Abstract......40

[1]H. Jelodar, Y. Wang, R. Orji, and S. Huang, "Deep Sentiment Classification and Topic Discovery on Novel Coronavirus or COVID-19 Online Discussions: NLP Using LSTM Recurrent Neural Network Approach," in IEEE Journal of Biomedical and Health Informatics, vol. 24, no. 10, pp. 2733-2742, Oct. 2020, doi: 10.1109/JBHI.2020.3001216.
[2]S. Poria, D. Hazarika, N. Majumder, and R. Mihalcea, "Beneath the Tip of the Iceberg: Current Challenges and New Directions in Sentiment Analysis Research," in IEEE Transactions on Affective Computing, vol. 14, no. 1, pp. 108-132, 1 Jan.-March 2023, doi: 10.1109/TAFFC.2020.3038167.
[3]HT. Duong, TA. Nguyen-Thi, A review: preprocessing techniques and data augmentation for sentiment analysis. Comput Soc Netw 8, 1 (2021), doi: 10.1186/s40649-020-00080-x.
[4]G. Xu, Y. Meng, X. Qiu, Z. Yu, and X. Wu, "Sentiment Analysis of Comment Texts Based on BiLSTM," in IEEE Access, vol. 7, pp. 51522-51532, 2019, doi: 10.1109/ACCESS.2019.2909919.
[5]C. Olah, Understanding LSTM Networks, colah’s blog, August 27, 2015. Retrieved from https://colah.github.io/posts/2015-08-Understanding-LSTMs/.
[6]S. Gupta, U. Vadgama, and T. R. Vedhavathy, "Identification and Labeling of Textual Cyberbullying using BiLSTM and BERT," 2023 International Conference on Networking and Communications (ICNWC), Chennai, India, 2023, pp. 1-5, doi: 10.1109/ICNWC57852.2023.10127308.
[7]J. Devlin, MW. Chang, K. Lee, and K. Toutanova, “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding,” in Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, vol. 1 (Long and Short Papers) pp. 4171-4186, 2019.
[8]Y. Zhu, R.Kiros, R.Zemel, R.Salakhutdinov, R.Urtasun, A.Torralba, and S.Fidler, "Aligning Books and Movies: Towards Story-Like Visual Explanations by Watching Movies and Reading Books," 2015 IEEE International Conference on Computer Vision (ICCV), Santiago, Chile, 2015, pp. 19-27, doi: 10.1109/ICCV.2015.11.
[9]C.J. Hutto and E. Gilbert, “VADER: A Parsimonious Rule-based Model for Sentiment Analysis of Social Media Text,” Proceedings of the 8th International Conference on Weblogs and Social Media, ICWSM 2014.
[10]Release Get OldTweets3 API in Python Release v3 0.0.11, November 2019, Retrieved from https://pypi.org/project/GetOldTweets3/.
[11]ShakesBeer(2021).重要冠狀病毒（COVID-19）術語須知 (Important Coronavirus (COVID-19) terms to know).Retrieved from https: //tw.voicetube.com/videos/119643 (January 14, 2021).