在充滿資訊的時代，網路聲量是一個重大議題。我們透過電腦自動分類多篇 日本動畫相關長評論中的情緒，進而擷取輿情。傳統的長文章總結及情感分類方 法是透過事先定義字典規則來判斷，但現實世界中的情況較為複雜。這樣的狀況 下，既不能完全列舉也無法每一種都定義。基於以上方法的缺點，我們設計一套 應用於深度學習的長文分析系統。當模型訓練完畢後，系統將會自動分類評論者 的正負向情緒。在我們收集大量的評論後，就可以對現今動畫作品市場反應的狀 況進行評斷，以達到與情擷取的目的。
本論文提出一個可應用於長篇日本動畫相關評論自動情緒分類的方法及一 份該領域自行蒐集的資料集。首先，我們使用爬蟲抓取大量帶有標籤的評論，並 透過前處理及跳躍式思維(Skip-thoughts)進行文章總結，再利用雙向 GRU 結合自 我注意力機制來進行深度遞歸神經網路的訓練，最後獲得一個情感分類模型。我 們再將模型整合進系統中，便能夠得到一個對長文做情感分類的系統。
我們所蒐集的資料集中的每一筆資料，除了包含標記與評論外，還有更詳細 的分類以及每一部動畫作品的資訊，這些資料將能夠成為日後研究的材料;另外， 在模型訓練的過程中，我們利用資料增強的方式對收集到的評論進行資料增加， 並再用模型進行學習;最後，它將作為判斷正負向情感的依據。
實驗的部份我們使用公開資料集進行評估並針對不同類型日本動畫作品的 評論進行分析，如動作、冒險、校園等等。我們提出的方法在大部分的情況可以 正常判斷，在 IMDB 資料集的情緒分類準確性是 89.9%，在 SST2 是 83.3%，在 MPQA 是 87.3%，在 MR 資料集是 86.0%，而在我們的資料集是 84.7%且整體 執行時間很短，平均每筆預測所花費的時間約 0.001 秒，能夠達到即時的預測。

In an era of information, Internet volume is a major issue. We automatically classify emotions from multiple long comments through a computer to extract sentiment. The traditional long article summary and sentiment classification methods define a dictionary of sentimental words in advance, but the situation in the real world is more complicated. Under the circumstances, neither the dictionary can fully enumerate every word nor totally define each word. Based on the shortcomings of the above methods, we design a long article analysis system with deep learning. When the model is completely trained, the system will automatically classify reviewers’ sentiment. After collecting lots of reviews, we can determinate the market reaction of today's animation works. According to these, we can achieve the target of extracting sentiment.
We propose an automatic sentimental classification method for long comments in the field of animation and a related dataset. First, we use crawlers to fetch a huge number of relevant labeled comments and summarize the articles through pre- processing and Skip-thoughts. Then we use the Bi-GRU combined with self-attention mechanism to train deep recurrent neural networks, and finally complete a sentiment classification model. By integrating the sentimental model into our long comment analysis classification system, we can obtain an ability of extracting emotions from long comments.
In addition to the labels and comments, each of the data we collect contains more detailed classifications and information about each of the animation works, which will be the material for future research. However, in the stage of model training, we adopt data enhancement to increase the collected comments and then employ the model to learn the features. Finally, the system can distinguish between positive and negative sentiments.
In the experiments, we use public open datasets to evaluate and analyze comments on different types of Japanese animation works, such as Action, Adventure, Comedy, and School. Our system based on the proposed methods can correctly predict in most cases of sentimental classification. In the open datasets, the accuracy in the sentimental classification for the IMDB, SST2, MPQA, and MR are 89.9%, 83.3%, 87.3%, and 86.0%, respectively. Additionally, for our dataset, the accuracy in sentimental classification reaches 84.7% and the overall execution time is very short. It spends about 0.001 seconds on an average per prediction. The experimental results reveal that our system can achieve real-time prediction.

[1] A. Go, R. Bhayani, and L. Huang, “Twitter sentiment classification using distant supervision,” in “CS224n: Natural Language Processing with Deep Learning Project Report,” Stanford, San Francisco, California, vol. 1, 2009. [Online] https://www- cs.stanford.edu/people/alecmgo/papers/TwitterDistantSupervision09.pdf
[2] Q. Tul et al., “Sentiment analysis using deep learning techniques: A review,” International Journal of Advanced Computer Science and Applications, vol. 8, no. 6, pp. 424-434, 2017.
[3] J. Tang et al., “Email data cleaning,” in Proceedings of the International Conference on Knowledge Discovery in Data Mining, Chicago, Illinois, pp. 489-498, 2005.
[4] J. J. Webster and C. Kit, “Tokenization as the initial phase in NLP,” in Proceedings of the 14th Conference on Computational Linguistics, Nantes, France, vol. 4, pp. 1106- 1110, 1992.
[5] P. Jin et al., “Bag-of-embeddings for text classification,” in Proceedings of the 25th International Joint Conference on Artificial Intelligence, New York, New York, pp. 2824-2830, 2016.
[6] T. Mikolov et al., “Efficient estimation of word representations in vector space,” in Proceedings of the International Conference on Learning Representations, Scottsdale, Arizona, pp. 1-12, 2013.
[7] J. Pennington, R. Socher, and C. D. Manning, “GloVe: Global vectors for word representation,” in Proceedings of the Conference on Empirical Methods in Natural Language Processing vol. 14, pp. 1532-1543, 2014.
[8] D. Das and A. F. T. Martins, “A survey on automatic text summarization,” in “Literature Survey for the Language and Statistics II course at CMU,” Carnegie Mellon University, Pittsburgh, Pennsylvania, vol. 4, 2007. [Online] https://www.cs.cmu.edu/~nasmith/LS2/das-martins.07.pdf
[9] H. Christian, M. Pramodana Agus, and D. Suhartono, “Single document Automatic text summarization using term frequency-inverse document frequency (TF- IDF),” ComTech: Computer, Mathematics and Engineering Applications, vol. 7, no. 4, p. 285, 2016.
[10]G. Salton and C. Buckley, “Term-weighting approaches in automatic text retrieval,” Information Processing and Management, vol. 24, no. 5, pp. 513-523, 1988.
[11] R. Kiros et al., “Skip-thought vectors,” in Proceedings of the 28th International Conference on Neural Information Processing Systems, Montreal, Canada, vol. 2, pp. 3294-3302, 2015.
[12]S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural Computation, vol. 9, no. 8, pp. 1735-1780, 1997.
[13] K. S. Tai, R. Socher, and C. D. Manning, “Improved semantic representations from tree-structured long short-term memory networks,” presented at the 53rd Annual Meeting of the Association for Computational Linguistics, Bejing, China, July, 1, 2015. [14] A. Vaswani et al., “Attention is all you need,” in Proceedings of the Conference on Neural Information Processing Systems, Long Beach, California, pp. 5998-6008, 2017.
[15] G. Lee et al., “Sentiment classification with word attention based on weakly supervised learning with a convolutional neural network,” in arXiv e-prints: 1709.09885, 2017.
[16] Y. Kim, “Convolutional neural networks for sentence classification,” in Proceedings of the Conference on Empirical Methods in Natural Language Processing, Doha, Qatar, pp. 1746-1751, 2014.
[17] A. L. Maas et al., “Learning word vectors for sentiment analysis,” presented at the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies, Portland, Oregon, 2011.
[18] T. V. Udapure, R. D. Kale, and R. Dharmik, “Study of web crawler and its different types,” IOSR Journal of Computer Engineering, vol. 16, pp. 1-5, 2014.
[19] J. Edwards, K. McCurley, and J. Tomlin, “An adaptive model for optimizing performance of an incremental web crawler,” in Proceedings of the 10th International Conference on World Wide Web, Hong Kong, China, pp. 106-113, 2001.
[20] J. Cho and H. Garcia-Molina, “Synchronizing a database to improve freshness,” Special Interest Group on Management of Data, vol. 29, no. 2, pp. 117-128, 2000. [21] C. Jr et al., “Optimal robot scheduling for web search engines,” Journal of Scheduling, vol. 1, no. 1, pp. 15-29, 1998.
[22] Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” nature, vol. 521, no. 7553, pp. 436-444, 2015.
[23] K. Cho et al., “Learning phrase representations using RNN encoder–decoder for statistical machine translation,” in Proceedings of the Conference on Empirical Methods in Natural Language Processing, Doha, Qatar, pp. 1724-1734, 2014.
[24] W. Wang, Yang, N., Wei, F., Chang, B., & Zhou, M., “R-NET: Machine reading comprehension with self-matching networks,” Natural Language Compute Group, Microsoft Research, Beijing, China, 5, 2017. [Online] https://www.microsoft.com/en- us/research/publication/mcr/
[25] U. G. Ayush Agrawal “Extraction based approach for text summarization using k- means clustering,” International Journal of Scientific and Research Publications, vol. 4, no. 11, pp. 3524-3527, 2014.
[26] G. n. Klambauer et al., “Self-normalizing neural networks,” presented at the Advances in Neural Information Processing Systems, Los Angeles, California, Jan, 24, 2017.
[27]D. Gannon, “TensorFlow meets microsoft's CNTK,” Microsoft, Indiana University, 2016. [Online] https://esciencegroup.com/2016/02/08/tensorflow-meets-microsofts- cntk/
[28] R. Socher et al., “Recursive deep models for semantic compositionality over a sentiment treebank,” in Proceedings of the Conference on Empirical Methods in Natural Language Processing, Seattle, Washington, pp. 1631-1642, 2013.
[29] T. Wilson, J. Wiebe, and P. Hoffmann, “Recognizing contextual polarity in phrase- level sentiment analysis,” presented at the Conference on Human Language Technology and Empirical Methods in Natural Language Processing, Vancouver, Canada, 2005. [30] B. Pang and L. Lee, “Seeing stars: exploiting class relationships for sentiment categorization with respect to rating scales,” in Proceedings of the 43rd Annual Meeting on Association for Computational Linguistics, Ann Arbor, Michigan, pp. 115-124, 2005. [31] T. Miyato, A. M. Dai, and I. Goodfellow, “Adversarial training methods for semi- supervised text classification,” in Proceedings of the 5th International Conference on Learning Representations, Toulon, France, 2017.
[32] J. Camacho-Collados and M. T. Pilevar, “On the role of text preprocessing in neural network architectures: An evaluation study on text categorization and sentiment analysis,” in arXiv e-prints: 1707.01780, 2017.
[33] Y. Zhang, Q. Liu, and L. Song, “Sentence-state LSTM for text representation,” in Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, Melbourne, Australia, vol. 1, pp. 317-327, 2018.
[34] D. Shen et al., “Baseline needs more love: On simple word-embedding-based models and associated pooling mechanisms,” in Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, Melbourne, Australia, vol. 1, pp. 440-450, 2018.
[35] P. Xu et al., “Emo2Vec: Learning generalized emotion representation by multi- task training,” in Proceedings of the 9th Workshop on Computational Approaches to Subjectivity, Sentiment & Social Media Analysis, Brussels, Belgium, pp. 292-298, 2018. [36] Q. Liu et al., “Task-oriented word embedding for text classification,” in Proceedings of the 27th International Conference on Computational Linguistics, Santa Fe, New Mexico, pp. 2023-2032, 2018.
[37] D. Cer et al., “Universal sentence encoder for English,” in Proceedings of the Conference on Empirical Methods in Natural Language Processing, Brussels, Belgium, pp. 169-174, 2018.
[38] A. Madasu and V. Anvesh Rao, “Effectiveness of self normalizing neural networks for text classification,” in arXiv e-prints: 1905.01338, 2019.
[39] Z. Lei et al., “A multi-sentiment-resource enhanced attention network for sentiment classification,” in Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, Melbourne, Australia, vol. 2, pp. 758-763, 2018.
[40] Y. Wang et al., “Sentiment analysis by capsules,” in Proceedings of the International World Wide Web Conferences Steering Committee, Lyon, France, pp. 1165-1174 2018.
[41] W. Zhao et al., “Investigating capsule networks with dynamic routing for text classification,” in Proceedings of the Conference on Empirical Methods in Natural Language Processing, Brussels, Belgium, pp. 3110-3119, 2018.
[42] M. E. Peters et al., “Deep contextualized word representations,” in arXiv e-prints: 1802.05365, 2018.
[43] J. Devlin et al., “Bert: Pre-training of deep bidirectional transformers for language understanding,” in arXiv e-prints: 1810.04805, 2018.
[44] K.-Y. Chen et al., “An information distillation framework for extractive summarization,” IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 26, no. 1, pp. 161-170, 2018.