透過圖形關係發展的情緒辨識整合框架

Towards a Unified Framework for Emotion Recognition

艾維斯(Saravia, Elvis)

指導教授 ： 陳宜欣

國立清華大學/電機資訊學院/資訊系統與應用研究所/博士(2019年)

摘要

社群網站為世界各地的使用者提供一個能透過短文字交流、分享見聞的平台。同時也給予研究者探索在不同文化背景下，人們所使用情緒語言差異的機會。然而，情緒往往藏在各種語言情境與枝微末節中，故使得欲從字裡行間擷取情緒表現、情緒特徵的研究者或運算系統面臨諸多挑戰。

在本篇研究中，我們提出一個情緒辨識的框架，它權衡傳統的資訊檢索方法與基於類神經的作法，並進一步整合為一。更確切地說，我們建構一個模型，它係以圖形理論作為基礎並延伸，能夠自情緒文本中萃取豐富的語意樣式；接著，透過能增強特徵之間的語意關係的類神經詞嵌入方法，豐富所提取之基於樣式的特徵。

我們所提出的框架，在兼容並蓄傳統的方法後，具備更完整良好的特質，例如模型的可解釋性、使用彈性以及概括性。此外，在本篇研究中，我們亦探究現階段自然語言處理的遷移式學習系統在情緒辨識任務上所扮演的角色；並且，我們提供了在不同情緒識別任務中，本篇模型與其他多種基準模型的實驗結果。最後，透過探討方法在未來的研究價值與改進方向作結，提供有興趣的研究者作為參考。

關鍵字

情感認知 ； 情緒分析 ； 情感計算 ； 文本分類

並列摘要

Social media platforms provide a means for people from all over the world to communicate and share opinions via short and concise text messages. This communication medium has provided a way for researchers to investigate the use of emotional language on social networks across different cultural groups. Emotional expressions are conveyed with all sorts of linguistic phenomena and nuances that present various challenges for computational systems that aim to extract emotion from textual information through various feature representations. In this work, we propose a unified framework for emotion recognition that leverages different traditional information retrieval methods and neural based approaches. Specifically, we propose a graph-based mechanism to extract rich syntactic patterns from an emotion corpus. Thereafter, the extracted pattern-based features are enriched with semantic information through a neural word embedding approach that aims to enhance the semantic relationship among the features. By combining these approaches, the proposed emotion recognition system offers desired characteristics such as explainability, flexibility, coverage, among others. Moreover, we explore and evaluate modern natural language processing transfer learning systems and discuss the role they play in emotion recognition. We include several baseline models, propose several benchmarks, and provide empirical results for several emotion recognition tasks. Lastly, we discuss future work and offer recommendations to further improve text-based emotion recognition systems.

並列關鍵字

emotion recognition ； sentiment analysis ； affective computing ； text classification

參考文獻

[1] Robert Plutchik. The nature of emotions human emotions have deep evolutionary roots, a fact that may explain their complexity and provide tools for clinical practice. American Scientist, 89(4):344–350, 2001.

[2] James Russell. A circumplex model of affect. Journal of Personality and Social Psychology, 39(6):1161–1178, 1980.

[3] Jeremy Howard and Sebastian Ruder. Universal language model fine-tuning for text classification. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, pages 328–339, 2018.

[4] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in Neural Information Processing Systems, pages 5998–6008, 2017.

[5] Bo Pang, Lillian Lee, and Shivakumar Vaithyanathan. Thumbs up?: sentiment classification using machine learning techniques. In Proceedings of Empirical Methods in Natural Language Processing, pages 79–86. ACL, 2002.