WPI – Computer Science
Prof. Elke Rundensteiner, WPI - Computer Science (Advisor)
Prof. Emanuel Agu, WPI - Computer Science
Prof. Kyumin Lee, WPI - Computer Science
Prof. Wei Ding, UMass Boston (External Member)
In recent years, there has been a great deal of interest in automatically identifying opinions, emotions, and sentiments in text. Supervised emotion classification is challenging due to limited number of labeled data resources. Moreover, it is further complicated due to involving a high-dimensional feature space and a large number of emotion categories.
In this dissertation, we study the problem of classifying and analyzing emotion in textual data using traditional machine learning and deep learning methods. First, we develop Emotex, a supervised emotion classification approach using static feature vectors. To solve the problem of the high-dimensional feature space, Emotex relies on hand-crafted features selected from lexicons for deriving word-emotion association. Emotex utilizes embedded hashtags to automatically label the emotions expressed in text messages. It builds a large corpus of emotion-labeled messages with no manual effort to train emotion classifiers. Hand-crafted features are time consuming to create and may be incomplete.
To solve this problem, we then develop a deep learning approach called DeepEmotex that learns dynamic features based on the input textual context instead of using static hand-crafted features. In particular, a sequential transfer learning framework is developed to fine-tune the pre-trained language models. DeepEmotex utilizes two state-of-the-art pre-trained models, known as Bidirectional Encoder Representations from Transformers (BERT) and Universal Sentence Encoder (USE). We evaluate the performance of DeepEmotex models in classifying emotion in a benchmark dataset. Evaluation results show that the proposed transfer learning model using BERT outperforms the state-of-the-art result to classify the benchmark dataset by 3%. Lastly, we deploy our emotion classification models to analyze emotion in streams of tweets. For this, we develop a framework called EmotexStream. An online method is proposed to measure public emotion and detect abrupt changes in emotion as emotion-burst moments in live text streams. Through a series of case studies we confirm that the proposed methods are able to detect emotion-critical moments during real-life events.