Prof. Elke A. Rundensteiner, WPI, Computer Science. Advisor
Prof. Emmanuel Agu, WPI, Computer Science
Prof. Kyumin Lee, WPI, Computer Science
Prof. Wei Ding, UMASS Boston, Computer Science
Detecting human emotion patterns plays an essential role in various application contexts from public health to marketing. In this proposal, the problem of online emotion classification in a stream of text messages is targeted. In particular, I address core research challenges related to emotion classification given a stream of text messages from Twitter.
First, as foundation, I design a supervised machine learning approach called Emotex that classifies emotion expressed in text messages using an offline training process. I engineer a set of features based on state-of-the-art emotion lexicons. The resulting emotion classifier Emotex is then plugged into a full-stack framework called EmotexStream to perform online emotion classification in a text stream. For this, EmotexStream exploits a binary classifier to separate messages with explicit emotion from messages without emotion. Then it utilizes Emotex’s emotion classification models for a fine-grained emotion classification of messages with explicit emotion. EmotexStream is deployed to measure public emotion and detect emotion-intensive moments during real-life events in live streams of Twitter messages.
Second, I research two challenges critical for transforming the Emotex classifier into an online deep learning approach. One, I propose to design and evaluate dynamic feature learning methods using deep convolutional neural network technologies to create emotion-specific word embeddings. Two, I will further extend Emotex into an online emotion classification system that dynamically selects a set of features during the prediction process.
Overall, I will conduct extensive experimental studies to evaluate the effectiveness, efficiency and practicality of the proposed methods on real-life social text streams such as Twitter messages, in particular, considering heavily debated themes on Twitter.