Abstract: 

Emotion detection is a challenging NLP task as textual data is missing facial expressions or audio streams which are usually direct representations of emotions. However, textual data is still reach in terms of the emotional content. Current approaches to emotion detection in NLP rely on sentiment analysis, which falls short when it comes to detecting different emotions from a given text. In this talk, we will review the shortcomings of traditional approaches to emotion detection and introduce our novel approach which uses Natural Language Inference.

Bio: 

Serdar Cellat is currently working as a Lead Machine Learning Engineer at Y Meadows, an early stage startup helping customer service teams with NLP solutions. Serdar's work includes a variety of NLP tasks such as Intent Classification, Named Entity Recognition, Emotion Detection, Topic Modeling, Question and Answering Systems, and Semantic Textual Similarity. Prior to that, Serdar was a Senior Data Scientist at Liberty Mutual Insurance Legal Department where he built NLP systems to detect excessive charges in legal invoices. Serdar has also served as an instructor and mentor for post graduate certificate programs in Machine Learning and Data Science offered by MIT and UT Austin. Serdar holds a PhD degree in Mathematics from Florida State University, and his research focus was on Machine Learning and Optimization.