Sentiment Analysis Using Language Models: A Study
Spraha Kumawat, Inna Yadav, Nisha Pahal, Deepti Goel
Abstract
Sentiment analysis is concerned with extracting sentiment to ascertain the attitudes, and emotions associated with the text. It is broadly applied to voice of the customer as they convey their experience and feelings more blatantly which is why comprehending customer's emotions is a must. To identify and classify these unstructured emotions natural language processing (NLP) and machine learning approaches have been adopted in recent times. The main issue with the existing techniques is the inability to deal with the correct interpretation of context owing to lack of labeled data. In this paper, we studied deep neural network based language models to interpret and classify textual sequences into positive, negative or neutral emotions which remove the bottleneck of explicit human labeling. These models were analyzed and evaluations were performed on the Twitter US Airline Sentiment dataset. We have observed a considerable amount of improvements with respect to prior state-of-the-art approaches which closes the gap with supervised feature learning.