Paper Title
Aspect Based Sentiment Analysis With Limited Training Data Using Transfer Learning

Abstract
Aspect Based Sentiment Analysis has become a hot trend in Natural Language Processing for brands to utilize and make effective business decisions. This approach helps brands identify the sentiment of customers towards different aspects or features of their products. However, the approach is not easy to implement as it requires the task of identifying phrases which express sentiment towards different aspects of the product. Additionally, obtaining huge amount of data for each aspect of the product to train the model in a short time may not be feasible. We look into the prospect of performing aspect-based sentiment analysis on mobile device data by considering different aspects such as audio, display, battery, and performance. This paper demonstrates the effectiveness of using pre-trained models with fine tuning in challenging situations where there may not be enough data to learn from. Further, a comparison between two class and three class classifier is performed to demonstrate the effectiveness of removing neutral statements prior to performing sentiment analysis. We also perform comparative analysis with deep learning and traditional machine learning models to demonstrate the usage of pre-trained models with limited data.. Keywords - BERT, LSTM, Natural Language Processing, Sentiment Analysis, Transfer Learning