Survey Paper: Domain Adaptation for Sentiment

Abstract — Online reviews make sentiment classification an interesting topic in industrial research. Given a review about a product – the goal is to classify whether it is positive or negative. Reviews are in different domain and it is difficult to collect data and train them for the entire domain. Domain adaptation is a fundamental problem in natural language processing (NLP). Transfer learning or domain adaptations are tools for sentiment analysis applications. In Sentiment classification the labeled training data comes from one distribution (source domain) and test data from other distribution (target domain). Such mismatches are considered different and it is usually very hard to measure and formulate these distribution differences. In order to build a sentiment classifier using text categorization methods it is necessary to take into account between these two distributions. The need for domain adaptation is widespread in many real world applications especially in computer vision this is the main challenge. Domain adaptation or Transfer learning is a key feature that requires a useful framework for sentiment classification. This paper provides an overview of literature dealing with challenges and techniques for domain adaptation with special focus on natural language processing