Domain Adaptation Approach for Sentiment Analysis

Customer satisfaction surveys, marketing research, and associated analysis are necessary but very costly procedures for companies to understand how their services are, determine how to develop their product better, and choose how to organize their marketing promotion strategies. Being able to mark sentiment for each review as positive or negative by machine learning is highly desirable for companies if sentiment analysis can be conducted on the information on social media such as Facebook, Yelp and Twitter, where people post their opinions on how they feel and what they think of certain products on a large scale. However, there is a wide range of domain variations across different fields, while it is not always to collect enough data for each individual domain. To account for such domain variations, this paper applies domain adaptation approach for sentiment analysis, where a system is trained from one source domain but deployed on another target domain. The method shows satisfactory effectiveness and efficiency when comparing to other methods. This work provides a framework for future data mining on understanding the sentiment across multiple domains using domain adaptation approach.