Text classification using semi-supervised approach for multi domain
Jyoti S Deshmukh, Amiya Kumar Tripathy · 2017
The rapid growth of Internet technologies has resulted in an increase in online user's content creation. User generated information, usually represents the people's opinions, thoughts, reviews and sentiments. Automatic sensing and analysis of opinions about products, brands, political publications, etc. is a challenging job. Opinions are expressed in different ways in different domains. Constructing and labeling corpus for every domain is a pricey thing. Words from source and target domains are not always similar; hence classifier trained with one domain to another domain leads into poor performance. Therefore the need of domain adaptation algorithms arises to diminish domain's reliance and tagging costs. Using adapted maximum entropy with bipartite clustering in the proposed work, opinionated words are classified in two categories as positive and negative. Results demonstrated that proposed approach performs fairly well compared to baseline method.