New techniques for sentiment analysis and classification in texts

Sentiment analysis aims to analyze the author's thoughts and feelings in a written text and sentiment classification focuses on classifying feelings as positive and negative. Supervised learning methods used for sentiment classification need several labeled data to train classifier. Moreover, since the sentiment classification is domain-dependent problem, it is necessary to collect labeled data for each domain in order to obtain high performance; however, this process is time consuming and costly. When a text whose domain is unknown is to be classified, it is not possible to collect labeled and unlabeled data. In this dissertation, a new approach called as domain classification-based sentiment classification is proposed to solve this problem. In the first phase of the proposed approach consisting of two phases, the domain with labeled data, which a text of an unknown domain belongs to, is found out. In the second phase, the text is classified using the sentiment classifier within the domain it belongs to. In the experiments, the cases that the domain of text exist and does not exist within the available domains were analyzed. For this purpose, two datasets including Turkish and English texts were employed. Thanks to the proposed method, classification performances, which are close to or even better than the ones expected, were attained. Also, it was verified that the proposed method is applicable for both languages as well.