On Mining Opinions from Social Media

The broad use of social media nowadays has led many users to express their opinions on various subjects through them. The need for these opinions to be automatically labeled and categorized according to their sentiment, has also arisen. In this paper, a novel sentiment analysis approach is introduced, which takes into account the total number of idiomatic expressions and emoticons that are used in the text, and simultaneously processes the original text in Greek along with its automatic translation in English as well. Moreover, the novelty of the proposed solution lies in the difficulty of Modern Greek language and the fact that the text in social media is mainly unstructured. The proposed methodology is tested on two distinct data sets of opinions regarding a certain matter, which have been collected from Facebook and Twitter respectively. Finally, we discuss the performance of various classiffication algorithms and we compare the extracted experimental results.