Opinion mining on non-english short text

Abstract

As the type and the number of such venues increase, automated analysis of sentiment on textual resources has become an essential data mining task. In this paper, we investigate the problem of mining opinions on the collection of informal short texts. Both positive and negative sentiment strength of texts are detected. We focus on a non-English language that has few resources for text mining. This approach would help enhance the sentiment analysis in languages where a list of opinionated words does not exist. We present a new method to automatically construct a list of words with their sentiment strengths. Then, we propose a new method projects the text into dense and low dimensional feature vectors according to the sentiment strength of the words. We detect the mixture of positive and negative sentiments on a multi-variant scale. Empirical evaluation of the proposed framework on Turkish tweets shows that our approach gets good results for opinion mining.