Eureka! Institutional Repository

In this master thesis our goal is to develop tools in order to be used in sentiment analysis applications for texts in Greek language. We have developed a Greek language dictionary using MySQL and MongoDB database systems, which contains words in different numbers, genders, tenses etc. A dataset, from users’ opinions about technology products characterized according to sentiment, has been developed. Two applications, one for automatic spelling correction and another for lemmatization have also been developed. A framework of semi-supervised recursive autoencoders, which are trained in vector space representations for multi-word phrases, is used. These representations in automatic assessment of text provision outperform other state-of–the-art approaches on commonly used datasets. This model is applied in on our users’ opinions dataset and the results were examined to see how they were affected by using our tools use, are tested. Using these tools for the pretreatment of the dataset, we have improved the results up to 24%.