Research and Development of Domain Independent Sentiment Classifier

The paper presents a method of constructing a sentiment classifier on two and three classes (positive and negative, positive, neutral and negative texts). It also presented the results of experiments showing the high accuracy of the proposed method on text which are not belong to any pre specified domains. The effectiveness of the presented method is confirmed by experiments' results on the text collection of blogs from ROMIP 2012 seminar. It was used following metrics for classifier evaluation: precision, recall, accuracy and F-measure. The value of F-measure of the proposed method for classification into 2 classes is up to 93%. In addition to blog collection ROMIP 2012 for experiments were used a collection of news and a collection of short-texts from social networks.