Feature selection for improving opinion identification from web authors' posts
Athanasia Koumpouri, Iosif Mporas, Vasileios Megalooikonomou · 2015
In the present article, we address the problem of automatic opinion identification of web users from movie reviews. Specifically, relying on six well-known machine learning algorithms, we investigate the effectiveness of feature selection in the improvement of the accuracy of opinion identification. The feature ranking is performed over a set of statistical, part-of-speech tagging and language model based features. In the experiments, we employed classification models based on decision trees, support vector machines and lazy-learning algorithms. The experimental evaluation performed on the publicly available Polarity Dataset v2.0 demonstrated that feature selection significantly improves the accuracy of opinion identification regardless of the type of machine learning algorithm used.