Opinion mining using an LVQ neural network

Abstract

Due to the increased use of social media in the past few years, a large volume of data has been accumulated which contains human sentiments and opinions. The field that deals with the automated extraction of opinions is named opinion mining. In this paper, we evaluate the performance of an LVQ neural network on document level analysis using a benchmark movie review dataset. Document-level opinion mining aims at classifying a text, usually as positive or negative based on its overall sentiment. In order to reduce the dimensions of the reviews' vector representations, we use the feature selection method Information Gain. We use an exhaustive grid search for hyperparameter tuning and two methods for performance evaluation: a nested cross validation and a non-nested 10-fold cross validation. We study the performance of our model for different numbers of selected features by Information-Gain.