Sentiment Analysis of Customer Reviews: Impact of Text Pre-Processing

In recent years, the number of customer reviews available online has grown rapidly. It is very important to give automatic support for analysis of such content. Whereas research in English language is quite covered, there is space for research of other morphologically more complex languages like Slovak. Such languages often necessitate more advanced processing. In our work, we proposed model for sentiment analysis of customer reviews based on a neural network architecture and studied impact of several text pre-processing techniques on overall sentiment classification. We showed text pre-processing has a significant impact on performance of sentiment analysis. Though utilizing a medium-sized dataset for model training, we achieved very promising results when compared to baseline SVM model.