Study of different statistical machine learning techniques for text sentiment classification

Abstract EN

Text classification is an important task in NLP for various applications from movie review classification to market analysis.

NLP as a tool provides the capability to process huge amount of text and come up with conclusions.

In this paper we inves-tigate statistical machine learning for NLP for document classification.

The target problem of choice is sentiment analysis, we explore various techniques for text pre-processing, feature selection and model selection to find a good fit model.

This paper acts as both a system proposal and also a primer for those who to start practicing NLP, we try to provide insight and intuition about modelling choices for text classi-fication that extend even beyond the task scope to general NLP.

In this paper we propose a feature based text sentiment analysis relying heavily of the BoN (Bag of N-grams) model and utilizing these features with a statistical ML classifier.

We use the IMDB movie review dataset (Maas et al.

2011) for benchmarking.