Decision Tree Based Feature Selection and Multilayer Perceptron for Sentiment Analysis

Sentiment analysis plays a big role in brand and product positioning, consumer attitude detection, market research and customer relationship management. Essential part of information-gathering for market research is to find the opinion of people about the product. With availability and popularity of like online review sites and personal blogs, more chances and challenges arise as people now can, and do use information technologies to understand others opinions. In this paper, a Multi-Layer Perceptron (MLP) is used to classify the features extracted from the movie reviews. A Decision Tree-based Feature Ranking is proposed for feature selection. The ranking is based on Manhattan Hierarchical Cluster Criterion In the proposed feature selection; a decision tree induction selects relevant features. Decision tree induction constructs a tree structure with internal nodes denoting an attribute test with the branch representing test outcome and external node denotes class prediction. In this paper, a hybrid algorithm based on Differential Evolution (DE) and Genetic Algorithm (GA) for weight optimization algorithm to optimize MLPNN is proposed. IMDb dataset is used to evaluate the proposed method. Experimental results showed that the MLP with proposed feature selection improves the performance of MLP significantly by 3.96% to 6.56%. Classification accuracy of 81.25% was achieved when 70 or 90 features were selected.