Deep Belief Networks with Feature Selection for Sentiment Classification
Patrawut Ruangkanokmas, Tiranee Achalakul, Khajonpong Akkarajitsakul · 2016
Due to the complexity of human languages, most of sentiment classification algorithms are suffered from a huge-scale dimension of vocabularies which are mostly noisy and redundant. Deep Belief Networks (DBN) tackle this problem by learning useful information in input corpus with their several hidden layers. Unfortunately, DBN is a time-consuming and computationally expensive process for large-scale applications. In this paper, a semi-supervised learning algorithm, called Deep Belief Networks with Feature Selection (DBNFS) is developed. Using our chi-squared based feature selection, the complexity of the vocabulary input is decreased since some irrelevant features are filtered which makes the learning phase of DBN more efficient. The experimental results of our proposed DBNFS shows that the proposed DBNFS can achieve higher classification accuracy and can speed up training time compared with others well-known semi-supervised learning algorithms.