An Enhanced Convolution Neural Network Based Approach for Classification of Sentiments

Abstract

Sentiment analysis is an approach to analyse the opinion and views of the people from the text or images posted by them on social media like Facebook and Twitter. Sentiment analysis is a challenging task because it is not easy to analyse the exact views, opinions, and feelings of the text. The way of expressing feelings varies with people in different contexts and topics. This issue can be resolved by combining the text and prior knowledge. This research work proposes a deep convolutional neural network that uses the character to sentence-level information to perform sentiment analysis of tweets. A new approach for the initialization of the weights of the convolutional neural network is suggested which helps to train the network efficiently and helps to find effective features. The model is further tuned by a deep learning model which reduces the classification error. It uses word vector features with feature engineering by means of a convolution neural network. Further, the process involves learning by the soft-max classifier. The experiments are performed using three different datasets with 3K,10K and 100K tweets. The proposed approach represents a significant improvement in accuracy, precision, and recall in comparison to existing approaches.