Aspect based Sentiment Analysis is one of the hot topics in Natural Language Processing (NLP). It is a process of breaking the input data into aspects that denotes the important features present in the dataset and allocates each aspect to its equivalent sentiment level(Positive, negative, or Neutral). Deep Learning techniques have been widely used to increase the effectiveness of the NLP task. We have chosen the Convolutional Neural Network (CNN) as a core methodology to perform Sentimental Analysis in the tweets obtained from Twitter. The dataset is constructed from a large number of raw tweets obtained from twitter and they are pre-processed through various procedures. An unsupervised graph-based ranking model CoRank is used to eliminate the redundant data after preprocessing which yields fast convergence for sentiment predictions in the CNN network. The experimental results show the proposed CNN model achieves an accuracy of 94.52% the sentiment analysis task. Depending on the constructed twitter dataset the comparative analysis is conducted for different models compared with our proposed model which shows the efficiency of the CNN framework. The results demonstrate that the proposed framework outperforms other Neural Networks and machine learning techniques compared.
Volume 12 | Issue 4
Pages: 106-114
DOI: 10.5373/JARDCS/V12I4/20201423