Comparison of VADER and LSTM for Sentiment Analysis
Adarsh R1, Ashwin Patil2, Shubham Rayar3, Veena K M4
1Adarsh R, Manipal Institute of Technology, Manipal Academy of Higher Education, Manipal (Karnataka), India.
2Ashwin Patil, Manipal Institute of Technology, Manipal Academy of Higher Education, Manipal (Karnataka), India.
3Shubham Rayar, Manipal Institute of Technology, Manipal Academy of Higher Education, Manipal (Karnataka), India.
4Veena K M, Manipal Institute of Technology, Manipal Academy of Higher Education, Manipal (Karnataka), India.
Manuscript received on 24 March 2019 | Revised Manuscript received on 05 April 2019 | Manuscript Published on 18 April 2019 | PP: 540-543 | Volume-7 Issue-6S March 2019 | Retrieval Number: F03040376S19/2019©BEIESP
Open Access | Editorial and Publishing Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: Sentiment analysis is one of the trending topics at present. It has a vast scope from analysing the mood of the person based on his tweet, to predicting the stock prices. But this field is quite challenging. It is not easy to make a machine understand what exactly the person is saying. In this paper, we are going to demonstrate two different methods that can be used in sentiment analysis and its comparison. The two methods used in this paper are: i) VADER-Valence Aware Dictionary for sEntiment Reasoning ii) LSTM model (Long Short-Term Memory). VADER uses a lexicon-based approach, where the lexicon contains the intensity of all the sentiment showing words. The intensities are fetched, the sentiment score is calculated and based on this sentiment score, the review is classified as either positive or negative. We used VADER from NLTK module of python for our study. Recurrent Neural Network has proved its results in a variety of problems like speech recognition, language modelling, and translation. We used LSTM which is an extension of RNN for our study. LSTM networks are very effective for sequential data like texts because they can relate the context of the sentence very well. We preferred LSTM over RNN as LSTM supports Long-term dependency which will help us predict our reviews better. We implemented the LSTM model using keras.
Keywords: GloVe, Lexicon Approach, LSTM, Sentiment Analysis, VADER.
Scope of the Article: Predictive Analysis