BI-LSTM and CNN hybrid approach for the improvement of twitter sentiment analysis Choudhary Prafful1*, Solanki Arun1** 1Lecturer, Department of Computer Science and Engineering, Gautam Buddha University, Greater Noida, Uttar Pradesh, India *E-mail: chaudhary.prafful@gmail.com
**asolanki@gbu.ac.in
Online published on 04 December, 2021. Abstract Sentiment Analysis provides the sentiments of Textual data. It is used widely by multinational companies to know customer sentiment to gain information about market trends and product reviews. In this field, we see a constant development and are still being extensively researched in different areas by top organizations and universities all around the globe. One significant flaw in their technique is that it is unable to account for the overall dependencies of sentences in a document. We look at sentiment analysis approaches based on deep learning, which have already shown promise in a variety of difficult problems in domains such as vision, speech, and text analytics. We recommend using the Bi-LSTM combined with CNN model, which operate like this: the CNN model is used for extraction of features, and the Bi-LSTM model receives input from the CNN model's output and retains the previous data. In this proposed work, we studied Machine Learning techniques and Deep Learning approaches for their uses in Sentiment Analysis. Our Proposed work is using a Hybrid approach with Glove embeddings, Bi-LSTM, and CNN layer in the deep learning model. Results show that the model we planned beats existing Naive Bayes, Support Vector Machine (SVM), Linear Regression, and LSTM and gives an accuracy of 82.74. Our CNN-Bi-LSTM model does 2.69% better than a regular LSTM model. On the other side, the Bi-LSTM-CNN model does 3% better than a normal LSTM model and 7.05% better than a Bernoulli NB model. Top Keywords Sentiment, Convolution Neural Network (CNN), LSTM, Bi-LSTM, Support Vector Machine (SVM), Linear Regression, Naive Bayes (NB), Random Forest. Top |