An Improved Model for Opinion Mining of Public Reviews using Recurrent Neural Network

Abstract

Mining of opinions are very crucial in all fields from e commerce websites to social media platforms. The products on any e commerce websites have thousands of reviews which helps customers to make a decision a product. Social media websites also have people with large number of opinions on a particular subject. Mining of opinions can be extensively used in the fields where opinions play a major role. This project caters to this need and classifies the opinions of people as positive and negative. This can further be used by movie recommendation systems and e commerce websites for evaluation of their product. It involved in the classification the opinions as positive opinions and negative opinions with the help of deep learning algorithms by achieving high accuracy. The procedures involved in this project will be of dataset selection, data preprocessing, data tokenization, and data cleansing and building a neural network. We have taken the dataset of reviews for this purpose. Data preprocessing and data cleansing is done so that deep learning algorithms can be easily applied on the data. Deep learning algorithms learn on their own and do not require guidance. The main objective of using deep learning model is for increasing efficiency, performance and accuracy. Here, we have applied three different neural network models to our dataset and compare the performances according to the testing and training accuracy obtained. Analysis of the three models concludes that Recurrent Neural Model (RNN) has least over fitting with considerable testing and training accuracy. Hence, it best suits the problem statement.