Sentiment analysis based on deep learning

Abstract

With the advent of the digital age, the ideas of our collective society have never been so easily accessible. Thousands of feelings and opinions are uploaded onto various social media platforms every minute of the day, whether they be in the form of a single sentence or a lengthy article. This results in institutions requiring techniques to classify this deluge of data, allowing for the analysis of the sentiments of the populace. Such techniques for the analysis of text data fall under the category of Natural Language Processing (NLP), of which Sentiment Analysis is a part of. Sentiment Analysis is the process in which the feelings of a writer of a piece of text are analysed using machine learning. One of the ways the feelings are categorised is into “positive”, “negative” and “neutral” tags, which can then be further analysed using more conventional statistical methods. This project aims to study the ways Sentiment Analysis can be improved based on deep learning methods, focusing on the differences in accuracy between different neural network models. These models include CNN, LSTM and BERT models, and methods used to improve on their accuracy.