Please use this identifier to cite or link to this item: https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4538
Title: Convolutional Neural Network based Multiclass Sentiment Analysis to Detect Human Emotions
Authors: Ramawickrama, R.H.
Issue Date: 12-Aug-2021
Abstract: Sentiment analysis is the area in computer science related to identifying and categorizing opinions implied through a piece of text. Opinions are the most important factor which drives social media, making it vital to understand the underlying emotions of different opinions. This thesis proposes an approach to perform multiclass sentiment analysis using deep convolutional neural networks. The sentiment classes are ‘Happiness’, ‘Sadness’, ‘Anger’, ‘Fear’ and ‘Surprise’. The thesis provides information on the design and the end to end implementation of the convolutional neural network which utilizes one-hot encoding, word embeddings and max pooling to improve the classifier accuracy and performance. It presents how the trained model is evaluated and the results are cross validated with four main performance metrics accuracy, precision, recall and f1-score for each sentiment class. With the proposed approach, the model well predicts four out of the five targeted sentiment classes with an overall model accuracy of 65-70%. Two algorithms are introduced. The first algorithm is to train and test a convolutional neural network which supports multiclass sentiment analysis and the second algorithm is to utilize the trained model for predicting the emotion of a single text and predict the emotions of multiple texts with a consolidated analysis.
URI: http://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4538
Appears in Collections:2020

Files in This Item:
File Description SizeFormat 
2017 MCS 063.pdf671.63 kBAdobe PDFView/Open


Items in UCSC Digital Library are protected by copyright, with all rights reserved, unless otherwise indicated.