COMPARATIVE STUDY OF RECURRENT NEURAL NETWORK AND SUPPORT VECTOR MACHINE FOR TEXT CLASSIFICATION

BAMGBOYE, PELUMI OYELAKIN (2021) COMPARATIVE STUDY OF RECURRENT NEURAL NETWORK AND SUPPORT VECTOR MACHINE FOR TEXT CLASSIFICATION. Masters thesis, Landmark University, Omu Aran, Kwara State.

Text BAMGBOYE PELUMI.pdf - Submitted Version Download (1MB)

Abstract

Text is constantly generated from our day to day use of the internet, and these large amounts of data generated are mostly unfiltered. In most cases, unstructured data needs to be classified to improve the rate at which a given text is understood. Text classification is a branch of Natural Language Processing that is used to create a distinction in an unstructured text data. As a result of this, Machine learning is widely used in the classification of textual data as a result of its ability to create complex prediction functions dynamically. Also, statistical models are commonly used to classify textual data because they are able to describe the relationship between two or more random variables. Text classification consists of various branches which include sentiment analysis, document categorization, product labelling, and information retrieval. In an ecommerce environment, customers review is a way in which customer’s opinion is given about a particular product and this can be negative, positive or neutral. However, there is always difficulty in classifying these opinions in ways in which they can be used to inform customer’s decision about a product. Despite the advancement of machine learning models, certain limitations are prevalent in the traditional techniques like Decision Tree and Naïve Bayes. In this study, a comparative study of Recurrent Neural Network (RNN) and Support Vector Machine (SVM) is done for a customer product review on whether they have positive or negative comments. The result of this work shows that RNN with an accuracy of 94.86% is better than the state of art SVM with an accuracy of 86.67%. The result of this work is not only better in terms if accuracy, but also other performance matrices.

Actions (login required)

View Item