Performance Evaluation of Multi-class Sentiment Classification Using Deep Neural Network Models Optimised for Binary Classification

Fiachra Merwick, Y Bi, Peter Nicholl

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Thispaperpresentsacomparativestudyontheperformanceofbinary- and multi-class Deep Neural Network classification models that have been trained with the optimized hyperparameters. Four sequential models are developed using Tensorflow and Keras to perform sentiment classification of product reviews posted on Amazon, IMDb, Twitter and Yelp. Preprocessing is carried out to cleanse text reviews and to extract informative features. 5-fold cross evaluation results demonstrate that the final multiclass model perform with an accuracy of 74.7% and 72.4%, whereas the binary model achieves of 80.0% and 69.5% indi- cating better performance. The conclusion of this study is that an optimized binary model architecture can be used to train multiclass classification models and save significant amounts of computing time.
Original languageEnglish
Title of host publicationKnowledge Science, Engineering and Management
Subtitle of host publicationKSEM 2021
Pages624–635
Number of pages12
ISBN (Electronic)978-3-030-82147-0
DOIs
Publication statusE-pub ahead of print - 7 Aug 2021
EventThe 14th International Conference on Knowledge Science, Engineering and Management (KSEM 2021) - Tokyo, Japan
Duration: 14 Aug 202116 Aug 2021
http://www.cloud-conf.net/ksem21/index.html

Publication series

NameLecture Notes in Computer Science
PublisherSpringer
Volume12816
ISSN (Print)0302-9743

Conference

ConferenceThe 14th International Conference on Knowledge Science, Engineering and Management (KSEM 2021)
Abbreviated titleKSEM 2021
CountryJapan
CityTokyo
Period14/08/2116/08/21
Internet address

Keywords

  • Sentiment analysis
  • Deep neural networks
  • Word embedding

Fingerprint

Dive into the research topics of 'Performance Evaluation of Multi-class Sentiment Classification Using Deep Neural Network Models Optimised for Binary Classification'. Together they form a unique fingerprint.

Cite this