Please use this identifier to cite or link to this item: https://hdl.handle.net/1959.11/61435
Title: Multi-PSO based Classifier Selection and Parameter Optimisation for Sentiment Polarity Prediction
Contributor(s): Budhi, Gregorius Satia (author); Chiong, Raymond  (author)orcid ; Hu, Zhongyi (author); Pranata, Ilung (author); Dhakal, Sandeep (author)
Publication Date: 2018
DOI: 10.1109/ICBDAA.2018.8629593
Handle Link: https://hdl.handle.net/1959.11/61435
Abstract: 

In the big data era, machine learning algorithms are extensively used for sentiment polarity prediction. However, identifying the correct machine learning algorithm and its parameter settings for the problem at hand can be a difficult task. We propose a system based on Particle Swarm Optimisation (PSO) to find the best machine learning algorithm and optimise its parameters for sentiment polarity prediction. The system's design consists of two layers, namely a multi-PSO layer and a knockout layer. From experimental results, we find that each PSO in the multi-PSO layer could optimise the parameters of the classifiers processed. Overall, the system is able to determine the best classifier from the collection of processed classifiers and also provide quasi-optimal parameters for the classifier to predict the sentiment polarity of customer reviews.

Publication Type: Conference Publication
Conference Details: IEEE ICBDA 2018: Conference on Big Data and Analytics, Langkawi, Malaysia, 21st - 22nd November, 2018
Source of Publication: 2018 IEEE Conference on Big Data and Analytics, ICBDA, p. 68-73
Publisher: IEEE
Place of Publication: United States of America
Fields of Research (FoR) 2020: 4602 Artificial intelligence
Peer Reviewed: Yes
HERDC Category Description: E1 Refereed Scholarly Conference Publication
Appears in Collections:Conference Publication
School of Science and Technology

Show full item record

SCOPUSTM   
Citations

4
checked on Oct 26, 2024

Page view(s)

98
checked on Aug 3, 2024
Google Media

Google ScholarTM

Check

Altmetric


Items in Research UNE are protected by copyright, with all rights reserved, unless otherwise indicated.