Empirical study of sentiment analysis tools and techniques on societal topics
- PDF / 1,480,402 Bytes
- 29 Pages / 439.642 x 666.49 pts Page_size
- 31 Downloads / 211 Views
Empirical study of sentiment analysis tools and techniques on societal topics Loitongbam Gyanendro Singh1
· Sanasam Ranbir Singh1
Received: 13 May 2020 / Revised: 14 August 2020 / Accepted: 17 August 2020 / © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract A surge in public opinions mining against various societal topics using publicly available off-the-shelf sentiment analysis tools is evident in recent times. Since sentiment analysis is a domain-dependent problem, and the majority of the tools are built for customer reviews, the suitability of using such existing off-the-the-shelf tools for a societal topic is subject to investigation. None of the existing studies has thoroughly investigated on societal issues. This paper systematically evaluates the performance of 10 popularly used off-the-shelf tools and 17 state-of-the-art machine learning techniques and investigates their strengths and weaknesses using various societal and non-societal topics. Keywords Sentiment analysis · Societal topics · Publicly available sentiment analysis tools · Machine learning techniques
1 Introduction With the increase in availability of public opinions on various social media platforms such as Twitter, Facebook, LinkedIn, Google Plus, YouTube, etc., a surge in attention of data scientists/agencies in understanding public opinions on various social issues such as social inequality (Cao et al. 2018; Singh et al. 2018; Lerman et al. 2016), public health (Karamibekr and Ghorbani 2012; Maynard and Bontcheva 2016; Garay et al. 2019), ¨ urk and Ayvaz 2018), election (Kuˇsen and Strembeck 2018; Saif et al. social unrest (Ozt¨ 2016; Mohammad et al. 2015; Tumasjan et al. 2010), disaster events (Neppalli et al. 2017; Chen et al. 2020), terror attack (Burnap et al. 2014), etc. is evident. Understanding public opinion on various social issues is vital for various communities like business associates, policymakers, law enforcement agencies, etc. One of the parameters often considered in such studies is public sentiment toward target policies or issues. As building a sentiment Loitongbam Gyanendro Singh
[email protected] Sanasam Ranbir Singh [email protected] 1
Indian Institute of Technology Guwahati, Assam, India
Journal of Intelligent Information Systems
analysis (SA) tool is an expensive task that potentially needs a large volume of annotated dataset and domain expertise, most of the studies that analyze public opinion use publicly available off-the-shelf tools. However, as observed in various studies (Ribeiro et al. 2016; Maynard and Bontcheva 2016; Giachanou and Crestani 2016; Silva et al. 2016; Zhou and Huang 2017), the task of SA is highly domain-dependent. A SA tool built for product reviews may not be suitable for finding sentiment of public opinions in the societal domain and vice versa. Therefore, the effectiveness of using off-the-shelf tools for SA on public opinion over various societal topics needs systematic investigation. Motivated by the above observations, this paper systematically ev
Data Loading...
