A deep neural network-based approach for fake news detection in regional language

• DOI: https://doi.org/10.1108/IJWIS-02-2022-0036
• Published date: 27 July 2022
• Date: 27 July 2022
• Pages: 286-309
• Subject Matter: Information & knowledge management,Information & communications technology,Information systems,Library & information science,Information behaviour & retrieval,Metadata,Internet
• Author: Piyush Katariya,Vedika Gupta,Rohan Arora,Adarsh Kumar,Shreya Dhingra,Qin Xin,Jude Hemanth

A deep neural network-based

approach for fake news detection

in regional language

Piyush Katariya

Department of Computer Science and Engineering,

Bharati Vidyapeeth’s College of Engineering, New Delhi, India

Vedika Gupta

Jindal Global Business School, O.P. Jindal Global University, Sonipat, India

Rohan Arora,Adarsh Kumar and Shreya Dhingra

Department of Computer Science and Engineering,

Bharati Vidyapeeth’s College of Engineering, New Delhi, India

Qin Xin

Faculty of Science and Technology, University of Faroe Islands,

Vestarabrygga, Faroe Islands, and

Jude Hemanth

Department of Electronics and Communication Engineering, Karunya University,

Coimbatore, India

Abstract

Purpose –The current natural language processingalgorithms are still lacking in judgment criteria, and

these approaches oftenrequire deep knowledge of political or social contexts. Seeingthe damage done by the

spreading of fake news in various sectors have attracted the attention of several low-level regional

communities.However, such methods are widely developedfor English language and low-resource languages

remain unfocused.This study aims to provide analysis of Hindi fake news and develop a referral system with

advancedtechniques to identify fake news in Hindi.

Design/methodology/approach –The technique deployed inthis model uses bidirectional long short-

term memory (B-LSTM) as comparedwith other models like naïve bayes, logistic regression, random forest,

support vector machine, decision tree classifier, kth nearest neighbor, gated recurrent unit and long short-

term models.

Findings –The deep learningmodel such as B-LSTM yields an accuracy of95.01%.

Originality/value –This study anticipates that this model will be a beneficial resource for building

technologiesto prevent the spreading of fake news and contribute to research with low resourcelanguages.

Keywords Natural language processing, Fake news, Machine learning, Gated recurrent unit,

Bidirectional LSTM (bi-LSTM), Hyperparameters, Fine tuning

Paper type Research paper

1. Introduction

In earlier times, people get their daily news and analysis from different sources like

newspapers, televisions, etc. But with the arrival of the internet, a new era of news sharing

came into existence. Many people nowadays get news updates from online sources and

IJWIS

18,5/6

286

Received19 February 2022

Revised4 May 2022

9 June2022

Accepted5 July 2022

InternationalJournal of Web

InformationSystems

Vol.18 No. 5/6, 2022

pp. 286-309

© Emerald Publishing Limited

1744-0084

DOI 10.1108/IJWIS-02-2022-0036

The current issue and full text archive of this journal is available on Emerald Insight at:

https://www.emerald.com/insight/1744-0084.htm

social media pages ,which is now in trendand most of the time it has become hard to decide

whether news read by us has any credibility or not, whether it is legitimate or not (Gupta

et al.,2016). In a democracy like India (Ludden, 2005), the right to be informed is a basic

human right. Seeking the right kind of information is essential before forming opinions.

False information or misleadingdetails can be disastrous in many aspects. News channels,

print media, broadcastmedia (Gaurav et al.,2020) and social media (Gaurav et al.,2020) play

a major role in the spread ofinformation and hence in forming the mindset of the nation.

Fake news (Quandt et al., 2019) refers to misleading information that could spread through

various modes and mediums. The news is called fake because it is spread generally todefame a

person or to spread out rumors to damage the reputation of an entity. The motive behind this is

to gain publicity or make money through advertising revenue. This can be disastrous to

society. Ensuring that everyone gets the right information is vital, as it is an important aspect

of the development of a healthy society. Fake news destroys credibility because it is formed on

misleading facts and figures (Sharma and Sharma, 2022a). Real news benefits a person

whereas fake news harms. One must be an informed consumer and an awar e citizen to help in

the overall progress of the nation. Fake news can cause chaos all over the nation. In a country

like India, only about 125 million people (Parshad et al.,2016) know how to communicate in

English which constitutes about 10% of the total population. The majority of people read and

interact in their regional languages. These people often fall prey to misleading information. In

the natural language processing area, various initiatives had been taken to detect fake news in

several ways, ranging from language-based approaches to content-based verification.

Although such techniques have been predominantly developed for English, research in low

resources languages like Hindi has been limited (Sharma and Garg, 2021).

About 57.09% of the total population of India (McKibben-Greene, 2020) considers themselves

native Hindi speakers. As a result of which lots of fake and manipulative news are now posing a

huge risk in regional languages. One of the major issues faced during the development of this

project was the lack of resources (Harish and Rangan, 2020) on Indian regional languages. This

was sorted by manual data set creation. Next, was to find the most accurate algorithmto use for

fake news detection. The data available on social media is enormous, but it is unlabeled a nd

hence it could not be used for training purposes. The need of the hour is to develop a technique for

managing the massive amount of fake and incorrect data present on the internet, which is

misleading the people to a great extent. Several computational methods (Reis et al.,2019)canbe

designed for detecting fake news in Indian regional languages. The analysis would have been

much easier if a labeled data set is accessible to us. As fake news propagation can be cross-

lingual, it is best to have data sets available in a wide variety of languages (Sharma et al., 2022b).

This paper provides an analysisof Hindi fake news on a manually created data set using

various Machine learning classification algorithms as well as using deep learning NLP

techniques like long short-term memory (LSTM) and gated recurrent unit (GRU). In this

work, our focus is shifted toward assembling feasible and reliable sources in the form of a

Hindi corpus for automatic fakenews detection. Social media (Wang et al.,2019) adds fuel to

the fire by spreading ill-conceived news during an important event where public opinion

matters the most. Hence,through the analysis, our aim is to find a possible Natural language

processing algorithm that can determine whether a source is trustworthy or politically

inclined with or withouthuman sentiments involved (Gupta et al.,2019). With large amounts

of data over the internet,checking the credibility of a source poses a huge challenge for us.

The main contributions of this work are the corpus for the Hindi languagewhich can be

used for research and analysison detecting fake news containing real news which is labeled

as ‘1’and fake news labeled as ‘0,’which is collected from several credible and legitimate

news agencies writtenby professional native Hindi-speaking journalists.

Deep neural

network-based

approach

287