Indirect aggression peaks on X during major events, study finds

Indirect aggression in Indian conversations on X is most common during major social, financial, sporting and political events, researchers at Goa Institute of Management and Goa Business School have found. Using machine learning models, they analysed more than 1.31 lakh posts, classifying them as overtly aggressive, covertly aggressive or non-aggressive. Overt aggression ranged from 11 to 37 per cent, peaking at 37 per cent during India's T20 World Cup loss. The study appeared in the journal Advances in Consumer Research.

Written by AI from Business Standard · not yet reviewed by an editor · Published 27 Sept 2026, 1:29 pm IST

Source

Business Standard · read the original report ↗

#social media#machine learning#research#online aggression#goa

Desk check · compared with the source

What the desk checked (5)

Researchers analysed more than 1.31 lakh X posts using machine learning models. — Figure appears in source; attributed to a study by Goa Institute of Management and Goa Business School.
Overtly aggressive responses ranged from 11 per cent to 37 per cent across four domains. — Figures appear in source and are internally consistent with the stated T20 World Cup peak.
Open aggression was highest at 37 per cent during India's T20 World Cup loss. — Attributed directly to Professor Lakshmi Vishnu Murthy Tunuguntla speaking to PTI.
Findings were published in the journal 'Advances in Consumer Research'. — Stated in source; journal name given but no issue or date details, unverifiable here.
Aggression detection differs from sentiment analysis as hostile language may not match negative sentiment. — Attributed to Associate Professor P Balasubramanyam of GIM.

Written by AI from Business Standard · not yet reviewed by an editor · How we use AI · Report an error

Analysts’ view opinion

AI Technology Analyst

The most interesting finding here is not that Indians get heated online during a T20 defeat or an election cycle — it is that the bulk of the hostility is indirect, and indirect hostility is exactly what today's moderation stacks are worst at catching. Keyword filters and sentiment scoring flag slurs and overt abuse; sarcasm, insinuation and coded jibes slip through as 'negative but permissible'. By separating aggression from sentiment across more than 1.31 lakh posts, the GIM–Goa Business School study points at a structural gap in how platforms police Indian-language conversation, rather than at any single platform's failure.

The researchers' core technical point — that aggression and sentiment are different signals — matters commercially, because most deployed moderation and brand-safety tools still lean on sentiment classification.
Overt aggression peaking at 37 per cent during India's T20 World Cup loss suggests event-driven spikes, which argues for moderation capacity that scales around scheduled flashpoints rather than staying flat.
The observed decline in aggression after each incident, at varying rates, implies platforms have a short intervention window in which throttling or friction would matter most.
India's multilingual, code-mixed feeds are a hard case for natural language processing, and locally trained models like this are the kind of groundwork global platforms have historically under-invested in.
For competitors and advertisers, aggression-aware classification is a plausible next layer in trust-and-safety tooling, though the study itself is research output, not a deployable product.

What to watch — Watch whether platforms, Indian regulators or trust-and-safety vendors begin treating covert aggression as a measurable category in moderation and compliance reporting, rather than folding it into generic 'negative sentiment'.

The story does not establish the study's model accuracy, how the events or the 1.31 lakh posts were sampled, the exact covert-aggression percentages, or whether X has responded or acted on the findings.

Deep dive

Research brief · 8 facts · 3 dates · exam-ready

The brief

Context

Researchers at Goa Institute of Management (GIM) and Goa Business School used machine learning models to study patterns of aggressive communication in Indian posts on the social media platform X (formerly Twitter). Their study, "Indian Aggression Detection through Multiple ML Models from Twitter Data", analysed more than 1.31 lakh posts and sorted them into overtly aggressive, covertly aggressive and non-aggressive categories. It found that indirect (covert) aggression is most common during major social, financial, sporting and political events. The findings were published in the journal "Advances in Consumer Research".

Key facts

The study analysed more than 1.31 lakh posts from the platform X to detect patterns of aggressive communication in Indian conversations.
Posts were classified into three categories: Overtly Aggressive (OAG), Covertly Aggressive (CAG) and Non-Aggressive (NAG).
Overtly aggressive responses ranged from 11 per cent to 37 per cent across incidents in four domains studied.
Open aggression was highest at 37 per cent during India's T20 World Cup loss.
Aggression levels showed a general declining trend over time after each incident, but the rate of decline varied across events.
The study, titled 'Indian Aggression Detection through Multiple ML Models from Twitter Data', was conducted by researchers at Goa Institute of Management and Goa Business School.
Findings were published in the journal 'Advances in Consumer Research'.
Researchers stressed the distinction between sentiment analysis (positive/negative/neutral) and aggression detection, which targets harmful or hostile language.

Timeline

Date not stated in the sourceIndia's T20 World Cup loss — the event during which open aggression peaked at 37 per cent in the dataset.
Date not stated in the sourceStudy 'Indian Aggression Detection through Multiple ML Models from Twitter Data' conducted by GIM and Goa Business School researchers.
Date not stated in the sourceFindings published in the journal 'Advances in Consumer Research'; researchers spoke to PTI about the results.

Who has a stake

Goa Institute of Management and Goa Business School researchers — Authors of the study; contributing to computational social science and culturally attuned AI applications in India.
Social media platform X and other platforms — Findings have practical implications for improving online moderation systems that must detect covert as well as overt aggression.
Government and policymakers — Study is positioned as guidance for responsible communication policies on digital platforms.
Indian social media users — Subjects of the analysis; exposed to hostile or harmful language during emotionally and politically charged events.
Lakshmi Vishnu Murthy Tunuguntla, Professor, GIM — Reported that overt aggression stayed relatively low and declined over time, with variation across events.
P Balasubramanyam, Associate Professor, GIM Big Data Analytics Department — Highlighted the sentiment-versus-aggression distinction and the NLP/ML contribution of the work.

Why it matters

Most automated content moderation relies on sentiment analysis, which can miss hostile language that is not overtly negative — exactly the covert aggression this study finds spikes around big sporting, political, financial and social events. Building aggression-detection models tuned to India's multilingual, context-heavy digital landscape could make moderation and communication policy more effective. It also offers empirical evidence that online hostility is event-driven and tends to fade over time, which matters for how platforms and governments time interventions.

UPSC angle

Prelims pointers

Study title: 'Indian Aggression Detection through Multiple ML Models from Twitter Data'; institutions: Goa Institute of Management and Goa Business School.
Published in the journal 'Advances in Consumer Research'; dataset: over 1.31 lakh posts on X.
Three classification categories used: Overtly Aggressive (OAG), Covertly Aggressive (CAG), Non-Aggressive (NAG).
Overt aggression range: 11 per cent to 37 per cent; peak of 37 per cent during India's T20 World Cup loss.
Aggression detection differs from sentiment analysis: it targets harmful or hostile language, not just negative sentiment.
Methods used: natural language processing and machine learning, applied within computational social science.

Mains framing

The GIM–Goa Business School study of over 1.31 lakh posts on X shows that aggressive online communication in India is event-triggered: overt aggression ranged from 11 to 37 per cent across four domains, peaking at 37 per cent during India's T20 World Cup loss, while indirect or covert aggression was common across several social, financial, sporting and political events. Two causes are visible in the findings — emotionally and politically charged events act as flashpoints, and hostility is often expressed covertly, escaping tools built only to detect negative sentiment. The implications are threefold: moderation systems calibrated to overt abuse may systematically under-detect covert hostility; India's multilingual digital environment needs culturally attuned natural language processing rather than imported models; and the observed decline of aggression over time, at varying rates, suggests interventions are most needed in the window immediately after an incident. The researchers frame the way forward as improving online moderation systems, guiding responsible government communication policies, and extending research into the psychology and sociology of digital interactions — while any such deployment must balance harm reduction against the risk of over-classifying speech.

Key terms

Overtly Aggressive (OAG): Posts expressing open, direct hostility; ranged from 11 to 37 per cent in the study.
Covertly Aggressive (CAG): Indirect or veiled aggression, found to be common across several major events studied.
Non-Aggressive (NAG): The third classification category, covering posts without aggressive content.
Sentiment analysis: Technique classifying communication as positive, negative or neutral — distinct from aggression detection.
Natural language processing (NLP): AI methods for analysing human language, used here to detect aggression in Indian posts on X.
Computational social science: Field applying computational tools such as ML to study social behaviour; the study's stated contribution area.

Practice questions

Discuss how machine learning based aggression detection differs from sentiment analysis, and why this distinction matters for content moderation in a multilingual country like India.
Evidence suggests online aggression in India spikes around major sporting, political and social events. Examine the implications for platform moderation and government communication policy.
What are the opportunities and risks of using AI to classify citizens' online speech as aggressive? Illustrate with reference to recent research on Indian social media data.

Grounded only in the source report — figures and dates are the source's, not inferred.

Next storyUS declines to intervene in Sheikh Hasina case, says officials →

← All stories