IIT Patna CSE team working on curbing hate speech online. First stop: Twitter

NEW DELHI: It works through a browser extension. Once installed on a web browser, it can scan a page to detect “hate speech” and prevent its propagation. The software module called Hate, Hyperpartisan, and Hyperpluralism Elicitation and Observer System – or, HELIOS – was built by the AI-NLP-ML Group in the department of computer science and engineering (CSE) at Indian Institute of Technology (IIT) Patna, in collaboration with Indraprastha Institute of Information Technology (IIIT) Delhi and University of Texas Austin.

Still being refined, the tool uses Artificial Intelligence (AI) Natural Language Processing (NLP) and Machine Learning (ML) – the terms from which the IIT Patna research group draws its name – to identify hate on a page of text, including social media posts, and sanitise it.

“With the rise in hate speech issues in India, how do you detect the offensive content and ensure it is stopped at that time? If you let hate speech propagate, there can be serious situations in maintaining law and order,” said Asif Ekbal who is leading the project at IIT Patna . “Lately, the amount of hate or offensive comments targeting certain people has increased tremendously. The question is, how do you detect this because 50 million tweets or posts are generated daily globally. Out of this, two-to-three percent is hate content. How do you filter it out so that you can control the offensive content on the platforms?”

Also read | Vedic Maths, Astrology, Anti-Vaxxers: This engineer calls out pseudoscience on YouTube

IIT Patna: HELIOS and hate

The IIT Patna team started planning the project in 2019 and was looking for collaborations. In 2020, it found a partner in IT major Wipro to build a commercially viable product to solve the problem of hate speech. “Hate speech is always targeted at someone or something and we wanted to identify this content by using AI-NLP-ML-based technique, developed at IIT P,” said Ekbal.

The definition of hate varies geographically, explained Ekbal. “It depends on the country, language, and other factors,” he said. “Currently, we are developing the research for Hindi and English. Our main goal is to develop solutions for Hindi and English that can automatically detect hate speech from social media.” A large amount of data – “offensive content” – is first entered into that system so that it learns to detect hate speech or offensive content in the browser. “This is done through an algorithm based on AI and NLP-ML technology,” said Eqbal. “Once detected, it automatically converts offensive into non-offensive content. This programme is applicable to websites or any social media platform, blog, etc. Currently, we are working on just tweets. It takes text as an input and then detects offensive content and converts automatically. It will work as an extension on the browser.”

The main purpose of HELIOS is to stop such content and hinder its further propagation. It is intended to stop the post from being forwarded. This also prevents propagation of fake news.

Also read | MBBS Admission: Why NEET cutoffs vary widely for medical colleges at close NIRF ranks

AI, machine learning

“In AI and machine learning models, machines are taught to bifurcate between hate and non-hate content. It cannot automatically detect such data and this data creation and further development on that data needs a lot of money,” said Ekbal. Wipro has funded the project, putting Rs 42 lakh into the consortium’s work for three years, according to Ekbal’s website.

The project consists of three parts: data creation , research, and building the product through engineering . Ekbal noted: “Wipro is giving us funding in terms of data creation, manpower and building the product. After we complete our research and prototype, Wipro will be fine-tuning the research and creating an application or product out of it.”

Furthermore, research and products that have a larger impact and solve a socially-relevant problem, are relevant to companies and industry because they want such solutions. “Their ultimate goal is commercialisation and through this project, they aspire to prevent not only hate speech but ensure commercialisation as well. Through our research, Wipro will be building a product for its client,” said Ekbal.

Also read | ‘To affirm constitutional politics’: What are DU prof Ratan Lal's posts on YouTube about?

Software engineering students

Supervised by Ekbal, the AI-NLP-ML is a thriving research group and this project has afforded training and exposure to a large number of students.

“This project includes client-interaction which exposes students to marketing and interaction. Also, students are empowered to solve a problem which is relevant to society and industry as such technologies have several applications,” said Ekbal.

Development of the technology also involves ideation and exchange of ideas. “Students can exchange their ideas with other researchers, and international organisations. Also, as students are solving the problem by being a part of the team, it makes them more professional,” said Ekbal.

Ekbal said the team plans to extend the project beyond December 23, when the three years of the project will be up, and develop the same solution for other languages.

Also read | 35% male dropouts ‘not interested in studies’: What NFHS 5 says about school education

“We started in 2020 and we have developed a prototype and research papers. We have used the core software of Wipro and they are building a kind of product by using our research submissions,” he said.