Scientists are developing an AI system to counter hate speech online


A group of scientists has developed an artificial intelligence system that can help counter hate speech to disadvantaged minorities such as the Rohingya community, where the system developed by researchers from Carnegie Mellon University in the United States can quickly analyze thousands of comments on social media, and researchers say that moderators On social media who have not been able to check many comments manually, they will have the option to highlight this “Help Letter” in the comments sections.

According to the Indian TOI website, a researcher at the Carnegie Mellon Institute of Language Technology at Carnegie Mellon University said: “Even if there is a lot of obnoxious content, we can still find positive comments.” The researchers said that finding and highlighting these positive comments may Much is being done to make the Internet a safer and healthier place like discovering and eliminating offensive content or banning responsible hunters.

They said that the Rohingya, who started fleeing Myanmar in 2017 to avoid ethnic cleansing, are helpless against hate speech on the Internet, and many of them have limited skill in global languages ​​like English, and they have little access to the Internet, and KhudaBukhsh said most of them Very busy trying to stay alive to spend a lot of time posting their content.

To find the relevant help letter, researchers used their search method in more than a quarter of a million YouTube comments on what they believe to be the first artificial intelligence-focused analysis of the Rohingya refugee crisis, said Jaime Carbonell, LTI director and co-author of the study: The ability to analyze such These large amounts of text for content and opinion is possible due to recent major improvements in language models.

These models learn from examples so that they can predict possible words in a particular sequence, and help machines understand what speakers and writers are trying to say.However, researchers have developed additional innovation that has enabled these models to be applied to short social media scripts in South Asia, as it is difficult for The machines interpret short parts of the text, often with spelling and grammatical errors.

It is also more difficult in South Asian countries, where people may speak several languages ​​and tend to “switch”, combining different parts of languages ​​and even different writing systems in the same statement.

The researchers have explained that the existing machine learning methods create a representation of words or word motifs, so that all words of similar meaning are represented in the same way, as this technique makes it possible to calculate the proximity of a word to others in a comment or publication, and to extend this technique to include South Asian texts Difficult, the team obtained new weddings that revealed linguistic groups or groups, and the researchers said this language-determining technique worked well or better than commercially available solutions.

Carbonell indicated that this innovation has become a favorable technique for social media arithmetic analysis in that region, samples from YouTube comments have shown that about 10 per cent of comments were positive, and when researchers used their method to search for help speech in the larger data set, the results were positive at 88 percent, which indicates that this method can significantly reduce the manual effort needed to find them, says Khouda Bakhsh.


Please enter your comment!
Please enter your name here