Breaking down Hate Speech

| Shikhar Mohan

Social media has percolated through our lives so much that it has become an everyday conduit of social interaction. But the pros aside, the accessibility of it all ends up being a double-edged sword, because, at the same time as giving people who didn’t have a platform to speak, a voice, it also gives the people who would have suppressed these voices a platform. This manifests itself in the form of hate speech, which is currently a huge problem in online discourse. It directly or indirectly attacks marginalized communities, affecting legitimate discourse at best and hurting people and inciting more hate at worst. The severity of this issue makes it important to analyze hate speech.

This needs to be done with multiple dimensions since these can give us a good idea regarding detection and large scale prevention of hate speech. One such research work is that of Dr. Mainack Mondal of IIT Kharagpur, presented in 25th ACM Conference on Hypertext and Social Media in Prague.

(You can find Dr. Mainack Mondal’s website here and this research work here)

So what did they find?

Dr. Mainack Mondal has worked on a multivariate analysis of hate speech as it exists in social media, namely four dimensions: targets, correlation with anonymity, the geography and context of hate speech on two platforms: Twitter and Whisper.

1. Targets: They usually tend to be strongly correlated with race, behavior, race, sexual orientation and other parameters which form the basis of hate and bigotry outside social media as well. It can be seen as a reminder about the severity of this issue, that hate speech targets have a strong correlation with hate crime targets, according to an FBI database from 2013 and 2014.
2. Correlation with anonymity: The general trend in twitter seems to be to preserve anonymity, but there seems to be a significantly larger proportion of tweets from anonymous users when considering hate speech, as compared to tweets in general.
3. Geography: An interesting observation was made regarding the targets of hate speech across countries. Different countries tend to have different general targets of hate speech, e.g. The US tends to have more hate speech regarding race compared to Canada or the UK.
4. Context of hate speech: It’s not just targets and the direct mechanics of hate speech that need to be considered. To properly analyze hate speech to the extent of being able to tangibly mitigate it requires an analysis of context as well.

Bottom line

We see that hate speech is a huge problem which needs tackling to improve online discourse and help to curb polarizing narratives which harm communities, and to be able to solve this problem we first need to thoroughly analyze how hate speech occurs. Easy access to these posts and tweets makes analysis easy and it gives us important directions in which work needs to be done to counter such activities.

This is one step forward for detection and early prevention of hate speech online and with some more steps, hate crimes in the offline world, and one small step towards a more harmonious world in general.