Anbot - Step 1
Week 6
For my thesis, I am working on community-building. In one of my research directions, I explored the need for ways people with different points of view can have productive arguments. Most popular social media platforms are designed to maximize engagement, which is not the ideal avenue to have these discussions. For my bot assignment, I wanted to explore how a bot could help create a more supportive environment for discussions among people with different points of view around an issue.
​
I started reading on the topic and found a few interesting points about arguments and heated debates on social media platforms. Some were:
-
People don't feel safe admitting to mistakes on pubic platforms
-
In-group and outgroup dynamics of the people participating and observing
-
The focus is on winning the argument than it is about listening and learning
-
Participants try to persuade each other rather than listen to each other
-
-
Human element being removed (people may say things on digital platforms that they might not say when face to face with someone)
​
While ideating for how the bot could help with creating a better environment for arguments on existing platforms, it was important to me to make sure that the bot does not intervene in ways that it becomes a part of the conversation and only intervenes in 'non-participating' ways.
​
While ideating, I came across this article that talks about the different kinds of disagreements on the internet. [link]
I have come across name-calling while reading arguments between people on different social issues online. I found it interesting how the author of the article talks about how some ways of disagreement are unproductive and maybe pointing it out might help people walk away from conversations where the participants are not putting in much thought and effort. But also, on the other hand, pointing out unproductive forms of disagreement to the participant's writing it, might lead them to rethink their response.
​
I thought that a bot that could detect name-calling in chats and call it out in a kind way could be an interesting first step. I also named the bot Anbot [Anbu meaning kind in Tamil which is my native language + bot]
​
My first step to creating Anbot is to detect a word (idiot) and writing a response to it. I thought this would be an easy way to start building Anbot.
​
My next step is to make Anbot detect name-calling. I couldn't find a dataset that had words generally used for name-calling on the internet. I tried to find datasets used to detect hate speech instead. Since these datasets have scores, I want to find a way it can be used to detect name-calling and unkind messages.