Anbot
For my thesis, I am focusing on community-building. One research direction explores how people with differing viewpoints can engage in productive arguments. Most popular social media platforms are designed to maximize engagement, which often undermines meaningful discussions. For my bot assignment, I sought to explore how a bot could help foster a more supportive environment for dialogue among individuals with opposing views.
​
In my research, I uncovered several key challenges surrounding arguments and heated debates on social media:
-
People feel unsafe admitting mistakes in public settings.
-
In-group and out-group dynamics shape both participants and observers.
-
The focus tends to be on "winning" the argument rather than listening and learning.
-
Participants aim to persuade rather than understand each other.
-
The human element is diminished, with users often saying things online they wouldn’t say face-to-face.
-
​
As I ideated on how the bot could create a healthier space for arguments on existing platforms, I emphasized ensuring that the bot doesn’t interfere by becoming part of the conversation but instead intervenes in subtle, non-participatory ways.
​
During my research, I also came across an article discussing various types of online disagreements, which further informed my thinking. [link]
While reading online arguments on social issues, I frequently encountered name-calling. The author of an article I came across highlighted how certain forms of disagreement are unproductive. I found it intriguing that pointing these out could either encourage people to disengage from unproductive conversations or prompt the offenders to rethink their responses.
​
This inspired me to create a bot that detects name-calling in conversations and gently calls it out. I named the bot Anbot—"Anbu" meaning "kind" in Tamil, my native language. My first step was to program Anbot to detect a word like "idiot" and respond appropriately, as a simple starting point.
​
Next, I focused on expanding Anbot’s capabilities to detect a broader range of name-calling. However, I couldn’t find an existing dataset specifically for name-calling. Instead, I explored datasets used to detect hate speech, aiming to adapt them for identifying name-calling and unkind messages based on scoring metrics.
​I also wanted to create a way for people in servers with many people to connect with each other. I made a Bot called ConnectBot that goes through the message history of the server and matches your last message with others in the server with similar interests that one could potentially connect with using embeddings.
