Kind Konversations

Week 6

The idea

For my finals this semester, I would like to make a digital space designed for people to have productive arguments about different topics.

For my programming from A-Z midterms, I created a bot that detects unkind messages on Discord and asks the sender to consider reframing their messages to convey the same idea or thoughts in a kinder way. [link]

For my research on online hate and arguments, I found two elements related to time.

1) Reacting vs Responding to a comment or a message on social media. Spending time to understand and reflect and respond to them rather than react to a message

2) How active is the thread/how frequently people are participating in the discussion and how many?

While I was reading about arguments on the internet, I found an interesting way different kinds arguments can be categorized

Name-calling was an interesting problem to target since it seems to be a reaction without much or any thought rather than a response.

I used the AFINN-111 Sentiment Analysis to detect name-calling. The main idea here is that when there are a lot of attack in a chat or too much name-calling, people might not be taking the time to try to understand and respond to each other. But they are only throwing their ideas and thought at each other.

This technique I used for the first step for my mid-terms has limitations. The 'negative words' are detected based on scores given to individual words and without understanding the context. For example, the phrase 'terribly cute' is detected as a negative comment. I hope to build on this further in the coming weeks to detect unkind messages or comments better.

For my mid-terms, I decided to use this method to detect name-calling to visually represent a topic being discussed (a thread or a comment section for example). Slower movement of letters would be less name-calling and faster would be more name-calling to represent the thoughtlessness behind these kinds of arguments.

It is currently set to a random speed but eventually, when I build something that can detect sentiment of multiple comments, I would try to understand how time can be mapped to these comments. And add another visualization to represent the frequency of responses on the thread or comment section.

The final visualization would be multiple nodes representing different topics. This visual representation could help a person potentially thinking of reading or participating in an argument or discussion to decide if they wish to join.