WhatTONE

Inspiration

Did you ever wonder how emotions evolved during a chat conversation? We wanted to solve this question by building a tool that does a sentiment analysis on provided chat dialogues.

What it does

The website allows the user to input a chat conversation and submit it for analysis. It runs a R script for data preprocessing and a python script for sentiment analysis in the background. Finally, it returns the detected sentiments in each message.

How we built it

We used R to preprocess the data (e.g. removing media messages and notifications), to cluster messages that were sent in short intervals and to collect these clustered messages for sentiment analysis. We used the vaderSentiment package in python to perform the sentiment analysis. This returns a probability for a given message to be positive, negative, neutral as well as a combination of these features.

Challenges we ran into

The biggest challenge was to find an analysis software that detects sentiments in common speech used on social media. After testing around 10 different APIs and packages in R and python we picked the one that throughout shows matching results for chat messages.

Accomplishments that we're proud of

We managed to combine different coding languages to form a fast framework for sentiment analysis and build a working platform that users can access.

What we learned

We learned a lot about the struggles with sentiment analysis which strongly depends on the training data. Most available software was trained on written language of well known books but cannot detect sentiments of slang.