Inspiration

-Several of our group members come off as the "heartless" friend in their friend groups, the one who says what they think and forgets to take into account the feelings of others. We recognize that this is a problem, but struggle with crafting more empathetic responses.

What it does

-EmPath seeks to remedy this problem by giving the user a concrete analysis of how their message relates to the emotional context of the conversation and how empathetic the message is. With this knowledge, a user can be better informed about how their message might affect the other people in the conversation.

How we built it

-EmPath uses an Android app interface called Flutter that connects to a Firebase database and runs Python code on Google Cloud Functions. -To analyze text the user inputs, we used the natural language processing library text2emotion, vector distance calculations, and conditional checks to reduce error.

Challenges we ran into

-Empathy is a quantity that is hard to measure or calculate, especially for us non-psychologists. -We spent time on a machine learning approach to calculate the empathy score using multi-class classification models, but the dataset, EmpatheticDialogues, contained primarily a score of 4 or 5 on an 1 to 5 empathy scale. This made it hard to get a meaningful model, since it was an unequal distribution.

Accomplishments that we're proud of

-We split the computational work to format our data set into a usable form for the machine learning algorithm and had testable models in less than 2 hours. -We coded our first Flutter app, without any prior experience with the platform

What we learned

-In machine learning, garbage in means garbage out. A skewed data set makes creating meaningful models almost impossible since it would be most successful

What's next for EmPath

-Finding a better data set to more effectively utilize a machine learning algorithm, which would likely provide more accurate empathy scores.

Built With

Share this project:

Updates