Inspiration
We all need a friend to help us keep calm about the ongoing crisis. With the advent of social distancing, there has been an increased need for companionship to help cope with the extreme isolation. Thus, we decided to create a program which would "talk" to the user, providing them some kind of companionship.
What It Does
Our project is a chatbot to help people cope with the isolation brought about by instances of social distancing, both for the present and the future. In its current state, it is capable of conversing to a user about their mental wellbeing, as well as their thoughts on the COVID-19 virus and how it is adversely affecting them.
How We Built It
To build our program, we started by utilizing the Tensorflow library for Python, which contains tools for creating a Deep Learning AI. We used Tensorflow to put our bot through a training loop to develop a neural network. In this loop, the bot would receive inputs from the user, then attempt to identify an appropriate response. To determine what prompts the AI would look out for, we had the training program refer to a .json file containing a list of topics, each with their own prompts and responses. Once we had completed the reference list, we trained the bot to develop its neural network by providing it inputs that were similar to the prompts on the .json file. Our initial version achieved a 50% success rate, and our final version is at a 98% success rate.
Challenges We Ran Into
We encountered a bug in Flask when trying to use TensorFlow. It may have been about the version and blocked the module from running. We fixed the issue after a lot of hair pulling and reinstalling all dependencies.
Accomplishments that we are proud of
Our crowning accomplishment was actually getting the neural network to properly function. As mentioned above, we ran into issues with our repository initially, causing Tensorflow to not be identified. Thus, when we finally got the program to work, it was quite satisfying.
What We Learned
For a start, we learned that neural networks are much more straightforward than initially expected. When our team initially began to work on this project, we were somewhat hesitant to create any kind of neural network, as we did not have any experience with such a system. However, we learned after some trial and error that neural networks are a lot more straightforward than expected to set up, as the only key componenents they need are a training program and a starting data set.
What's Next For COVID Companion
There are a multitude of changes we have in mind for the COVID Companion in the future. We want to expand the scope of our chatbot to cover a wider variety of topics. Another cool next step would be to pull stats in real time from Stats Canada to give people an easy way to keep track of the current situation. Finally, we would expand on our website, maybe adding a pledge page where people could donate to a local charity that helps people in their community survive through the pandemic.
Log in or sign up for Devpost to join the conversation.