Inspiration
After learning the LSTM, we are very interested in developing a chat bot that can automatically reply to people's question. We found a lot of questions that people asked in the slack of the SwampHacks are repeated, so we decided to build a chat bot that could automatically reply basic questions that are related to SwampHacks.
What it does
The chat bot will identify your intent and give you proper responses. It can greatly help people who are not familiar with Hackathon and give them immediate reply.
How we built it
We created a GUI leveraging Python tkinter to interact with user; we built a deep neural network by using TensorFlow; we manually created a json file to work as training data.
Challenges we ran into
Because of the limit training examples, our LSTM model tend to generate inaccurate response. However, we modified our model by replacing LSTM with DNN models, the correct rate increase a large amount.
Accomplishments that we're proud of
Even though we experienced a tough time to get correct result from our training data, we went through it and we finally did it. Also, we utilize the google-speech-to-text api to make our project have a more modern look and become more user-friendly, which is a brand-new experience for us during our hacking life.
What we learned
We have learned some new stuff rather than some skills we already have, like we learned some new knowledge about implementations of Tensorflow and training DNN models. Moreover, we learned how to work as a team together.
What's next for Chat Dancer
Our team believe there will be many scenes that could utilize the "Chatbot Dancer". So we think there is more room for our project to improve and if we have some more time, it would tend to be a complete project to implement.
Log in or sign up for Devpost to join the conversation.