Help new Ubuntu users to get familiar with the environment with a chatbot

What it does

Ubuntu Helpdesk is a virtual chatbot assistant that can answer users' question about the Ubuntu system.

How we built it

We trained the chatbot using a dialogue corpus of dialogues related to Ubuntu technical support and deployed it to Heroku. We used the Seq2seq RNN model from tensorflow for training the bot and the Compute Engine on the Google Cloud Platform to facilitate the training process. Finally, we made the interface for the chatbot on a Flask app and deployed it to Heroku.

Challenges we ran into

There were too much slang in the corpus, which caused our bot to use 'lol', 'thanks', and emojis very often. We faced a lot of technical errors while setting up the compute engine and the heroku app. It was our first time creating a chatbot purely by training from corpus data, so we had to learn a lot. We couldn't use GPU for training the model, so trainings were extremely slow and was with limited data, causing the results to be not as ideal.

Accomplishments that we're proud of

We have a bot that can talk! (And we got it working through machine learning!!)

What we learned

How to use tensorflow for chatbot tasks How to set up and use a virtual machine on the google cloud platform

What's next for Ubuntu Helpdesk

More training time and data so that the chatbot can perform better

Share this project: