ChatPDT

• 

  capture

Inspiration

We found this topic really interesting after reading through the write-up. We always use chatbots in life and have our preference, but we never think deeply into this question- what make the chatbots' response good.

What it does

We built a machine learning model to learn from users' judgments on which is a better response to predict users' preferences.

How we built it

We started with solid data cleaning with the nltk library and Pandas. We creatively analyzed the data and figured out some important metrics useful for indicating users' preferences. We made use of metrics, dataset given, and various Python machine learning libraries like Hugging Face, sklearn, and transformer to train our models.

Challenges we ran into

We had trouble figuring out the correct workflow and the correct way to collaborate since we are all new to Datathon. Our work once got stuck because of slow progress in figuring out the metrics. We also faced some hardware bottleneck too.

Accomplishments that we're proud of

Even though we are all new to Datathon and to collaborating with each other, we nailed down all the work, and our project works decently now.

What we learned

We learned how to work in data project in general, how to collaborate efficiently, and how to face difficulties.

What's next for ChatPDT

We are going to deploy a better tensorflow model which jumps out of 3-class classification problem and does a pairwise-rank solution instead.

Built With

• python