BotWise

Inspiration

Have you experienced a situation where you sent the message and later wish you could go back in time and not send it? Everybody was in this situation before!

What it does

The bot would help you make rational decisions rather than being controlled by your emotions, by analyzing the conversion environment and telling the user if the message should be sent. The bot will prevent you from burning relationship bridges and avoiding arguments with your friends, collegues, and family members.

How I built it

The bot itself was build in Node JS using Messenger API. To analyze the conversation and decide whether or not to send the message, we used BERT(Bidirectional Transformer) Model and Binary Sentiment Classification. The reason why our Sentimental Classification Model is Different from the others is because we trained the model ourselves, making sure it is tailored down to our needs.

Challenges I ran into

Finding the appropriate dataset to train our model. Linking together the Node JS file responsible for the bot and the Machine Learning model to predict the outcome. Saving the checkpoints and loading them to the model.

Accomplishments that I'm proud of

The Accuracy of over 93%. The time the model takes to analyze the general mood of the conversation. The quality of the teamwork.

What I learned

Became familiar with PyTorch and its models as well as learning the Messenger API and the process of creating a ChatBot

What's next for BotWise

Training it with a data set that has a broader range of emotions. The Bot would be able to not only tell if the message is suitable but also to modify it.