Kaiwa Sensei is a Sentiment Analysis Facebook Messenger bot that displays the conversation's mood during the conversation.
It was developed as part of the Facebook Developer Circles London: Building Conversational AI for Messenger Hackathon (07/12/19-08/12/19), by UCL students Shubham, Dylan, Tsz, Gabriel, Ana.
See the DevPost for more background on our project!
The project is split into 2 parts: a web chat interface and a Pytorch-backed ML model for the Sentiment Analysis.
The web chat interface and Messenger chat extension bot are developed in Node.JS and JavaScript.
The Sentiment Analysis is done in Python using Pytorch and Deepmoji. Note that the Python/Pytorch implementation library has been patched within this repo to make it work for the latest version of Python 3.
- The
./chat-webviewdirectory contains the client-side code for the chat application. - The
./nlp-emotionsdirectory contains the patched version of TorchMoji. The./nlp-emotions/examples/single_text_emotions.pyfile contains the code used by the bot to determine a sentence's mood - The
./word-datasetsdirectory contains word datasets used in the project
The ./nlp-emotions/examples/single_text_emotions.py Python script uses the emojis results to determine what the overall mood of a sentence is. The script returns a JSON string containing the fields positive (the positive words in the sentence), negative (the negative words in the sentence), user_emotion (positive and negative), and main_vibe (an emoji that describes the overall mood of the conversation).
The index.js file is the main server file that manages the Node.JS/Socket.IO HTTP server/websocket server for the client to talk to.
The code is not complete due to the nature of a Hackathon. It should not be used in production or sensitive environments. The bot will be disabled after the Hackathon so any tokens in the repo will be obsolete.