Happy2Help Chatbot

Samsung Innovation Campus IA bootcamp final project. Málaga University.

Team Members:

• Miriam Andrés: @miriandres
• Jesica Míguez: @jemrosas
• Sandra Orengo: @sof1508

Built with 🛠️

Project:

Happy2Help is a conversational interface with automated text interactions that recognizes behavioral patterns and gives advice for those who are struggling. Built as a natural language understanding app in Flask API, using a jinja2 template and JavaScript.

Tweets related to depression issues categorized from a labeled dataset gathered from Twitter API’s service using. TF-IDF weighting to classify their polarity.

User Interface (UI)

Workflow

System Diagram

- app.py                    Graphical user interface
- chat.py                   Building a chatbot framework
- intents.json              Contains every reply under each intent in JSON format
- model.py                  Deep learning components for chat interaction
- model_sentiments.h5       Deep learning model for sentiments
- nltk_utils.py             Natural language processing with nltk
- diccionario.json          Contains the dictionary of words
- train.py                  Training

Flowchart

Training Methodology

Model

Initial Setup:

a. Basic version

Clone repo and create a virtual environment

$ git clone https://github.com/miriandres/Happy2Help-chatbot.git
$ cd HAPPY2HELP-CHATBOT
$ python3 -m venv venv
$ . venv/bin/activate

Install dependencies

$ (venv) pip install Flask torch torchvision tensorflow
$ (venv) pip install numpy nltk

Install nltk packages

$ (venv) python
>>> import nltk
>>> nltk.download('punkt')
>>> nltk.download('wordnet')

Run

$ (venv) python train.py

This will dump data.pth file. And then run the following command to test it in the console.

$ (venv) python app.py

b. Conda version

$ git clone https://github.com/miriandres/Happy2Help-chatbot.git
$ cd HAPPY2HELP-CHATBOT
$ conda env create -f environment.yml