Skip to content

mparoca/chatbot_D-590

Repository files navigation

Chatbot Created for D-590 FA2021

Chatbot for D-590 FA2021

screenshot

Description

QA chatbot is trained with roberta-base-squad2 model fine-tuned on WikiQA dataset. Code to prepare the data and train the model can be found in training_model_on_WikiQA.ipynb and is partly based on these two notebooks:

When the user enters a query, the wikipedia library is used to extract the initial paragraphs of a Wikipedia entry that matches the user query. The fine-tuned RoBERTa model is then used to select the location of the response that best matches the query given the context.

Deployment

App is currently deployed here

Plotly Dash

The app was built on Dash, which is a simple and effective way to bind a user interface around Python code.

GitHub Integration (Heroku GitHub Deploys)

The app is currently deployed using Heorku through GitHub integration on the Heroku website. GitHub integration is configured for a Heroku app and manually deployed, which creates an immediate deployment of the master branch from the current GitHub repo.


Deployed app is using pre-trained minimal model bert-medium-squad2-distilled. To use the model fine-tuned in training_model_on_WikiQA.ipynb or your own fine-tuned model change get_answer.py

Chatbot front-end partly based on: https://github.com/plotly/dash-sample-apps/tree/main/apps/dash-chatbot

Theme: https://bootswatch.com/vapor/ Icons: https://fontawesome.com/v5.15/icons?d=gallery&p=2

About

Chatbot code for D-590 FA2021

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published