The project investigates using a Mixture Model to build an IR4QA system that recommends an answer to a natural language question from precompiled list of question-answer pairs.
Our dataset for training is Yahoo! Answers Comprehensive Questions and Answers version 1.03 which consists of more than 4 million question-answer pairs provided by Yahoo Labs on the WebScope site.
We describe the system, results, and some of the key takeaway lessons.
- paper.pdf - Results and resources to download the dataset.
- ir4qa.ipynb - Jupyter notebook for exploration.
- server.py - Flask server.
- gui/ - HTML/CSS/JS that invokes RESTful APIs exposed by
server.py
Sample questions to ask:
- How do I get rid of stomach ache?
- What's the meaning of life?
- How to lose weight?
- I have long thick hair. Is that pretty?
- One side of my body freezes. What should I do?