Following the memory network and bAbI dataset introduced in Lesson 11, Part 2, I found an interesting machine comprehension dataset SQuAD. Most of the SQuAD models are too complicated to implement within a week, but I found this paper Reading Wikipedia to Answer Open-Domain Questions and a partial implementation in Facebook/ParlAI. This fascinating model is conceptually simpler than most others but still has strong performance. So I decided to turn the partial implementation into a full one and upload the project to GitHub (hitvoice/DrQA). I think it’s a good choice for student projects if you want to try something in machine comprehension. Besides, it’s my first GitHub repo and I’m not sure whether I set everything right. Any suggestion is welcomed!
Awesome - will clone and try it out when I get a sec