Given a Wikipedia article, generate N "good" questions and answer N questions. We won the 1st place in question-answering competition! See how we make it at Animorphemes FinalVideo or QA system - Question Generation
This is a semester-long project for CMU 11611. We have a team of 4 people. The original code is on https://github.com/hexiaoyuhaha/Wiki-Question-Answering-System.
It can generate and answer yes-no questions and wh-questions like what/when/where/how/how many(much)/why.
I'm still working on it and trying to improve the performance.
- python==2.7
- spaCy==1.7.5
- pattern==2.6
- textblob==0.12.0
- nltk==3.2.2
- requests==2.13.0
- scipy==0.18.1
- sklearn==0.18.1
- numpy==1.11.3
- stanford-corenlp-full-2016-10-31
-
stanford-corenlp-full-2016-10-31/: I'm working on python wrapper now. Hopefully it will be released within 2 weeks.Currently, using the following command to run the server on port 3456,
java -mx4g -cp "*" edu.stanford.nlp.pipeline.StanfordCoreNLPServer -port 3456 -timeout 15000
-
S10/: contains sample wikipedia dataset
article folder contains the sample articles, questions and answers
data folder contains only articles in htm, txt format
-
data/: contains training data and model information for answer type detection
-
.py files will be reorganized in the near future
- install the stuff the requirement mentions, and run the stanford corenlp server on port 3456.
-
./run-ask.sh filename nquestions
-
./run-answer.sh filename questions