Skip to content

Latest commit

 

History

History
12 lines (10 loc) · 833 Bytes

README.md

File metadata and controls

12 lines (10 loc) · 833 Bytes

Question-Answering

This project is aiming to find an answer for questions when a document is given.

Files in the data directory

  • documents.jason : A list of documents. Each item is a tuple of a document_id and a list of paragraphs.
  • training.jason & devel.jason : A list of questions along with the answers. Each item is in the format of (question, answer, answer_paragraph, document_id).
  • testing.jason : A list of questions only. Each item is in the format of (question, document_id, id).
  • random.csv : Answer for the questions in the testing.jason file.

Methods

  • Infomation Retrival: find answers based on NER.
  • Seq2Seq models: one with GRU, one with LSTM. Both locate the answer paragraph/sentence first then predict the answer by approxiamting the function (answer para/sents, question)->answer.