/WODQA-CHT

WODQA-CHT : A Dataset and Baselines for Traditional Chinese Wikipedia Open Domain QA

Primary LanguagePython

WODQA-CHT

WODQA-CHT and its Generator for Traditional Chinese Open Domain Question Answering

The dataset download at here

Col Description
1 question Question
2 qitm Question keywords
3 article Wiki Article
4 aitm Article keywords
5 answer Answer

setup

pip install -q -r requirement.txt

Inverted Index for wiki Corpus

python -m pyserini.index.lucene \\ <br>
  --collection JsonCollection \\  <br>
  --input corpus \\  <br>
  --language zh \\  <br>
  --index Index/Wiki_Chinese \\  <br>
  --generator DefaultLuceneDocumentGenerator \\  <br>
  --threads 1 \\  <br>
  --storePositions --storeDocvectors --storeRaw

Running

Naive QG without filter

python main.py -t naive

Entity pass filter-1

python main.py -t f1

Question pass filter-1&2

python main.py -t f2

Full filter passed

python main.py -t all

Framework

Baseline Sore

Dataset EM F1 Avg.Len(q)
Without filter 29.30 % 26.90 % 21.58
F1 filter 32.23 % 30.77 % 21.17
F1+F2 filter 35.40 % 34.35 % 26.25
F1+F2+F3 filter 61.80 % 57.30 % 26.73

EM : Exact Match
F1 : F1 score
Avg. Len(q) : The avg length of questions