bigshanedogg / survey

2 stars 0 forks source link

[DPR] Dense Passage Retrieval for Open-Domain Question Answering #16

Open bigshanedogg opened 1 year ago

bigshanedogg commented 1 year ago

Problem statement

  1. 적은 양의 (question, passage) pair로 retrieval 성능을 낼 수 있는 training scheme을 찾는다.
  2. question과 passage의 내적으로 유사도를 비교할 수 있는 low-dimensional & continuous space에 임베딩한다.

Baseline

Data details

name abbr type format source size description remark related tasks
Natural Questions NQ text Wikipedia open-domain question answering
TriviaQA text Web(scraped) question answering
WebQuestions WQ text Freebase question answering
CuratedTREC TREC text Web open-domain question answering
SQuAD v1.1 text Wikipedia question answering

Approach

Evaluation

image

Ablation & Tricks

image

Limitations

참고 & 개인적인 생각

bigshanedogg commented 1 year ago

2004.04906.pdf