Entity-Based Knowledge Conflicts in Question Answering.

Overview

Entity-Based Knowledge Conflicts in Question Answering

Run Instructions | Paper | Citation | License

This repository provides the Substitution Framework described in Section 2 of our paper Entity-Based Knowledge Conflicts in Question Answering. Given a quesion answering dataset, we derive a new dataset where the context passages have been modified to have new answers to their question. By training on the original examples and evaluating on the derived examples, we simulate a parametric-contextual knowledge conflict --- useful for understanding how model's employ sources of knowledge to arrive at a decision.

Our dataset derivation follows two steps: (1) identifying named entity answers, and (2) replacing all occurrences of the answer in the context with a substituted entity, effectively changing the answer. The answer substitutions depend on the chosen substitution policy.

Run Instructions

1. Setup

Setup requirements and download SpaCy and WikiData dependencies.

bash setup.sh

2. (Optional) Download and Process Wikidata

This optional stage reproduces wikidata/entity_info.json.gz, downloaded during Setup.

Download the Wikidata dump from October 2020 here and the Wikipedia pageviews from June 2, 2020 here.

NOTE: We don't use the newest Wikidata dump because Wikidata doesn't keep old dumps so reproducibility is an issue. If you'd like to use the newest dump, it is available here. Wikipedia pageviews, on the other hand, are kept around and can be found here. Be sure to download the *-user.bz2 file and not the *-automatic.bz2 or the *-spider.bz2 files.

To extract out Wikidata information, run the following (takes ~8 hours)

python extract_wikidata_info.py --wikidata_dump wikidata-20201026-all.json.bz2 --popularity_dump pageviews-20210602-user.bz2 --output_file entity_info.json.gz

The output file of this step is available here.

3. Load and Preprocess Dataset

PYTHONPATH=. python src/load_dataset.py -d MRQANaturalQuestionsTrain -w wikidata/entity_info.json.gz
PYTHONPATH=. python src/load_dataset.py -d MRQANaturalQuestionsDev -w wikidata/entity_info.json.gz

4. Generate Substitutions

PYTHONPATH=. python src/generate_substitutions.py --inpath datasets/normalized/MRQANaturalQuestionsTrain.jsonl --outpath datasets/substitution-sets/MRQANaturalQuestionsTrain
   
    .jsonl 
    
      -n 1 ...
PYTHONPATH=. python src/generate_substitutions.py --inpath datasets/normalized/MRQANaturalQuestionsDev.jsonl --outpath datasets/substitution-sets/MRQANaturalQuestionsDev
     
      .jsonl 
      
        -n 1 ...

      
     
    
   

See descriptions of the substitution policies (substitution-commands) we provide here. Inspect the argparse and substitution-specific subparsers in generate_substitutions.py to see additional arguments.

Our Substitution Functions

Here we define the the substitution functions we provide. These functions ingests a QADataset, and modifies the context passage, according to defined rules, such that there is now a new answer to the question, according to the context. Greater detail is provided in our paper.

  • Alias Substitution (sub-command: alias-substitution) --- Here we replace an answer with one of it's wikidata aliases. Since the substituted answer is always semantically equivalent, answer type preservation is naturally maintained.
  • Popularity Substitution (sub-command: popularity-substitution) --- Here we replace answers with a WikiData answer of the same type, with a specified popularity bracket (according to monthly page views).
  • Corpus Substitution (sub-command: corpus-substitution) --- Here we replace answers with other answers of the same type, sampled from the same corpus.
  • Type Swap Substitution (sub-command: type-swap-substitution) --- Here we replace answers with other answers of different type, sampled from the same corpus.

How to Add Your own Dataset / Substitution Fn / NER Models

Use your own Dataset

To add your own dataset, create your own subclass of QADataset (in src/classes/qadataset.py).

  1. Overwrite the read_original_dataset function, to read your dataset, creating a List of QAExample objects.
  2. Add your class and the url/filepath to the DATASETS variable in src/load_dataset.py.

See MRQANaturalQuetsionsDataset in src/classes/qadataset.py as an example.

Use your own Substitution Function

We define 5 different substitution functions in src/generate_substitutions.py. These are described here. Inspect their docstrings and feel free to add your own, leveraging any of the wikidata, derived answer type, or other info we populate for examples and answers. Here are the steps to create your own:

  1. Add a subparser in src/generate_substitutions.py for your new function, with any relevant parameters. See alias_sub_parser as an example.
  2. Add your own substitution function to src/substitution_fns.py, ensuring the signature arguments match those specified in the subparser. See alias_substitution_fn as an example.
  3. Add a reference to your new function to SUBSTITUTION_FNS in src/generate_substitutions.py. Ensure the dictionary key matches the subparser name.

Use your own Named Entity Recognition and/or Entity Linking Model

Our SpaCy NER model is trained and used mainly to categorize answer text into answer types. Only substitutions that preserve answer type are likely to be coherent.

The functions which need to be changed are:

  1. run_ner_linking in utils.py, which loads the NER model and populates info for each answer (see function docstring).
  2. Answer._select_answer_type() in src/classes/answer.py, which uses the NER answer type label and wikidata type labels to cateogrize the answer into a type category.

Citation

Please cite the following if you found this resource or our paper useful.

@misc{longpre2021entitybased,
      title={Entity-Based Knowledge Conflicts in Question Answering}, 
      author={Shayne Longpre and Kartik Perisetla and Anthony Chen and Nikhil Ramesh and Chris DuBois and Sameer Singh},
      year={2021},
      eprint={2109.05052},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

License

The Knowledge Conflicts repository, and entity-based substitution framework are licensed according to the LICENSE file.

Contact Us

To contact us feel free to email the authors in the paper or create an issue in this repository.

Owner
Apple
Apple
Extreme Rotation Estimation using Dense Correlation Volumes

Extreme Rotation Estimation using Dense Correlation Volumes This repository contains a PyTorch implementation of the paper: Extreme Rotation Estimatio

Ruojin Cai 29 Nov 18, 2022
Repository for the paper : Meta-FDMixup: Cross-Domain Few-Shot Learning Guided byLabeled Target Data

1 Meta-FDMIxup Repository for the paper : Meta-FDMixup: Cross-Domain Few-Shot Learning Guided byLabeled Target Data. (ACM MM 2021) paper News! the rep

Fu Yuqian 44 Nov 18, 2022
Spam your friends and famly and when you do your famly will disown you and you will have no friends.

SpamBot9000 Spam your friends and family and when you do your family will disown you and you will have no friends. Terms of Use Disclaimer: Please onl

DJ15 0 Jun 09, 2022
📖 Deep Attentional Guided Image Filtering

📖 Deep Attentional Guided Image Filtering [Paper] Zhiwei Zhong, Xianming Liu, Junjun Jiang, Debin Zhao ,Xiangyang Ji Harbin Institute of Technology,

9 Dec 23, 2022
RMTD: Robust Moving Target Defence Against False Data Injection Attacks in Power Grids

RMTD: Robust Moving Target Defence Against False Data Injection Attacks in Power Grids Real-time detection performance. This repo contains the code an

0 Nov 10, 2021
Code for CVPR 2021 paper: Anchor-Free Person Search

Introduction This is the implementationn for Anchor-Free Person Search in CVPR2021 License This project is released under the Apache 2.0 license. Inst

158 Jan 04, 2023
The code for our paper CrossFormer: A Versatile Vision Transformer Based on Cross-scale Attention.

CrossFormer This repository is the code for our paper CrossFormer: A Versatile Vision Transformer Based on Cross-scale Attention. Introduction Existin

cheerss 238 Jan 06, 2023
Dataloader tools for language modelling

Installation: pip install lm_dataloader Design Philosophy A library to unify lm dataloading at large scale Simple interface, any tokenizer can be inte

5 Mar 25, 2022
Code for "Steerable Pyramid Transform Enables Robust Left Ventricle Quantification"

Code for "Steerable Pyramid Transform Enables Robust Left Ventricle Quantification" This is an end-to-end framework for accurate and robust left ventr

2 Jul 09, 2022
Trading Strategies for Freqtrade

Freqtrade Strategies Strategies for Freqtrade, developed primarily in a partnership between @werkkrew and @JimmyNixx from the Freqtrade Discord. Use t

Bryan Chain 242 Jan 07, 2023
Authors implementation of LieTransformer: Equivariant Self-Attention for Lie Groups

LieTransformer This repository contains the implementation of the LieTransformer used for experiments in the paper LieTransformer: Equivariant self-at

35 Oct 18, 2022
An essential implementation of BYOL in PyTorch + PyTorch Lightning

Essential BYOL A simple and complete implementation of Bootstrap your own latent: A new approach to self-supervised Learning in PyTorch + PyTorch Ligh

Enrico Fini 48 Sep 27, 2022
Joint Unsupervised Learning (JULE) of Deep Representations and Image Clusters.

Joint Unsupervised Learning (JULE) of Deep Representations and Image Clusters. Overview This project is a Torch implementation for our CVPR 2016 paper

Jianwei Yang 278 Dec 25, 2022
(CVPR2021) ClassSR: A General Framework to Accelerate Super-Resolution Networks by Data Characteristic

ClassSR (CVPR2021) ClassSR: A General Framework to Accelerate Super-Resolution Networks by Data Characteristic Paper Authors: Xiangtao Kong, Hengyuan

Xiangtao Kong 308 Jan 05, 2023
Learning Lightweight Low-Light Enhancement Network using Pseudo Well-Exposed Images

Learning Lightweight Low-Light Enhancement Network using Pseudo Well-Exposed Images This repository contains the implementation of the following paper

Seonggwan Ko 9 Jul 30, 2022
Benchmarks for Model-Based Optimization

Design-Bench Design-Bench is a benchmarking framework for solving automatic design problems that involve choosing an input that maximizes a black-box

Brandon Trabucco 43 Dec 20, 2022
PSTR: End-to-End One-Step Person Search With Transformers (CVPR2022)

PSTR (CVPR2022) This code is an official implementation of "PSTR: End-to-End One-Step Person Search With Transformers (CVPR2022)". End-to-end one-step

Jiale Cao 28 Dec 13, 2022
DLWP: Deep Learning Weather Prediction

DLWP: Deep Learning Weather Prediction DLWP is a Python project containing data-

Kushal Shingote 3 Aug 14, 2022
The implementation of 'Image synthesis via semantic composition'.

Image synthesis via semantic synthesis [Project Page] by Yi Wang, Lu Qi, Ying-Cong Chen, Xiangyu Zhang, Jiaya Jia. Introduction This repository gives

DV Lab 71 Jan 06, 2023
A bunch of random PyTorch models using PyTorch's C++ frontend

PyTorch Deep Learning Models using the C++ frontend Gettting started Clone the repo 1. https://github.com/mrdvince/pytorchcpp 2. cd fashionmnist or

Vince 0 Jul 13, 2021