Skip to content
This repository provides the implementation for the paper "Combining Fact Extraction and Verification withNeural Semantic Matching Networks".
Python Shell
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.


This repository provides the implementation for the paper Combining Fact Extraction and Verification with Neural Semantic Matching Networks (AAAI 2019 and EMNLP-FEVER Shared Task Rank-1 System).


  • Python 3.6
  • pytorch 0.4.1
  • allennlp 0.7.1
  • sqlitedict
  • wget
  • flashtext
  • pexpect
  • fire
  • inflection

Try to install the package as the order above. Previous version of pytorch can be find at legacy pytorch.


  1. Setup the python environment and download the required package listed above.
  2. Run the preparation script.
bash ./scripts/

The script will download all the required data, the auxiliary packages and files.

  1. Tokenize the dataset and build wiki document database for easy and fast access and query.
python src/pipeline/ tokenization        # Tokenization
python src/pipeline/ build_database      # Build document database. (This might take a while)

After preparation, the following folder should contain similar files as listed below.

├── fever
│   ├── license.html
│   ├── shared_task_dev.jsonl
│   ├── shared_task_test.jsonl
│   └── train.jsonl
├── fever.db
├── id_dict.jsonl
├── license.html
├── sentence_tokens.json
├── tokenized_doc_id.json
├── tokenized_fever
│   ├── shared_task_dev.jsonl
│   └── train.jsonl
├── vocab_cache
│   └── nli_basic
│       ├── labels.txt
│       ├── non_padded_namespaces.txt
│       ├── tokens.txt
│       ├── unk_count_namespaces.txt
│       └── weights
│           └── glove.840B.300d
├── wiki-pages
│   ├── wiki-001.jsonl
│   ├── ... ...
│   └── wiki-109.jsonl
└── wn_feature_p
    ├── ant_dict
    ├── em_dict
    ├── em_lemmas_dict
    ├── hyper_lvl_dict
    ├── hypernym_stems_dict
    ├── hypo_lvl_dict
    └── hyponym_stems_dict
├── DrQA
└── stanford-corenlp-full-2017-06-09
└── chaonan99
├── saved_nli_m
├── nn_doc_selector
└── saved_sselector

Automatic pipeline procedure.

Running the pipeline system on the dev set with the code below:

python src/pipeline/

Note that this pipeline is the (SotA) model in the AAAI paper. For EMNLP-FEVER Shared Task version, please refer to src/nli/ and src/pipeline/


If you find this implementation helpful, please consider citing:

  title={Combining Fact Extraction and Verification with Neural Semantic Matching Networks},
  author={Yixin Nie and Haonan Chen and Mohit Bansal},
  booktitle={Association for the Advancement of Artificial Intelligence ({AAAI})},
You can’t perform that action at this time.