tmpFever | NLP Project

FEVER Challenge

The FEVER challenge is a natural language processing task where claims are automatically verified against facts in Wikipedia articles. Refer to the following paper for more information: https://arxiv.org/pdf/1803.05355.pdf

Some files were taken from the original baseline code, which can be found here

-Data setup Download the dataset by running the script download-data.sh. Create the wikipedia db by either downloading the pages and processing them to create a database, or download the preprocessed .db file by running download-preprocessed.sh

The NER model and MongoDB dump can be found here

The Siamese Network weights can be found here

The ESIM model weights can be found here

The final report can be found here

Wikipedia

Download pre-processed Wiki Dump

bash scripts/download-processed-wiki.sh

Dataset

bash scripts/download-data.sh

-Required modules The following modules will be required to run some of the backend components that support the database and some of the components involved in the document and sentence retrieval stages. In order to run, these components require the following python modules

• fever-drqa
• tqdm
• orator
• PyMySQL
• torch
• argparse
• json
• numpy
• pandas
• sklearn
• logging
• allennlp
• spacy
• pyglot
• django

-Scoring The code for the scoring program can be found at this repo

• scorer: https://github.com/sheffieldnlp/fever-scorer.git

The following breaks down each of the main files and briefly explains what each directory holds:

1. pipeline: contains the scripts that integrate the different parts of the system together
2. doc_ret: contains utility scripts and preprocessing files and models related to document retrieval
3. rte: contains the models, training loops and utility functions for the ESIM and Aggregator modules which collectively make up the RTE stage
4. scripts: contains general setup scripts
5. training: contains some scripts that perform some preprocessing on the dataset in preparation for training
6. common: contains some baseline code scripts and database classes that are essential for running the FEVER baseline model
7. fever-frontend: contains a basic one-page django web application that exposes the functionality of the system through the pipeline scripts. To run the frontend, run the pipeline/start_modules.sh script which initializes the system, then start the application by running manage.py runserver