EA2N: Evidence-based AMR Attention Network for Fake News Detection

Code for the above paper

Environment Setup

Install the required dependencies

pip install -r requirements.txt

Setup Conda Environment

conda env create -f environment.yml

Data Preparation

In this module we'll download the dataset and generate AMR linked with WikiData (referred as WikiAMR)

Data Preprocessing

Download FakeNewsNet Dataset by following the steps mentioned here and place it in 'dataset' folder
Download the WikiData5m from here and place it in 'wikinet' folder
Download Artifacts for glove embeddings and other utilities
```
./scripts/download_artifacts.sh
```
cd in 'dataset' and run
```
python preprocess.py
```

Prepare data for train, dev and test:

cd ..
./scripts/prepare_data.sh -v 2 -p [project_path]

We use Stanford CoreNLP (version 3.9.2) for tokenizing.
- Start a CoreNLP server.
- Annotate news article sentences:
```
./scripts/annotate_features.sh amr_data/amr_2.0/csqa
```
Preprocess this data:
```
./scripts/preprocess_2.0.sh
```
Generate AMR for news articles using:
```
cd prepare/
python amr_generation.py {fin.txt} {fout.txt}
```
fin is train, dev or test and fout is the output file
Or use the stog model from here to generate it

Affective Features

Create Affective Dataloader

cd affective_features/
python aff_dataloader.py {dataset} {data_file}

Generate affFeatures

python read_data.py {dataset} {aff_embeddings}

WikiAMR Integration Module

Dividing our data into bathches to attach wikidata

It will takes some time as we write all the paths of the WikiAMR graph. This will need enoguh space to save the data (mnt folder would be fine choice)

cd prepare
python generate_batch.py 50 10 10 <Dataset Name># train/dev/test
cd ..
bash wikidata_integration.sh

Creating Vocab

The vocab will be created from AMR entities and the linked Wikidata paths

sh ./scrips/prepare.sh

Combine the divided files

python generate_prepare.py combine 50 10 10 # train/dev/test

Train

Train the model by running run.py with mode train. This will prompt custom pipeline which the user wants to train the model on

python run.py {gpu} train

Evaluate

Evaluate the model by running run.py with mode eval. This will prompt custom pipeline which the user wants to evaluate the model on

python run.py {gpu} eval

shubhamgpt007/EA2N-Fake-News-Detection-Framework