Model for text classification regarding veracity.

Baseline uses W2V embeddings trained on google corpus for fake news

Lux proposes the usage of Linguistic Aspects as Features.

Lux

INSTALLATION

This repository uses bert. in order to use BERT properly:

1)Clone bert repo inside Lux/res:

-- git clone https://github.com/google-research/bert

2)Download the pre-trained model from bert:

-- wget https://storage.googleapis.com/bert_models/2018_10_18/uncased_L-12_H-768_A-12.zip

3)Unzip the model inside bert folder:

-- unzip uncased_L-12_H-768_A-12.zip

You should have 3 files, the model, bert_config.json and vocab.txt.

4)Set env variable BERT_BASE_DIR:

-- export BERT_BASE_DIR=/path/to/Lux/res/bert/uncased_L-12_H-768_A-12

in our case: export BERT_BASE_DIR=~/Lux/res/bert/uncased_L-12_H-768_A-12

5)Start bert-as-a-service server for requests in another session/screen tab

-- bert-serving-start -model_dir $BERT_BASE_DIR -max_seq_len 512 -mask_cls_sep

Install Specificity model

1)Download DASSP.zip inside res/specificity

-- wget https://www.dropbox.com/s/41uw7wm2bbgoff4/DASSP.zip

2)Unzip its contents

-- unzip DASSP.zip

3)Go into folder, download and unzip glove:

-- cd Domain-Agnostic-Sentence-Specificity-Prediction/
-- wget https://www.dropbox.com/s/0g880op64chjw4b/glove.840B.300d.zip
-- unzip glove.840B.300d.zip

+)Check the README.md inside the folder, if modifications have to be done

Create an virtual environment with python3 and activate it

1)Back to Lux

-- virtualenv envLux-p python3

-- source envLux/bin/activate

Install requirements

-- pip install -r requirements.txt

Download and extract GoogleNews-vectors-negative300.bin into data/

-- cd data/
-- wget -c "https://s3.amazonaws.com/dl4j-distribution/GoogleNews-vectors-negative300.bin.gz"
-- gunzip GoogleNews-vectors-negative300.bin.gz

Running

-- bash run.sh

OR

-- sudo -E python3 lux.py

if 'True' is passed as first argument, force_reload will receive its value and new bert models as well as new features will be generated.

---

Papers

Please cite the published articles related to this work:

Azevedo, Lucas, et al. "LUX (Linguistic aspects Under eXamination): Discourse Analysis for Automatic Fake News Classification." Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021. 2021.

@inproceedings{azevedo2021lux,
  title={LUX (Linguistic aspects Under eXamination): Discourse Analysis for Automatic Fake News Classification},
  author={Azevedo, Lucas and d’Aquin, Mathieu and Davis, Brian and Zarrouk, Manel},
  booktitle={Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021},
  pages={41--56},
  year={2021}
}

Azevedo, Lucas, and Mohamed Moustafa. "Veritas annotator: Discovering the origin of a rumour." Proceedings of the Second Workshop on Fact Extraction and VERification (FEVER). 2019.

@inproceedings{azevedo2019veritas,
  title={Veritas annotator: Discovering the origin of a rumour},
  author={Azevedo, Lucas and Moustafa, Mohamed},
  booktitle={Proceedings of the Second Workshop on Fact Extraction and VERification (FEVER)},
  pages={90--98},
  year={2019}
}

Azevedo, Lucas. "Truth or lie: Automatically fact checking news." Companion Proceedings of the The Web Conference 2018. 2018.

@inproceedings{azevedo2018truth,
  title={Truth or lie: Automatically fact checking news},
  author={Azevedo, Lucas},
  booktitle={Companion Proceedings of the The Web Conference 2018},
  pages={807--811},
  year={2018}
}