This contains the code for detecting the false article detection using weakly supervised learning.
Due to storage constraint, we are not able to load many files such as embedding.model, glove, etc. We have kept them in Google drive
For our paper, we use the following datasets:
- FakeNewsNet contains news articles in politics and entertainment.
- FakeHealth contains news articles in health.
- NELA, NELA-2017 and NELA-2018 are used for unlabeled samples. From the NELA-2019 dataset, we use the aggregated labels.
Run the bash code in order to get experiment folds.
You are supposed to see the processed files in Data folder
The Data\Input2 is the dataset on which overall comparision have been done and it is default dataset for model.The reports also generated from dataset2
Python compiler is 3.7.9.
Install libraries in requirements.txt:
pip3 install -r requirements.txtFrom the command prompt, change your present working directory to folder where your downloaded code is present. Change the directory to Fakenewdetection folder execute the below command
python Datapreprocessing\Datapreprocessing.pyPlace the label.csv and unlabel.csv under Data\Preprocessing\Input1 folder. We have created 4 different types of datasets each of them are placed under Input1, Input2, Input3, Input4 folders respectively.
Summary of the columns present in the data file.
| Column | Description |
|---|---|
| Article | contains true and fake articles |
| Label | For True news = 1, Fake news = 0 and unlabeled = -1. (numeric value) |
Once the data preprocessing is completed, train data and test data files will be available under Data\Input folder.
Default values
As the preprocessing is done, the npy files will be available in Data\Input2, which will be input for Mean Teacher and Pseduo Label for training.
For example:-
python Datapreprocessing\Datapreprocessing.py --inputPath C:\fake_news_detection\Data\Preprocessing\Input4\ --outputPath C:\fake_news_detection\Data\Input4\| Parameter | Description |
|---|---|
| --inputPath | input path for un-processed data |
| --outputPath | output path for processed data, taken as input for ML models |
| --labeldataname | file name for label data |
| --unlabeldataname | file name for unlabel data |
| --xTrain | npy xTrain file name |
| --xTest | npy xTest file name |
| --yTrain | npy yTrain file name |
| --yTest | npy yTest file name |
| --xUnlabel | npy xUnlabel file name |
| Parameter | Default Value |
|---|---|
| --inputPath | os.path.abspath(os.getcwd())+"\Data\\Preprocessing\Input2\ |
| --outputPath | os.path.abspath(os.getcwd())+"\Data\Input2\ |
| --labeldataname | label.csv |
| --unlabeldataname | unlabel.csv |
| --xTrain | xtr_shuffled.npy |
| --xTest | xte_shuffled.npy |
| --yTrain | ytr_shuffled.npy |
| --yTest | yte_shuffled.npy |
| --xUnlabel | xun_shuffled.npy |
We have completed preprocessing and placed all those data files in respective folders such as Data\Input1,Data\Input2,Data\Input3,Data\Input4. From these folder, preprocessed data can take for further training the meanteacher and pseduolabel models
From the analysis,we found that
Mean Teacher model for false article classification.
It will take test and train data from Data\Input2 folder as default, and once training complete it will place the report and plot in Data\Output1 folder as default Please Go to folder path where code is available and run like below Mean Teacher, we have created different method of noises which can be called by parameters
| Parameter | Description |
|---|---|
| --meanteacher MT_syn_unlabel_diff | Mean teacher |
| --meanteacher MT_syn_noise_diff | Mean teacher with synonym noise |
| --meanteacher MT_dropout | Mean teacher with dropout |
| --meanteacher MT_syn_dropout | Mean teacher with synonym and dropout |
Once you call these method as parameter, automatically code regarding to this function and train the data in Data\Input2 folder (default) and the report will be placed in Data\Output2 folder (default) , default folders also changed by parameter passing
python MeanTeacher\main.py --meanteacher MT_dropout --lr 0.0005 --epochs 5 --batch_size 64 --alpha 0.95other parameters are also available
| Parameter | Description |
|---|---|
| --ratio | |
| --threashold | |
| --inputPath | Input path to the train model ( /Users/tmp/Testing/) |
| --outputPath | Output path to place report and files ( /Users/tmp/Testing/) |
Default values
| Parameter | Default Value |
|---|---|
| --lr | 0.0001 |
| --epochs | 10 |
| --batchSize | 64 |
| --ratio | 0.5 |
| --alpha | 0.97 |
| --dropout | 0.2 |
| --synonym_noise_b1 | 0.5 |
| --synonym_noise_b2 | 0.3 |
| --inputPath | os path where you placed the code in system + \Data\Input2\ |
| --outputPath | os.path.here you placed the code in system + \Data\Output2\ |
| --xTrain | "xtr_shuffled.npy" |
| --xTest | "xte_shuffled.npy" |
| --yTrain | "ytr_shuffled.npy" |
| --yTest | "yte_shuffled.npy" |
| --xUnlabel | "xun_shuffled.npy" |
| --reportIterate | "report_iterate_mean_teacher.csv" |
Pseduo Label model for false article classification.
It will take test and train data from Data\Input2 folder as default, and once training complete it will place the report and plot in Data\Output2 folder as default Please Go to folder path where code is available and run like below
python PseudoLabel\main.py --inputPath C:\fake_news_detection\Data\Input4\ --epochs 30 --batch_size 32--inputPath - input path to the train model (Eg: /Users/tmp/Testing/) --outputPath - output path to place report and files (Eg: /Users/tmp/Testing/)
| Parameter | Default Value |
|---|---|
| --lr | 0.0001 |
| --Epochs | 14 |
| --batch_size | 64 |
| --Thresh | 0.90 |
| --n | 10 |
| --max_len | 128 |
| --inputPath | os path where you placed the code in system + \Data\Input2\ |
| --outputPath | os.path.here you placed the code in system + \Data\Output2\ |
| --xTrain | "xtr_shuffled.npy" |
| --xTest | "xte_shuffled.npy" |
| --yTrain | "ytr_shuffled.npy" |
| --yTest | "yte_shuffled.npy" |
| --xUnlabel | "xun_shuffled.npy" |
VAT used the spacy for the preprocessing the code and its has its own folder strutcure to the run the VAT main code.
Raw data was placed in the VAT\Input1\raw -- which is not processed data (label.csv and unlable.csv files), the processed files will be placed in the temp inside same folder
we have created 4 different types datasets each data sets placed in the different folder like Input1,Input2,Input3,Input4 in the same way.
Here preprocessing done in four stages
Stage 1 : Here raw (default-VAT\Input1\raw) file wil be processed and placed in processed folder ( default - VAT\Input1\processed)
Example:
python VAT\preprocessing\text_spacy.py --inputPath C:\fake_news_detection\VAT\Input4\raw\ --outputPath C:\fake_news_detection\VAT\Input4\processed\| Parameter | Default Value |
|---|---|
| --inputPath | \VAT\Input1\raw\ |
| --labelfilename | label.csv |
| --unlabelfilename | unlabel.csv |
| --outputPath | \VAT\Input1\processed\ |
| --Remain | label.csv |
| --xUnlabel | "xun_shuffled.npy" |
Stage 2 : split the label data and place it in the processed folder
Example:
python VAT\preprocessing\split_data.py --Path C:\fake_news_detection\VAT\Input2\processed\| Parameter | Default Value |
|---|---|
| --Path | \VAT\Input1\processed\ |
| --filename | label.csv |
| --xUnlabel | xun_shuffled.npy |
| --xTrain | x_train.npy |
| --xTest | x_test.npy |
| --yTrain | y_train.npy |
| --yTest | y_test.npy |
Stage 3 : Tokenize the splitted data., here word picket file will be placed in the meta folder
Example:
python VAT\preprocessing\tokenization_vat.py --inputPath C:\fake_news_detection\VAT\Input2\processed\ --outputPath C:\fake_news_detection\VAT\Input2\temp\ --pickelPath C:\fake_news_detection\VAT\Input2\meta\Parameters with default values:
| Parameter | Default Value |
|---|---|
| --inputPath | \VAT\Input1\processed\ |
| --outputPath | \VAT\Input1\temp\ |
| --pickelPath | \VAT\Input1\meta\ |
| --filename | label.csv |
| --xUnlabel | xun_shuffled.npy |
| --xTrain | x_train.npy |
| --xTest | x_test.npy |
| --yTrain | y_train.npy |
| --yTest | y_test.npy |
Stage 4 : Emedding, for this golve file is need, this file you can download from the site https://nlp.stanford.edu/projects/glove/ and place it in the path \VAT\ because this files is common to all input files. Once emedding its created and it will place it in the meta folder which is default.
Example :
python VAT\preprocessing\embedding.py --pickelPath C:\fake_news_detection\VAT\Input2\meta\ Parameters with default values:
| Parameter | Default Value |
|---|---|
| --pickelPath | \VAT\Input1\meta |
| --glovePath | \VAT |
| --EMBEDDING_DIM | 300 |
Once all processing completed the final processed file will be placed inside the VAT\Input1\temp\ folder for each data. We have completed the preprocessing and those files are already in location.
Once all files are available in temp directery, then we have to call only the main\vat.py
Example:
python VAT\main\vat.py --n_epochs 30 --batchSize 64 --lr 0.0001 Parameters with default values:
| Parameter | Default Value |
|---|---|
| --lr | 0.0001 |
| --n_epochs | 10 |
| --batch_size | 32 |
| --emb_dim | 300 |
| --doc_length | 100 |
| --n_splits | 10 |
| --epsilon | 0.01 |
| --act_fn | 'tanh' |
| --model_type | 'VAT' |
| --comment | 'Unknown embeddings 6210/38028' |
| --inputPath | os.path.abspath(os.getcwd())+"\VAT\Input1\" |
| --outputPath | os.path.abspath(os.getcwd())+"\VAT\Output1\" |
Bhupender kumar Saini
Swathi Chidipothu Hare
Mayur Rameshbhai Waghela
Lokesh Sharma
Ravineesh Goud
Ipek Baris