Identifying Machine-Paraphrased Plagiarism

README.txt

Title: Identifying Machine-Paraphrased Plagiarism
Authors: Jan Philip Wahle, Terry Ruas, Tomas Foltynek, Norman Meuschke, and  Bela Gipp
contact email: wahle@gipplab.org; ruas@gipplab.org;
Venue: iConference
Year: 2022
================================================================
Dataset Description:

Training:
200,767 paragraphs (98,282 original, 102,485paraphrased) extracted from 8,024 Wikipedia (English) articles (4,012 original, 4,012 paraphrased using the SpinBot API).

Testing:
SpinBot: 
    arXiv         - Original - 20,966;    Spun - 20,867
    Theses        - Original - 5,226;        Spun - 3,463
    Wikipedia    - Original - 39,241;    Spun - 40,729
    
SpinnerChief-4W: 
    arXiv         - Original - 20,966;    Spun - 21,671
    Theses        - Original - 2,379;        Spun - 2,941
    Wikipedia    - Original - 39,241;    Spun - 39,618
    
SpinnerChief-2W: 
    arXiv         - Original - 20,966;    Spun - 21,719
    Theses        - Original - 2,379;        Spun - 2,941
    Wikipedia    - Original - 39,241;    Spun - 39,697

================================================================
Dataset Structure:

[human_evaluation] folder: human evaluation to identify human-generated text and machine-paraphrased text. It contains the files (original and spun) as for the answer-key for the survey performed with human subjects (all data is anonymous for privacy reasons).

NNNNN.txt - whole document from which an extract was taken for human evaluation
    key.txt.zip - information about each case (ORIG/SPUN)
    results.xlsx - raw results downloaded from the survey tool (the extracts which humans judged are in the first line)
    results-corrected.xlsx - at the very beginning, there was a mistake in one question (wrong extract). These results were excluded.

[automated_evaluation]: contains all files used for the automated evaluation considering [spinbot] (https://spinbot.com/API) and [spinnerchief] (http://developer.spinnerchief.com/API_Document.aspx).

• Each paraphrase tool folder contains:
• [corpus] and [vectors] sub-folders.
• For [spinnerchief], two variations are included, with 4-word-chaging ratio (default) and 2-word-chaging ratio. 

[vectors] sub-folder contains the average of all word vectors for each paragraph. Each line has the number of dimensions of the word embeddings technique used (see paper for more details) followed by its respective class (i.e., label mg or og). Each file belongs to one class, either "mg" or "og". The values are comma-separated (.csv). The extension is .arff can be read as a normal .txt file.

• The word embedding technique used is described in the file name with the following structure: <technique>-<type>-mean-<data>.arff . Where

<technique> -     d2v - doc2vec
                google - word2vec
                fasttextnw - fastText without subwording
                fasttextsw - fastText with subwording
                glove - Glove
                
Details for each technique used can be found in the paper.
                
<type> -     arxivp  - arXiv paragraph split
            thesisp - Theses paragraph split
            wikip   - Wikipedia paragraph split (wikipedia_paragraph_vector_train are the vectors used for training. It follows the same wikip structure)     

Details for each technique used can be found in the paper referenced at the start of this README file.

[corpus] sub-folder: contains de raw text (No pre-processing) used for train and test at a paragraph level.

• The Spun paragraphs used for training are only generated using the SpinBot tool. For test both SpinBot and SpinnerChief are used. 
• The paragraph split is generated by selecting paragraphs from the original documents with 3 or more sentences. Each folder is divided in mg (i.e., machine-generated through SpinBot and SpinnerChief) and og (i.e., original-generated file). the document split is not avaiable since our experiments only use the paragraph level.
• Machine Learning models: SVM, Naive Bayes, and Logistic Regression. The grid search for hyperparameter adjustments for the machine learning classifiers is described in the paper.

@incollection{WahleRFM22,
  title = {Identifying {{Machine-Paraphrased Plagiarism}}},
  booktitle = {Information for a {{Better World}}: {{Shaping}} the {{Global Future}}},
  author = {Wahle, Jan Philip and Ruas, Terry and Folt{\’y}nek, Tom{\’a}{\v s} and Meuschke, Norman and Gipp, Bela},
  editor = {Smits, Malte},
  year = {2022},
  volume = {13192},
  pages = {393--413},
  publisher = {{Springer International Publishing}},
  address = {{Cham}},
  doi = {10.1007/978-3-030-96957-8_34},
  isbn = {978-3-030-96956-1 978-3-030-96957-8},
}

For our previous publication using only SpinBot and Wikipedia articles for document and paragraph split, please see the following publication. The dataset used is hosted in DeepBlue