Authorship Attribution in Russian with New High-Performing and Fully Interpretable Morpho-Syntactic Features

This project contain implementation the system, proposed in the "Authorship Attribution in Russian with New High-Performing and Fully Interpretable Morpho-Syntactic Features" (E. Pimonova, O. Durandin, A. Malafeev), and will be presented in the AIST-2019 Conference 17-19 July 2019.

Structure of project:

• Texts should be placed into "Texts" (we didn't shared this folder due to licence restrictions).
• Result of processing dependency trees with rules will be placed in "OutData" folder (represent in repository).

Initially you should place target texts into Text folder (name format: "Author_Title.txt", i.e. Author must be separated from title with subscript _ ). Script preprocessor.py processed all text files from "Text" folder, through SpaCy analyzer. Please, pay attention that we use Model of Russian Language from https://github.com/buriy/spacy-ru

Dependency trees will be saved into pickle format and will be saved in ProcessedTexts folder. NB: Change value of variable PATH_TO_RUS_LANG_MODEL with path to spaCy model of russian language.

datasetBuilder.py read pickle files from ProcessedTexts and proceed it through processor.py (that implement morpho and syntactic rules). datasetBuilder.py build a few csv files, that contain different text representations (simple/complex morphology, simple/complex syntax, see the paper for additional info).

In the first step you must to run preprocessor.py script to build pickle files. Another runs of this script will check if file already processed. Only new files will be processed.