StyloMetrix

Zakład Inżynierii Lingwistycznej i Analizy Tekstu, NASK PIB

📌 Quick

💡 Stylometry tool in beta version for English, German, Polish, Russian and Ukrainian language, distributed as a Python package

💡 Tutorial notebook

💡 List of built-in metrics for Polish, English, German, Ukrainian, Russian

🔖 Citation

Please cite this article when referring to StyloMetrix:

Okulska, I., Stetsenko, D., Kołos, A., Karlińska, A., Głąbińska, K., & Nowakowski, A. (2023). StyloMetrix: An Open-Source Multilingual Tool for Representing Stylometric Vectors. arXiv preprint arXiv:2309.12810.

🔔 About

StyloMetrix is a tool for creating text representations as StyloMetrix vectors. Each metric in vector quantifies a linguistic feature in text. Therefore a detailed information of the style of text can be translated to numeric values and used for - whatever you want!

The metrics are:

• interpretable - each metric represents an aspect of linguistic knowledge
• normalized - metrics express number of ocurrences of given feature per number of tokens in text, which lets us escape scaling effect in texts of different lengths
• reproducible - values of metrics can be recalculated or even counted manually giving always the same output. The representation doesn't depend on any random factor or seeding
• customizable - if your needs exceed the scope of built-in metrics, create your own! Don't forget to share your work and contribute to the community of StyloMetrix!

A StyloMetrix vector can be used as:

• stylometric signature that encodes the writing style of the author and the genre
• input for classifiers of supervised or unsupervised learning, for example Random Forest classifier or feature selection algorithms
• values for statistical analyses in science
• set of linguistic data for manual reference

The tool offers customization of vectors by selecting from built-in metrics or creating new metrics according to user's needs. We provide a user-friendly interface to support these tasks. See instructions below! ⬇

Currently StyloMetrix is available for English, German, Polish, Russian and Ukrainian language.

📢 Release

Our most recent release is:

v0.1.0

• Changing the structure of StyloMetrix
• Works much faster!
• New metrics and categories in Polish and English language
• German language in beta version
• Russian language in beta version
• Ukrainian language in beta version
• Possibility to define metrics to use / categories of metrics as list of strings containing names.
• Possibility to save intermediate steps so even if something crashes, you still have some of work saved.

Please notice that support for Russian and Ukrainian languages will no longer be available.

🔨 Installation

1. Install spaCy

Install spacy according to spaCy install instructions

2. Install model

▶ For Polish:

Download and install model pl_nask v0.0.7

📍 pl_nask is the new HerBERT based model from IPI PAN, requires spacy==3.3

python -m pip install <PATH_TO_MODEL/pl_nask-0.0.7.tar.gz> 

▶ For other languages:

• For English install en_core_web_trf from spaCy install instructions
• For German install de_core_news_lg from spaCy install instructions
• For Russian install ru_core_news_lg from spaCy install instructions
• For Ukrainian install uk_core_web_trf from spaCy install instructions

3. Install StyloMetrix

pip install stylo_metrix

🪁 How to use

1. Get your texts and import StyloMetrix:

import stylo_metrix as sm

texts = ['Panno święta, co Jasnej bronisz Częstochowy I w Ostrej świecisz Bramie!',
        'Ofiarowany, martwą podniosłem powiekę; I zaraz mogłem pieszo, do Twych świątyń progu...',
        'W ludziach straty nie było. Ale wszystkie ławy Miały zwichnione nogi;']

2. Use StyloMetrix object for this texts:

stylo = sm.StyloMetrix('pl')
metrics = stylo.transform(texts)
print(metrics)

3. Your results is now in metrics object.

That's it! Find out about more usages and customization options in notebook tutorial.

Find out about using StyloMetrix in classification or in clustering in example notebook

📈 Metrics

We have put care into creating a set of powerful built-in metrics. See the list below ⬇. However, since flexibility is strength, we provide an easy way to create new metrics.

Polish (see full list)

English (see full list)

German (see full list)

Russian (see full list)

Ukrainian (see full list)

📚 We use

• spaCy (MIT License)
• spacy-syllables (MIT License)
• pl_nask model (GNU GPL 3.0 License), Ryszard Tuora and Łukasz Kobyliński, "Integrating Polish Language Tools and Resources in spaCy". In: Proceedings of PP-RAI'2019 Conference, 16-18.10.2019, Wrocław, Poland.
• experimental data from Imbir, K. K. (2016). Affective Norms for 4900 Polish Words Reload (ANPW_R): Assessments for valence, arousal, dominance, origin, significance, concreteness, imageability and, age of acquisition. Frontiers in Psychology, 7, Article 1081. https://doi.org/10.3389/fpsyg.2016.01081

📪 Contact

Zakład Inżynierii Lingwistycznej i Analizy Tekstu, Naukowa i Akademicka Sieć Komputerowa – Państwowy Instytut Badawczy

Inez Okulska inez.okulska@nask.pl | ziliat@nask.pl

Copyright (C) 2024 NASK PIB