Skip to content
maxvonlancasterPublic

About

Research into Discrete Mathematics, NLP, FL etc.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

Science Repository

This is a repository for scientific ideas and research on the topics of fuzzy logic, NLP, ML and other interesting areas of computer sciences.

Notebooks

  1. NLP Direction

  2. Math Econ

  3. Discrete Mathematics

  4. Interesting Projects


Featured Research: Text Summarization

Publication-Ready Evaluation Framework

A comprehensive research framework for extractive text summarization using unsupervised machine learning has been developed, transforming initial exploratory work into a publication-ready empirical study.

Key Features:

  • ✅ Large-scale evaluation: 1000+ documents (MTSamples medical corpus)
  • ✅ Strong baseline comparisons: Lead-N, Random, TextRank, LexRank, LSA
  • ✅ Statistical rigor: Bootstrap confidence intervals, paired t-tests
  • ✅ Comprehensive analysis: Ablation studies, error analysis, case studies
  • ✅ Publication-ready: Suitable for COLING, EMNLP, NAACL conferences

Methods Evaluated:

  • K-Means Clustering: Semantic grouping with centroid-based sentence selection
  • Kohonen Self-Organizing Maps (SOM): Topological mapping preserving neighborhood relationships
  • TextRank: Graph-based ranking with PageRank
  • LexRank: Graph-based with transformer embeddings
  • LSA: Latent Semantic Analysis with SVD

Evaluation Metrics:

  • ROUGE-1, ROUGE-2, ROUGE-L (F-measure)
  • Statistical significance testing (paired t-test, α=0.05)
  • 95% confidence intervals via bootstrap resampling
  • Per-document error analysis and failure pattern detection

Framework Components:

  • src/summarization_evaluation.ipynb - Main evaluation notebook (11 sections)
  • src/helpers/ - Modular evaluation infrastructure
    • dataset_loader.py - Multi-corpus support (MTSamples, CNN/DailyMail, Wiki Movies)
    • baselines.py - Standard comparison methods
    • evaluation.py - ROUGE metrics with statistical analysis
    • analysis.py - Error analysis and variance explanation
    • ablation.py - Component contribution testing
    • visualization.py - Publication-ready figures
    • case_studies.py - Qualitative analysis generation
  • thesis/ - Complete documentation
    • 2025-summarization.md - Main research paper
    • methodology_detailed.md - Full experimental protocol
    • algorithms.md - Pseudocode and complexity analysis
    • results_summary.md - Comprehensive results template

Quick Start:

# Install dependencies
pip install -r requirements.txt

# Run the full evaluation (60-90 minutes for 1000 documents)
jupyter notebook src/summarization_evaluation.ipynb

# Results are exported to:
# - results/tables/ (CSV and LaTeX)
# - results/figures/ (PNG at 300 DPI)
# - results/evaluation_results.json (complete data)

Research Status: Framework complete and tested. Full 1000-document evaluation can be executed to generate final publication results.

Citation:

Melnyk, H. (2025). Unsupervised Machine Learning Methods for Text Summarization:
A Comprehensive Evaluation of Clustering and Topological Approaches.
[Details to be added upon publication]

Our Team

Avatar Name GitHub Profile Areas of Interest Role
Andrii Lazoryk @12-deadblue Generating functions, NLP Hedgehog
Valentyn Vikovan @cryme666 NLP, Time Series Forecasting Hedgehog
Anastasia Rarenko @wanesssssty RAG Systems Hedgehog
Oleksandr Mykchailyk @Olexandr26 NLP, FL Hedgehog
TBA Liza Kvasnytska TBA NLP Hedgehog
TBA Dmytro TBA Text Summarization Hedgehog
TBA Kyryllo Kravtsov TBA TBA Hedgehog
TBA Nicu TBA TBA Hedgehog
Vasyl Melnyk @maxvonlancaster Owl

Ideas (TODO):

  • Sentiment (emotion in text, energicity of text, ...)

  • disinformation

  • sarcasm (hard)

  • medicine?

  • movies?

  • text professionality

  • manipulation in text

  • shortness/brevity of text

  • financial markers

  • marketing: is product represented on market?

  • cv parsing (with genetic alg and fl)

  • market analysis (i.e. dollar / crypto value based on news)

  • air attack prognosis (based on tg chanels / radio)

  • fuzzy timeseries prognosis (may be + genetic algrthms)

  • RAG system (Nastya)

  • text summarizing (may be local news, etc.) (Dmitro)

  • Also: web parsing, + web app for text summarization


About

Research into Discrete Mathematics, NLP, FL etc.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages