VM

Selected research

← All research

Paper explained

Master's thesis Wikipedia

Vandalism or Knowledge Manipulation? Detecting Narratives in Wikipedia Edits

Viktoriia Makovska

What can manipulation signals add to a vandalism detector?

The question

Why this work

The thesis develops separate manipulation and vandalism signals, then combines them with revision features. A Russian government-backed Wikipedia fork supplies candidate manipulation examples; Russian and Ukrainian Wikipedia provide the main evaluation setting.

The approach

A visual guide

Two signals, one combined prediction

  1. Manipulation signal

    Learn from text changes in the fork.

  2. Vandalism signal

    Learn from revision and revert patterns.

  3. Combine features

    Add the scores to metadata and semantic features.

  4. Evaluate

    Compare predictions, agreement, and calibration.

Ukrainian news experiment · F1

Reported F1, from 0 to 1. Higher is better.

Vandalism features
0.89
Manipulation features
0.82
Combined features
0.91
Schematic of the combined approach. The result chart reproduces Table 5.3 for the separate Ukrainian news dataset; it does not compare Wikipedia test sets. Chapters 3, 5–6; Table 5.3

What the work contributes

Combined signals improved aspects of agreement and calibration in the Wikipedia experiments. On a separate Ukrainian-language news dataset labeled for propaganda, the combined classifier reached F1 0.91, compared with 0.89 using vandalism features and 0.82 using manipulation features alone.

Chapters 3, 5–6; Table 5.3

Original paper & resources

Master's thesis, Ukrainian Catholic University, Faculty of Applied Sciences (2025).

Repository
Citation 
Makovska, V. (2025). Vandalism or Knowledge Manipulation? Detecting Narratives in Wikipedia Edits (Master's thesis). Ukrainian Catholic University.