Automatic Classification of “Gross Negligence” in Serbian Court Decisions Using TF-IDF and Windowing

View at publisher Summarise with VIRAT
Authors: Dušan Kuzmanović, Ana Đilas
Cited by: 0
References: 8

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This paper presents an interpretable, resourceefficient baseline for detecting a concrete legal standard in Serbian judicial reasoning: whether a court decision affirms gross negligence or explicitly negates it. Using a small, expertlabeled corpus of decisions, we compare three input representations under leave-one-out evaluation: (i) manually extracted “gold” paragraphs, (ii) whole-document reasoning blocks, and (iii) window-based processing with document-level aggregation. Across experiments, the gold-paragraph setting achieves the highest performance, while the full-reasoning baseline outperforms the windowing approach on the current dataset. Windowing remains useful as an evidence-localizing representation that can support auditability and error analysis, even when it does not improve accuracy. The study is positioned as a pilot methodology and an engineering blueprint suitable for data-constrained e-Justice analytics.

Fields of Study & Subjects

Artificial Intelligence in Law Text and Document Classification Technologies Topic Modeling

Publication details

Pages
1–4
Publication date
24 February 2026
Language
EN