contestant001 at SemEval-2026 Task 13 Stylometric and TF-IDF-Based Detection of Machine-Generated Code
Year: 2026
Type: conference-paper
Abstract: Reliable detection of machine-generated code has become increasingly important for academic integrity and software quality as code generation is largely being undertaken by large language models.This paper presents our approach to SemEval-2026 Task 13, Subtask A: detecting machine-generated code in a binary classification setting, where we propose an ensemble approach combining TF-IDF lexical representations with 23 hand-crafted stylometric features for binary classification of AI-generated code.Our system aims to address the challenge of cross-language generalization by extracting language-agnostic patterns and combining them with TF-IDF.While we observed that transformer-based models (Code-BERT, UniXcoder) noticeably underperformed under distribution shift, our analysis revealed that AI-generated code exhibits distinct stylometric patterns and our TF-IDF ensemble achieved 0.5175 Macro F1 on the task submission.
Author: Bora Ozaylar
Institution: University of Tübingen
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7166805156