B. Koloski

Exploring Social Bias in Slovenia: The EEC-SL Dataset

Evaluating fine-tuned, embedding-based, and zero-shot models for aspect-based sentiment analysis in South Slavic news

This work introduces AspectBench, a new benchmark dataset for document-level ABSA, comprising real-world online news articles in Slovenian and Serbo-Croatian, designed to test generalization to unseen aspects, and offers practical insights into the trade-offs between model complexity, data characteristics, and the deve...

VLM-guided retrieval augmented generation (RAG) for robot action prediction

A VLM-guided retrieval-augmented generation (RAG) framework that crops task-relevant image regions, retrieves examples from a structured local database, re-ranks candidates, and expands the search when needed enables accurate, data-efficient adaptation to new device types without fine-tuning, at the cost of increased a...