XLM-RoBERTa for Multi-Label Hotel Review Topic Detection
Fine-tuned XLM-RoBERTa-base for cross-lingual multi-label topic classification
of hotel reviews. Developed for the NLP Shared Task at the University of Antwerp.
Details
- 20 topic labels (hotel-general, staff, room-cleanliness, etc.)
- Training languages: German, Spanish, French, English
- Test languages: English, Dutch, Italian
- Best checkpoint: epoch 6, validation macro-F1 = 0.289
- Final test macro-F1 = 0.296 (with per-label threshold tuning)
Authors
Yukiko Tominaga and Ghadir Slim
University of Antwerp, 2025-2026
