VN-MSASar: A Multilingual Aspect-Based Sentiment Analysis Dataset for Vietnamese Hospitality

VN-MSASar is an aspect-based sentiment analysis (ABSA) dataset for the hotel and restaurant domain in Vietnam, covering four languages (Vietnamese, English, Chinese, Japanese). It contains 83,009 (sentence, aspect) pairs over nine service aspects, each with an ordinal sentiment label on a [-3,+3] scale and a binary sarcasm label. Labels were produced by a two-step LLM pipeline and validated against two independent reviewers (Krippendorff's alpha = 0.982) and an external gold set of 1,500 pairs re-annotated by three independent annotators. The deposit includes both the aspect-pair version (vn_msa_sar) and the derived multi-label aspect-detection version (vn_hotel_res_aspect) with train/validation/test splits, the annotation codebook v1.0, the human-validated gold set, and the collection scripts.