IndoHotelABSA: A large-scale aspect-based sentiment analysis dataset of Indonesian hotel reviews for hospitality CRM.

IndoHotelABSA is a large-scale dataset for aspect-based sentiment analysis (ABSA) of Indonesian hotel reviews. It comprises 14,988 unique, deduplicated Indonesian-language reviews collected on 10 July 2026 through the official Google Places API from 3,075 establishments across 55 cities spanning all six major regions of Indonesia. A class-balanced sample of 3,000 reviews was annotated for seven CRM-meaningful aspect categories (Location, Cleanliness, Service/Staff, Room & Facilities, Price/Value, Food & Beverage, and Public Facilities), each with three-way polarity (positive, negative, neutral), using a two-tier protocol: all 3,000 reviews received large-language-model (LLM) draft labels, forming the silver tier, and a 500-review subset was checked and corrected independently by three human annotators and adjudicated by majority vote, forming the gold tier. The gold subset reaches almost-perfect agreement (Fleiss' κ = 0.894 ) and yields 1,506 gold aspect annotations. The release explicitly distinguishes the 2,480 unverified silver training records from the 500 human-verified gold records and provides fixed train/validation/test splits. The dataset fills the absence of a public, quality-controlled Indonesian hotel ABSA benchmark and can be reused for low-resource NLP, LLM-annotation research, and hospitality customer relationship management (CRM) analytics. All files are provided as UTF-8 JSON Lines with a datasheet and annotation guidelines.