Olegator47/hotel-query-description-ru

tourism-hotel-search-ru

Dataset Overview

tourism-hotel-search-ru is a specialized Russian-language dataset designed for training and evaluating Information Retrieval and Sentence Similarity models in the tourism and hospitality domain. It includes 14689 pairs query->hotel's description for 977 hotels.

This dataset is optimized for training models using loss functions such as MultipleNegativesRankingLoss (MNRL). The core objective is to match a natural language user search query with the most relevant hotel or tour package description. It serves as an excellent foundation for building smart AI travel assistants and semantic tour search systems.

Methodology

A hybrid approach combining real-world data collection and Synthetic Data Generation was used to create this dataset:

1. Target Description Collection (Parsing): Real text descriptions of hotels, infrastructure, beach types, and amenities were parsed from specialized travel portals, tour operator websites, and aggregators. This ensured the preservation of authentic terminology, specific hospitality markers, and the natural style of property descriptions.
2. Search Query Synthesis (AI Generation): Based on the parsed text descriptions, large language models (LLMs) were used to generate a pool of relevant search queries. The prompts were engineered to mimic real user behavior, ranging from simple keyword phrases to complex, detailed queries with fuzzy phrasing.

Geographical Focus & Biases

The data has a strong regional specificity and focuses on the most popular resort and tourist destinations:

• Turkey
• Egypt
• Thailand
• Georgia
• China

Data Bias: Due to the geographical composition of the dataset, it contains a high density of vocabulary related to beach, sightseeing, and exotic vacations (e.g., "all-inclusive", "first coastline", "reef", "transfer", "bungalow"). Models trained on this data will demonstrate high ranking quality for resort areas but may require additional fine-tuning when handling queries about European urban tourism or ski resorts.

Dataset Structure

The dataset is provided in a tabular format (CSV) and contains the following main fields (columns):

• query (string): The user's search query, synthesized by the AI based on the context.
• positive (string): The parsed original text (hotel or tour description) which serves as the true and most relevant answer to the query.