Modeling Price-Demand Relationships for Online Hotel Booking: Demand Functions, Causal Inference, and Multi-Scenario Joint Modeling

Online hotel booking platforms face varying demand and occupancy rates across room types and time, making accurate occupancy estimation challenging.

This article examines the relationship between supply‑chain pricing and sales in the travel industry through four parts: background, demand‑function based quantity‑price modeling, causal‑inference based modeling, and multi‑scenario joint modeling.

First, the difficulty of quantity‑price modeling is discussed, including lack of ground‑truth, complex demand functions, data sparsity, monotonicity constraints, and offline‑online evaluation consistency.

Second, a demand‑function approach introduces an elastic demand function and a Price Elasticity Prediction Model (PEM) that incorporates competition, temporal, and feature factors via a multi‑task learning framework with modules such as a Competition Representation Module (CRM) and a Multi‑Sequence Fusion Module (MSFM). Experiments on two proprietary datasets (high‑season and low‑season) show PEM achieves the best trade‑off among metrics like MAPE, WMAPE, PDR, PDP, PIR, PIP and BR.

Third, causal‑inference modeling treats price as a treatment and occupancy as an outcome, addressing challenges of sparse multi‑value treatments, monotonicity, and treatment bias. The proposed CANDY method outperforms correlation‑based baselines on causal metrics.

Finally, a multi‑scenario joint model (MSP) combines demand representation (DRE) and price‑competition representation (PCRE) to share information across platforms while preserving scenario‑specific signals, achieving superior performance on offline and online evaluations.

The presentation concludes with acknowledgments of the speakers and organizers.