Real Estate Rental Platform: True Listing Model and Credit System Construction

Yan Yan, head of the Data Strategy Department at Beike Rental, presented at the 2018 WOT Global AI Summit, sharing the platform's practice of using big data and machine‑learning methods to control listing authenticity, enhance service quality, and establish a comprehensive rental credit system.

The Chinese rental market, now a trillion‑yuan industry, suffers from a supply‑demand mismatch and low rental penetration (only about 12% of the population), creating significant efficiency challenges that demand data‑driven solutions.

Non‑authentic listings—whether deliberately false or outdated—cause trust issues across pre‑rental, rental, and post‑rental stages; the platform therefore defines four true‑listing standards: physical existence, current availability, accurate visual perception, and truthful pricing.

Given the short transaction cycle (5‑7 days) and brand‑related price variance, the team built a data pipeline (collection, cleaning, feature engineering) and evaluated several tree‑ensemble models (XGBoost, LightGBM, CatBoost) with hyper‑parameter tuning, addressing brand premium effects and uneven data distribution.

Image analysis models were also deployed to classify room types, assess renovation level, and score image quality, handling the massive volume of photos per listing and achieving up to 30× efficiency gains.

A merchant credit scoring system was designed, combining true‑listing scores, fulfillment behavior, service quality, and historical performance; the score drives incentives, anti‑cheating checks, and differentiated platform privileges for high‑scoring merchants.

Since implementation, problem listings have declined, merchant engagement has risen, and user experience has improved; future work includes refining house‑status models, reducing verification costs, and promoting an industry‑wide open credit data set.

The talk concluded with a call for collaborative efforts to standardize data and models across the rental sector, emphasizing that domain knowledge is essential for successful AI deployment.