Application of Big Data and Algorithms in the Real‑Estate Internet

On November 24, the Shanghai Computer Society Annual Meeting was held at the Shanghai Science Hall, where Dr. Yang Yun, Director of the Data and Traffic Operations Department of the Real‑Estate Business Group, delivered a presentation titled “Application of Big Data and Algorithms in the Real‑Estate Internet”.

The rapid development of big‑data technologies and algorithms provides a powerful engine for enterprises to innovate and improve efficiency. In the mobile‑internet era, home‑buyers face an expanding choice set and more complex demands, raising expectations for personalized services and high‑quality information on real‑estate platforms.

58.com and Anjuke, the nation’s largest housing platforms, have integrated data from 58.com, Anjuke, Ganji and other subsidiaries, creating a comprehensive real‑estate data ecosystem.

Real‑Estate Big Data Status

Real‑estate data includes community/estate information, user data, and agent data. The “58 Housing Source Holographic Dictionary” launched in 2017 now contains over 800 fields across more than 50 categories, covering price, floor plans, area, age, etc., and spans 640 cities, 550,000 communities, and nearly 200 million listings.

58.com & Anjuke serve over 30 million daily house‑search users. By analyzing click, favorite, subscription and other behaviors, they construct multi‑dimensional user profiles covering basic info, device info, behavior tags, interest tags, and demographic groups, enabling fine‑grained operation and personalized services.

Agent data includes over one million agents nationwide. Comprehensive analysis of posting, Q&A, micro‑chat, calls, and viewings, combined with user evaluations, yields rich agent tags such as basic info, listing quality, service level, user feedback, and professional knowledge, facilitating the selection of high‑quality agents.

Big Data and Algorithm Practice

1. Personalized Recommendation – Real‑estate search is a low‑frequency, sparse‑data scenario. User interest graphs are built from basic info and behavior features, distinguishing early‑stage buyers (broad interests) from late‑stage buyers (focused on specific communities). Recommendations combine rule‑based statistics, LDA topic modeling, and deep neural networks with attention mechanisms for long‑short‑term interest modeling. Candidate recall uses collaborative filtering, location‑based methods, interest‑based recall, and item2vec.

Ranking employs large‑scale sparse features with Logistic Regression, continuous features with XGBoost, and deep models such as FNN and Wide&Deep. Re‑ranking applies rule‑based quality and authenticity checks, resulting in significant improvements in conversion and connection efficiency.

2. Intelligent Matching – With tens of millions of daily users and over 1 million agents, the platform matches high‑intent buyers with top agents using behavior signals (micro‑chat, calls, browsing, subscriptions) and interest graphs. Matching models evolved from rule‑based similarity (cosine, Euclidean) in 2015, to feature‑engineered streaming pipelines in 2016, and to GBDT/XGBoost and deep networks combined with NLP features since 2017, dramatically boosting matching efficiency.

3. Video Quality Assessment – Launched in 2016, video tours enhance user experience and help verify listing authenticity. A quality‑assessment system combines deep learning, machine learning, and classic image algorithms to evaluate shooting quality, content completeness, motion smoothness, and audio relevance. Video frames are analyzed for motion vectors, and a shallow neural network predicts overall quality, leading to higher watch times and better connection rates.

Conclusion – Big data and AI offer vast opportunities in the real‑estate internet domain. Beyond the discussed applications, 58.com has also deployed VR tours, intelligent customer service, and AI‑driven advertising, positioning the company at the forefront of domestic and international innovation.