Algorithmic Role in Building the 58 User Profiling Platform

Introduction: User profiling platforms are the foundation for precise marketing, refined operations and personalized recommendation; accurate depiction of user behavior, interests and needs is crucial.

Background: Traditional profiling relies on data warehouse modeling, integrating multi‑business data, and requires both data mining and platform capabilities for storage, query and sharing.

58’s platform addresses business demands such as personalized recommendation, fine‑grained operation and user value growth, using a UA+CDP+MA solution built on OneID data, traffic and audience insights, and algorithm‑driven audience generation.

Algorithmic contributions: (1) Construction of a comprehensive tag system (over 1,500 tags across social attributes, geography, behavior, preferences, user segmentation) with fact‑based tags generated via SQL and algorithmic tags via data mining; (2) Production of algorithmic tags such as gender, age, business inclination, rental purpose, and content‑preference tags using models like XGBoost, DeepFM, LightGCN, DSSM, CTR ranking, and BERT/M3E embeddings; (3) Offline recommendation pipelines that combine recall (popularity, rules, collaborative filtering, graph neural networks) and ranking (CTR models) to deliver Top‑N personalized content; (4) Smart operation tools such as traffic maps, intelligent audience generation (Look‑alike, minHash, LSH, clustering), and automated feature selection with AutoML.

Application cases: personalized resource placement, push notifications, and search recommendation across 58 App services, leveraging the tag system to drive thousand‑person‑one‑face experiences.

Outlook: Continue deep collaboration with business units, explore new scenarios, iterate and innovate the platform, and enhance AI‑driven capabilities for greater user and enterprise value.