Architecture and Components of an Intelligent Recommendation Platform

With the growing adoption of intelligent recommendation scenarios, the platform now supports over 40 company scenarios, improving user experience, user stickiness, click‑through rate, and conversion rate.

The overall architecture consists of a unified gateway service, AB testing service, recommendation engine service, engineering engine, profiling service, and related resource services, evolving from a single service to a micro‑service architecture to decouple services and provide platform operation capabilities.

The AB testing service follows Google’s overlapping experiment architecture, offering hierarchical experiments, traffic splitting via hash modulo based on region, client, network conditions, or user profile, and uses BitMap and Base64 to store experiment parameters and effect metrics such as request exposure, visible exposure, clicks, and CTR.

The recommendation engine service comprises three key modules:

Index module : Handles multi‑level index queries (offline and real‑time) using an in‑memory index covering fields such as title, tags, category, keywords, vehicle series, model, brand, level, city, author, and other auxiliary attributes; employs multi‑shard indexing to achieve ~30 ms latency for 3,000 candidates.

Recall module : Assembles user and vehicle profiles, generates recall intents based on collaborative filtering, topic models, and content‑based models, queries the index to obtain a coarse‑rank candidate set, and allows real‑time adjustment of recall strategies via experiment parameters.

Filter module : Applies various filters (exposure, negative feedback, timeliness, city, blacklist, etc.) per channel during the recall stage to ensure high‑quality resources, all configurable through the platform.

Future plans aim to further mature the platform into a middle‑platform architecture with streamlined integration, standardized processes, unified input/output, and open recommendation development tools to encourage broader participation.