Real-Time Computation Architecture for Non-Timeline Feed Ranking

The article describes a scenario where users follow popular accounts and the feed content is scored and sorted based on factors such as post count, repost count, and activity index within a limited time window (e.g., the last 72 hours). Unlike traditional timeline feeds, this non‑timeline feed requires complex ranking calculations.

Real‑time computation is chosen to meet three requirements: (1) data must be limited to a sliding time window, (2) rankings are refreshed every n minutes, and (3) scoring rules and factor weights may change dynamically.

Input architecture: user actions (posts, activity, reposts) are sent asynchronously to a real‑time computation platform built on Alibaba Cloud Blink. These actions become scoring factors.

Output architecture: the platform calculates scores and stores results in a key‑value structure, using Redis ZSET where key = fansId, score = totalScore, member = extraUserId . This enables fast retrieval of a sorted list of followed users for each fan.

The scoring process aggregates factor counts per user, multiplies by weights, and sums them to obtain a total score. A sliding‑window (hop) operator implements the periodic recomputation.

Optimization strategies include adjusting the recomputation interval (e.g., from 5 to 10 minutes) and offloading the join‑and‑rank step to an external system (such as a search service) for large‑scale deployments.

Conclusion: Real‑time computation effectively supports non‑timeline feed ranking with time‑window constraints and frequent updates, but scaling challenges arise when the user base and join operations become massive.