RealtimeBoost Events Design
Topics: Entity based search, News and Discover, Probably in use, Ranking, Reranking, Semantic Search
Realtime Boost is a Google system designed to detect breaking-news events and trending topics in under a minute by indexing freshly published documents into Hivemind and tracking real-time spikes in their volume. It uses multiple token classes—unigrams, half-hour time buckets, knowledge-graph entities, geographic S2 cells, and quality signals like Freshbox Article Score and NSR—to calculate a “lift” metric (spike > median + 3 × IQR) and identify queries that are suddenly surging. The real-time spike signal and its correlated entities and terms are then exposed through Superroot and QRewrite to boost fresh, high-quality content in search results.
