HN Trends

Transparent by design

How HN Trends measures attention

The system measures the breadth of Hacker News conversation around a concept. It does not treat raw keyword repetition, votes, or a single large comment thread as equivalent to broad community interest.

Collection

New and changed HN items are ingested continuously from the public Firebase API. Every batch is transactional and restart-safe. The historical corpus is bulk imported at lower confidence while a high-fidelity API backfill progressively replaces it.

Topics and aliases

Canonical topics combine known names, acronyms, and spelling variants. Exact matching uses token boundaries, preserving distinctions such as PostgreSQL/postgres/psql while avoiding substring matches inside unrelated words.

Normalized interest

Story-title matches receive a weight of 3, story text 1.5, and comments at most 1 per author and root story. Weighted breadth is divided by all active HN items in the bucket and expressed per 10,000 items.

Confidence and rising scores

Confidence requires distinct stories and participants. Rising scores compare normalized interest with previous and hour-of-week baselines, with minimum-support rules preventing small denominators from dominating rankings.

Event clustering

Recent HN stories become candidates for the same event when their titles substantially overlap, supported by shared canonical topics, close publication time, or a common domain. Placeholder or unresolved titles are never clustered.

Topic labels require either a story-title match or evidence from at least two distinct items in the thread. Event pages list every member story so groupings remain inspectable.

Current data coverage

107,136

Indexed items

08/12/2025

Earliest item

01/09/2026, 12:22:06

Methodology

How HN Trends measures attention