Common Ground · Trends in the making
knowledge graph
Established · tracked since 2 September 2026 · first seen 4 August 2026
Established Established: present through the month at a steady pace.
Counted every morning across Hacker News, Google News, GitHub, arXiv and Reddit, from each platform’s own search interface, with the documents the counts were read from listed below. This page is rewritten with every run; the run it stands on is .
- Also counted as: knowledge graphs
- Entered the watchlist as an editorial decision — Editorial seed, 2026-09-02: an older idea returning as the grounding layer under retrieval.
- Carried in the last thirty days by: Hacker News · Google News · GitHub
- Wikipedia: Knowledge_graph — 7,859 page views in thirty days, counted apart from the mentions
Mentions per platform
The counts per platform, as a table
| Platform | 1 day | 7 days | 30 days | Note |
|---|---|---|---|---|
| Hacker News | 1 | 1 | 5 | |
| Google News | 1 | 12 | 29 | may undercount |
| GitHub | 1 | 20 | 100 | may undercount |
| arXiv | 0 | 0 | 0 | |
| 0 | 0 | 0 | ||
| Wikipedia page views | 164 | 1,652 | 7,859 | |
| Total (page views excluded) | 3 | 33 | 134 |
The term and each of its declared aliases are searched as one quoted phrase through every platform's own search interface, once a day; matching documents are deduplicated by URL, and each document counts once however often the phrase occurs in it. The three windows end at the run, so seven days means the seven days before this morning. A platform marked as possibly undercounting returns at most a fixed number of items per query, so its thirty-day figure is a floor rather than a count. Page views are shown where the watchlist names a Wikipedia article and are never added to a total — they are readers, not documents.
May undercount: Google News, GitHub — the feed returns at most a fixed number of items per query, so the thirty-day figure is a floor.
The arc
The arc, day by day, as a table
| Run | Mentions that day |
|---|---|
| 2026-10-19 | 3 |
| 2026-10-18 | 2 |
| 2026-10-17 | 2 |
| 2026-10-16 | 4 |
| 2026-10-15 | 2 |
| 2026-10-14 | 6 |
| 2026-10-13 | 5 |
| 2026-10-12 | 4 |
| 2026-10-11 | 6 |
| 2026-10-10 | 4 |
| 2026-10-09 | 1 |
| 2026-10-08 | 2 |
| 2026-10-07 | 3 |
| 2026-10-06 | 9 |
| 2026-10-05 | 0 |
| 2026-10-04 | 2 |
| 2026-10-03 | 0 |
| 2026-10-02 | 3 |
| 2026-10-01 | 2 |
| 2026-09-30 | 5 |
| 2026-09-29 | 8 |
| 2026-09-28 | 5 |
| 2026-09-27 | 4 |
| 2026-09-26 | 1 |
| 2026-09-25 | 5 |
| 2026-09-24 | 3 |
| 2026-09-23 | 3 |
| 2026-09-22 | 3 |
| 2026-09-21 | 1 |
| 2026-09-20 | 0 |
One point per committed run, oldest on the left, read out of the dated files in src/data/trending/terms rather than from any stored series: a day the machine did not run has no bar, and a day that counted nothing is drawn as a hairline, because a measured zero is still a measurement. The strip is as long as the archive, which is why a term added recently has a short one.
Receipts
The newest documents the counts were read from — title, platform and date, nothing else. The bodies are not stored anywhere: the receipt exists so a reader can open the source and check the count themselves.
- Franz Inc. Announces AllegroGraph 9.1 with GraphTalker 2.0, Advancing Context-Aware AI for Enterprise Knowledge Graphs — EIN News Google News · 2026-10-19
- Show HN: Rubi – A knowledge graph that cuts how much Claude Code re-reads Hacker News · 2026-10-18
- manish11111111/movielens-knowledge-graph GitHub · 2026-10-18
- isha-arjun/Ideavanta GitHub · 2026-10-18
- AgilePemdigi-2026/sage GitHub · 2026-10-18
- AI Learns to Play the Accordion by Encoding Music Theory into a Knowledge Graph — Bioengineer.org Google News · 2026-10-17
- AI Knowledge Graph Matches University Curricula to Real-World Job Skills — Bioengineer.org Google News · 2026-10-17
- eldeucebag/Memento GitHub · 2026-10-17
- Video: Going Meta S04E01 – a Series on Semantics, Knowledge Graphs and All Things AI — Neo4j Google News · 2026-10-16
- Knowledge graphs ground agentic AI at Raymond James — SiliconANGLE Google News · 2026-10-16
- bshravancmu/knowledge-graphs GitHub · 2026-10-16
- Have Knowledge Graphs Fallen Out Of Favor With AI Builders? — TV News Check Google News · 2026-10-15
How this is measured
- A term is a phrase, not a subject: it is counted when it or one of its declared aliases occurs in the title of a document a platform returns for that quoted phrase.
- Each document counts once. Duplicates are removed by URL, so one story syndicated to five outlets counts as the number of distinct URLs, not as five mentions of a topic.
- The windows end at the run, not at midnight: one day, seven days and thirty days back from the moment the machine asked.
- The status compares the seven-day count with the three weeks before it against thresholds committed in the pipeline. No editorial judgement enters, and no model is called.
- The first sighting is the earliest receipt across this run and every committed file, so it can only move backwards — never forwards, and never to today because the archive is young.
Method sheet, section seven → This term as JSON → The whole watchlist as JSON →
No language model writes anything on this page: the count is a platform’s own answer to a quoted phrase, the status is a threshold in the pipeline, and the sentence above the numbers is assembled from them. Data CC0 1.0; cite as Frank Bültge, Common Ground — trends in the making, https://frankbueltge.de/trending/topics/knowledge-graph/