How NewsNinja groups and ranks related news stories
Understand NewsNinja Scientific grouping, Tversky, Overlap, Cosine, and the recency and score signals used to rank stories.
Several publications may cover the same event with slightly different headlines. NewsNinja’s Scientifically option groups similar titles so you can explore related coverage together.
How grouping works
NewsNinja compares meaningful words in titles, removing common words such as “the” and “and.” Recent items with sufficient similarity form groups. On aggregate Home screens, groups containing multiple items start collapsed and can be expanded.
In Settings → News & Articles, choose the comparison method:
- Tversky balances shared words against words unique to each title.
- Overlap measures how much of the shorter title’s wording is shared.
- Cosine compares weighted word patterns, giving distinctive terms more influence.
The Threshold controls strictness. Lower values allow looser matches; higher values require closer matches. The default is 0.300. Headline similarity can miss differently worded coverage or group unrelated stories.
How ranking works
Scientific mode ranks items before grouping, then orders groups by their leading item. Ordinary items receive a recency boost that fades over 18 hours. Reddit items with engagement metadata and Hacker News items combine their score with a freshness factor that fades over nine hours. Score influence grows gradually, rather than directly in proportion to vote totals.
Comment counts are not a direct ranking signal in this code. Pinned X posts receive priority. Items older than 36 hours remain standalone rather than entering similarity groups.
Choose Newest first, Oldest first, or By feed when you prefer another arrangement. Reddit-only views use Reddit’s own sorting options.