msoukhomlinov/n8n-nodes-hudu v2.10.0

on GitHub

Added

• Confidence signal for hudu_articles name lookups (issue #42). getAll responses for name-intent calls now include bestMatchScore and noConfidentMatch, so callers know to retry with a shorter fragment or search instead of trusting a weak top hit. Confidence fires when the top-ranked title either contains the full query as a substring OR contains every distinctive (non-stopword) query token when there are at least two of them — so a correctly-ranked reworded/reordered title (all distinctive words present, different order) is treated as confident rather than falsely flagged, while a partial overlap or a lone common-word coincidence still reports noConfidentMatch: true.

Fixed

• hudu_articles name fuzzy matching diluted by common words in long titles (issue #42). Tier-2 title-word-overlap scoring now strips function words (how, to, on, up, etc.) before ranking, so a long multi-word title's distinctive terms aren't drowned out by shared stopwords. All-stopword queries fall back to raw tokens so they never zero out. it is excluded from the stopword list — Hudu is an IT documentation tool, so IT is a common content word in real titles (e.g. "IT Glue Import") that shouldn't be stripped.
• name and search description text corrected. Both parameters already shared one matcher (title + body, 100-candidate re-rank, exact-title substring boosted to top) — the descriptions previously implied they used different logic. Text now says so explicitly, recommends a short distinctive fragment over a full title, and notes possible indexing lag for very recently created articles.
• Tier-2 token matching now requires a whole-word hit, not a bare substring. titleMatchScore and isConfidentTitleMatch previously used plain String.includes() for each query token, so a short distinctive token (e.g. it) could spuriously match inside an unrelated longer word in the title (e.g. split), undermining both the ranking and the new confidence signal. Token checks are now boundary-aware (word-boundary regex).
• Tier-1 full-query substring check now requires a whole-word boundary too. The tier-1 check still used a bare substring test, so a single-token query (e.g. IT) could match mid-word inside an unrelated title (e.g. Digital) and report noConfidentMatch: false before the tier-2 two-token floor got a chance to block it. Both titleMatchScore's ranking boost and isConfidentTitleMatch's tier-1 check now reuse the same whole-word boundary helper as tier 2.
• Whole-word boundary check now treats underscores as a boundary too. Query tokens are split on _ (among other delimiters), but the whole-word regex previously excluded _ from its boundary class, so an underscore-separated title (e.g. MFA_Office365) never matched tokens from a space-separated query (MFA Office365) and scored 0 despite being the correct match. Underscore now counts as a boundary character on both sides, matching the query-side tokenization.
• Whole-word boundary check now tolerates a pluralized title word. The underscore-boundary fix above made the boundary check too strict for plain English pluralization: a singular query word/phrase (e.g. Reset Password) no longer matched a pluralized title (e.g. Reset Passwords), losing both the tier-1 substring boost and the tier-2 token hit, which could leave isConfidentTitleMatch reporting false for an obvious match. The right-hand boundary now also accepts a single trailing literal s before the real boundary, so password/passwords, article/articles, vlan/vlans, etc. still match — while a bare mid-word substring (e.g. it inside split or items) still correctly fails, since only a lone trailing s is special-cased, not arbitrary trailing characters.
• Plural allowance no longer applies to 2-letter tokens. The trailing-s allowance above let a bare 2-letter acronym query (e.g. IT) falsely whole-word-match an unrelated title containing the distinct word its (a possessive pronoun, not the plural of it), reintroducing the original acronym-collision bug for short tokens specifically. The allowance now only applies to tokens of 3+ characters, where this collision doesn't occur in practice.