Choose profiles and verb handling

Start with standard for rule-based analysis without a lexicon import. Use full when you need the bundled keep list, known broken plurals, and additional word or verb mappings.

import { createAnalyzer } from "fa-search-kit";
import { lexicon } from "fa-search-kit/lexicon";
const light = createAnalyzer({ profile: "light" });
const standard = createAnalyzer();
const full = createAnalyzer({ profile: "full", lexicon, verbs: "lemma" });
const stem = createAnalyzer({ profile: "full", lexicon, verbs: "stem" });
console.log(JSON.stringify(light.analyze("کتاب‌ها", { mode: "query" }))); // => ["کتابها"]
console.log(JSON.stringify(standard.analyze("کتاب‌ها", { mode: "query" }))); // => ["کتاب"]
console.log(JSON.stringify(full.analyze("می‌شود شد نمی‌کند", { mode: "query" }))); // => ["شد","شد","نکرد"]
console.log(JSON.stringify(stem.analyze("می‌شود شد نمی‌کند", { mode: "query" }))); // => ["شود","شد","نمیکن"]
console.log(JSON.stringify(full.analyze("کتب", { mode: "query" }))); // => ["کتاب"]

What each profile enables

light normalizes and tokenizes, rejoins recognized spaced affixes, and removes ZWNJ from final terms. It does not stem or use a lexicon. By default it still adds madda-less terms and half-space segments in index mode; it is not just normalize().

standard adds Snowball Persian stemming and wrapper rules: handling closed suffixes after ZWNJ, guarded joined می, some possessive clitics, and preservation of derivational suffixes. It needs no lexicon.

full requires an explicitly imported lexicon; otherwise construction throws. It adds keep words, known broken plurals, lemma edits, recognized verbs, and more aggressive joined-clitic stripping. The lexicon is finite and analysis has no sentence context. Passing lexicon directly to createAnalyzer() does not select full: set profile: "full". Adapters do infer full from a supplied lexicon unless you explicitly choose another profile.

Lemma versus stem

With full and verbs: "lemma", recognized conjugations map to a past-stem term, such as «می‌شود» and «شد» both yielding «شد». This broadens cross-tense matching but can reduce ranking distinctions. It is not a grammatical lemma extractor for arbitrary text.

With verbs: "stem", verb lemma lookup is disabled, while other full-profile word lists and clitic rules remain active. Outputs are search stems and need not be dictionary words. Negation defaults to keep; merge can conflate positive and negative forms and should be evaluated for your content.

Direct createAnalyzer() defaults to lemma behavior when full is selected. Adapters default to lemma for Pagefind, FlexSearch, and MiniSearch configured with AND; to stem for Orama, Lunr, and MiniSearch with OR. Supplying your own analyzer bypasses these adapter defaults.

Options and tradeoffs

rejoin: true, zwnj: "both", alefMadda: true, and hamzaYeh: true are analyzer defaults. spellings defaults to true for standard/full and false for light. zwnj: "keep" suppresses index-only compound parts; it does not retain ZWNJ in final terms. Disabling alternatives narrows matches and can reduce false matches.

Analyzer stem defaults are closedSplit: true, joinedMi: "rule", negation: "keep", and derivational: "keep". Standard trusts clitics after ZWNJ or plural suffixes; full also sets joinedMin: 3 and singleMin: 3. A supplied clitics object replaces this object rather than merging its fields.

Use the core reference for lower-level options and the current joinedMi: "lexicon" caveat. Rebuild your index when any term-producing setting or lexicon changes.