@wcs-colab/plugin-fuzzy-phrase 3.1.16-custom.newbase.2 → 3.1.16-custom.newbase.20

package/dist/index.js

@@ -48,9 +48,6 @@ function fuzzyMatch(word, queryToken, tolerance) {
   if (word === queryToken) {
     return { matches: true, distance: 0, score: 1 };
   }
-  if (word.startsWith(queryToken)) {
-    return { matches: true, distance: 0, score: 0.95 };
-  }
   const result = boundedLevenshtein(word, queryToken, tolerance);
   if (result.isBounded) {
     const score = 1 - result.distance * 0.2;
@@ -180,23 +177,53 @@ function filterCandidatesByScore(candidatesMap, minScore) {
 }
 
 // src/scoring.ts
-function findPhrasesInDocument(documentTokens, candidatesMap, config, documentFrequency, totalDocuments) {
+function findPhrasesInDocument(documentTokens, candidatesMap, config, documentFrequency, totalDocuments, originalQueryTokens, docPositions) {
   const phrases = [];
-  const queryTokens = Array.from(candidatesMap.keys());
+  const queryTokens = originalQueryTokens;
   const wordMatches = [];
-  for (let i = 0; i < documentTokens.length; i++) {
-    const docWord = documentTokens[i];
+  if (docPositions) {
+    const candidateLookup = /* @__PURE__ */ new Map();
     for (const [queryToken, candidates] of candidatesMap.entries()) {
       for (const candidate of candidates) {
-        if (candidate.word === docWord) {
-          wordMatches.push({
-            word: docWord,
-            queryToken,
-            position: i,
-            type: candidate.type,
-            distance: candidate.distance,
-            score: candidate.score
-          });
+        if (!candidateLookup.has(candidate.word)) {
+          candidateLookup.set(candidate.word, []);
+        }
+        candidateLookup.get(candidate.word).push({ queryToken, candidate });
+      }
+    }
+    for (const [docWord, positions] of Object.entries(docPositions)) {
+      const matches = candidateLookup.get(docWord);
+      if (matches) {
+        for (const { queryToken, candidate } of matches) {
+          for (const position of positions) {
+            wordMatches.push({
+              word: docWord,
+              queryToken,
+              position,
+              type: candidate.type,
+              distance: candidate.distance,
+              score: candidate.score
+            });
+          }
+        }
+      }
+    }
+    wordMatches.sort((a, b) => a.position - b.position);
+  } else {
+    for (let i = 0; i < documentTokens.length; i++) {
+      const docWord = documentTokens[i];
+      for (const [queryToken, candidates] of candidatesMap.entries()) {
+        for (const candidate of candidates) {
+          if (candidate.word === docWord) {
+            wordMatches.push({
+              word: docWord,
+              queryToken,
+              position: i,
+              type: candidate.type,
+              distance: candidate.distance,
+              score: candidate.score
+            });
+          }
         }
       }
     }
@@ -209,47 +236,76 @@ function findPhrasesInDocument(documentTokens, candidatesMap, config, documentFr
       config,
       documentFrequency,
       totalDocuments,
-      wordMatches
-      // Pass all word matches for density calculation
+      wordMatches,
+      documentTokens
+      // Pass document tokens to extract gap words
     );
     if (phrase && phrase.words.length > 0) {
       phrases.push(phrase);
     }
   }
-  return deduplicatePhrases(phrases);
+  const minTokensRequired = queryTokens.length >= 3 ? 2 : 1;
+  const filteredPhrases = phrases.filter((p) => p.words.length >= minTokensRequired);
+  return deduplicatePhrases(filteredPhrases);
 }
-function buildPhraseFromPosition(wordMatches, startIndex, queryTokens, config, documentFrequency, totalDocuments, allWordMatches) {
+function buildPhraseFromPosition(wordMatches, startIndex, queryTokens, config, documentFrequency, totalDocuments, allWordMatches, documentTokens) {
   const startMatch = wordMatches[startIndex];
   const phraseWords = [startMatch];
-  const coveredTokens = /* @__PURE__ */ new Set([startMatch.queryToken]);
+  const queryTokenCounts = /* @__PURE__ */ new Map();
+  for (const token of queryTokens) {
+    queryTokenCounts.set(token, (queryTokenCounts.get(token) || 0) + 1);
+  }
+  const matchedCounts = /* @__PURE__ */ new Map();
+  matchedCounts.set(startMatch.queryToken, 1);
+  const gapWords = [];
+  let totalGapUsed = 0;
+  let totalMatchedTokens = 1;
   for (let i = startIndex + 1; i < wordMatches.length; i++) {
     const match = wordMatches[i];
-    const gap = match.position - phraseWords[phraseWords.length - 1].position - 1;
+    const lastPos = phraseWords[phraseWords.length - 1].position;
+    const gap = match.position - lastPos - 1;
     if (gap > config.maxGap) {
       break;
     }
-    if (!coveredTokens.has(match.queryToken)) {
+    const neededCount = queryTokenCounts.get(match.queryToken) || 0;
+    const currentCount = matchedCounts.get(match.queryToken) || 0;
+    if (currentCount < neededCount) {
+      for (let pos = lastPos + 1; pos < match.position; pos++) {
+        totalGapUsed++;
+        gapWords.push({
+          word: documentTokens[pos],
+          position: pos,
+          gapIndex: totalGapUsed
+        });
+      }
       phraseWords.push(match);
-      coveredTokens.add(match.queryToken);
-    }
-    if (coveredTokens.size === queryTokens.length) {
-      break;
+      matchedCounts.set(match.queryToken, currentCount + 1);
+      totalMatchedTokens++;
+      if (totalMatchedTokens === queryTokens.length) {
+        break;
+      }
     }
   }
   if (phraseWords.length > 0) {
+    const coverage = phraseWords.length / queryTokens.length;
+    const span = phraseWords[phraseWords.length - 1].position - phraseWords[0].position + 1;
     const { score, breakdown } = calculatePhraseScore(
       phraseWords,
       queryTokens,
       config,
       documentFrequency,
       totalDocuments,
-      allWordMatches
+      allWordMatches,
+      coverage
     );
     return {
       words: phraseWords,
+      gapWords,
+      gapUsed: totalGapUsed,
+      coverage,
       startPosition: phraseWords[0].position,
       endPosition: phraseWords[phraseWords.length - 1].position,
-      gap: phraseWords[phraseWords.length - 1].position - phraseWords[0].position,
+      span,
       inOrder: isInOrder(phraseWords, queryTokens),
       score,
       scoreBreakdown: breakdown
@@ -257,7 +313,7 @@ function buildPhraseFromPosition(wordMatches, startIndex, queryTokens, config, d
   }
   return null;
 }
-function calculatePhraseScore(phraseWords, queryTokens, config, documentFrequency, totalDocuments, allWordMatches) {
+function calculatePhraseScore(phraseWords, queryTokens, config, documentFrequency, totalDocuments, allWordMatches, coverage) {
   let baseScore = 0;
   for (const word of phraseWords) {
     const weight = word.type === "exact" ? config.weights.exact : word.type === "fuzzy" ? config.weights.fuzzy : config.weights.fuzzy * 0.8;
@@ -266,14 +322,16 @@ function calculatePhraseScore(phraseWords, queryTokens, config, documentFrequenc
   baseScore /= phraseWords.length;
   const inOrder = isInOrder(phraseWords, queryTokens);
   const orderScore = inOrder ? 1 : 0.5;
-  const span = phraseWords[phraseWords.length - 1].position - phraseWords[0].position + 1;
-  const proximityScore = Math.max(0, 1 - span / (queryTokens.length * 5));
+  let proximityScore = 0;
+  if (config.maxGap > 0 && config.weights.proximity > 0 && queryTokens.length > 1) {
+    const span = phraseWords[phraseWords.length - 1].position - phraseWords[0].position + 1;
+    const proximityWindow = queryTokens.length * config.proximitySpanMultiplier;
+    proximityScore = Math.max(0, 1 - span / proximityWindow);
+  }
   let densityScore = 0;
   if (queryTokens.length === 1) {
     const totalOccurrences = allWordMatches.length;
-    densityScore = totalOccurrences / queryTokens.length;
-  } else {
-    densityScore = phraseWords.length / queryTokens.length;
+    densityScore = Math.min(1, totalOccurrences / 10);
   }
   const semanticScore = calculateSemanticScore(
     phraseWords,
@@ -287,8 +345,13 @@ function calculatePhraseScore(phraseWords, queryTokens, config, documentFrequenc
   const weightedDensity = densityScore * weights.density;
   const weightedSemantic = semanticScore * weights.semantic;
   const totalScore = weightedBase + weightedOrder + weightedProximity + weightedDensity + weightedSemantic;
-  const maxPossibleScore = 1 + weights.order + weights.proximity + weights.density + weights.semantic;
-  const score = totalScore / maxPossibleScore;
+  const canHaveFuzzyMatches = config.tolerance > 0 && weights.fuzzy > 0;
+  const maxBaseWeight = canHaveFuzzyMatches ? Math.max(weights.exact, weights.fuzzy) : weights.exact;
+  const effectiveProximityWeight = config.maxGap > 0 && weights.proximity > 0 && queryTokens.length > 1 ? weights.proximity : 0;
+  const maxPossibleScore = maxBaseWeight + weights.order + effectiveProximityWeight + weights.density + weights.semantic;
+  const normalizedScore = totalScore / maxPossibleScore;
+  const coverageMultiplier = queryTokens.length > 1 ? coverage : 1;
+  const score = normalizedScore * coverageMultiplier;
   const base = weightedBase / maxPossibleScore;
   const order = weightedOrder / maxPossibleScore;
   const proximity = weightedProximity / maxPossibleScore;
@@ -301,18 +364,27 @@ function calculatePhraseScore(phraseWords, queryTokens, config, documentFrequenc
       order,
       proximity,
       density,
-      semantic
+      semantic,
+      coverage: coverageMultiplier
+      // Show coverage multiplier in breakdown
     }
   };
 }
 function isInOrder(phraseWords, queryTokens) {
-  const tokenOrder = new Map(queryTokens.map((token, index) => [token, index]));
-  for (let i = 1; i < phraseWords.length; i++) {
-    const prevOrder = tokenOrder.get(phraseWords[i - 1].queryToken) ?? -1;
-    const currOrder = tokenOrder.get(phraseWords[i].queryToken) ?? -1;
-    if (currOrder < prevOrder) {
+  const tokenPositions = queryTokens.map((token, index) => ({ token, index }));
+  let lastMatchedIndex = -1;
+  for (const phraseWord of phraseWords) {
+    let foundIndex = -1;
+    for (const pos of tokenPositions) {
+      if (pos.token === phraseWord.queryToken && pos.index > lastMatchedIndex) {
+        foundIndex = pos.index;
+        break;
+      }
+    }
+    if (foundIndex === -1) {
       return false;
     }
+    lastMatchedIndex = foundIndex;
   }
   return true;
 }
@@ -355,7 +427,8 @@ function deduplicatePhrases(phrases) {
 
 // src/index.ts
 var DEFAULT_CONFIG = {
-  textProperty: "content",
+  textProperty: "normalized_content",
+  // Must match server's field name
   tolerance: 1,
   adaptiveTolerance: true,
   enableSynonyms: false,
@@ -370,7 +443,10 @@ var DEFAULT_CONFIG = {
     semantic: 0.15
   },
   maxGap: 5,
-  minScore: 0.1
+  minScore: 0.1,
+  enableFinalScoreMinimum: false,
+  finalScoreMinimum: 0.3,
+  proximitySpanMultiplier: 5
 };
 var pluginStates = /* @__PURE__ */ new WeakMap();
 function pluginFuzzyPhrase(userConfig = {}) {
@@ -390,7 +466,10 @@ function pluginFuzzyPhrase(userConfig = {}) {
       semantic: userConfig.weights?.semantic ?? DEFAULT_CONFIG.weights.semantic
     },
     maxGap: userConfig.maxGap ?? DEFAULT_CONFIG.maxGap,
-    minScore: userConfig.minScore ?? DEFAULT_CONFIG.minScore
+    minScore: userConfig.minScore ?? DEFAULT_CONFIG.minScore,
+    enableFinalScoreMinimum: userConfig.enableFinalScoreMinimum ?? DEFAULT_CONFIG.enableFinalScoreMinimum,
+    finalScoreMinimum: userConfig.finalScoreMinimum ?? DEFAULT_CONFIG.finalScoreMinimum,
+    proximitySpanMultiplier: userConfig.proximitySpanMultiplier ?? DEFAULT_CONFIG.proximitySpanMultiplier
   };
   const plugin = {
     name: "fuzzy-phrase",
@@ -403,7 +482,8 @@ function pluginFuzzyPhrase(userConfig = {}) {
         synonymMap: {},
         config,
         documentFrequency: /* @__PURE__ */ new Map(),
-        totalDocuments: 0
+        totalDocuments: 0,
+        vocabulary: /* @__PURE__ */ new Set()
       };
       if (config.enableSynonyms && config.supabase) {
         try {
@@ -420,6 +500,23 @@ function pluginFuzzyPhrase(userConfig = {}) {
         state.documentFrequency = calculateDocumentFrequencies(docs, config.textProperty);
         console.log(`\u{1F4CA} Calculated document frequencies for ${state.totalDocuments} documents`);
       }
+      try {
+        const indexData = orama.data?.index;
+        let radixNode = null;
+        if (indexData?.indexes?.[config.textProperty]?.node) {
+          radixNode = indexData.indexes[config.textProperty].node;
+        } else if (indexData?.[config.textProperty]?.node) {
+          radixNode = indexData[config.textProperty].node;
+        }
+        if (radixNode) {
+          state.vocabulary = extractVocabularyFromRadixTree(radixNode);
+          console.log(`\u{1F4DA} Cached ${state.vocabulary.size} vocabulary words (eliminates per-query radix traversal)`);
+        } else {
+          console.warn("\u26A0\uFE0F  Could not find radix tree for vocabulary caching");
+        }
+      } catch (error) {
+        console.error("\u26A0\uFE0F  Failed to cache vocabulary:", error);
+      }
       pluginStates.set(orama, state);
       console.log("\u2705 Fuzzy Phrase Plugin initialized");
       setImmediate(() => {
@@ -441,7 +538,7 @@ async function searchWithFuzzyPhrase(orama, params, language) {
     console.error("\u274C Plugin state not initialized");
     throw new Error("Fuzzy Phrase Plugin not properly initialized");
   }
-  const { term, properties } = params;
+  const { term, properties, tokenCache } = params;
   if (!term || typeof term !== "string") {
     return { elapsed: { formatted: "0ms", raw: 0 }, hits: [], count: 0 };
   }
@@ -452,32 +549,31 @@ async function searchWithFuzzyPhrase(orama, params, language) {
   }
   const tolerance = state.config.adaptiveTolerance ? calculateAdaptiveTolerance(queryTokens, state.config.tolerance) : state.config.tolerance;
   console.log(`\u{1F50D} Fuzzy phrase search: "${term}" (${queryTokens.length} tokens, tolerance: ${tolerance})`);
-  let vocabulary;
-  try {
-    const indexData = orama.data?.index;
-    if (!indexData) {
-      console.error("\u274C No index data found in orama.data.index");
-      return { elapsed: { formatted: "0ms", raw: 0 }, hits: [], count: 0 };
-    }
-    console.log("\u{1F50D} DEBUG: Index data keys:", Object.keys(indexData || {}));
-    let radixNode = null;
-    if (indexData.indexes?.[textProperty]?.node) {
-      radixNode = indexData.indexes[textProperty].node;
-      console.log("\u2705 Found radix via QPS-style path (data.index.indexes)");
-    } else if (indexData[textProperty]?.node) {
-      radixNode = indexData[textProperty].node;
-      console.log("\u2705 Found radix via standard path (data.index[property])");
-    }
-    if (!radixNode) {
-      console.error("\u274C Radix tree not found for property:", textProperty);
-      console.error("   Available properties in index:", Object.keys(indexData));
+  let vocabulary = state.vocabulary;
+  if (vocabulary.size === 0) {
+    console.log("\u{1F4DA} Vocabulary cache empty - extracting on first search...");
+    try {
+      const indexData = orama.data?.index;
+      let radixNode = null;
+      if (indexData?.indexes?.[textProperty]?.node) {
+        radixNode = indexData.indexes[textProperty].node;
+      } else if (indexData?.[textProperty]?.node) {
+        radixNode = indexData[textProperty].node;
+      }
+      if (radixNode) {
+        state.vocabulary = extractVocabularyFromRadixTree(radixNode);
+        vocabulary = state.vocabulary;
+        console.log(`\u{1F4DA} Cached ${vocabulary.size} vocabulary words (will be reused for subsequent queries)`);
+      } else {
+        console.error("\u274C Radix tree not found for vocabulary extraction");
+        return { elapsed: { formatted: "0ms", raw: 0 }, hits: [], count: 0 };
+      }
+    } catch (error) {
+      console.error("\u274C Failed to extract vocabulary:", error);
       return { elapsed: { formatted: "0ms", raw: 0 }, hits: [], count: 0 };
     }
-    vocabulary = extractVocabularyFromRadixTree(radixNode);
-    console.log(`\u{1F4DA} Extracted ${vocabulary.size} unique words from index`);
-  } catch (error) {
-    console.error("\u274C Failed to extract vocabulary:", error);
-    return { elapsed: { formatted: "0ms", raw: 0 }, hits: [], count: 0 };
+  } else {
+    console.log(`\u{1F4DA} Using cached vocabulary (${vocabulary.size} words)`);
   }
   const candidatesMap = findAllCandidates(
     queryTokens,
@@ -486,10 +582,7 @@ async function searchWithFuzzyPhrase(orama, params, language) {
     state.config.enableSynonyms ? state.synonymMap : void 0,
     state.config.synonymMatchScore
   );
-  const filteredCandidates = filterCandidatesByScore(
-    candidatesMap,
-    state.config.minScore
-  );
+  const filteredCandidates = tolerance === 0 ? candidatesMap : filterCandidatesByScore(candidatesMap, state.config.minScore);
   console.log(`\u{1F3AF} Found candidates: ${Array.from(filteredCandidates.values()).reduce((sum, c) => sum + c.length, 0)} total`);
   const documentMatches = [];
   console.log("\u{1F50D} DEBUG orama.data structure:", {
@@ -516,22 +609,48 @@ async function searchWithFuzzyPhrase(orama, params, language) {
       dataDocsDocsCount: orama.data?.docs?.docs ? Object.keys(orama.data.docs.docs).length : 0
     });
   }
-  console.log(`\u{1F4C4} Searching through ${Object.keys(docs).length} documents`);
+  const cacheHits = tokenCache ? tokenCache.size : 0;
+  let hasPositionalIndex = false;
+  if (tokenCache && tokenCache.size > 0) {
+    const firstEntry = tokenCache.values().next().value;
+    hasPositionalIndex = !!(firstEntry && !Array.isArray(firstEntry) && firstEntry.positions);
+  }
+  console.log(`\u{1F4C4} Searching through ${Object.keys(docs).length} documents (${hasPositionalIndex ? "\u26A1 positional index" : cacheHits > 0 ? "tokens cached" : "no cache"})`);
   for (const [docId, doc] of Object.entries(docs)) {
     const text = doc[textProperty];
     if (!text || typeof text !== "string") {
       continue;
     }
-    const docTokens = tokenize(text);
+    let docTokens;
+    let docPositions;
+    if (tokenCache && tokenCache.has(docId)) {
+      const cached = tokenCache.get(docId);
+      if (Array.isArray(cached)) {
+        docTokens = cached;
+      } else if (cached.tokens && cached.positions) {
+        docTokens = cached.tokens;
+        docPositions = cached.positions;
+      } else {
+        docTokens = tokenize(text);
+      }
+    } else {
+      docTokens = tokenize(text);
+    }
     const phrases = findPhrasesInDocument(
       docTokens,
       filteredCandidates,
       {
         weights: state.config.weights,
-        maxGap: state.config.maxGap
+        maxGap: state.config.maxGap,
+        proximitySpanMultiplier: state.config.proximitySpanMultiplier,
+        tolerance
       },
       state.documentFrequency,
-      state.totalDocuments
+      state.totalDocuments,
+      queryTokens,
+      // Original tokens with duplicates preserved
+      docPositions
+      // Positional index for O(matches) lookup
     );
     if (phrases.length > 0) {
       const docScore = Math.max(...phrases.map((p) => p.score));
@@ -544,8 +663,15 @@ async function searchWithFuzzyPhrase(orama, params, language) {
     }
   }
   documentMatches.sort((a, b) => b.score - a.score);
-  const limit = params.limit ?? documentMatches.length;
-  const limitedMatches = documentMatches.slice(0, limit);
+  let filteredMatches = documentMatches;
+  if (state.config.enableFinalScoreMinimum && state.config.finalScoreMinimum > 0) {
+    const threshold = state.config.finalScoreMinimum;
+    const beforeCount = filteredMatches.length;
+    filteredMatches = filteredMatches.filter((m) => m.score >= threshold);
+    console.log(`\u{1F39A}\uFE0F Final score filter: ${beforeCount} \u2192 ${filteredMatches.length} (threshold: ${threshold})`);
+  }
+  const limit = params.limit ?? filteredMatches.length;
+  const limitedMatches = filteredMatches.slice(0, limit);
   const hits = limitedMatches.map((match) => ({
     id: match.id,
     score: match.score,