@absolutejs/absolute 0.19.0-beta.496 → 0.19.0-beta.497

package/dist/ai/index.js

@@ -2165,6 +2165,31 @@ var collectMetadataStrings = (value) => {
 };
 var normalizeSourceForLexical = (source) => source.replace(/[#/_.-]+/g, " ").replace(/\bmd\b/g, "markdown").replace(/\bpptx\b/g, "presentation").replace(/\bxlsx\b/g, "spreadsheet workbook sheet").replace(/\bmp3\b/g, "audio transcript media").replace(/\bmp4\b/g, "video transcript media").replace(/\bzip\b/g, "archive bundle");
 var toFieldText = (value) => collectMetadataStrings(value).filter(Boolean).join(" ");
+var normalizeLooseText = (value) => value.toLowerCase().replace(/[^a-z0-9]+/g, " ").trim().replace(/\s+/g, " ");
+var scoreLoosePhraseMatch = (query, text) => {
+  const normalizedQuery = normalizeLooseText(query);
+  const normalizedText = normalizeLooseText(text ?? "");
+  if (normalizedQuery.length === 0 || normalizedText.length === 0) {
+    return 0;
+  }
+  if (normalizedText.includes(normalizedQuery)) {
+    return 1;
+  }
+  const words = normalizedQuery.split(" ").filter(Boolean);
+  for (let size = Math.min(5, words.length);size >= 2; size -= 1) {
+    for (let index = 0;index <= words.length - size; index += 1) {
+      const phraseWords = words.slice(index, index + size);
+      if (phraseWords.every((word) => STOP_WORDS.has(word))) {
+        continue;
+      }
+      const phrase = phraseWords.join(" ");
+      if (normalizedText.includes(phrase)) {
+        return Math.min(1, size / 4);
+      }
+    }
+  }
+  return 0;
+};
 var scoreTokenCoverage = (queryTokens, text) => {
   const normalizedText = (text ?? "").toLowerCase();
   if (normalizedText.length === 0) {
@@ -2181,10 +2206,8 @@ var scoreTokenCoverage = (queryTokens, text) => {
 var scorePhraseMatch = (query, text) => {
   const normalizedQuery = tokenize(query).join(" ");
   const normalizedText = tokenize(text ?? "").join(" ");
-  if (normalizedQuery.length === 0 || normalizedText.length === 0) {
-    return 0;
-  }
-  return normalizedText.includes(normalizedQuery) ? 1 : 0;
+  const tokenPhraseMatch = normalizedQuery.length > 0 && normalizedText.length > 0 ? normalizedText.includes(normalizedQuery) ? 1 : 0 : 0;
+  return Math.max(tokenPhraseMatch, scoreLoosePhraseMatch(query, text ?? ""));
 };
 var scoreWeightedField = ({
   coverageWeight,
@@ -2774,7 +2797,36 @@ var collectMetadataStrings2 = (value) => {
   }
   return [];
 };
-var scoreHeuristicMatch = (queryTokens, result) => {
+var normalizeLooseText2 = (value) => value.toLowerCase().replace(/[^a-z0-9]+/g, " ").trim().replace(/\s+/g, " ");
+var scoreLoosePhraseMatch2 = (query, text) => {
+  const normalizedQuery = normalizeLooseText2(query);
+  const normalizedText = normalizeLooseText2(text);
+  if (normalizedQuery.length === 0 || normalizedText.length === 0) {
+    return 0;
+  }
+  if (normalizedText.includes(normalizedQuery)) {
+    return 1;
+  }
+  const words = normalizedQuery.split(" ").filter(Boolean);
+  for (let size = Math.min(5, words.length);size >= 2; size -= 1) {
+    for (let index = 0;index <= words.length - size; index += 1) {
+      const phraseWords = words.slice(index, index + size);
+      if (phraseWords.every((word) => STOP_WORDS3.has(word))) {
+        continue;
+      }
+      const phrase = phraseWords.join(" ");
+      if (normalizedText.includes(phrase)) {
+        return Math.min(1, size / 4);
+      }
+    }
+  }
+  return 0;
+};
+var scoreHeuristicMatch = ({
+  query,
+  queryTokens,
+  result
+}) => {
   if (queryTokens.length === 0) {
     return result.score;
   }
@@ -2783,8 +2835,7 @@ var scoreHeuristicMatch = (queryTokens, result) => {
   const haystackSet = new Set(haystack);
   const overlap = queryTokens.filter((token) => haystackSet.has(token)).length;
   const overlapBoost = overlap / queryTokens.length;
-  const normalizedQuery = queryTokens.join(" ");
-  const exactPhraseBoost = normalizeText([result.title, result.source, result.chunkText, ...metadataValues].filter(Boolean).join(" ")).includes(normalizedQuery) ? 1 : 0;
+  const exactPhraseBoost = Math.max(normalizeText([result.title, result.source, result.chunkText, ...metadataValues].filter(Boolean).join(" ")).includes(queryTokens.join(" ")) ? 1 : 0, scoreLoosePhraseMatch2(query, [result.title, result.source, result.chunkText, ...metadataValues].filter(Boolean).join(" ")));
   const sourcePathBoost = typeof result.source === "string" && queryTokens.some((token) => result.source?.toLowerCase().includes(token)) ? 0.5 : 0;
   const metadataBoost = metadataValues.length > 0 ? queryTokens.filter((token) => metadataValues.some((value) => value.toLowerCase().includes(token))).length / queryTokens.length : 0;
   return result.score + overlapBoost + exactPhraseBoost + sourcePathBoost + metadataBoost;
@@ -2812,7 +2863,11 @@ var createHeuristicRAGReranker = (options = {}) => createRAGReranker({
     return [...results].map((result, index) => ({
       index,
       result,
-      score: scoreHeuristicMatch(queryTokens, result)
+      score: scoreHeuristicMatch({
+        query,
+        queryTokens,
+        result
+      })
     })).sort((left, right) => {
       if (right.score !== left.score) {
         return right.score - left.score;
@@ -8900,5 +8955,5 @@ export {
   aiChat
 };
 
-//# debugId=7CB5678D47F7B89564756E2164756E21
+//# debugId=55FD05298CEAFBDB64756E2164756E21
 //# sourceMappingURL=index.js.map