fix: harden related concept suggestions

Author: ieduer

backend/main.py

@@ -1556,7 +1556,7 @@ def related(
     q: str = Query(..., min_length=1, max_length=200),
     limit: int = Query(8, ge=1, le=20),
 ):
-    """Find concepts that co-occur with the query term."""
+    """Find recognized concepts that co-occur with the query term."""
     con = get_db()
     try:
         clean_q = re.sub(r'[^\w\u4e00-\u9fff\s]', '', q).strip()
@@ -1565,7 +1565,7 @@ def related(
 
         # Get text chunks matching the query
         rows = con.execute("""
-            SELECT c.text
+            SELECT c.subject, c.text
             FROM chunks c
             JOIN chunks_fts f ON c.id = f.rowid
             WHERE chunks_fts MATCH ?
@@ -1575,26 +1575,26 @@ def related(
         if not rows:
             return []
 
-        # Extract Chinese word candidates (2-4 char sequences) from matching chunks
+        # Aggregate only exact concept hits already recognized by concept_map.
         word_counter = Counter()
         query_chars = set(clean_q)
         for r in rows:
             text = r["text"] or ""
-            # Find Chinese word-like sequences (2-4 chars)
-            words = re.findall(r'[\u4e00-\u9fff]{2,4}', text)
-            for w in words:
-                # Skip if the word is part of the query or too generic
+            subject = r["subject"] or ""
+            for concept in _match_concepts(text, subject, con):
+                w = concept["concept"]
                 if w == clean_q or w in clean_q or clean_q in w:
                     continue
+                if w in GRAPH_GENERIC_TERMS:
+                    continue
                 if len(w) < 2:
                     continue
                 word_counter[w] += 1
 
-        # Return top co-occurring terms (appearing in multiple chunks)
         candidates = [
             {"term": term, "count": count}
             for term, count in word_counter.most_common(limit * 3)
-            if count >= 2  # must appear in at least 2 chunks
+            if count >= 2
         ][:limit]
 
         return candidates

frontend/assets/app.js

@@ -609,7 +609,7 @@ async function doSearch(q) {
         currentData = await res.json();
         renderResults(currentData);
         // Load related concepts
-        loadRelated(q);
+        loadRelated(currentData, q);
         // Refresh trending after search (new query logged)
         setTimeout(() => loadTrending(), 500);
         // Show concept subgraph for the search term
@@ -620,10 +620,29 @@ async function doSearch(q) {
 }
 
 // ── Related Concepts ──────────────────────────────────────
-async function loadRelated(q) {
+async function loadRelated(searchData, q) {
     try {
-        const res = await fetch(`${API}/api/related?q=${encodeURIComponent(q)}&limit=10`);
-        const data = await res.json();
+        const conceptCounts = new Map();
+        const groups = Array.isArray(searchData?.groups) ? searchData.groups : [];
+        for (const item of groups) {
+            const concepts = Array.isArray(item?.matched_concepts) ? item.matched_concepts : [];
+            for (const concept of concepts) {
+                const term = String(concept || '').trim();
+                if (!term || term === q || term.includes(q) || q.includes(term)) continue;
+                conceptCounts.set(term, (conceptCounts.get(term) || 0) + 1);
+            }
+        }
+
+        let data = Array.from(conceptCounts.entries())
+            .sort((a, b) => b[1] - a[1] || a[0].length - b[0].length)
+            .slice(0, 10)
+            .map(([term, count]) => ({ term, count }));
+
+        if (data.length === 0) {
+            const res = await fetch(`${API}/api/related?q=${encodeURIComponent(q)}&limit=10`);
+            data = await res.json();
+        }
+
         if (data.length > 0) {
             relatedBarEl.innerHTML = `
                 <span class="related-label">🔗 相关概念：</span>

frontend/assets/version.json

@@ -1,7 +1,12 @@
 {
-  "frontend_refactor_version": "2026.03.06-r9",
+  "frontend_refactor_version": "2026.03.06-r10",
   "updated_at": "2026-03-06",
   "history": [
+    {
+      "version": "2026.03.06-r10",
+      "date": "2026-03-06",
+      "summary": "search related concepts now aggregate recognized matched_concepts instead of broken 2-4 character fragments"
+    },
     {
       "version": "2026.03.06-r9",
       "date": "2026-03-06",

frontend/index.html

@@ -23,7 +23,7 @@
     <script defer src="https://unpkg.com/katex@0.16.9/dist/contrib/auto-render.min.js"></script>
     <script src="https://d3js.org/d3.v7.min.js"></script>
 
-    <link rel="stylesheet" href="assets/style.css?v=20260306h">
+    <link rel="stylesheet" href="assets/style.css?v=20260306i">
 </head>
 
 <body>
@@ -286,7 +286,7 @@ <h3>🔓 开源与反馈</h3>
             <p id="footer-version-line">AI 高中教材 · 开源项目 · MIT License · 前端版本加载中…</p>
         </footer>
     </div>
-    <script src="assets/app.js?v=20260306h"></script>
+    <script src="assets/app.js?v=20260306i"></script>
 </body>
 
 </html>