yumoqing/rag
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: 纯标签检索支持非文本文档（视频/图片等）

Browse Source 
- tag-only 检索从 documents 表查询而非仅查 document_chunks
- 对无 chunks 的文档显示文件名和类型图标
- 有 chunks 的文本文档显示内容预览
This commit is contained in:
ymq 2026-08-10 16:56:54 +08:00
parent 3425c09188
commit f181de9a10
1 changed files with 41 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

49
wwwroot/knowledge_bases_list/search_result.dspy
View File

@ -153,24 +153,57 @@ if query:
hits = hits[:top_k]
else:
# Tag-only search: direct DB lookup for chunks of tagged documents
# Tag-only search: query documents table directly (handles text + non-text docs)
if tag_doc_ids:
try:
async with get_sor_context(env, 'rag') as sor:
kb_cond = ""
placeholders2 = []
nsq2 = {}
kb_cond = ""
if kb_id and kb_id != "all":
kb_cond = "kb_id=${kb_id}$ AND "
nsq2["kb_id"] = kb_id
placeholders2 = []
for i, did in enumerate(tag_doc_ids):
placeholders2.append("${did_" + str(i) + "}$")
nsq2["did_" + str(i)] = did
sql2 = ("SELECT id, doc_id, content FROM document_chunks WHERE " + kb_cond +
"doc_id IN (" + ",".join(placeholders2) + ") ORDER BY updated_at DESC LIMIT " + str(top_k))
crecs = await sor.sqlExe(sql2, nsq2)
for r in crecs:
hits.append({"id": r.id, "doc_id": r.doc_id or "", "score": 1.0, "text": r.content or '', "kw": False})
# Get documents
sql2 = ("SELECT id, file_name, kb_id FROM documents WHERE " + kb_cond +
"id IN (" + ",".join(placeholders2) + ") ORDER BY updated_at DESC LIMIT " + str(top_k * 2))
docs = await sor.sqlExe(sql2, nsq2)
doc_list = [(r.id, r.file_name or '', r.kb_id or '') for r in docs]
if doc_list:
doc_ids = [d[0] for d in doc_list]
# Try to get chunks for these docs
chunk_placeholders = []
nsq_c = {}
for i, did in enumerate(doc_ids):
chunk_placeholders.append("${cdid_" + str(i) + "}$")
nsq_c["cdid_" + str(i)] = did
chunks_sql = ("SELECT id, doc_id, content FROM document_chunks WHERE doc_id IN (" +
",".join(chunk_placeholders) + ") ORDER BY updated_at DESC LIMIT " + str(top_k))
chunk_recs = await sor.sqlExe(chunks_sql, nsq_c)
chunks_by_doc = {}
for cr in chunk_recs:
if cr.doc_id not in chunks_by_doc:
chunks_by_doc[cr.doc_id] = []
chunks_by_doc[cr.doc_id].append((cr.id, cr.content or ''))
for did, fname, kbid in doc_list[:top_k]:
chs = chunks_by_doc.get(did, [])
if chs:
for cid, ctext in chs[:2]: # up to 2 chunks per doc
hits.append({"id": cid, "doc_id": did, "score": 1.0, "text": ctext, "file_name": fname, "kw": False})
else:
# Non-text document: use file_name as display
ext = fname.rsplit('.', 1)[-1] if '.' in fname else 'file'
icon_map = {'png':'🖼', 'jpg':'🖼', 'jpeg':'🖼', 'gif':'🖼', 'webp':'🖼',
'mp4':'🎬', 'avi':'🎬', 'mov':'🎬', 'mkv':'🎬',
'mp3':'🎵', 'wav':'🎵', 'ogg':'🎵',
'pdf':'📄', 'doc':'📝', 'docx':'📝', 'txt':'📝'}
icon = icon_map.get(ext.lower(), '📎')
display_text = f"{icon} {fname} (非文本文档)"
hits.append({"id": did, "doc_id": did, "score": 1.0, "text": display_text, "file_name": fname, "kw": False, "is_media": True})
except Exception as e:
info(f'[search_result] tag-only lookup error: {e}')