Added semi-automatic mode to the metadata filter (#11886)

### What problem does this PR solve?

Retrieval metadata filtering adds semi-automatic mode, and users can
manually check the metadata key that participates in LLM to generate
filter conditions.
### Type of change

- [x] New Feature (non-breaking change which adds functionality)
This commit is contained in:
TeslaZY
2025-12-11 10:45:21 +08:00
committed by GitHub
parent a6afb7dfe2
commit c610bb605a
12 changed files with 246 additions and 3 deletions

View File

@ -327,10 +327,44 @@ async def retrieval_test():
local_doc_ids.extend(meta_filter(metas, filters["conditions"], filters.get("logic", "and")))
if not local_doc_ids:
local_doc_ids = None
elif meta_data_filter.get("method") == "semi_auto":
selected_keys = meta_data_filter.get("semi_auto", [])
if selected_keys:
filtered_metas = {key: metas[key] for key in selected_keys if key in metas}
if filtered_metas:
chat_mdl = LLMBundle(user_id, LLMType.CHAT, llm_name=search_config.get("chat_id", ""))
filters: dict = gen_meta_filter(chat_mdl, filtered_metas, question)
local_doc_ids.extend(meta_filter(metas, filters["conditions"], filters.get("logic", "and")))
if not local_doc_ids:
local_doc_ids = None
elif meta_data_filter.get("method") == "manual":
local_doc_ids.extend(meta_filter(metas, meta_data_filter["manual"], meta_data_filter.get("logic", "and")))
if meta_data_filter["manual"] and not local_doc_ids:
local_doc_ids = ["-999"]
else:
meta_data_filter = req.get("meta_data_filter")
if meta_data_filter:
metas = DocumentService.get_meta_by_kbs(kb_ids)
if meta_data_filter.get("method") == "auto":
chat_mdl = LLMBundle(user_id, LLMType.CHAT)
filters: dict = gen_meta_filter(chat_mdl, metas, question)
local_doc_ids.extend(meta_filter(metas, filters["conditions"], filters.get("logic", "and")))
if not local_doc_ids:
local_doc_ids = None
elif meta_data_filter.get("method") == "semi_auto":
selected_keys = meta_data_filter.get("semi_auto", [])
if selected_keys:
filtered_metas = {key: metas[key] for key in selected_keys if key in metas}
if filtered_metas:
chat_mdl = LLMBundle(user_id, LLMType.CHAT)
filters: dict = gen_meta_filter(chat_mdl, filtered_metas, question)
local_doc_ids.extend(meta_filter(metas, filters["conditions"], filters.get("logic", "and")))
if not local_doc_ids:
local_doc_ids = None
elif meta_data_filter.get("method") == "manual":
local_doc_ids.extend(meta_filter(metas, meta_data_filter["manual"], meta_data_filter.get("logic", "and")))
if meta_data_filter["manual"] and not local_doc_ids:
local_doc_ids = ["-999"]
tenants = UserTenantService.query(user_id=user_id)
for kb_id in kb_ids:

View File

@ -984,10 +984,45 @@ async def retrieval_test_embedded():
local_doc_ids.extend(meta_filter(metas, filters["conditions"], filters.get("logic", "and")))
if not local_doc_ids:
local_doc_ids = None
elif meta_data_filter.get("method") == "semi_auto":
selected_keys = meta_data_filter.get("semi_auto", [])
if selected_keys:
filtered_metas = {key: metas[key] for key in selected_keys if key in metas}
if filtered_metas:
chat_mdl = LLMBundle(tenant_id, LLMType.CHAT, llm_name=search_config.get("chat_id", ""))
filters: dict = gen_meta_filter(chat_mdl, filtered_metas, _question)
local_doc_ids.extend(meta_filter(metas, filters["conditions"], filters.get("logic", "and")))
if not local_doc_ids:
local_doc_ids = None
elif meta_data_filter.get("method") == "manual":
local_doc_ids.extend(meta_filter(metas, meta_data_filter["manual"], meta_data_filter.get("logic", "and")))
if meta_data_filter["manual"] and not local_doc_ids:
local_doc_ids = ["-999"]
else:
meta_data_filter = req.get("meta_data_filter")
if meta_data_filter:
metas = DocumentService.get_meta_by_kbs(kb_ids)
if meta_data_filter.get("method") == "auto":
chat_mdl = LLMBundle(tenant_id, LLMType.CHAT)
filters: dict = gen_meta_filter(chat_mdl, metas, question)
local_doc_ids.extend(meta_filter(metas, filters["conditions"], filters.get("logic", "and")))
if not local_doc_ids:
local_doc_ids = None
elif meta_data_filter.get("method") == "semi_auto":
selected_keys = meta_data_filter.get("semi_auto", [])
if selected_keys:
filtered_metas = {key: metas[key] for key in selected_keys if key in metas}
if filtered_metas:
chat_mdl = LLMBundle(tenant_id, LLMType.CHAT)
filters: dict = gen_meta_filter(chat_mdl, filtered_metas, question)
local_doc_ids.extend(meta_filter(metas, filters["conditions"], filters.get("logic", "and")))
if not local_doc_ids:
local_doc_ids = None
elif meta_data_filter.get("method") == "manual":
local_doc_ids.extend(meta_filter(metas, meta_data_filter["manual"], meta_data_filter.get("logic", "and")))
if meta_data_filter["manual"] and not local_doc_ids:
local_doc_ids = ["-999"]
tenants = UserTenantService.query(user_id=tenant_id)
for kb_id in kb_ids:

View File

@ -425,6 +425,15 @@ async def async_chat(dialog, messages, stream=True, **kwargs):
attachments.extend(meta_filter(metas, filters["conditions"], filters.get("logic", "and")))
if not attachments:
attachments = None
elif dialog.meta_data_filter.get("method") == "semi_auto":
selected_keys = dialog.meta_data_filter.get("semi_auto", [])
if selected_keys:
filtered_metas = {key: metas[key] for key in selected_keys if key in metas}
if filtered_metas:
filters: dict = gen_meta_filter(chat_mdl, filtered_metas, questions[-1])
attachments.extend(meta_filter(metas, filters["conditions"], filters.get("logic", "and")))
if not attachments:
attachments = None
elif dialog.meta_data_filter.get("method") == "manual":
conds = dialog.meta_data_filter["manual"]
attachments.extend(meta_filter(metas, conds, dialog.meta_data_filter.get("logic", "and")))
@ -834,6 +843,15 @@ async def async_ask(question, kb_ids, tenant_id, chat_llm_name=None, search_conf
doc_ids.extend(meta_filter(metas, filters["conditions"], filters.get("logic", "and")))
if not doc_ids:
doc_ids = None
elif meta_data_filter.get("method") == "semi_auto":
selected_keys = meta_data_filter.get("semi_auto", [])
if selected_keys:
filtered_metas = {key: metas[key] for key in selected_keys if key in metas}
if filtered_metas:
filters: dict = gen_meta_filter(chat_mdl, filtered_metas, question)
doc_ids.extend(meta_filter(metas, filters["conditions"], filters.get("logic", "and")))
if not doc_ids:
doc_ids = None
elif meta_data_filter.get("method") == "manual":
doc_ids.extend(meta_filter(metas, meta_data_filter["manual"], meta_data_filter.get("logic", "and")))
if meta_data_filter["manual"] and not doc_ids:
@ -909,6 +927,15 @@ async def gen_mindmap(question, kb_ids, tenant_id, search_config={}):
doc_ids.extend(meta_filter(metas, filters["conditions"], filters.get("logic", "and")))
if not doc_ids:
doc_ids = None
elif meta_data_filter.get("method") == "semi_auto":
selected_keys = meta_data_filter.get("semi_auto", [])
if selected_keys:
filtered_metas = {key: metas[key] for key in selected_keys if key in metas}
if filtered_metas:
filters: dict = gen_meta_filter(chat_mdl, filtered_metas, question)
doc_ids.extend(meta_filter(metas, filters["conditions"], filters.get("logic", "and")))
if not doc_ids:
doc_ids = None
elif meta_data_filter.get("method") == "manual":
doc_ids.extend(meta_filter(metas, meta_data_filter["manual"], meta_data_filter.get("logic", "and")))
if meta_data_filter["manual"] and not doc_ids: