From a0dfab482f4cd93605628719c04568e6678ff46d Mon Sep 17 00:00:00 2001
From: YumemiDream <1803068130@qq.com>
Date: Thu, 4 Jun 2026 19:41:19 +0800
Subject: [PATCH 1/6] fix: prevent plugin from overwriting manually edited
jargon meanings
- Add meaning_edited flag to Jargon ORM and dataclass
- Set meaning_edited=True when user edits meaning via dashboard
- _should_infer_meaning skips jargon with meaning_edited=True
- Auto-migration will add the new column on startup
---
models/jargon.py | 2 ++
models/orm/jargon.py | 2 ++
services/database/facades/jargon_facade.py | 4 ++++
services/jargon/jargon_miner.py | 3 +++
webui/services/jargon_service.py | 1 +
5 files changed, 12 insertions(+)
diff --git a/models/jargon.py b/models/jargon.py
index a3272c5f..109b1d0b 100644
--- a/models/jargon.py
+++ b/models/jargon.py
@@ -20,6 +20,7 @@ class Jargon:
count: int = 1 # 出现次数
last_inference_count: int = 0 # 上次推断时的count值
is_complete: bool = False # 是否完成所有推断 (count>=100)
+ meaning_edited: bool = False # 用户是否手动编辑过释义
is_global: bool = False # 是否全局黑话
chat_id: str = "" # 群组ID
created_at: Optional[datetime] = None # 创建时间
@@ -36,6 +37,7 @@ def to_dict(self) -> dict:
'count': self.count,
'last_inference_count': self.last_inference_count,
'is_complete': self.is_complete,
+ 'meaning_edited': self.meaning_edited,
'is_global': self.is_global,
'chat_id': self.chat_id,
'created_at': self.created_at.isoformat() if self.created_at else None,
diff --git a/models/orm/jargon.py b/models/orm/jargon.py
index b8cf462f..cf313279 100644
--- a/models/orm/jargon.py
+++ b/models/orm/jargon.py
@@ -19,6 +19,7 @@ class Jargon(Base):
count = Column(Integer, default=1)
last_inference_count = Column(Integer, default=0)
is_complete = Column(Boolean, default=False)
+ meaning_edited = Column(Boolean, default=False)
is_global = Column(Boolean, default=False)
chat_id = Column(String(255), nullable=False, index=True)
# 使用 BigInteger 存储 Unix 时间戳（自动迁移会将 DATETIME 转换为 BIGINT）
@@ -46,6 +47,7 @@ def to_dict(self):
'count': self.count,
'last_inference_count': self.last_inference_count,
'is_complete': self.is_complete,
+ 'meaning_edited': self.meaning_edited,
'is_global': self.is_global,
'chat_id': self.chat_id,
'created_at': self.created_at,
diff --git a/services/database/facades/jargon_facade.py b/services/database/facades/jargon_facade.py
index f978fd59..448e1d20 100644
--- a/services/database/facades/jargon_facade.py
+++ b/services/database/facades/jargon_facade.py
@@ -164,6 +164,9 @@ async def update_jargon(self, jargon_data: Dict[str, Any]) -> bool:
record.meaning = json.dumps(meaning_val, ensure_ascii=False)
else:
record.meaning = str(meaning_val) if meaning_val is not None else None
+ # Only mark meaning_edited when explicitly set (not from inference)
+ if jargon_data.get('meaning_edited'):
+ record.meaning_edited = True
if 'is_jargon' in jargon_data:
record.is_jargon = jargon_data['is_jargon']
if 'count' in jargon_data:
@@ -476,6 +479,7 @@ async def search_jargon(
'is_jargon': r.is_jargon,
'count': r.count or 0,
'is_complete': r.is_complete,
+ 'meaning_edited': r.meaning_edited or False,
'is_global': r.is_global or False,
'chat_id': r.chat_id,
'created_at': r.created_at,
diff --git a/services/jargon/jargon_miner.py b/services/jargon/jargon_miner.py
index 97ae33a2..b30002f1 100644
--- a/services/jargon/jargon_miner.py
+++ b/services/jargon/jargon_miner.py
@@ -299,6 +299,8 @@ def _should_infer_meaning(self, jargon: Jargon) -> bool:
"""
if jargon.is_complete:
return False
+ if jargon.meaning_edited:
+ return False
count = jargon.count or 0
last_inference = jargon.last_inference_count or 0
@@ -512,6 +514,7 @@ async def save_or_update_jargon(
count=existing_dict.get('count', 1),
last_inference_count=existing_dict.get('last_inference_count', 0),
is_complete=existing_dict.get('is_complete', False),
+ meaning_edited=existing_dict.get('meaning_edited', False),
is_global=existing_dict.get('is_global', False),
chat_id=existing_dict.get('chat_id', ''),
created_at=existing_dict.get('created_at'),
diff --git a/webui/services/jargon_service.py b/webui/services/jargon_service.py
index ef4badb9..2f82192e 100644
--- a/webui/services/jargon_service.py
+++ b/webui/services/jargon_service.py
@@ -352,6 +352,7 @@ async def update_jargon(
payload["content"] = content
if meaning is not None:
payload["meaning"] = meaning
+ payload["meaning_edited"] = True
if len(payload) <= 1:
return False, "没有需要更新的字段", self._format_jargon_for_frontend(current)
From 0194dd411170b5c17afdc1fcc51f3ea3654d694f Mon Sep 17 00:00:00 2001
From: YumemiDream <1803068130@qq.com>
Date: Thu, 4 Jun 2026 19:58:42 +0800
Subject: [PATCH 2/6] fix: change jargon count to actual chat occurrence
frequency
- Add sync_jargon_counts to facade: bulk-update count from statistical
filter's term frequency table
- mine_jargon syncs filter frequencies to DB before inference
- Remove manual count+1 in save_or_update_jargon (count now managed
by frequency sync)
- Inference thresholds [3,6,10,20,40,60,100] now reflect actual chat
occurrences, not LLM validation pass count
---
services/database/facades/jargon_facade.py | 46 +++++++++++++++++++
.../database/sqlalchemy_database_manager.py | 3 ++
services/jargon/jargon_miner.py | 3 +-
services/learning/message_pipeline.py | 10 ++++
4 files changed, 60 insertions(+), 2 deletions(-)
diff --git a/services/database/facades/jargon_facade.py b/services/database/facades/jargon_facade.py
index 448e1d20..cba48273 100644
--- a/services/database/facades/jargon_facade.py
+++ b/services/database/facades/jargon_facade.py
@@ -195,6 +195,52 @@ async def update_jargon(self, jargon_data: Dict[str, Any]) -> bool:
self._logger.error(f"[JargonFacade] 更新黑话失败: {e}", exc_info=True)
return False
+ # 3b. sync_jargon_counts
+ async def sync_jargon_counts(
+ self, chat_id: str, term_frequencies: Dict[str, int]
+ ) -> int:
+ """Sync occurrence counts for existing jargon in a group.
+
+ Only updates jargon terms that already exist in DB for *chat_id*.
+ Returns the number of records updated.
+
+ Args:
+ chat_id: Group ID.
+ term_frequencies: ``{term: actual_chat_count}`` from the
+ statistical filter.
+ """
+ if not term_frequencies:
+ return 0
+
+ updated = 0
+ try:
+ async with self.get_session() as session:
+ stmt = select(Jargon).where(
+ Jargon.chat_id == chat_id,
+ Jargon.content.in_(list(term_frequencies.keys())),
+ )
+ result = await session.execute(stmt)
+ records = result.scalars().all()
+
+ now = int(time.time())
+ for record in records:
+ new_count = term_frequencies.get(record.content, 0)
+ if new_count > (record.count or 0):
+ record.count = new_count
+ record.updated_at = now
+ updated += 1
+
+ if updated:
+ await session.commit()
+ self._logger.debug(
+ f"[JargonFacade] 同步黑话出现次数: "
+ f"chat={chat_id}, updated={updated}"
+ )
+ except Exception as e:
+ self._logger.error(f"[JargonFacade] 同步黑话出现次数失败: {e}", exc_info=True)
+
+ return updated
+
# 4. get_jargon_statistics
async def get_jargon_statistics(self, group_id: str = None) -> Dict[str, Any]:
"""获取黑话学习统计信息
diff --git a/services/database/sqlalchemy_database_manager.py b/services/database/sqlalchemy_database_manager.py
index 82fc5b4a..e13af6bc 100644
--- a/services/database/sqlalchemy_database_manager.py
+++ b/services/database/sqlalchemy_database_manager.py
@@ -960,6 +960,9 @@ async def insert_jargon(self, jargon_data: Dict[str, Any]) -> Optional[int]:
async def update_jargon(self, jargon_data: Dict[str, Any]) -> bool:
return await self._call_jargon("update_jargon", False, jargon_data)
+ async def sync_jargon_counts(self, chat_id: str, term_frequencies: Dict[str, int]) -> int:
+ return await self._call_jargon("sync_jargon_counts", 0, chat_id, term_frequencies)
+
async def get_jargon_statistics(self, group_id: str = None) -> Dict[str, Any]:
return await self._call_jargon(
"get_jargon_statistics",
diff --git a/services/jargon/jargon_miner.py b/services/jargon/jargon_miner.py
index b30002f1..589d59d2 100644
--- a/services/jargon/jargon_miner.py
+++ b/services/jargon/jargon_miner.py
@@ -521,8 +521,7 @@ async def save_or_update_jargon(
updated_at=existing_dict.get('updated_at')
)
- # 更新现有记录
- existing.count = (existing.count or 0) + 1
+ # 更新现有记录（count 由频率同步管理，此处不递增）
# 合并 raw_content
existing_list = safe_parse_llm_json(existing.raw_content) or []
diff --git a/services/learning/message_pipeline.py b/services/learning/message_pipeline.py
index 6e8be3a2..3be8d547 100644
--- a/services/learning/message_pipeline.py
+++ b/services/learning/message_pipeline.py
@@ -256,6 +256,16 @@ async def mine_jargon(self, group_id: str) -> None:
if not statistical_candidates:
statistical_candidates = None
+ # Sync actual chat occurrence counts to DB
+ term_freq = self._jargon_statistical_filter._group_term_freq.get(group_id)
+ if term_freq:
+ try:
+ await self._db_manager.sync_jargon_counts(
+ group_id, dict(term_freq)
+ )
+ except Exception as e:
+ logger.debug(f"[JargonMining] Frequency sync failed: {e}")
+
await jargon_miner.run_once(
chat_messages,
len(recent_messages),
From 4ef77bf7761bf310e2d9ee9f201be93d3181d0bb Mon Sep 17 00:00:00 2001
From: YumemiDream <1803068130@qq.com>
Date: Thu, 4 Jun 2026 20:08:37 +0800
Subject: [PATCH 3/6] Revert "fix: change jargon count to actual chat
occurrence frequency"
This reverts commit 0194dd411170b5c17afdc1fcc51f3ea3654d694f.
---
services/database/facades/jargon_facade.py | 46 -------------------
.../database/sqlalchemy_database_manager.py | 3 --
services/jargon/jargon_miner.py | 3 +-
services/learning/message_pipeline.py | 10 ----
4 files changed, 2 insertions(+), 60 deletions(-)
diff --git a/services/database/facades/jargon_facade.py b/services/database/facades/jargon_facade.py
index cba48273..448e1d20 100644
--- a/services/database/facades/jargon_facade.py
+++ b/services/database/facades/jargon_facade.py
@@ -195,52 +195,6 @@ async def update_jargon(self, jargon_data: Dict[str, Any]) -> bool:
self._logger.error(f"[JargonFacade] 更新黑话失败: {e}", exc_info=True)
return False
- # 3b. sync_jargon_counts
- async def sync_jargon_counts(
- self, chat_id: str, term_frequencies: Dict[str, int]
- ) -> int:
- """Sync occurrence counts for existing jargon in a group.
-
- Only updates jargon terms that already exist in DB for *chat_id*.
- Returns the number of records updated.
-
- Args:
- chat_id: Group ID.
- term_frequencies: ``{term: actual_chat_count}`` from the
- statistical filter.
- """
- if not term_frequencies:
- return 0
-
- updated = 0
- try:
- async with self.get_session() as session:
- stmt = select(Jargon).where(
- Jargon.chat_id == chat_id,
- Jargon.content.in_(list(term_frequencies.keys())),
- )
- result = await session.execute(stmt)
- records = result.scalars().all()
-
- now = int(time.time())
- for record in records:
- new_count = term_frequencies.get(record.content, 0)
- if new_count > (record.count or 0):
- record.count = new_count
- record.updated_at = now
- updated += 1
-
- if updated:
- await session.commit()
- self._logger.debug(
- f"[JargonFacade] 同步黑话出现次数: "
- f"chat={chat_id}, updated={updated}"
- )
- except Exception as e:
- self._logger.error(f"[JargonFacade] 同步黑话出现次数失败: {e}", exc_info=True)
-
- return updated
-
# 4. get_jargon_statistics
async def get_jargon_statistics(self, group_id: str = None) -> Dict[str, Any]:
"""获取黑话学习统计信息
diff --git a/services/database/sqlalchemy_database_manager.py b/services/database/sqlalchemy_database_manager.py
index e13af6bc..82fc5b4a 100644
--- a/services/database/sqlalchemy_database_manager.py
+++ b/services/database/sqlalchemy_database_manager.py
@@ -960,9 +960,6 @@ async def insert_jargon(self, jargon_data: Dict[str, Any]) -> Optional[int]:
async def update_jargon(self, jargon_data: Dict[str, Any]) -> bool:
return await self._call_jargon("update_jargon", False, jargon_data)
- async def sync_jargon_counts(self, chat_id: str, term_frequencies: Dict[str, int]) -> int:
- return await self._call_jargon("sync_jargon_counts", 0, chat_id, term_frequencies)
-
async def get_jargon_statistics(self, group_id: str = None) -> Dict[str, Any]:
return await self._call_jargon(
"get_jargon_statistics",
diff --git a/services/jargon/jargon_miner.py b/services/jargon/jargon_miner.py
index 589d59d2..b30002f1 100644
--- a/services/jargon/jargon_miner.py
+++ b/services/jargon/jargon_miner.py
@@ -521,7 +521,8 @@ async def save_or_update_jargon(
updated_at=existing_dict.get('updated_at')
)
- # 更新现有记录（count 由频率同步管理，此处不递增）
+ # 更新现有记录
+ existing.count = (existing.count or 0) + 1
# 合并 raw_content
existing_list = safe_parse_llm_json(existing.raw_content) or []
diff --git a/services/learning/message_pipeline.py b/services/learning/message_pipeline.py
index 3be8d547..6e8be3a2 100644
--- a/services/learning/message_pipeline.py
+++ b/services/learning/message_pipeline.py
@@ -256,16 +256,6 @@ async def mine_jargon(self, group_id: str) -> None:
if not statistical_candidates:
statistical_candidates = None
- # Sync actual chat occurrence counts to DB
- term_freq = self._jargon_statistical_filter._group_term_freq.get(group_id)
- if term_freq:
- try:
- await self._db_manager.sync_jargon_counts(
- group_id, dict(term_freq)
- )
- except Exception as e:
- logger.debug(f"[JargonMining] Frequency sync failed: {e}")
-
await jargon_miner.run_once(
chat_messages,
len(recent_messages),
From 28a9de16ac82ac1ba39d83a8683014400710efc5 Mon Sep 17 00:00:00 2001
From: YumemiDream <1803068130@qq.com>
Date: Thu, 4 Jun 2026 20:09:16 +0800
Subject: [PATCH 4/6] fix: remove jargon sort by occurrences (count not
reliable)
---
web_res/static/html/dashboard.html | 3 ---
1 file changed, 3 deletions(-)
diff --git a/web_res/static/html/dashboard.html b/web_res/static/html/dashboard.html
index 71368b6b..011c1c87 100644
--- a/web_res/static/html/dashboard.html
+++ b/web_res/static/html/dashboard.html
@@ -2489,7 +2489,6 @@ 黑话与批次
-
@@ -6399,8 +6398,6 @@ ${escapeHtml(item.title)}
items.reverse();
} else if (state.jargon.sort === 'name') {
items.sort((a, b) => (a.term || a.word || '').localeCompare(b.term || b.word || ''));
- } else if (state.jargon.sort === 'occurrences') {
- items.sort((a, b) => (b.occurrences || 0) - (a.occurrences || 0));
}
state.jargon.items = items;
From 6a9b68f71698b3a8aa5af4642f5bdff96f6dcbdf Mon Sep 17 00:00:00 2001
From: YumemiDream <1803068130@qq.com>
Date: Thu, 4 Jun 2026 20:26:39 +0800
Subject: [PATCH 5/6] fix: prioritize real dialogue pairs over LLM-generated
patterns
Style learning now extracts user->bot pairs from actual chat history
first (chronologically matched), falling back to LLM-generated
expression patterns only when no real pairs are found.
---
services/core_learning/progressive_learning.py | 15 ++++++++-------
1 file changed, 8 insertions(+), 7 deletions(-)
diff --git a/services/core_learning/progressive_learning.py b/services/core_learning/progressive_learning.py
index c220cb97..0087c013 100644
--- a/services/core_learning/progressive_learning.py
+++ b/services/core_learning/progressive_learning.py
@@ -1206,13 +1206,9 @@ async def _save_style_learning_record(self, group_id: str, style_analysis: Dict[
logger.debug(f"群组 {group_id} 没有风格分析结果且没有消息，跳过风格学习记录保存")
return
- # 1. 保存表达模式到 expression_patterns 表
- expression_patterns = style_analysis_dict.get('expression_patterns', [])
- expression_patterns = self._filter_expression_patterns(expression_patterns)
-
- # 在 fewshot 模式下，style_analysis 可能不包含 expression_patterns。
- # 此时从数据库获取 bot 消息与用户消息合并，提取 user->bot 对话对。
- if not expression_patterns and messages:
+ # 1. 优先从真实对话中提取 user->bot 对话对（逻辑连贯）
+ expression_patterns = []
+ if messages:
try:
merged = await self._merge_bot_messages_for_pairs(group_id, messages)
if merged:
@@ -1220,6 +1216,11 @@ async def _save_style_learning_record(self, group_id: str, style_analysis: Dict[
except Exception as pair_err:
logger.debug(f"提取 fewshot 对话对失败: {pair_err}")
+ # 真实对话对不足时，回退到 LLM 生成的表达模式
+ if not expression_patterns:
+ expression_patterns = style_analysis_dict.get('expression_patterns', [])
+ expression_patterns = self._filter_expression_patterns(expression_patterns)
+
if expression_patterns:
await self._save_expression_patterns(group_id, expression_patterns)
From adaffe25d88bb210ea9deeff9ec348d4591cdf69 Mon Sep 17 00:00:00 2001
From: YumemiDream <1803068130@qq.com>
Date: Thu, 4 Jun 2026 20:39:15 +0800
Subject: [PATCH 6/6] fix: deduplicate style learning dialogue pairs at
extraction and injection
- Extraction: _extract_fewshot_pairs_from_merged deduplicates by
(situation, expression) content within a single batch
- Extraction: _extract_style_dialog_pairs deduplicates learned_patterns
and few-shot pairs
- Injection: _build_style_begin_dialogs checks existing begin_dialogs
for matching user messages before appending, preventing duplicates
across multiple approved reviews
---
.../core_learning/progressive_learning.py | 7 +++++
webui/services/persona_review_service.py | 27 +++++++++++++++----
2 files changed, 29 insertions(+), 5 deletions(-)
diff --git a/services/core_learning/progressive_learning.py b/services/core_learning/progressive_learning.py
index 0087c013..3a551598 100644
--- a/services/core_learning/progressive_learning.py
+++ b/services/core_learning/progressive_learning.py
@@ -1373,8 +1373,10 @@ def _extract_fewshot_pairs_from_merged(
Mirrors the logic of ExpressionPatternLearner._extract_few_shot_pairs
but operates on plain dicts and returns expression pattern dicts.
+ Deduplicates by (situation, expression) content.
"""
pairs = []
+ seen = set()
current_time = time.time()
for i in range(len(merged) - 1):
@@ -1400,6 +1402,11 @@ def _extract_fewshot_pairs_from_merged(
if '@' in msg_text or '@' in nxt_text:
continue
+ key = (msg_text[:50], nxt_text[:100])
+ if key in seen:
+ continue
+ seen.add(key)
+
pairs.append({
'situation': msg_text[:50],
'expression': nxt_text[:100],
diff --git a/webui/services/persona_review_service.py b/webui/services/persona_review_service.py
index f6f06348..ce8bad5d 100644
--- a/webui/services/persona_review_service.py
+++ b/webui/services/persona_review_service.py
@@ -176,19 +176,27 @@ def _build_change_payload(
@staticmethod
def _extract_style_dialog_pairs(review: Dict[str, Any]) -> List[Tuple[str, str]]:
- """Extract style review dialog pairs from structured patterns or few-shot text."""
+ """Extract style review dialog pairs from structured patterns or few-shot text. Deduplicates by content."""
dialog_pairs = []
+ seen = set()
learned_patterns = review.get('learned_patterns', [])
for pattern in learned_patterns:
situation = pattern.get('situation', '') if isinstance(pattern, dict) else ''
expression = pattern.get('expression', '') if isinstance(pattern, dict) else ''
if situation and expression:
- dialog_pairs.append((str(situation), str(expression)))
+ key = (str(situation), str(expression))
+ if key not in seen:
+ seen.add(key)
+ dialog_pairs.append(key)
if not dialog_pairs:
- dialog_pairs = PersonaReviewService._parse_few_shots_to_pairs(
+ for user_msg, assistant_msg in PersonaReviewService._parse_few_shots_to_pairs(
review.get('few_shots_content', '') or ''
- )
+ ):
+ key = (user_msg, assistant_msg)
+ if key not in seen:
+ seen.add(key)
+ dialog_pairs.append(key)
return dialog_pairs
def _dialog_pairs_for_style_review(self, review: Dict[str, Any]) -> List[Tuple[str, str]]:
@@ -199,10 +207,19 @@ def _build_style_begin_dialogs(
current_begin_dialogs: List[str],
dialog_pairs: List[Tuple[str, str]]
) -> List[str]:
- """Append style examples and keep only latest style example pairs."""
+ """Append style examples, skipping pairs already present in begin_dialogs."""
updated_dialogs = PersonaReviewService._normalize_begin_dialogs(current_begin_dialogs)
+ # Build set of existing user messages for dedup
+ existing_user_msgs = set()
+ for d in updated_dialogs:
+ if isinstance(d, str) and d.startswith(STYLE_BEGIN_DIALOG_PREFIX):
+ existing_user_msgs.add(d[len(STYLE_BEGIN_DIALOG_PREFIX):].strip())
+
for user_msg, assistant_msg in dialog_pairs:
+ if user_msg.strip() in existing_user_msgs:
+ continue
+ existing_user_msgs.add(user_msg.strip())
updated_dialogs.append(f"{STYLE_BEGIN_DIALOG_PREFIX}{user_msg}")
updated_dialogs.append(str(assistant_msg))