成语接龙接入 LLM 和 MarkDown、LaTeX 接入

konabot/plugins/idiomgame/__init__.py

@@ -21,10 +21,13 @@ from nonebot_plugin_alconna import (
 from konabot.common.longtask import DepLongTaskTarget
 from konabot.common.path import ASSETS_PATH
 
-DATA_FILE_PATH = (
-    Path(__file__).parent.parent.parent.parent / "data" / "idiom_banned.json"
-)
+from konabot.common.llm import get_llm
 
+DATA_DIR = Path(__file__).parent.parent.parent.parent / "data"
+
+DATA_FILE_PATH = (
+    DATA_DIR / "idiom_banned.json"
+)
 
 def load_banned_ids() -> list[str]:
     if not DATA_FILE_PATH.exists():
@@ -75,6 +78,27 @@ class TryVerifyState(Enum):
     BUT_NO_NEXT = 5
     GAME_END = 6
 
+class IdiomGameLLM:
+    @classmethod
+    async def verify_idiom_with_llm(cls, idiom: str) -> bool:
+        if len(idiom) != 4:
+            return False
+        llm = get_llm()
+        system_prompt = "请判断用户的输入是否为一个合理的成语，或者这四个字在中文环境下是否说得通。如果是请回答「T」，否则回答「F」。请注意，即使这个词不是成语，如果说得通（也就是能念起来很通顺），你也该输出「T」。请不要包含任何解释，也不要包含任何标点符号。"
+        message = await llm.chat([{"role": "system", "content": system_prompt}, {"role": "user", "content": idiom}])
+        answer = message.content
+        logger.info(f"LLM 对成语 {idiom} 的判断结果是 {answer}")
+        if answer == "T":
+            await cls.storage_idiom(idiom)
+        return answer == "T"
+    
+    @classmethod
+    async def storage_idiom(cls, idiom: str):
+        # 将 idiom 存入本地文件以备后续分析
+        with open(DATA_DIR / "idiom_llm_storage.txt", "a", encoding="utf-8") as f:
+            f.write(idiom + "\n")
+        IdiomGame.append_into_word_list(idiom)
+        
 
 class IdiomGame:
     ALL_WORDS = []  # 所有四字词语
@@ -101,6 +125,17 @@ class IdiomGame:
         self.idiom_history: list[list[str]] = [] # 成语使用历史记录，多个数组以存储不同成语链
         IdiomGame.INSTANCE_LIST[group_id] = self
 
+    @classmethod
+    def append_into_word_list(cls, word: str):
+        '''
+        将一个新词加入到词语列表中
+        '''
+        if word not in cls.ALL_WORDS:
+            cls.ALL_WORDS.append(word)
+            if word[0] not in cls.IDIOM_FIRST_CHAR:
+                cls.IDIOM_FIRST_CHAR[word[0]] = []
+            cls.IDIOM_FIRST_CHAR[word[0]].append(word)
+
     def be_able_to_play(self) -> bool:
         if self.last_play_date != datetime.date.today():
             self.last_play_date = datetime.date.today()
@@ -186,7 +221,7 @@ class IdiomGame:
         用户发送成语
         """
         async with self.lock:
-            state = self._verify_idiom(idiom, user_id)
+            state = await self._verify_idiom(idiom, user_id)
         return state
 
     def is_nextable(self, last_char: str) -> bool:
@@ -218,16 +253,24 @@ class IdiomGame:
             result.append(" -> ".join(chain))
         return result
 
-    def _verify_idiom(self, idiom: str, user_id: str) -> list[TryVerifyState]:
+    async def _verify_idiom(self, idiom: str, user_id: str) -> list[TryVerifyState]:
         state = []
         # 新成语的首字应与上一条成语的尾字相同
         if idiom[0] != self.last_char:
             state.append(TryVerifyState.WRONG_FIRST_CHAR)
             return state
         if idiom not in IdiomGame.ALL_IDIOMS and idiom not in IdiomGame.ALL_WORDS:
-            self.add_score(user_id, -0.1)
-            state.append(TryVerifyState.NOT_IDIOM)
-            return state
+            logger.info(f"用户 {user_id} 发送了未知词语 {idiom}，正在使用 LLM 进行验证")
+            try:
+                if not await IdiomGameLLM.verify_idiom_with_llm(idiom):
+                    self.add_score(user_id, -0.1)
+                    state.append(TryVerifyState.NOT_IDIOM)
+                    return state
+            except Exception as e:
+                logger.error(f"LLM 验证成语 {idiom} 时出现错误：{e}")
+                self.add_score(user_id, -0.1)
+                state.append(TryVerifyState.NOT_IDIOM)
+                return state
         # 成语合法，更新状态
         self.add_history_idiom(idiom)
         score_k = 0.5 ** self.get_already_used_num(idiom) # 每被使用过一次，得分减半
@@ -335,6 +378,16 @@ class IdiomGame:
         logger.debug(f"Loaded {len(THUOCL_WORDS)} words from THUOCL txt files")
         logger.debug(f"Sample words: {THUOCL_WORDS[:5]}")
 
+        # 读取本地的 idiom_llm_storage.txt 文件，补充词语表
+        LOCAL_LLM_WORDS = []
+        if (DATA_DIR / "idiom_llm_storage.txt").exists():
+            with open(DATA_DIR / "idiom_llm_storage.txt", "r", encoding="utf-8") as f:
+                for line in f:
+                    word = line.strip()
+                    if len(word) == 4:
+                        LOCAL_LLM_WORDS.append(word)
+        logger.debug(f"Loaded additional {len(LOCAL_LLM_WORDS)} words from idiom_llm_storage.txt")
+
         # 只有成语的大表
         cls.ALL_IDIOMS = [idiom["word"] for idiom in ALL_IDIOMS_INFOS] + THUOCL_IDIOMS
         cls.ALL_IDIOMS = list(set(cls.ALL_IDIOMS))  # 去重
@@ -344,6 +397,7 @@ class IdiomGame:
             [word for word in cls.ALL_WORDS if len(word) == 4]
             + THUOCL_WORDS
             + COMMON_WORDS
+            + LOCAL_LLM_WORDS
         )
         cls.ALL_WORDS = list(set(cls.ALL_WORDS))  # 去重