增加多模态模型开放集识别。

exp_multimodal/dish_name_cleaner.py

@@ -0,0 +1,102 @@
+"""
+菜品名清洗工具
+用于将VLM输出的菜品名进行标准化处理，提升向量匹配准确率
+"""
+import re
+from typing import Dict
+
+
+# 同义词映射表（可根据实际情况扩展）
+SYNONYM_MAP: Dict[str, str] = {
+    "西红柿": "番茄",
+    "土豆": "马铃薯",
+    "洋芋": "马铃薯",
+    "青椒": "柿子椒",
+    # 可继续添加...
+}
+
+
+def clean_dish_name(name: str) -> str:
+    """
+    清洗菜品名：去除括号注释、前缀、英文等干扰信息
+    
+    示例：
+        "宫保鸡丁（川菜）" -> "宫保鸡丁"
+        "川菜-麻婆豆腐" -> "麻婆豆腐"
+        "红烧肉 Braised Pork" -> "红烧肉"
+    """
+    if not isinstance(name, str):
+        return ""
+    
+    # 1. 去除括号及内容（中英文括号）
+    name = re.sub(r"[（(].*?[)）]", "", name)
+    name = re.sub(r"\[.*?\]", "", name)
+    
+    # 2. 去除常见前缀（菜系、地域等）
+    prefixes = ["川菜", "粤菜", "鲁菜", "苏菜", "浙菜", "闽菜", "湘菜", "徽菜",
+                "东北", "西北", "西南", "华南", "华北"]
+    for prefix in prefixes:
+        if name.startswith(prefix):
+            name = name[len(prefix):]
+            break
+    
+    # 3. 去除分隔符后的前缀（如 "川菜-宫保鸡丁"）
+    name = re.sub(r"^[^-—]*[-—]", "", name)
+    
+    # 4. 去除英文部分（保留中文）
+    name = re.sub(r"[a-zA-Z\s]+", "", name)
+    
+    # 5. 去除多余空格和标点
+    name = re.sub(r"[,，、·\s]+", "", name)
+    
+    # 6. 去除可能的烹饪方式后缀（如果VLM违规输出）
+    cooking_suffixes = ["炒制", "烹饪", "料理", "做法"]
+    for suffix in cooking_suffixes:
+        if name.endswith(suffix):
+            name = name[:-len(suffix)]
+    
+    return name.strip()
+
+
+def normalize_dish_name(name: str) -> str:
+    """
+    标准化菜品名：应用同义词映射
+    
+    示例：
+        "西红柿炒鸡蛋" -> "番茄炒鸡蛋"
+    """
+    cleaned = clean_dish_name(name)
+    
+    # 应用同义词替换
+    for synonym, standard in SYNONYM_MAP.items():
+        cleaned = cleaned.replace(synonym, standard)
+    
+    return cleaned
+
+
+def extract_main_dish_name(text: str) -> str:
+    """
+    从VLM返回文本中提取主要菜品名（兼容多种输出格式）
+    
+    示例：
+        "这是宫保鸡丁" -> "宫保鸡丁"
+        "菜品：红烧肉" -> "红烧肉"
+    """
+    if not text:
+        return ""
+    
+    # 尝试匹配常见模式
+    patterns = [
+        r"菜品[:：]\s*([^，,。\n]+)",
+        r"识别为[:：]\s*([^，,。\n]+)",
+        r"这是\s*([^，,。\n]+)",
+        r"应该是\s*([^，,。\n]+)",
+    ]
+    
+    for pattern in patterns:
+        match = re.search(pattern, text)
+        if match:
+            return match.group(1).strip()
+    
+    # 如果没有匹配到，返回清洗后的整个文本
+    return clean_dish_name(text)