Update app.py

app.py

@@ -486,11 +486,10 @@ class TextToSQLSystem:
 
     def _validate_and_fix_sql(self, question: str, raw_response: str) -> Tuple[Optional[str], str]:
         """
-        (V22 / ID 識別版)
+        (V23 / 统一实体识别版)
         一個全面、多層次的 SQL 驗證與生成引擎。
-        極大地增強了實體識別能力。新增了一個獨立的 ID 識別模塊，
-        能夠主動從問題中捕捉並分類各種格式化的 ID (如 'C0761N', 'M1044N')，
-        並將其用於後續的模板生成中。
+        引入了全新的、统一的实体识别引擎，能够准确解析 "买家 Gap", "c0761n", 
+        "买家ID c0761n" 等多种复杂的实体提问模式。
         """
         q_lower = question.lower()
         
@@ -498,41 +497,49 @@ class TextToSQLSystem:
         #  第一層：高價值意圖識別與模板覆寫 (Intent Recognition & Templating)
         # ==============================================================================
         
-        # --- 預先檢測所有可能的意圖和實體 ---
-        job_no_match = re.search(r"(?:工單|jobno)\s*'\"?([A-Z]{2,3}\d+)'\"?", question, re.IGNORECASE)
-        
-        # **新增的、更強大的實體識別模塊**
+        # --- **全新的统一实体识别引擎** ---
         entity_match_data = None
         
-        # 步驟 1: 優先識別格式化的 ID (例如 M1234N 或 C5678N)
-        id_match = re.search(r'\b([A-Z]\d{4}[A-Z])\b', question, re.IGNORECASE)
-        if id_match:
-            entity_id = id_match.group(1).upper()
-            column_name = 'sd.ApplicantID' # 默認值
-            entity_type_log = 'ID'
-            # 根據用戶問題中的上下文關鍵詞來判斷 ID 類型
-            if 'buyer' in q_lower or '買家' in q_lower:
-                column_name, entity_type_log = 'sd.BuyerID', '買家ID'
-            elif 'applicant' in q_lower or '申請' in q_lower:
-                column_name, entity_type_log = 'sd.ApplicantID', '申請方ID'
-            elif 'invoice' in q_lower or '付款' in q_lower:
-                column_name, entity_type_log = 'sd.InvoiceToID', '付款方ID'
-            elif 'agent' in q_lower or '代理' in q_lower:
-                column_name, entity_type_log = 'sd.AgentID', '代理商ID'
+        # 定义多种识别模式，【优先级从高到低】
+        entity_patterns = [
+            # 模式1: 匹配 "类型 + ID" (e.g., "买家ID C0761N") - 最高优先级
+            {'pattern': r"(买家|buyer)\s*(?:id|代號|代码)\s*'\"?\b([A-Z]\d{4}[A-Z])\b'\"?", 'column': 'sd.BuyerID', 'type': '買家ID'},
+            {'pattern': r"(申请方|申请厂商|applicant)\s*(?:id|代號|代码)\s*'\"?\b([A-Z]\d{4}[A-Z])\b'\"?", 'column': 'sd.ApplicantID', 'type': '申請方ID'},
+            {'pattern': r"(付款方|付款厂商|invoiceto)\s*(?:id|代號|代码)\s*'\"?\b([A-Z]\d{4}[A-Z])\b'\"?", 'column': 'sd.InvoiceToID', 'type': '付款方ID'},
+            {'pattern': r"(代理商|agent)\s*(?:id|代號|代码)\s*'\"?\b([A-Z]\d{4}[A-Z])\b'\"?", 'column': 'sd.AgentID', 'type': '代理商ID'},
             
-            entity_match_data = {"type": entity_type_log, "name": entity_id, "column": column_name}
-        
-        # 步驟 2: 如果沒有找到 ID，再識別文本名稱 (舊邏輯)
-        if not entity_match_data:
-            ENTITY_TO_COLUMN_MAP = {'申請廠商':'sd.ApplicantName','申請方':'sd.ApplicantName','applicant':'sd.ApplicantName','付款廠商':'sd.InvoiceToName','付款方':'sd.InvoiceToName','invoiceto':'sd.InvoiceToName','代理商':'sd.AgentName','agent':'sd.AgentName','買家':'sd.BuyerName','buyer':'sd.BuyerName','客戶':'sd.BuyerName','品牌':'tsr.BuyerName'}
-            for keyword, column in ENTITY_TO_COLUMN_MAP.items():
-                if keyword in q_lower:
-                    match = re.search(fr"{re.escape(keyword)}[\s:;\'\"-]*([a-zA-Z0-9&.\s-]+?)(?:\s*的|\s+|為|$)", question, re.IGNORECASE)
-                    if match: entity_match_data = {"type": keyword, "name": match.group(1).strip(), "column": column}; break
+            # 模式2: 匹配 "类型 + 名称" (e.g., "买家 Gap")
+            {'pattern': r"(买家|buyer|客戶)\s*'\"?([a-zA-Z0-9&.\s-]+?)(?:\s*的|\s+|$|有)", 'column': 'sd.BuyerName', 'type': '買家'},
+            {'pattern': r"(申请方|申请厂商|applicant)\s*'\"?([a-zA-Z0-9&.\s-]+?)(?:\s*的|\s+|$|有)", 'column': 'sd.ApplicantName', 'type': '申請方'},
+            {'pattern': r"(付款方|付款厂商|invoiceto)\s*'\"?([a-zA-Z0-9&.\s-]+?)(?:\s*的|\s+|$|有)", 'column': 'sd.InvoiceToName', 'type': '付款方'},
+            {'pattern': r"(代理商|agent)\s*'\"?([a-zA-Z0-9&.\s-]+?)(?:\s*的|\s+|$|有)", 'column': 'sd.AgentName', 'type': '代理商'},
+            
+            # 模式3: 单独匹配一个 ID (e.g., "c0761n") - 较低优先级
+            {'pattern': r"\b([A-Z]\d{4}[A-Z])\b", 'column': 'sd.ApplicantID', 'type': 'ID'} # 默认为 ApplicantID，可以根据业务调整
+        ]
         
-        # --- 判斷邏輯: 依優先級進入對應的模板 ---
+        for p in entity_patterns:
+            match = re.search(p['pattern'], question, re.IGNORECASE)
+            if match:
+                entity_value = match.group(2) if len(match.groups()) > 1 else match.group(1)
+                entity_match_data = {
+                    "type": p['type'], 
+                    "name": entity_value.strip().upper(),
+                    "column": p['column']
+                }
+                break
+
+        # --- 预先检测其他意图 ---
+        job_no_match = re.search(r"(?:工單|jobno)\s*'\"?([A-Z]{2,3}\d+)'\"?", question, re.IGNORECASE)
+        lab_group_match_data = None
+        LAB_GROUP_MAP = {'A':'TA','B':'TB','C':'TC','D':'TD','E':'TE','Y':'TY','TA':'TA','TB':'TB','TC':'TC','TD':'TD','TE':'TE','TY':'TY','WC':'WC','EO':'EO','GCI':'GCI','GCO':'GCO','MI':'MI'}
+        lab_group_match = re.findall(r"([A-Z]+)\s*組", question, re.IGNORECASE)
+        if lab_group_match:
+            codes = [LAB_GROUP_MAP.get(g.upper()) for g in lab_group_match if LAB_GROUP_MAP.get(g.upper())]
+            if codes: lab_group_match_data = {"codes": codes, "identifiers": lab_group_match}
+
+        # --- 判断逻辑: 依优先级进入对应的模板 ---
 
-        # 意图 1: 报告列表查询 (高优先级)
         if any(kw in q_lower for kw in ['報告號碼', '報告清單', '列出報告', 'report number', 'list of reports']):
             year_match = re.search(r'(\d{4})\s*年?', question)
             month_match = re.search(r'(\d{1,2})\s*月', question)
@@ -554,8 +561,7 @@ class TextToSQLSystem:
             if entity_match_data:
                 entity_name, column_name = entity_match_data["name"], entity_match_data["column"]
                 if "JOIN TSR53SampleDescription" not in from_clause: from_clause = "FROM JobTimeline AS jt JOIN TSR53SampleDescription AS sd ON jt.JobNo = sd.JobNo"
-                # ID 查詢使用精確匹配，名稱查詢使用模糊匹配
-                match_operator = "=" if entity_match_data["type"].endswith("ID") else "LIKE"
+                match_operator = "=" if column_name.endswith("ID") else "LIKE"
                 entity_value = f"'{entity_name}'" if match_operator == "=" else f"'%{entity_name}%'"
                 where_conditions.append(f"{column_name} {match_operator} {entity_value}")
                 log_parts.append(entity_name)
@@ -584,7 +590,18 @@ class TextToSQLSystem:
         fixed_sql = " " + parsed_sql.strip() + " "
         fixes_applied_fallback = []
         
-        # ... (后备修正字典和循环)
+        dialect_corrections = {r'YEAR\s*\(([^)]+)\)': r"strftime('%Y', \1)"}
+        for pattern, replacement in dialect_corrections.items():
+            if re.search(pattern, fixed_sql, re.IGNORECASE):
+                fixed_sql = re.sub(pattern, replacement, fixed_sql, flags=re.IGNORECASE)
+                fixes_applied_fallback.append(f"修正方言: {pattern}")
+        
+        schema_corrections = {'TSR53Report':'TSR53SampleDescription', 'TSR53InvoiceReportNo':'JobNo', 'TSR53ReportNo':'JobNo', 'TSR53InvoiceNo':'JobNo', 'TSR53InvoiceCreditNoteNo':'InvoiceCreditNoteNo', 'TSR53InvoiceLocalAmount':'LocalAmount', 'Status':'OverallRating', 'ReportStatus':'OverallRating'}
+        for wrong, correct in schema_corrections.items():
+            pattern = r'\b' + re.escape(wrong) + r'\b'
+            if re.search(pattern, fixed_sql, re.IGNORECASE):
+                fixed_sql = re.sub(pattern, correct, fixed_sql, flags=re.IGNORECASE)
+                fixes_applied_fallback.append(f"映射 Schema: '{wrong}' -> '{correct}'")
         
         log_msg = "AI 生成並成功修正" if fixes_applied_fallback else "AI 生成且無需修正"
         return self._finalize_sql(fixed_sql, log_msg)