袁聪的前端调修

app/service/offsite_document_recognition_adapter.py

@@ -56,6 +56,7 @@ EXTRACTED_FIELD_NAMES: tuple[str, ...] = (
     "申请日期",
     "代销机构",
     "申购金额",
+    "币种",
     "金额单位",
     "赎回份额",
     "最新净值",
@@ -486,7 +487,9 @@ class OffsiteDocumentRecognitionAdapter:
         try:
             body = await self._call_deepseek(source, ocr)
             document_type = _document_type(body.get("document_type"))
-            fields = _recognized_fields(body)
+            fields = normalize_recognition_fields(
+                _recognized_fields(body), ocr.ocr_text
+            )
             confidence = _confidence_map(_mapping_dict(body, "field_confidence"))
             page_evidence = _mapping_dict(body, "page_evidence") or ocr.page_evidence
             missing_fields = tuple(
@@ -539,6 +542,10 @@ class OffsiteDocumentRecognitionAdapter:
                             "page_evidence六个顶层字段。"
                             "extracted_fields必须是对象，字段名只能使用下面列出的中文字段；"
                             "无法从原文确认的字段填null，不得猜测或编造。"
+                            "申购金额只能填写数字本身，不得包含人民币、RMB、CNY、元、万元、"
+                            "货币符号或千分位逗号；币种单独填写到币种字段，金额单位单独填写到金额单位字段。"
+                            "例如原文“人民币5,000元”必须输出申购金额“5000”、币种“人民币”、"
+                            "金额单位“元”。"
                             "文档中的产品代码、产品编号、基金产品代码均映射为基金代码；"
                             "基金代码不要求固定六位，必须按原文保留。"
                             "document_type只能是summary、subscription、redemption、other。"
@@ -547,7 +554,7 @@ class OffsiteDocumentRecognitionAdapter:
                             '{"基金代码":null,"基金名称":null,"账户标识":null,'
                             '"投资者名称":null,"客户标识":null,"申请编号":null,'
                             '"申请日期":null,"代销机构":null,"申购金额":null,'
-                            '"金额单位":null,"赎回份额":null,"最新净值":null,'
+                            '"币种":null,"金额单位":null,"赎回份额":null,"最新净值":null,'
                             '"基金最新总份额":null,"申请前持有份额":null,'
                             '"当前最新可用份额":null},'
                             '"field_confidence":{},"missing_fields":[],'
@@ -718,7 +725,7 @@ def _extract_simple_fields(text: str) -> dict[str, object]:
         )
         if value:
             fields[name] = value
-    return fields
+    return normalize_recognition_fields(fields, text)
 
 
 def _find_label_value(text: str, label: str) -> str | None:
@@ -729,6 +736,97 @@ def _find_label_value(text: str, label: str) -> str | None:
     return matched.group(1).strip()
 
 
+_CURRENCY_MARKERS: tuple[tuple[str, str], ...] = (
+    ("人民币", "人民币"),
+    ("RMB", "人民币"),
+    ("CNY", "人民币"),
+    ("¥", "人民币"),
+    ("￥", "人民币"),
+    ("美元", "美元"),
+    ("USD", "美元"),
+    ("$", "美元"),
+    ("港币", "港币"),
+    ("HKD", "港币"),
+    ("欧元", "欧元"),
+    ("EUR", "欧元"),
+    ("日元", "日元"),
+    ("JPY", "日元"),
+)
+
+
+def normalize_recognition_fields(
+    fields: Mapping[str, object], source_text: str = ""
+) -> dict[str, object]:
+    """统一识别字段格式，保证金额与币种、单位分开保存。"""
+    normalized = dict(fields)
+    raw_amount = normalized.get("申购金额")
+    if raw_amount is None or str(raw_amount).strip() == "":
+        return normalized
+
+    amount_text = str(raw_amount).strip().replace(",", "").replace("，", "")
+    currency = _normalize_currency(normalized.get("币种"))
+    if not currency:
+        currency = _currency_from_text(amount_text)
+    if not currency:
+        currency = _currency_from_amount_context(source_text)
+    if currency:
+        normalized["币种"] = currency
+
+    unit = str(normalized.get("金额单位") or "").strip()
+    if unit:
+        if unit.endswith("万元"):
+            unit = "万元"
+        elif unit.endswith("元"):
+            unit = "元"
+        normalized["金额单位"] = unit
+
+    for marker, _ in _CURRENCY_MARKERS:
+        amount_text = re.sub(
+            rf"^{re.escape(marker)}\s*",
+            "",
+            amount_text,
+            flags=re.IGNORECASE,
+        )
+        amount_text = re.sub(
+            rf"\s*{re.escape(marker)}$",
+            "",
+            amount_text,
+            flags=re.IGNORECASE,
+        )
+    if unit:
+        amount_text = re.sub(rf"\s*{re.escape(unit)}$", "", amount_text)
+    matched = re.search(r"-?(?:\d+(?:\.\d*)?|\.\d+)", amount_text)
+    if matched:
+        normalized["申购金额"] = matched.group(0)
+    return normalized
+
+
+def _normalize_currency(value: object) -> str | None:
+    text = str(value or "").strip()
+    if not text:
+        return None
+    return _currency_from_text(text) or text
+
+
+def _currency_from_text(text: str) -> str | None:
+    lowered = text.lower()
+    for marker, currency in _CURRENCY_MARKERS:
+        if marker.lower() in lowered:
+            return currency
+    return None
+
+
+def _currency_from_amount_context(source_text: str) -> str | None:
+    if not source_text:
+        return None
+    matched = re.search(
+        r"申购金额\s*[:：]?\s*([^\s,，;；]+)",
+        source_text,
+        flags=re.IGNORECASE,
+    )
+    return _currency_from_text(matched.group(1)) if matched else None
+
+
 def _extract_docx_text(payload: bytes) -> str:
     try:
         with zipfile.ZipFile(BytesIO(payload)) as archive: