YG-Soft/X-Financial
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: 完善差旅票据行程提取与费用明细回填逻辑

Browse Source 
增强文档智能识别的票据场景关键词和字段提取能力，优化
会话关联草稿报销单的解析路径，修复费用明细合并和票据
去重边界问题，前端改进报销创建和审批详情交互，补充单
元测试覆盖。
This commit is contained in:
caoxiaozhu
2026-05-21 14:24:51 +08:00
parent b183b0bd5e
commit f28d7e6d16
24 changed files with 1565 additions and 433 deletions
Download Patch File Download Diff File Expand all files Collapse all files

241
server/src/app/services/document_intelligence.py
View File

@@ -184,6 +184,7 @@ AMOUNT_PATTERNS = (
re.compile(r"([0-9]+(?:[.,][0-9]{1,2})?)\s*元"),
)
DATE_PATTERN = re.compile(r"((?:20\d{2}|19\d{2})[-/年.](?:1[0-2]|0?[1-9])[-/月.](?:3[01]|[12]\d|0?[1-9])日?)")
TIME_PATTERN = re.compile(r"(?<!\d)([01]?\d|2[0-3])[:：]([0-5]\d)(?!\d)")
INVOICE_NUMBER_PATTERN = re.compile(r"(?:发票号码|票号|单号|订单号)[：:\s]*([A-Za-z0-9-]{6,24})")
INVOICE_CODE_PATTERN = re.compile(r"(?:发票代码)[：:\s]*([A-Za-z0-9-]{6,24})")
TRIP_NO_PATTERN = re.compile(r"(?:车次|航班(?:号)?)[：:\s]*([A-Za-z0-9]{2,12})")
@@ -192,6 +193,58 @@ MERCHANT_PATTERNS = (
re.compile(r"(?:销售方(?:名称)?|商户(?:名称)?|开票方(?:名称)?|收款方(?:名称)?)[：:\s]*([A-Za-z0-9\u4e00-\u9fa5（）()·&\\-]{2,40})"),
re.compile(r"([A-Za-z0-9\u4e00-\u9fa5（）()·&\\-]{2,40}(?:酒店|宾馆|饭店|酒楼|餐厅|航空|铁路|滴滴出行|停车场|服务区))"),
)
DATE_FIELD_KEYS = {
"date",
"time",
"issued_at",
"invoice_date",
"issue_date",
"travel_date",
"trip_date",
"journey_date",
"departure_date",
"departure_time",
"depart_date",
"depart_time",
"boarding_date",
"boarding_time",
"train_date",
"train_time",
"train_departure_time",
"scheduled_departure_time",
"flight_date",
"flight_time",
"ride_date",
"ride_time",
"pickup_time",
"start_time",
}
TRIP_DATE_LABEL_BY_DOCUMENT_TYPE = {
"train_ticket": "列车出发时间",
"flight_itinerary": "起飞日期",
"taxi_receipt": "乘车时间",
"transport_receipt": "乘车时间",
"parking_toll_receipt": "通行日期",
}
TRIP_DATE_FIELD_LABEL_TOKENS = (
"日期",
"时间",
"开票日期",
"发生时间",
"行程日期",
"出发日期",
"出发时间",
"列车出发时间",
"发车日期",
"发车时间",
"开车时间",
"乘车日期",
"乘车时间",
"起飞日期",
"航班日期",
"上车时间",
"用车时间",
)
class DocumentIntelligenceService:
@@ -212,7 +265,10 @@ class DocumentIntelligenceService:
compact = re.sub(r"\s+", "", raw_text).lower()
rule_match = _match_document_rule(compact)
base_rule = rule_match.rule or DEFAULT_RULE
fields = tuple(_extract_document_fields(raw_text))
fields = _apply_document_type_field_labels(
tuple(_extract_document_fields(raw_text, base_rule.document_type)),
base_rule.document_type,
)
rule_insight = DocumentInsight(
document_type=base_rule.document_type,
document_type_label=base_rule.document_type_label,
@@ -275,7 +331,10 @@ class DocumentIntelligenceService:
for item in parsed.evidence
if str(item or "").strip()
][:4]
normalized_fields = _normalize_llm_document_fields(parsed.fields)
normalized_fields = _apply_document_type_field_labels(
tuple(_normalize_llm_document_fields(parsed.fields)),
normalized_type,
)
return LlmDocumentClassification(
document_type=normalized_type,
@@ -312,7 +371,10 @@ class DocumentIntelligenceService:
scene_code=rule_insight.scene_code,
scene_label=rule_insight.scene_label,
expense_type=rule_insight.expense_type,
fields=merged_fields,
fields=_apply_document_type_field_labels(
merged_fields,
rule_insight.document_type,
),
classification_source=rule_insight.classification_source,
classification_confidence=rule_insight.classification_confidence,
evidence=rule_insight.evidence,
@@ -337,7 +399,10 @@ class DocumentIntelligenceService:
scene_code=rule_insight.scene_code,
scene_label=rule_insight.scene_label,
expense_type=rule_insight.expense_type,
fields=merged_fields,
fields=_apply_document_type_field_labels(
merged_fields,
rule_insight.document_type,
),
classification_source=rule_insight.classification_source,
classification_confidence=rule_insight.classification_confidence,
evidence=rule_insight.evidence,
@@ -354,7 +419,7 @@ class DocumentIntelligenceService:
scene_code=rule.scene_code if parsed.scene_code == "other" else parsed.scene_code,
scene_label=rule.scene_label if parsed.scene_label == "其他票据" else parsed.scene_label,
expense_type=rule.expense_type if parsed.expense_type == "other" else parsed.expense_type,
fields=merged_fields,
fields=_apply_document_type_field_labels(merged_fields, rule.document_type),
classification_source=source,
classification_confidence=max(parsed.confidence, rule_insight.classification_confidence),
evidence=tuple(parsed.evidence or rule_insight.evidence),
@@ -464,8 +529,49 @@ def _normalize_llm_document_field_key(key: str, label: str) -> str:
token in compact_label for token in ("金额", "价税合计", "合计", "总额", "总计", "票价", "支付金额", "实付金额", "实收金额")
):
return "amount"
if compact_key in {"date", "time", "issued_at", "invoice_date"} or any(
token in compact_label for token in ("日期", "时间", "开票日期", "发生时间")
if compact_key in {
"travel_date",
"trip_date",
"journey_date",
"departure_date",
"departure_time",
"depart_date",
"depart_time",
"boarding_date",
"boarding_time",
"train_date",
"train_time",
"train_departure_time",
"scheduled_departure_time",
"flight_date",
"flight_time",
"ride_date",
"ride_time",
"pickup_time",
"start_time",
} or any(
token in compact_label
for token in (
"行程日期",
"出发日期",
"出发时间",
"列车出发时间",
"发车日期",
"发车时间",
"开车时间",
"乘车日期",
"乘车时间",
"起飞日期",
"航班日期",
"上车时间",
"用车时间",
)
):
return "trip_date"
if compact_key in {"issued_at", "issue_date", "invoice_date"} or "开票日期" in compact_label:
return "invoice_date"
if compact_key in {"date", "time"} or any(
token in compact_label for token in ("日期", "时间", "发生时间")
):
return "date"
if compact_key in {"merchant_name", "merchant", "seller_name", "vendor_name"} or any(
@@ -504,7 +610,7 @@ def _normalize_llm_document_field_value(key: str, value: str) -> str:
return ""
text_value = format(candidate.quantize(Decimal("0.01")), "f").rstrip("0").rstrip(".")
return f"{text_value}元"
if key == "date":
if key in {"date", "time", "invoice_date", "trip_date"}:
return _extract_date(raw_value) or _clean_field_value(raw_value)
if key == "route":
return _extract_route(raw_value) or _clean_field_value(
@@ -517,6 +623,8 @@ def _llm_document_field_label(key: str) -> str:
return {
"amount": "金额",
"date": "日期",
"invoice_date": "开票日期",
"trip_date": "行程日期",
"merchant_name": "商户",
"invoice_number": "票据号码",
"invoice_code": "发票代码",
@@ -525,6 +633,35 @@ def _llm_document_field_label(key: str) -> str:
}.get(key, key)
def _apply_document_type_field_labels(
fields: tuple[DocumentField, ...],
document_type: str,
) -> tuple[DocumentField, ...]:
date_label = TRIP_DATE_LABEL_BY_DOCUMENT_TYPE.get(
str(document_type or "").strip().lower()
)
if not date_label:
return fields
adjusted: list[DocumentField] = []
for field in fields:
compact_key = str(field.key or "").strip().lower()
compact_label = str(field.label or "").replace(" ", "")
if compact_key in {"issued_at", "issue_date", "invoice_date"} or any(
token in compact_label for token in ("开票日期", "发票日期")
):
adjusted.append(field)
continue
is_date_field = compact_key in DATE_FIELD_KEYS or any(
token in compact_label for token in TRIP_DATE_FIELD_LABEL_TOKENS
)
if is_date_field:
adjusted.append(DocumentField(key=field.key, label=date_label, value=field.value))
continue
adjusted.append(field)
return tuple(adjusted)
def _merge_document_fields(
base_fields: tuple[DocumentField, ...],
override_fields: tuple[DocumentField, ...],
@@ -540,13 +677,13 @@ def _merge_document_fields(
return tuple(merged[key] for key in order if key in merged)
def _extract_document_fields(text: str) -> list[DocumentField]:
def _extract_document_fields(text: str, document_type: str = "") -> list[DocumentField]:
fields: list[DocumentField] = []
amount = _extract_amount(text)
if amount:
fields.append(DocumentField(key="amount", label="金额", value=amount))
date_value = _extract_date(text)
date_value = _extract_date(text, document_type=document_type)
if date_value:
fields.append(DocumentField(key="date", label="日期", value=date_value))
@@ -594,10 +731,33 @@ def _extract_amount(text: str) -> str:
return f"{text_value}元"
def _extract_date(text: str) -> str:
match = DATE_PATTERN.search(text)
if not match:
def _extract_date(text: str, *, document_type: str = "") -> str:
matches = list(DATE_PATTERN.finditer(text))
if not matches:
return ""
normalized_type = str(document_type or "").strip().lower()
if normalized_type in TRIP_DATE_LABEL_BY_DOCUMENT_TYPE:
candidates: list[tuple[int, int, bool, str]] = []
for index, match in enumerate(matches):
value = _format_date_match_with_time(text, match)
if not value:
continue
invoice_context = _is_invoice_date_context(text, match)
score = _score_trip_date_context(text, match, value, invoice_context)
candidates.append((score, index, invoice_context, value))
non_invoice_candidates = [candidate for candidate in candidates if not candidate[2]]
if non_invoice_candidates:
return max(non_invoice_candidates, key=lambda candidate: (candidate[0], -candidate[1]))[3]
if candidates:
return ""
return ""
return _format_date_match_with_time(text, matches[0])
def _format_date_match_with_time(text: str, match: re.Match[str]) -> str:
raw_value = str(match.group(1) or "").strip()
normalized = raw_value.replace("年", "-").replace("月", "-").replace("日", "")
normalized = normalized.replace("/", "-").replace(".", "-")
@@ -605,7 +765,60 @@ def _extract_date(text: str) -> str:
if len(parts) != 3:
return raw_value
year, month, day = parts
return f"{year.zfill(4)}-{month.zfill(2)}-{day.zfill(2)}"
date_value = f"{year.zfill(4)}-{month.zfill(2)}-{day.zfill(2)}"
surrounding = str(text or "")[max(0, match.start() - 18): match.end() + 24]
time_match = TIME_PATTERN.search(surrounding)
if time_match:
hour = str(time_match.group(1) or "").zfill(2)
minute = str(time_match.group(2) or "").zfill(2)
return f"{date_value} {hour}:{minute}"
return date_value
def _is_invoice_date_context(text: str, match: re.Match[str]) -> bool:
window = str(text or "")[max(0, match.start() - 12): match.end() + 8]
compact = window.replace(" ", "")
return any(token in compact for token in ("开票日期", "发票日期", "开票时间", "开票"))
def _score_trip_date_context(
text: str,
match: re.Match[str],
value: str,
invoice_context: bool,
) -> int:
window = str(text or "")[max(0, match.start() - 32): match.end() + 32]
compact = window.replace(" ", "")
score = -20 if invoice_context else 0
if ":" in value or "：" in value:
score += 8
if any(
token in compact
for token in (
"行程日期",
"出发日期",
"出发时间",
"列车出发时间",
"发车日期",
"发车时间",
"开车时间",
"乘车日期",
"乘车时间",
"起飞日期",
"起飞时间",
"航班日期",
"上车时间",
"用车时间",
)
):
score += 6
if any(token in compact for token in ("车次", "检票", "二等座", "一等座", "商务座", "软卧", "硬卧")):
score += 3
if re.search(r"[A-Z]\d{1,4}", compact):
score += 2
if re.search(r"[\u4e00-\u9fa5A-Za-z0-9（）()·]{2,20}(?:至|到|→|->|—|–|-)[\u4e00-\u9fa5A-Za-z0-9（）()·]{2,20}", compact):
score += 2
return score
def _extract_merchant(text: str) -> str: