"""纯正则 PDF 数据提取模块 — 不调用 LLM，零延迟。 从保险计划书 PDF 的原始文本中，用正则 + 启发式规则提取结构化 JSON。 提取率目标 > 80%（按保险演示表行数计算）。 用法： from insurance.ppt.regex_extractor import extract_insurance_regex data = extract_insurance_regex(pdf_text) """ import re import logging logger = logging.getLogger(__name__) # ─── 金额解析 ────────────────────────────────────────────── def _parse_money(s: str) -> float | None: """解析各种格式的金额字符串为 float。 支持: "1,234,567.89" "1234567" "1,234,567" "US$1,234" "HK$1234" """ if not s: return None # 去掉货币符号和空白 s = re.sub(r'[Uu][Ss]\$|[Hh][Kk]\$|[Cc][Nn][Yy]?¥|¥|\$|\s', '', s.strip()) # 去掉逗号 s = s.replace(',', '') try: v = float(s) return v if v > 0 else None except (ValueError, TypeError): return None # ─── 产品信息提取 ────────────────────────────────────────── def _extract_product_name(text: str) -> str | None: """提取产品名称。""" patterns = [ r'(?:产品名称|产品名|产品|保单名称)\s*[：:]\s*([^\n,，]+)', r'(?:Product(?:\s+Name)?|Plan(?:\s+Name)?)\s*[：:]\s*([^\n,，]+)', r'(?:建议书|建議書|計劃書|計劃名稱|保障計劃|保障计划)\s*[：:]\s*([^\n,，]+)', r'(?:Proposal)\s*[：:]\s*([^\n,，]+)', ] for p in patterns: m = re.search(p, text, re.IGNORECASE) if m: name = m.group(1).strip() # 过滤掉太短或纯数字的结果 if len(name) >= 2 and not name.isdigit(): return name return None def _extract_currency(text: str) -> str: """提取保单货币，支持多币种（优先匹配保费行中的币种符号）。""" # 按优先级搜索 patterns = [ (r'(?:保单货币|保费货币|Currency)\s*[：:]\s*(USD|HKD|CNY|RMB|SGD|GBP|EUR)', 'direct'), (r'\b(USD)\b', 'usd'), (r'\b(HKD)\b', 'hkd'), (r'\b(CNY|RMB)\b', 'cny'), (r'\b(SGD)\b', 'sgd'), (r'U\s*S\s*\$', 'us_symbol'), (r'H\s*K\s*\$', 'hk_symbol'), ] for pattern, tag in patterns: m = re.search(pattern, text, re.IGNORECASE) if m: if tag == 'us_symbol': return 'USD' if tag == 'hk_symbol': return 'HKD' return m.group(1).upper() return 'USD' # 默认 def _extract_annual_premium(text: str) -> float | None: """提取年缴保费。""" patterns = [ r'(?:每年(?:缴付)?保费|年缴保费|Annual\s+Premium|年保费)\s*[：:]\s*[\$UuSsHhKk]*\s*([\d,]+(?:\.\d+)?)', r'(?:每年(?:缴付)?保费|年缴保费|Annual\s+Premium)\s*[:：]?\s*(?:[A-Z]{2,3}\$?\s*)?([\d,]+(?:\.\d+)?)', ] for p in patterns: m = re.search(p, text, re.IGNORECASE) if m: val = _parse_money(m.group(1)) if val and val >= 100: # 排除太小的数字（可能是年期等） return val return None def _extract_insured_info(text: str) -> dict: """提取被保人信息（年龄、性别）。""" info = {'age': None, 'gender': None} # 年龄 age_patterns = [ r'(?:受保人|被保人|投保时)?年龄\s*[：:]\s*(\d{1,3})\s*岁?', r'(?:Age)\s*[：:]\s*(\d{1,3})', r'(?:受保人|被保人)\s*[：:]?\s*(?:[^\n,，]{0,10})?(\d{1,3})\s*岁', ] for p in age_patterns: m = re.search(p, text, re.IGNORECASE) if m: age = int(m.group(1)) if 0 <= age <= 120: info['age'] = age break # 性别 gender_patterns = [ r'(?:受保人|被保人|投保人)?性[别別]?\s*[：:]\s*(男|女|Male|Female|M|F)', r'(?:Gender|Sex)\s*[：:]\s*(Male|Female|M|F)', r'(男|女)\s*(?:性|士|仕)', ] for p in gender_patterns: m = re.search(p, text, re.IGNORECASE) if m: raw = m.group(1).strip().upper() if raw in ('男', 'M', 'MALE'): info['gender'] = '男' elif raw in ('女', 'F', 'FEMALE'): info['gender'] = '女' break return info def _extract_premium_payment_period(text: str) -> str | None: """提取保费缴付年期。""" patterns = [ r'(?:保费缴付年期|缴费年期|缴付年期|Premium\s+Payment\s+(?:Period|Term|Years?))\s*[：:]\s*(\d+)\s*(?:年|years?|yrs?)?', r'(?:整付|趸缴|Single\s+Premium)', r'(?:缴费|缴付)\s*(?:期限|年期)\s*[：:]\s*(\d+)\s*年', ] for p in patterns: m = re.search(p, text, re.IGNORECASE) if m: if '整付' in m.group(0) or '趸缴' in m.group(0) or 'Single' in m.group(0).title(): return '整付' year_str = m.group(1) try: years = int(year_str) if 1 <= years <= 100: return f'{years}年' except ValueError: pass return None def _extract_coverage_period(text: str) -> str | None: """提取保障年期。""" patterns = [ r'(?:保障年期|保障期|Coverage\s+(?:Period|Term))\s*[：:]\s*([^\n,，]+)', r'(?:终身保障|保障至终身|终身|Whole\s+Life)', r'(?:保障至|保至)\s*(\d+)\s*岁', ] for p in patterns: m = re.search(p, text, re.IGNORECASE) if m: if '终身' in m.group(0) or 'Whole' in m.group(0).title(): return '终身' return m.group(1).strip() if m.lastindex else None return None # ─── 演示表提取 ────────────────────────────────────────── # 列名映射：各种 PDF 表头 → 标准字段名 _COLUMN_ALIASES = { # 保单年度 '保单年度': 'policy_year', '保單年度': 'policy_year', 'policy year': 'policy_year', 'policyyear': 'policy_year', 'year': 'policy_year', '年度': 'policy_year', '保单': 'policy_year', # 简称 '受保人年龄': 'policy_year', # 有些表用年龄代替年度 # 已缴总保费 '缴付保费总额': 'total_premium_paid', '已缴保费': 'total_premium_paid', '已付保费总额': 'total_premium_paid', '已缴付保费': 'total_premium_paid', 'total premium paid': 'total_premium_paid', 'premium paid': 'total_premium_paid', '累积已付保费': 'total_premium_paid', '累计已缴': 'total_premium_paid', '已缴保费总额': 'total_premium_paid', '缴付保费': 'total_premium_paid', # 保证现金价值 '保证现金价值': 'guaranteed_cash_value', '保證現金價值': 'guaranteed_cash_value', 'guaranteed cash value': 'guaranteed_cash_value', 'gcv': 'guaranteed_cash_value', '保证退保价值': 'guaranteed_cash_value', '保证价值': 'guaranteed_cash_value', '保证现金': 'guaranteed_cash_value', # 归原红利 '归原红利': 'reversionary_bonus', '歸原紅利': 'reversionary_bonus', 'reversionary bonus': 'reversionary_bonus', '累积归原红利': 'reversionary_bonus', '累计归原红利': 'reversionary_bonus', '周年红利': 'reversionary_bonus', '复归红利': 'reversionary_bonus', '歸原紅利/復歸紅利': 'reversionary_bonus', # 终期分红 '终期分红': 'terminal_dividend', '終期分紅': 'terminal_dividend', 'terminal dividend': 'terminal_dividend', '终期红利': 'terminal_dividend', '特别红利': 'terminal_dividend', # 退保发还总额 '退保发还总额': 'total_surrender_value', '退保發還總額': 'total_surrender_value', 'total surrender value': 'total_surrender_value', 'cash surrender value': 'total_surrender_value', '退保价值': 'total_surrender_value', '退保总值': 'total_surrender_value', '退保金额': 'total_surrender_value', '退保总额': 'total_surrender_value', '保证利益总额': 'total_surrender_value', # 身故赔偿 '身故赔偿': 'death_benefit', '身故賠償': 'death_benefit', 'death benefit': 'death_benefit', 'death': 'death_benefit', '身故保障': 'death_benefit', '身故保险金': 'death_benefit', # IUL 特有 '账户价值': 'account_value', '非保证账户价值': 'non_guaranteed_account_value', '保证账户价值': 'guaranteed_account_value', 'non-guaranteed account value': 'non_guaranteed_account_value', 'guaranteed account value': 'guaranteed_account_value', '保险成本': 'cost_of_insurance', 'cost of insurance': 'cost_of_insurance', 'coi': 'cost_of_insurance', # 繁体中文别名 '保單': 'policy_year', '保證現金': 'guaranteed_cash_value', '歸原紅利': 'reversionary_bonus', '復歸紅利': 'reversionary_bonus', '週年紅利': 'reversionary_bonus', '終期分紅': 'terminal_dividend', '特別紅利': 'terminal_dividend', '退保發還總額': 'total_surrender_value', '退保總值': 'total_surrender_value', '退保價值': 'total_surrender_value', '身故賠償': 'death_benefit', '已付保費總額': 'total_premium_paid', '已繳保費總額': 'total_premium_paid', '累積已付保費': 'total_premium_paid', '保證價值': 'guaranteed_cash_value', '保證現金價值': 'guaranteed_cash_value', # ─── 提领表列名 ──────────────────────────────────────── # 年度提取金额 '提取金额': 'annual_withdrawal', '提取金額': 'annual_withdrawal', '提款金额': 'annual_withdrawal', '提款金額': 'annual_withdrawal', '现金提取': 'annual_withdrawal', '現金提取': 'annual_withdrawal', '款项提取': 'annual_withdrawal', '款項提取': 'annual_withdrawal', '每年提取': 'annual_withdrawal', '每年提款': 'annual_withdrawal', 'annual withdrawal': 'annual_withdrawal', 'withdrawal': 'annual_withdrawal', 'cash withdrawal': 'annual_withdrawal', # 累计提取 '累计提取': 'total_withdrawn', '累計提取': 'total_withdrawn', '累计提款': 'total_withdrawn', '累計提款': 'total_withdrawn', '累积提取': 'total_withdrawn', '累積提取': 'total_withdrawn', 'total withdrawn': 'total_withdrawn', 'cumulative withdrawal': 'total_withdrawn', # 提取前退保价值 '提取前退保金额': 'surrender_value_before', '提取前退保價值': 'surrender_value_before', '提款前退保金额': 'surrender_value_before', '提款前退保價值': 'surrender_value_before', 'surrender value before': 'surrender_value_before', # 提取后退保价值 '提取后退保金额': 'surrender_value_after', '提取後退保金額': 'surrender_value_after', '提款后退保金额': 'surrender_value_after', '提款後退保金額': 'surrender_value_after', '退保金额（提取后）': 'surrender_value_after', 'surrender value after': 'surrender_value_after', } def _normalize_column_name(name: str) -> str | None: """将 PDF 表头规范化为标准字段名。""" normalized = name.strip().lower() # 去掉括号内容和多余空格 normalized = re.sub(r'\(.*?\)|（.*?）', '', normalized).strip() normalized = re.sub(r'\s+', ' ', normalized) if normalized in _COLUMN_ALIASES: return _COLUMN_ALIASES[normalized] # 模糊匹配：检查是否包含关键词 for alias, field in _COLUMN_ALIASES.items(): if alias in normalized or normalized in alias: return field return None def _detect_columns(header_line: str) -> list[tuple[int, str]]: """检测表头行的列位置和字段名。返回 [(start_pos, field_name), ...]。""" columns = [] # 尝试按多个空格或制表符分割 parts = re.split(r'\s{2,}|\t', header_line) pos = 0 for part in parts: field = _normalize_column_name(part) if field: columns.append((pos, field)) pos += len(part) + 2 # 近似位置 # 如果没有检测到列，尝试逐个关键词搜索 if len(columns) < 2: for alias, field in _COLUMN_ALIASES.items(): idx = header_line.lower().find(alias) if idx >= 0: # 避免重复 if not any(f == field for _, f in columns): columns.append((idx, field)) columns.sort(key=lambda x: x[0]) return columns def _extract_benefit_rows(text: str) -> list[dict]: """从 PDF 文本中提取利益演示表数据行。 策略： 1. 找到包含"保单年度"等关键词的表头行 2. 解析表头列位置 3. 提取后续数字行 """ rows = [] lines = text.split('\n') in_table = False columns: list[tuple[int, str]] = [] consecutive_non_data = 0 MAX_NON_DATA_LINES = 5 # 允许的最大连续非数据行数 for line in lines: stripped = line.strip() if not stripped: continue # 检测表头行 if not in_table: header_fields = ['保单年度', '保單年度', 'policy year', '年度'] has_header = any(f in stripped.lower() for f in header_fields) # 需要至少有两个数值相关列名 value_cols = ['价值', '保费', '红利', '分红', '赔偿', 'surrender', 'premium', 'cash', 'bonus', 'benefit', 'value', '退保', '保证'] has_value = sum(1 for v in value_cols if v in stripped.lower()) >= 1 if has_header and has_value: columns = _detect_columns(stripped) if len(columns) >= 2: in_table = True consecutive_non_data = 0 logger.debug(f"检测到演示表头: {[(c[1], c[0]) for c in columns]}") continue # 跳过子表头行（如"保证 非保证 总额"等） subheader_patterns = ['保证', '非保证', '总额', 'guaranteed', 'non-guaranteed', 'total'] if sum(1 for p in subheader_patterns if p in stripped.lower()) >= 2: continue # 尝试解析数据行 numbers = re.findall(r'[\d,]+(?:\.\d+)?', stripped) if not numbers: consecutive_non_data += 1 if consecutive_non_data >= MAX_NON_DATA_LINES: break # 表格结束 continue consecutive_non_data = 0 row = _parse_data_row(stripped, columns, numbers) if row and row.get('policy_year') is not None: rows.append(row) return rows def _parse_data_row( line: str, columns: list[tuple[int, str]], numbers: list[str] ) -> dict | None: """解析单行数据。""" row = {} # 提取第一个数字作为保单年度（通常是行首的小数字） year_match = re.match(r'^\s*(\d{1,3})\b', line) if year_match: year_val = int(year_match.group(1)) if 0 < year_val <= 100: row['policy_year'] = year_val # 用列位置映射数值 if columns: # 获取非 policy_year 的列（按原始顺序） value_columns = [(pos, f) for pos, f in columns if f != 'policy_year'] # 确定从哪个数字开始分配（跳过已识别的年度） start_idx = 0 if 'policy_year' in row: # 验证第一个数字确实是年度 first_num = _parse_money(numbers[0]) if numbers else None if first_num is not None and first_num == row['policy_year']: start_idx = 1 # 跳过年度对应的数字 # 如果第一个数字是大的金额（如50000），说明行首没有年度列 elif first_num is not None and first_num > 100: start_idx = 0 # 从头开始分配给 value_columns for i, num_str in enumerate(numbers[start_idx:], start=start_idx): val = _parse_money(num_str) if val is None: continue col_idx = i - start_idx if col_idx < len(value_columns): _, field = value_columns[col_idx] row[field] = val else: # 没有列位置信息，按顺序猜测 all_fields = ['policy_year', 'total_premium_paid', 'guaranteed_cash_value', 'reversionary_bonus', 'terminal_dividend', 'total_surrender_value', 'death_benefit'] for i, num_str in enumerate(numbers): if i >= len(all_fields): break val = _parse_money(num_str) if val is not None: row[all_fields[i]] = val # 如果没有识别到保单年度，检查是否有"年龄"列推算 if 'policy_year' not in row: age_match = re.search(r'(\d{1,3})\s*岁', line) if age_match: age = int(age_match.group(1)) row['_age'] = age # 留给上层推算 return row if row else None def _fill_missing_premiums(rows: list[dict], annual_premium: float | None) -> list[dict]: """填充缺失的 total_premium_paid。""" if not annual_premium or annual_premium <= 0: return rows for row in rows: if row.get('total_premium_paid') is None and row.get('policy_year') is not None: row['total_premium_paid'] = annual_premium * row['policy_year'] return rows def _ensure_required_fields(row: dict) -> dict: """确保每行有所有必需字段。""" required = { 'policy_year': None, 'total_premium_paid': None, 'guaranteed_cash_value': None, 'reversionary_bonus': None, 'terminal_dividend': None, 'total_surrender_value': None, 'death_benefit': None, } result = dict(required) result.update(row) return result # ─── 提领表提取 ────────────────────────────────────────── _WITHDRAWAL_HEADER_KEYWORDS = [ '提取', '提款', '现金提取', '現金提取', '款项提取', '款項提取', 'withdrawal', 'cash withdrawal', ] _WITHDRAWAL_REQUIRED_COLS = [ '价值', '價值', 'surrender', 'cash', 'value', '退保', ] def _extract_withdrawal_rows(text: str) -> list[dict]: """从 PDF 文本中提取提领/提款演示表。 注意：只提取"总额"列，不取"保证"子列。 """ rows = [] lines = text.split('\n') in_table = False columns: list[tuple[int, str]] = [] consecutive_non_data = 0 MAX_NON_DATA_LINES = 5 for line in lines: stripped = line.strip() if not stripped: continue if not in_table: has_header = any(kw in stripped for kw in _WITHDRAWAL_HEADER_KEYWORDS) has_value = any(kw in stripped.lower() for kw in _WITHDRAWAL_REQUIRED_COLS) if has_header and has_value: columns = _detect_columns(stripped) if len(columns) >= 2: in_table = True consecutive_non_data = 0 logger.debug(f"检测到提领表头: {[(c[1], c[0]) for c in columns]}") continue # 跳过子表头（"保证 非保证 总额"等） subheader_pats = ['保证', '保證', '非保证', '非保證', '总额', '總額', 'guaranteed', 'non-guaranteed', 'total'] if sum(1 for p in subheader_pats if p in stripped.lower()) >= 2: continue numbers = re.findall(r'[\d,]+(?:\.\d+)?', stripped) if not numbers: consecutive_non_data += 1 if consecutive_non_data >= MAX_NON_DATA_LINES: break continue consecutive_non_data = 0 row = _parse_withdrawal_row(stripped, columns, numbers) if row and row.get('policy_year') is not None: rows.append(row) return rows def _parse_withdrawal_row( line: str, columns: list[tuple[int, str]], numbers: list[str] ) -> dict | None: """解析提领表单行。""" row = {} year_match = re.match(r'^\s*(\d{1,3})\b', line) if year_match: year_val = int(year_match.group(1)) if 0 < year_val <= 100: row['policy_year'] = year_val # 提领表标准字段（按优先级） withdrawal_fields = [ 'annual_withdrawal', 'total_withdrawn', 'surrender_value_before', 'surrender_value_after', ] if columns: value_columns = [(pos, f) for pos, f in columns if f != 'policy_year'] start_idx = 0 if 'policy_year' in row: first_num = _parse_money(numbers[0]) if numbers else None if first_num is not None and first_num == row['policy_year']: start_idx = 1 for i, num_str in enumerate(numbers[start_idx:], start=start_idx): val = _parse_money(num_str) if val is None: continue col_idx = i - start_idx if col_idx < len(value_columns): _, field = value_columns[col_idx] row[field] = val else: for i, num_str in enumerate(numbers): if i == 0 and 'policy_year' in row: continue field_idx = i - 1 if 'policy_year' in row else i if field_idx >= len(withdrawal_fields): break val = _parse_money(num_str) if val is not None: row[withdrawal_fields[field_idx]] = val return row if row else None def _ensure_withdrawal_fields(row: dict) -> dict: """确保提领行有所有必需字段。""" required = { 'policy_year': None, 'annual_withdrawal': None, 'total_withdrawn': None, 'surrender_value_before': None, 'surrender_value_after': None, } result = dict(required) result.update(row) return result # ─── 产品类型识别 ───────────────────────────────────────── def _detect_product_type(text: str, benefit_rows: list[dict]) -> str: """从文本和提取数据推断产品类型：savings / ci / iul。""" text_lower = text.lower() # 文本关键词判断 ci_keywords = ['危疾', '重疾', '重大疾病', 'critical illness', 'ci plan', '严重疾病', '危疾保障', 'dread disease'] iul_keywords = ['iul', 'index universal', 'universal life', '指数型万用', '万用寿险', '萬用壽險', 'index account', 'index_accounts'] ci_score = sum(1 for kw in ci_keywords if kw in text_lower) iul_score = sum(1 for kw in iul_keywords if kw in text_lower) # 数据特征判断 for row in benefit_rows: if row.get('account_value') or row.get('non_guaranteed_account_value'): iul_score += 3 if row.get('death_benefit') and not row.get('guaranteed_cash_value'): ci_score += 1 # policy 字段判断 if re.search(r'(?:保额|投保额|sum\s*insured|保額)\s*[：:]\s*[\d,]+', text_lower): ci_score += 1 if iul_score > ci_score and iul_score > 0: return 'iul' if ci_score > 0: return 'ci' return 'savings' # ─── 主提取函数 ────────────────────────────────────────── def extract_insurance_regex(pdf_text: str) -> dict: """从 PDF 原始文本中用纯正则提取结构化保险数据。 Args: pdf_text: PDF 提取的原始文本 Returns: 符合 extraction schema 的 dict，字段缺失填 null """ product_name = _extract_product_name(pdf_text) or 'unknown' currency = _extract_currency(pdf_text) annual_premium = _extract_annual_premium(pdf_text) insured_info = _extract_insured_info(pdf_text) premium_period = _extract_premium_payment_period(pdf_text) coverage_period = _extract_coverage_period(pdf_text) # 提取利益演示表 benefit_rows = _extract_benefit_rows(pdf_text) # 提取提领表 withdrawal_rows = _extract_withdrawal_rows(pdf_text) # 产品类型由上层 infer_plan_type() 推断（基于实际提取数据），此处不重复判断 # 填充缺失保费 benefit_rows = _fill_missing_premiums(benefit_rows, annual_premium) # 确保字段完整 benefit_rows = [_ensure_required_fields(r) for r in benefit_rows] withdrawal_rows = [_ensure_withdrawal_fields(r) for r in withdrawal_rows] # 按年度排序并去重 seen_years = set() deduped = [] for row in sorted(benefit_rows, key=lambda r: r.get('policy_year') or 0): year = row.get('policy_year') if year is not None and year not in seen_years: seen_years.add(year) deduped.append(row) benefit_rows = deduped # 提领表去重 seen_wd_years = set() deduped_wd = [] for row in sorted(withdrawal_rows, key=lambda r: r.get('policy_year') or 0): year = row.get('policy_year') if year is not None and year not in seen_wd_years: seen_wd_years.add(year) deduped_wd.append(row) withdrawal_rows = deduped_wd # 提取保额（如果有） sum_insured = None si_match = re.search( r'(?:保额|投保额|保額|sum\s*insured)\s*[：:]\s*[\$UuSsHhKk]*\s*([\d,]+(?:\.\d+)?)', pdf_text, re.IGNORECASE, ) if si_match: sum_insured = _parse_money(si_match.group(1)) logger.info( f"[RegexExtractor] 提取完成: product={product_name}, " f"currency={currency}, premium={annual_premium}, " f"age={insured_info.get('age')}, benefit_rows={len(benefit_rows)}, " f"withdrawal_rows={len(withdrawal_rows)}" ) data = { 'product_name': product_name, 'product_type': 'savings', # 由上层 infer_plan_type() 覆盖 'insured': { 'name': None, 'age': insured_info.get('age'), 'gender': insured_info.get('gender'), 'relation': None, 'smoker': None, }, 'policy': { 'product_name': product_name, 'currency': currency, 'sum_insured': sum_insured, 'basic_sum_insured': None, 'annual_premium': annual_premium, 'premium_payment_period': premium_period, 'coverage_period': coverage_period, }, 'benefit_illustration': benefit_rows, 'withdrawal_illustration': withdrawal_rows, 'sales_insights': None, } # CI 产品：提取保障项目（如果文本包含 CI 关键词） coverage_items = _extract_coverage_items(pdf_text) if coverage_items: data['coverage_items'] = coverage_items # IUL 产品：提取指数账户（如果文本包含 IUL 关键词） index_accounts = _extract_index_accounts(pdf_text) if index_accounts: data['index_accounts'] = index_accounts return data def _extract_coverage_items(text: str) -> list[dict]: """提取危疾保险的保障项目。""" items = [] # 匹配 "保障项目 赔付金额" 格式的表 lines = text.split('\n') in_section = False for line in lines: stripped = line.strip() if not stripped: continue if '保障项目' in stripped or '保障範圍' in stripped or 'coverage' in stripped.lower(): in_section = True continue if in_section: # 尝试匹配 "项目名 金额" 格式 m = re.match(r'^(.{2,20})\s+[\$UuSsHhKk]*\s*([\d,]+(?:\.\d+)?)', stripped) if m: label = m.group(1).strip() amount = _parse_money(m.group(2)) if amount and label: items.append({'label': label, 'amount': amount, 'percentage': None, 'description': None}) elif not re.search(r'\d', stripped): break # 非数字行，可能到了下一节 return items def _extract_index_accounts(text: str) -> list[dict]: """提取 IUL 指数账户信息。""" accounts = [] # 匹配 "账户名 配置比例 利率" 格式 patterns = [ r'(S&P\s*500|Hang\s*Seng|恒生指数|Global\s*index|指数\d?)\s+([\d.]+)%?\s+([\d.]+)%?\s+([\d.]+)%?', r'(S&P\s*500|Hang\s*Seng|恒生指数|Global\s*index)\s+([\d.]+)%', ] for p in patterns: for m in re.finditer(p, text, re.IGNORECASE): account = { 'name': m.group(1).strip(), 'allocation': float(m.group(2)), 'current_rate': float(m.group(3)) if m.lastindex >= 3 else None, 'guaranteed_floor': float(m.group(4)) if m.lastindex >= 4 else None, } accounts.append(account) return accounts def count_benefit_rows(data: dict) -> int: """统计提取到的利益演示行数。""" rows = data.get('benefit_illustration', []) if not isinstance(rows, list): return 0 return len(rows)