430 lines
17 KiB
Python
430 lines
17 KiB
Python
#!/usr/bin/env python3
|
||
"""
|
||
大医网OCR/文本杂质清洗脚本
|
||
-----------------------------
|
||
清洗7个核心库的JSON源文件中的OCR形近错字和格式杂质。
|
||
直接修改源文件（明确OCR校正，不改变语义）。
|
||
生成清洗报告到 04_分析报告/数据质量/OCR清洗报告.md
|
||
"""
|
||
|
||
import json
|
||
import os
|
||
import re
|
||
from collections import defaultdict, OrderedDict
|
||
|
||
# ============================================================
|
||
# 配置
|
||
# ============================================================
|
||
BASE_DIR = "/home/songyi/Documents/ai_agent_scraper_study/data/大医网"
|
||
SOURCE_DIR = os.path.join(BASE_DIR, "01_来源数据")
|
||
SUBDIRS = ["疾病", "方剂", "中药材", "针灸穴位", "症状", "术语", "药膳食疗"]
|
||
REPORT_DIR = os.path.join(BASE_DIR, "04_分析报告", "数据质量")
|
||
REPORT_FILE = os.path.join(REPORT_DIR, "OCR清洗报告.md")
|
||
|
||
# ============================================================
|
||
# 清洗规则定义（按优先级）
|
||
# 每条规则: (pattern, replacement, description, is_regex)
|
||
# ============================================================
|
||
CLEAN_RULES = [
|
||
# --- OCR形近错字（注意负向零宽断言）---
|
||
(re.compile(r'(?<!酸)枣仁'), '酸枣仁', '枣仁→酸枣仁', True),
|
||
(re.compile(r'(?<!耳)白木(?!耳)'), '白术', '白木→白术 (排除白木耳)', True),
|
||
(re.compile(r'(?<!薏)苡仁'), '薏苡仁', '苡仁→薏苡仁', True),
|
||
(re.compile(r'枝子'), '栀子', '枝子→栀子', True), # 仅匹配"枝子"整体词
|
||
(re.compile(r'构杞'), '枸杞', '构杞→枸杞', True),
|
||
(re.compile(r'构杞子'), '枸杞子', '构杞子→枸杞子', True),
|
||
(re.compile(r'当妇'), '当归', '当妇→当归', True),
|
||
(re.compile(r'意苡仁'), '薏苡仁', '意苡仁→薏苡仁', True),
|
||
(re.compile(r'获苓'), '茯苓', '获苓→茯苓', True),
|
||
(re.compile(r'川萼'), '川芎', '川萼→川芎', True),
|
||
(re.compile(r'黄茂'), '黄芪', '黄茂→黄芪', True),
|
||
(re.compile(r'甘革'), '甘草', '甘革→甘草', True),
|
||
(re.compile(r'灵艺'), '灵芝', '灵艺→灵芝', True),
|
||
(re.compile(r'大赉'), '大枣', '大赉→大枣', True),
|
||
(re.compile(r'干娈'), '干姜', '干娈→干姜', True),
|
||
# --- 格式杂质 ---
|
||
(re.compile(r'\u3000'), ' ', '全角空格→半角', True),
|
||
(re.compile(r'[\u200b\u200c\u200d\ufeff]'), '', '零宽字符删除', True),
|
||
(re.compile(r' +'), ' ', '多空格合并', True),
|
||
]
|
||
|
||
|
||
def apply_clean_rules(text, stats_counter, file_stat):
|
||
"""对单个字符串应用所有清洗规则，记录每类错误的修正次数"""
|
||
if not isinstance(text, str):
|
||
return text, 0
|
||
|
||
total_changes = 0
|
||
for pattern, replacement, desc, is_regex in CLEAN_RULES:
|
||
if is_regex:
|
||
new_text, count = pattern.subn(replacement, text)
|
||
if count > 0:
|
||
stats_counter[desc] += count
|
||
file_stat[desc] += count
|
||
total_changes += count
|
||
text = new_text
|
||
else:
|
||
# 普通字符串替换（保留备用）
|
||
count = text.count(pattern)
|
||
if count > 0:
|
||
text = text.replace(pattern, replacement)
|
||
stats_counter[desc] += count
|
||
file_stat[desc] += count
|
||
total_changes += count
|
||
return text, total_changes
|
||
|
||
|
||
def walk_and_clean(data, stats_counter, file_stat):
|
||
"""递归遍历嵌套的dict/list结构，清洗所有字符串字段和键名"""
|
||
total_changes = 0
|
||
if isinstance(data, dict):
|
||
# 先处理键名：可能含有零宽字符等杂质
|
||
cleaned_keys = {}
|
||
for key in list(data.keys()):
|
||
if isinstance(key, str):
|
||
new_key, changes = apply_clean_rules(key, stats_counter, file_stat)
|
||
if changes > 0:
|
||
cleaned_keys[key] = new_key
|
||
total_changes += changes
|
||
# 重建dict（如果键名有变化）
|
||
if cleaned_keys:
|
||
for old_key, new_key in cleaned_keys.items():
|
||
data[new_key] = data.pop(old_key)
|
||
# 处理值
|
||
for key in list(data.keys()):
|
||
val = data[key]
|
||
if isinstance(val, str):
|
||
new_val, changes = apply_clean_rules(val, stats_counter, file_stat)
|
||
if changes > 0:
|
||
data[key] = new_val
|
||
total_changes += changes
|
||
elif isinstance(val, (dict, list)):
|
||
total_changes += walk_and_clean(val, stats_counter, file_stat)
|
||
elif isinstance(data, list):
|
||
for i in range(len(data)):
|
||
val = data[i]
|
||
if isinstance(val, str):
|
||
new_val, changes = apply_clean_rules(val, stats_counter, file_stat)
|
||
if changes > 0:
|
||
data[i] = new_val
|
||
total_changes += changes
|
||
elif isinstance(val, (dict, list)):
|
||
total_changes += walk_and_clean(val, stats_counter, file_stat)
|
||
return total_changes
|
||
|
||
|
||
def clean_file(filepath, stats_counter, file_changes_log):
|
||
"""
|
||
清洗单个JSON文件：
|
||
1. json.load 读取
|
||
2. 递归遍历清洗
|
||
3. json.dump 写回
|
||
返回 (修改文件数:0或1, 修正总次数)
|
||
"""
|
||
try:
|
||
with open(filepath, 'r', encoding='utf-8') as f:
|
||
data = json.load(f)
|
||
except (json.JSONDecodeError, UnicodeDecodeError) as e:
|
||
print(f" [!] 跳过无法解析的JSON: {filepath} — {e}")
|
||
return 0, 0
|
||
|
||
file_stat = defaultdict(int)
|
||
changes = walk_and_clean(data, stats_counter, file_stat)
|
||
|
||
if changes == 0:
|
||
return 0, 0
|
||
|
||
# 记录此文件的修改详情
|
||
file_changes_log.append({
|
||
'filepath': filepath,
|
||
'changes': changes,
|
||
'detail': {k: v for k, v in file_stat.items() if v > 0}
|
||
})
|
||
|
||
# 写回
|
||
try:
|
||
with open(filepath, 'w', encoding='utf-8') as f:
|
||
json.dump(data, f, ensure_ascii=False, indent=2)
|
||
return 1, changes
|
||
except Exception as e:
|
||
print(f" [!!] 写回失败: {filepath} — {e}")
|
||
return 0, 0
|
||
|
||
|
||
def scan_for_residual(data):
|
||
"""扫描数据中是否仍有任何未清洗的匹配项"""
|
||
residuals = defaultdict(int)
|
||
def _scan(d):
|
||
if isinstance(d, dict):
|
||
for v in d.values():
|
||
_scan(v)
|
||
elif isinstance(d, list):
|
||
for v in d:
|
||
_scan(v)
|
||
elif isinstance(d, str):
|
||
for pattern, replacement, desc, is_regex in CLEAN_RULES:
|
||
if is_regex:
|
||
matches = pattern.findall(d)
|
||
if matches:
|
||
residuals[desc] += len(matches)
|
||
else:
|
||
cnt = d.count(pattern)
|
||
if cnt:
|
||
residuals[desc] += cnt
|
||
_scan(data)
|
||
return residuals
|
||
|
||
|
||
def validate_residuals():
|
||
"""清洗后重新扫描所有JSON文件，验证零残留"""
|
||
residual_counts = defaultdict(int)
|
||
residual_files = defaultdict(list)
|
||
|
||
for subdir in SUBDIRS:
|
||
dirpath = os.path.join(SOURCE_DIR, subdir)
|
||
if not os.path.isdir(dirpath):
|
||
continue
|
||
for fname in os.listdir(dirpath):
|
||
if not fname.endswith('.json'):
|
||
continue
|
||
fpath = os.path.join(dirpath, fname)
|
||
try:
|
||
with open(fpath, 'r', encoding='utf-8') as f:
|
||
data = json.load(f)
|
||
except:
|
||
continue
|
||
res = scan_for_residual(data)
|
||
if res:
|
||
for desc, cnt in res.items():
|
||
residual_counts[desc] += cnt
|
||
residual_files[desc].append(fname)
|
||
|
||
return residual_counts, residual_files
|
||
|
||
|
||
def get_dataset_name(subdir):
|
||
"""映射子目录名到中文数据集名"""
|
||
mapping = {
|
||
'疾病': '疾病库',
|
||
'方剂': '方剂库',
|
||
'中药材': '中药材库',
|
||
'针灸穴位': '针灸穴位库',
|
||
'症状': '症状库',
|
||
'术语': '术语库',
|
||
'药膳食疗': '药膳食疗库',
|
||
}
|
||
return mapping.get(subdir, subdir)
|
||
|
||
|
||
def relative_path(filepath):
|
||
"""将绝对路径转为相对路径（用于报告）"""
|
||
return filepath.replace(BASE_DIR, '.../大医网')
|
||
|
||
|
||
def main():
|
||
print("=" * 70)
|
||
print(" 大医网OCR/文本杂质清洗")
|
||
print("=" * 70)
|
||
|
||
# 全局统计
|
||
# stats_counter: { desc: total_count }
|
||
stats_counter = defaultdict(int)
|
||
# 按数据集的统计: { subdir: { 'files_cleaned': N, 'total_changes': N, 'detail': {desc: count} } }
|
||
dataset_stats = {}
|
||
# 所有修改的文件日志
|
||
all_file_changes_log = []
|
||
|
||
for subdir in SUBDIRS:
|
||
dirpath = os.path.join(SOURCE_DIR, subdir)
|
||
if not os.path.isdir(dirpath):
|
||
print(f"\n [!] 目录不存在: {dirpath}")
|
||
continue
|
||
|
||
print(f"\n{'─'*50}")
|
||
print(f" 处理: {subdir}/")
|
||
print(f"{'─'*50}")
|
||
|
||
json_files = sorted([f for f in os.listdir(dirpath) if f.endswith('.json')])
|
||
total_files = len(json_files)
|
||
files_cleaned = 0
|
||
total_changes_subdir = 0
|
||
subdir_detail = defaultdict(int)
|
||
|
||
for idx, fname in enumerate(json_files):
|
||
fpath = os.path.join(dirpath, fname)
|
||
file_changes_log_entry = []
|
||
cleaned_flag, changes = clean_file(fpath, stats_counter, file_changes_log_entry)
|
||
files_cleaned += cleaned_flag
|
||
total_changes_subdir += changes
|
||
|
||
if file_changes_log_entry:
|
||
all_file_changes_log.extend(file_changes_log_entry)
|
||
for desc, cnt in file_changes_log_entry[0]['detail'].items():
|
||
subdir_detail[desc] += cnt
|
||
|
||
# 进度显示
|
||
if (idx + 1) % 200 == 0 or idx == 0 or (idx + 1) == total_files:
|
||
print(f" 进度: {idx+1}/{total_files} | 已清洗: {files_cleaned} 个文件 | 修正: {total_changes_subdir} 处")
|
||
|
||
dataset_stats[subdir] = {
|
||
'total_files': total_files,
|
||
'files_cleaned': files_cleaned,
|
||
'total_changes': total_changes_subdir,
|
||
'detail': dict(subdir_detail),
|
||
}
|
||
|
||
ds_name = get_dataset_name(subdir)
|
||
print(f" ✓ {ds_name}: {files_cleaned}/{total_files} 个文件被修改，共修正 {total_changes_subdir} 处")
|
||
|
||
# ============================================================
|
||
# 汇总
|
||
# ============================================================
|
||
print("\n" + "=" * 70)
|
||
print(" 清洗完成 - 汇总统计")
|
||
print("=" * 70)
|
||
|
||
total_files_all = sum(s['total_files'] for s in dataset_stats.values())
|
||
total_cleaned_all = sum(s['files_cleaned'] for s in dataset_stats.values())
|
||
total_changes_all = sum(s['total_changes'] for s in dataset_stats.values())
|
||
|
||
print(f" 总文件数: {total_files_all}")
|
||
print(f" 被修改文件: {total_cleaned_all}")
|
||
print(f" 总修正处: {total_changes_all}")
|
||
print(f"\n 错误类型分布:")
|
||
for desc, cnt in sorted(stats_counter.items(), key=lambda x: -x[1]):
|
||
print(f" {desc}: {cnt}")
|
||
|
||
# ============================================================
|
||
# 清洗后验证 - 零残留
|
||
# ============================================================
|
||
print("\n" + "=" * 70)
|
||
print(" 清洗后残留验证")
|
||
print("=" * 70)
|
||
residual_counts, residual_files = validate_residuals()
|
||
|
||
if residual_counts:
|
||
print(f" ⚠ 发现 {sum(residual_counts.values())} 处残留:")
|
||
for desc, cnt in sorted(residual_counts.items(), key=lambda x: -x[1]):
|
||
print(f" {desc}: {cnt} 处 (共 {len(residual_files.get(desc, []))} 个文件)")
|
||
residual_status = "有残留"
|
||
else:
|
||
print(" ✓ 零残留！所有规则均已正确应用。")
|
||
residual_status = "零残留 ✓"
|
||
|
||
# ============================================================
|
||
# 生成清洗报告
|
||
# ============================================================
|
||
print(f"\n 生成报告 → {REPORT_FILE}")
|
||
os.makedirs(REPORT_DIR, exist_ok=True)
|
||
|
||
lines = []
|
||
lines.append("# OCR/文本杂质清洗报告")
|
||
lines.append("")
|
||
lines.append(f"**数据目录：** `{SOURCE_DIR}`")
|
||
lines.append(f"**子目录：** 疾病/ 方剂/ 中药材/ 针灸穴位/ 症状/ 术语/ 药膳食疗")
|
||
lines.append(f"**清洗时间：** {__import__('datetime').datetime.now().strftime('%Y-%m-%d %H:%M:%S')}")
|
||
lines.append("")
|
||
lines.append("---")
|
||
lines.append("")
|
||
lines.append("## 一、汇总统计")
|
||
lines.append("")
|
||
lines.append("| 数据集名 | 总文件数 | 清洗文件数 | 总修正处 | 错误类型分布 |")
|
||
lines.append("|----------|----------|------------|----------|--------------|")
|
||
|
||
for subdir in SUBDIRS:
|
||
if subdir not in dataset_stats:
|
||
continue
|
||
s = dataset_stats[subdir]
|
||
ds_name = get_dataset_name(subdir)
|
||
detail_str = "; ".join([f"{d}: {c}" for d, c in sorted(s['detail'].items(), key=lambda x: -x[1])])
|
||
if not detail_str:
|
||
detail_str = "-"
|
||
lines.append(f"| {ds_name} | {s['total_files']} | {s['files_cleaned']} | {s['total_changes']} | {detail_str} |")
|
||
|
||
lines.append(f"| **合计** | **{total_files_all}** | **{total_cleaned_all}** | **{total_changes_all}** | **{'; '.join([f'{d}: {c}' for d, c in sorted(stats_counter.items(), key=lambda x: -x[1])])}** |")
|
||
lines.append("")
|
||
|
||
lines.append("## 二、清洗规则详情")
|
||
lines.append("")
|
||
lines.append("| 优先级 | 规则 | 模式 | 替换为 | 修正次数 |")
|
||
lines.append("|--------|------|------|--------|----------|")
|
||
for i, (pattern, replacement, desc, is_regex) in enumerate(CLEAN_RULES, 1):
|
||
pat_str = pattern.pattern if is_regex else pattern
|
||
cnt = stats_counter.get(desc, 0)
|
||
lines.append(f"| {i} | {desc} | `{pat_str}` | `{replacement}` | {cnt} |")
|
||
lines.append("")
|
||
|
||
lines.append("## 三、清洗后验证")
|
||
lines.append("")
|
||
if residual_counts:
|
||
lines.append(f"### ⚠ 残留检测结果：{sum(residual_counts.values())} 处残留")
|
||
lines.append("")
|
||
lines.append("| 规则 | 残留数 | 涉及文件数 |")
|
||
lines.append("|------|--------|------------|")
|
||
for desc, cnt in sorted(residual_counts.items(), key=lambda x: -x[1]):
|
||
lines.append(f"| {desc} | {cnt} | {len(residual_files.get(desc, []))} |")
|
||
lines.append("")
|
||
lines.append("**注意：** 残留可能存在于非字符串字段（如JSON key）或特殊格式中，需人工复核。")
|
||
else:
|
||
lines.append("**验证结果：零残留 ✓** — 所有规则均已正确应用，无未处理的匹配项。")
|
||
lines.append("")
|
||
|
||
lines.append("## 四、错误类型分布（全部）")
|
||
lines.append("")
|
||
lines.append("| 错误类型 | 修正次数 | 占比 |")
|
||
lines.append("|----------|----------|------|")
|
||
total = sum(stats_counter.values())
|
||
for desc, cnt in sorted(stats_counter.items(), key=lambda x: -x[1]):
|
||
pct = f"{cnt/total*100:.1f}%" if total > 0 else "-"
|
||
lines.append(f"| {desc} | {cnt} | {pct} |")
|
||
lines.append(f"| **总计** | **{total}** | **100%** |")
|
||
lines.append("")
|
||
|
||
lines.append("## 五、被修改文件列表")
|
||
lines.append("")
|
||
lines.append(f"共 {len(all_file_changes_log)} 个文件被修改：")
|
||
lines.append("")
|
||
|
||
# 按数据集分组显示
|
||
by_dataset = defaultdict(list)
|
||
for entry in all_file_changes_log:
|
||
fpath = entry['filepath']
|
||
for subdir in SUBDIRS:
|
||
if f'/01_来源数据/{subdir}/' in fpath:
|
||
by_dataset[subdir].append(entry)
|
||
break
|
||
|
||
for subdir in SUBDIRS:
|
||
entries = by_dataset.get(subdir, [])
|
||
if not entries:
|
||
continue
|
||
ds_name = get_dataset_name(subdir)
|
||
lines.append(f"### {ds_name}（{len(entries)} 个文件）")
|
||
lines.append("")
|
||
lines.append("| 文件名 | 修正总数 | 修正详情 |")
|
||
lines.append("|--------|----------|----------|")
|
||
for entry in sorted(entries, key=lambda x: -x['changes']):
|
||
fname = os.path.basename(entry['filepath'])
|
||
detail_str = "; ".join([f"{d}: {c}" for d, c in sorted(entry['detail'].items(), key=lambda x: -x[1])])
|
||
lines.append(f"| {fname} | {entry['changes']} | {detail_str} |")
|
||
lines.append("")
|
||
|
||
# 写入报告
|
||
report_content = "\n".join(lines)
|
||
with open(REPORT_FILE, 'w', encoding='utf-8') as f:
|
||
f.write(report_content)
|
||
|
||
print(f"\n{'='*70}")
|
||
print(f" 报告已生成: {REPORT_FILE}")
|
||
print(f" 总文件: {total_files_all}")
|
||
print(f" 被修改: {total_cleaned_all}")
|
||
print(f" 总修正: {total_changes_all}")
|
||
print(f" 残留验证: {residual_status}")
|
||
print(f"{'='*70}")
|
||
|
||
return total_files_all, total_cleaned_all, total_changes_all, residual_status
|
||
|
||
|
||
if __name__ == "__main__":
|
||
main()
|