Yuzhiran Dev
f1dc29f439
- nvidia 激活、模型改为 step-3.7-flash → 用于日常生成/采集/优化等 - compliance_optimizer.py 显式指定 provider=sensenova → 合规审查走 deepseek-v4-flash - _FALLBACK 同步更新为 step-3.7-flash - 管理后台标签同步更新
364 lines
15 KiB
Python
364 lines
15 KiB
Python
#!/usr/bin/env python3
|
||
"""
|
||
合规审查：文章合规检查 → LLM迭代修复
|
||
从articles表读取待审文章，进行合规评分；不合格文章由LLM修复（最多3次），通过后更新选题状态为待发布
|
||
"""
|
||
import json, datetime, logging, sys, re
|
||
from pathlib import Path
|
||
from typing import Dict, List, Optional, Tuple
|
||
from dataclasses import dataclass, asdict
|
||
|
||
PROJECT_ROOT = Path(__file__).resolve().parent.parent
|
||
sys.path.insert(0, str(PROJECT_ROOT))
|
||
sys.path.insert(0, str(PROJECT_ROOT / "platform" / "backend"))
|
||
|
||
from scripts.compliance_checker import check_article
|
||
try:
|
||
from app.core.nvidia_client import call_llm
|
||
HAVE_LLM = True
|
||
except ImportError:
|
||
HAVE_LLM = False
|
||
|
||
from db_helper import get_topic_by_id, update_topic_status, get_active_llm_config, get_articles_by_topic, save_article
|
||
from content_cleaner import strip_thinking, strip_ai_preface, strip_thinking_html, clean_html_content
|
||
from prompt_loader import get_prompt, get_prompt_params
|
||
|
||
DATA_DIR = PROJECT_ROOT / "automation" / "data"
|
||
DRAFTS_DIR = DATA_DIR / "drafts"
|
||
LOGS_DIR = PROJECT_ROOT / "automation" / "logs"
|
||
TODAY = datetime.datetime.now().strftime("%Y-%m-%d")
|
||
|
||
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s',
|
||
handlers=[logging.FileHandler(LOGS_DIR / f"optimizer_{TODAY}.log"), logging.StreamHandler()])
|
||
logger = logging.getLogger(__name__)
|
||
|
||
DEFAULT_PLATFORM_TAGS = {
|
||
"zhihu": ["科技", "职场"],
|
||
"xiaohongshu": ["AI", "可持续", "生活方式"]
|
||
}
|
||
|
||
_cached_platform_tags = None
|
||
|
||
def _load_platform_tags():
|
||
global _cached_platform_tags
|
||
if _cached_platform_tags is not None:
|
||
return _cached_platform_tags
|
||
try:
|
||
from app.database import SessionLocal
|
||
from app.models import PlatformConfig
|
||
db = SessionLocal()
|
||
try:
|
||
configs = db.query(PlatformConfig).filter(PlatformConfig.is_active == True).all()
|
||
if configs:
|
||
_cached_platform_tags = {}
|
||
for c in configs:
|
||
tags = c.to_dict().get("allowed_tags", [])
|
||
if tags:
|
||
_cached_platform_tags[c.platform] = tags
|
||
if _cached_platform_tags:
|
||
logger.info(f"从DB加载 {len(_cached_platform_tags)} 个平台的标签")
|
||
return _cached_platform_tags
|
||
finally:
|
||
db.close()
|
||
except Exception as e:
|
||
logger.warning(f"从DB加载 platform_tags 失败: {e}")
|
||
_cached_platform_tags = DEFAULT_PLATFORM_TAGS
|
||
return _cached_platform_tags
|
||
|
||
def get_platform_tags():
|
||
return _load_platform_tags()
|
||
|
||
_llm_config_cache = None
|
||
|
||
def get_llm_config():
|
||
global _llm_config_cache
|
||
if _llm_config_cache is None:
|
||
_llm_config_cache = get_active_llm_config()
|
||
return _llm_config_cache
|
||
|
||
@dataclass
|
||
class OptimizationResult:
|
||
file: str
|
||
platform: str
|
||
topic_id: str
|
||
title: str
|
||
original_issues: int
|
||
fixed_issues: int
|
||
final_score: int
|
||
status: str
|
||
|
||
def load_topic_map():
|
||
from db_helper import export_topics_to_json
|
||
topics = export_topics_to_json()
|
||
return {t['id']: t for t in topics}
|
||
|
||
def get_articles_from_db(topic_ids: Optional[List[str]] = None, today_only: bool = False) -> List[Tuple[str, str, str]]:
|
||
"""从 articles 表读取 HTML 内容
|
||
|
||
Returns: [(html_content, platform, topic_id), ...]
|
||
"""
|
||
from db_helper import get_articles_by_topic
|
||
results = []
|
||
seen_topics = set()
|
||
if topic_ids:
|
||
for tid in topic_ids:
|
||
articles = get_articles_by_topic(tid)
|
||
for a in articles:
|
||
if a.get("html_content"):
|
||
results.append((a["html_content"], a["platform"], a["topic_id"]))
|
||
seen_topics.add(a["topic_id"])
|
||
else:
|
||
from app.database import SessionLocal
|
||
from app.models import Article
|
||
from sqlalchemy import func
|
||
db = SessionLocal()
|
||
try:
|
||
query = db.query(Article).filter(Article.html_content.isnot(None))
|
||
if today_only:
|
||
cutoff = datetime.datetime.now() - datetime.timedelta(hours=24)
|
||
query = query.filter(Article.created_at >= cutoff)
|
||
all_articles = query.all()
|
||
for a in all_articles:
|
||
results.append((a.html_content, a.platform, a.topic_id))
|
||
finally:
|
||
db.close()
|
||
return results
|
||
|
||
def fix_wechat_title(html: str, title: str) -> str:
|
||
suffix = f" - {TODAY} - 微信公众号"
|
||
max_base_len = 32 - len(suffix)
|
||
title_tag = re.search(r'<title>([^<]+)</title>', html)
|
||
if title_tag:
|
||
full_title = title_tag.group(1)
|
||
if full_title.endswith(suffix):
|
||
base = full_title[:-len(suffix)]
|
||
else:
|
||
base = full_title.split(" - ")[0]
|
||
if len(base) > max_base_len:
|
||
base = base[:max_base_len-3] + "..."
|
||
new_full = base + suffix
|
||
html = html.replace(full_title, new_full)
|
||
h1_match = re.search(r'<h1[^>]*>([^<]+)</h1>', html)
|
||
if h1_match:
|
||
current_h1 = h1_match.group(1)
|
||
base_h1 = current_h1.split(" - ")[0] if " - " in current_h1 else current_h1
|
||
if len(base_h1) > 32:
|
||
base_h1 = base_h1[:29] + "..."
|
||
html = html.replace(current_h1, base_h1)
|
||
return html
|
||
|
||
def fix_tags(html: str, platform: str) -> str:
|
||
tags_map = get_platform_tags()
|
||
if platform == "zhihu":
|
||
tags_str = " ".join(f"#{t}" for t in tags_map.get("zhihu", []))
|
||
if '<div class="tags">' in html:
|
||
old = html.split('<div class="tags">')[1].split('</div>')[0]
|
||
html = html.replace(f'<div class="tags">{old}</div>', f'<div class="tags">{tags_str}</div>')
|
||
elif platform == "xiaohongshu":
|
||
tags_str = " ".join(f"#{t}" for t in tags_map.get("xiaohongshu", []))
|
||
if '<div class="hashtags">' in html:
|
||
old = html.split('<div class="hashtags">')[1].split('</div>')[0]
|
||
html = html.replace(f'<div class="hashtags">{old}</div>', f'<div class="hashtags">{tags_str}</div>')
|
||
return html
|
||
|
||
def polish_with_llm(html: str, platform: str, remaining_issues: Optional[List[Dict]] = None) -> Tuple[str, Optional[str]]:
|
||
"""用 LLM 优化文章内容，返回 (html, log_message_or_None)
|
||
如果指定 remaining_issues，则针对性修复合规问题
|
||
LLM 失败时自动重试一次
|
||
固定使用 opencode-go (deepseek-v4-flash) — 审查用更好的模型
|
||
"""
|
||
if not HAVE_LLM:
|
||
return html, None
|
||
for attempt in range(2):
|
||
try:
|
||
temperature = 0.5
|
||
max_tokens = 4000
|
||
system_prompt = "你是一个专业的内容合规与优化助手，擅长在保持文章质量和可读性的前提下修复合规问题。"
|
||
if remaining_issues:
|
||
issues_desc = "\n".join(
|
||
f"- [{i['type']}] {i.get('category','')}: {i.get('detail','')}"
|
||
for i in remaining_issues
|
||
)
|
||
prompt = get_prompt("compliance_fix", issues_desc=issues_desc, html=html)
|
||
else:
|
||
prompt = get_prompt("compliance_polish", html=html)
|
||
polished = call_llm(prompt, provider="sensenova", temperature=temperature, max_tokens=max_tokens, system_prompt=system_prompt)
|
||
polished = clean_html_content(polished)
|
||
polished = strip_ai_preface(polished)
|
||
polished = strip_thinking_html(polished)
|
||
if '<h2' in polished or '<p>' in polished:
|
||
if len(polished) > len(html) * 0.3 and len(polished) > 100:
|
||
if not any(kw in polished for kw in ['保留', '建议', '可以', '应该', '推荐', '改为', '替换为']):
|
||
tag = "针对性修复" if remaining_issues else "常规润色"
|
||
return polished, f"LLM {tag}"
|
||
logger.warning(f"LLM 优化输出异常（过短或含建议性文字），保留原文 (len={len(polished)})")
|
||
except Exception as e:
|
||
logger.warning(f"LLM 优化失败 (尝试 {attempt+1}/2): {e}")
|
||
if attempt == 0:
|
||
logger.info(f"重试 LLM 调用...")
|
||
return html, None
|
||
|
||
def optimize_article(html: str, platform: str, topic_data: Dict, remaining_issues: Optional[List[Dict]] = None) -> Tuple[str, List[str]]:
|
||
logs = []
|
||
html = strip_thinking_html(html)
|
||
if platform == "wechat":
|
||
html = fix_wechat_title(html, topic_data.get("title", ""))
|
||
logs.append("标题截断（含后缀）")
|
||
if platform in ["zhihu", "xiaohongshu"]:
|
||
before = html
|
||
html = fix_tags(html, platform)
|
||
if html != before:
|
||
tags_map = get_platform_tags()
|
||
logs.append(f"标签标准化为{tags_map.get(platform, [])}")
|
||
polished, pol_log = polish_with_llm(html, platform, remaining_issues)
|
||
if pol_log:
|
||
html = polished
|
||
logs.append(pol_log)
|
||
return html, logs
|
||
|
||
def _load_platform_configs() -> Dict[str, Dict]:
|
||
"""从 DB 加载所有平台配置"""
|
||
from app.database import SessionLocal
|
||
from app.models import PlatformConfig
|
||
db = SessionLocal()
|
||
try:
|
||
configs = db.query(PlatformConfig).all()
|
||
return {c.platform: c.to_dict() for c in configs}
|
||
finally:
|
||
db.close()
|
||
|
||
def main(topic_ids: List[str] = None, today_only: bool = False):
|
||
logger.info("=== 合规审查与优化开始 ===")
|
||
logger.info("LLM 配置: opencode-go (model=deepseek-v4-flash) — 固定用于合规审查")
|
||
|
||
platform_configs = _load_platform_configs()
|
||
logger.info(f"已加载 {len(platform_configs)} 个平台配置")
|
||
|
||
if today_only:
|
||
logger.info("仅处理当天创建的选题文章")
|
||
|
||
articles = get_articles_from_db(topic_ids, today_only)
|
||
if not articles:
|
||
logger.warning("未找到任何文章（可能尚未创作或同步到 DB）")
|
||
report_file = DRAFTS_DIR / TODAY / "optimization_report.json"
|
||
report_file.parent.mkdir(parents=True, exist_ok=True)
|
||
with open(report_file, 'w', encoding='utf-8') as f:
|
||
json.dump({
|
||
"date": TODAY,
|
||
"summary": {"total_articles": 0, "passed_auto": 0, "average_score": 0},
|
||
"details": []
|
||
}, f, ensure_ascii=False, indent=2)
|
||
print("OPTIMIZATION_COMPLETE: 0 articles found")
|
||
sys.exit(0)
|
||
|
||
topic_map = load_topic_map()
|
||
results = []
|
||
force_passed_topics = set()
|
||
|
||
for html, platform_dir, topic_id in articles:
|
||
topic_data = topic_map.get(topic_id)
|
||
if not topic_data:
|
||
logger.warning(f"未找到选题: {topic_id}")
|
||
continue
|
||
|
||
pc = platform_configs.get(platform_dir, {})
|
||
check_result = check_article(html, platform_dir, topic_data, platform_config=pc)
|
||
issues = check_result['issues']
|
||
score = check_result['score']
|
||
label = f"{platform_dir}/{topic_id}"
|
||
|
||
if issues:
|
||
current_html = html
|
||
current_issues = list(issues)
|
||
current_score = score
|
||
for attempt in range(3):
|
||
optimized_html, opt_logs = optimize_article(current_html, platform_dir, topic_data, current_issues)
|
||
recheck = check_article(optimized_html, platform_dir, topic_data)
|
||
if recheck['passed'] and opt_logs and recheck['score'] > current_score:
|
||
save_article(topic_id, platform_dir, optimized_html)
|
||
logger.info(f"✅ {label} 已修复并通过审查 (第{attempt+1}次修复) 分数: {current_score}→{recheck['score']}")
|
||
results.append(OptimizationResult(
|
||
file=f"db:{platform_dir}_{topic_id}",
|
||
platform=platform_dir,
|
||
topic_id=topic_id,
|
||
title=topic_data.get('title',''),
|
||
original_issues=len(issues),
|
||
fixed_issues=len(issues),
|
||
final_score=recheck['score'],
|
||
status="passed"
|
||
))
|
||
break
|
||
if optimized_html == current_html and not opt_logs:
|
||
logger.warning(f"{label} LLM 未输出有效修改（第{attempt+1}次），继续重试...")
|
||
continue
|
||
current_issues = recheck['issues']
|
||
current_html = optimized_html
|
||
current_score = recheck['score']
|
||
else:
|
||
save_article(topic_id, platform_dir, current_html)
|
||
logger.warning(f"⚠️ {label} 仍有 {len(current_issues)} 个问题未修复，已强制通过（当前分 {current_score}）")
|
||
force_passed_topics.add(topic_id)
|
||
results.append(OptimizationResult(
|
||
file=f"db:{platform_dir}_{topic_id}",
|
||
platform=platform_dir,
|
||
topic_id=topic_id,
|
||
title=topic_data.get('title',''),
|
||
original_issues=len(issues),
|
||
fixed_issues=len(issues) - len(current_issues),
|
||
final_score=current_score,
|
||
status="force_passed"
|
||
))
|
||
else:
|
||
results.append(OptimizationResult(
|
||
file=f"db:{platform_dir}_{topic_id}",
|
||
platform=platform_dir,
|
||
topic_id=topic_id,
|
||
title=topic_data.get('title',''),
|
||
original_issues=0,
|
||
fixed_issues=0,
|
||
final_score=score,
|
||
status="passed"
|
||
))
|
||
logger.info(f"✅ {label} 合规检查通过 ({score}分)")
|
||
|
||
passed_scores = {}
|
||
for res in results:
|
||
if res.topic_id not in passed_scores:
|
||
passed_scores[res.topic_id] = []
|
||
passed_scores[res.topic_id].append(res.final_score)
|
||
|
||
for tid, scores in passed_scores.items():
|
||
if tid in force_passed_topics:
|
||
logger.warning(f"选题 {tid} 仍有未修复问题，跳过状态更新（保留当前状态）")
|
||
continue
|
||
avg_score = sum(scores) // len(scores)
|
||
update_topic_status(tid, 'ready', compliance_score=avg_score, reviewed_at=datetime.datetime.now())
|
||
logger.info(f"选题 {tid} 状态 → ready（待发布）, 合规分={avg_score}")
|
||
|
||
report_file = DRAFTS_DIR / TODAY / "optimization_report.json"
|
||
report_file.parent.mkdir(parents=True, exist_ok=True)
|
||
report = {
|
||
"date": TODAY,
|
||
"summary": {
|
||
"total_articles": len(results),
|
||
"passed_auto": len(results),
|
||
"average_score": sum(r.final_score for r in results) / len(results) if results else 0
|
||
},
|
||
"details": [asdict(r) for r in results]
|
||
}
|
||
with open(report_file, 'w', encoding='utf-8') as f:
|
||
json.dump(report, f, ensure_ascii=False, indent=2)
|
||
|
||
logger.info(f"✅ 合规审查完成: {len(results)} 篇文章, 其中 {len([r for r in results if r.status=='force_passed'])} 篇强制通过")
|
||
print(f"OPTIMIZATION_COMPLETE: {len(results)} articles, {len(force_passed_topics)} topics skipped status update")
|
||
sys.exit(0)
|
||
|
||
if __name__ == "__main__":
|
||
import argparse
|
||
parser = argparse.ArgumentParser(description='合规审查与优化任务')
|
||
parser.add_argument('--topic-ids', help='逗号分隔的选题ID列表，例如: A01,B02')
|
||
parser.add_argument('--today-only', action='store_true', help='仅处理当天创建的选题文章')
|
||
args = parser.parse_args()
|
||
topic_ids = args.topic_ids.split(',') if args.topic_ids else None
|
||
main(topic_ids, today_only=args.today_only)
|