feat: 添加 L4 AI 诊断报告（后端 DeepSeek 端点 + 前端摘要块与详情页）

2026-07-03 21:04:08 +08:00
parent f37679530a
commit 3dd718f53d
10 changed files with 917 additions and 7 deletions
@@ -6,11 +6,19 @@
  GET /api/analytics/episodes?year=2026 → 指定年份所有期次的收视数据 + 年度目标
 """

+import json
+import hashlib
+from pathlib import Path
+from datetime import datetime
+
 from fastapi import APIRouter, Depends, Query
+from pydantic import BaseModel
 from sqlalchemy import extract
 from sqlalchemy import distinct
 from sqlmodel import Session, select
+from openai import OpenAI

+from app.core.config import settings
 from app.core.deps import require_role
 from app.db.session import get_session
 from app.models.episode import Episode
@@ -19,6 +27,13 @@ from app.models.user import UserRole

 router = APIRouter(prefix="/api/analytics", tags=["收视分析"])

+# 诊断报告缓存（内存，重启清空）
+_report_cache = {}
+
+# prompt5 文件路径
+_PROJECT_ROOT = Path(__file__).parent.parent.parent.parent
+_PROMPT5_PATH = _PROJECT_ROOT / "ai-labeling" / "prompts" / "prompt5_diagnosis_report.md"
+

 def _require_read():
    """三角色都可读"""
@@ -111,4 +126,206 @@ def get_analytics_episodes(
        "year": year,
        "yearly_target": yearly_target,
        "episodes": ep_list,
-    }
+    }
+
+
+# ── AI 诊断报告 ──
+
+class DiagnosisRequest(BaseModel):
+    year: int
+    ep_start: int
+    ep_end: int
+    force: bool = False
+
+
+def _build_user_message(episodes, base_target, stretch_target, avg_share, pass_count, max_ep, min_ep):
+    """组装给 DeepSeek 的 user message，格式对齐 prompt5 的输入规范。"""
+    first_ep = episodes[0]
+    last_ep = episodes[-1]
+    count = len(episodes)
+
+    # 判色函数
+    def judge(share):
+        if share >= stretch_target:
+            return "优秀"
+        elif share >= base_target:
+            return "达标"
+        else:
+            return "待提升"
+
+    # 摸高完成率
+    stretch_pct = round(avg_share / stretch_target * 100, 1) if stretch_target > 0 else 0
+
+    lines = []
+    lines.append("请根据以下数据，撰写收视诊断分析报告。\n")
+
+    # 分析范围
+    lines.append("## 分析范围\n")
+    lines.append(
+        f"第{first_ep.episode_number}期《{first_ep.program_name}》至 "
+        f"第{last_ep.episode_number}期《{last_ep.program_name}》（共{count}期），"
+        f"{first_ep.air_date}至{last_ep.air_date}播出"
+    )
+    lines.append(f"年度目标：基础目标 {base_target}，摸高目标 {stretch_target}\n")
+
+    # 整体统计
+    lines.append("## 整体统计\n")
+    lines.append(f"- 平均份额：{avg_share}（摸高完成率 {stretch_pct}%）")
+    lines.append(f"- 达标期数：{pass_count}/{count}")
+    lines.append(f"- 最高份额：{float(max_ep.audience_share)}（第{max_ep.episode_number}期《{max_ep.program_name}》）")
+    lines.append(f"- 最低份额：{float(min_ep.audience_share)}（第{min_ep.episode_number}期《{min_ep.program_name}》）\n")
+
+    # 逐期数据表格
+    lines.append("## 逐期数据\n")
+    lines.append("| 播出期号 | 节目名 | 份额 | 判定 | 题材类型 | 叙事结构 | 钩子强度 | 装备领域 |")
+    lines.append("|---------|-------|------|------|---------|---------|---------|---------|")
+    for ep in episodes:
+        share = float(ep.audience_share)
+        domain_str = "、".join(ep.equipment_domain) if ep.equipment_domain else "-"
+        lines.append(
+            f"| 第{ep.episode_number}期 | {ep.program_name} | {share} | {judge(share)} "
+            f"| {ep.program_format or '-'} | {ep.narrative_structure or '-'} "
+            f"| {ep.opening_hook or '-'} | {domain_str} |"
+        )
+    lines.append("")
+
+    # 各期内容摘要卡
+    lines.append("## 各期内容摘要卡\n")
+    for ep in episodes:
+        share = float(ep.audience_share)
+        lines.append(f"### 第{ep.episode_number}期《{ep.program_name}》（份额 {share}）")
+        digest = ep.content_digest
+        if digest:
+            lines.append(f"- 核心切口：{digest.get('核心切口', '-')}")
+            # 叙事亮点可能是数组
+            highlights = digest.get('叙事亮点', [])
+            if isinstance(highlights, list):
+                lines.append(f"- 叙事亮点：{'；'.join(highlights)}")
+            else:
+                lines.append(f"- 叙事亮点：{highlights}")
+            lines.append(f"- 观众门槛：{digest.get('观众门槛', '-')}")
+            # 话题性是嵌套结构
+            topic = digest.get('话题性', {})
+            if isinstance(topic, dict):
+                lines.append(
+                    f"- 话题性：{topic.get('总评', '-')} — "
+                    f"大众认知度：{topic.get('大众认知度', '-')}；"
+                    f"降维切口：{topic.get('降维切口', '-')}；"
+                    f"惊奇密度：{topic.get('惊奇密度', '-')}"
+                )
+            else:
+                lines.append(f"- 话题性：{topic}")
+            # 潜在弱点可能是数组
+            weaknesses = digest.get('潜在弱点', [])
+            if isinstance(weaknesses, list):
+                lines.append(f"- 潜在弱点：{'；'.join(weaknesses)}")
+            else:
+                lines.append(f"- 潜在弱点：{weaknesses}")
+            lines.append(f"- 时效关联：{digest.get('时效关联', '-')}")
+        else:
+            lines.append("- （无文稿摘要）")
+        lines.append("")
+
+    return "\n".join(lines)
+
+
+@router.post("/diagnosis-report")
+def generate_diagnosis_report(
+    req: DiagnosisRequest,
+    session: Session = Depends(get_session),
+    current_user=Depends(_require_read()),
+):
+    """生成 AI 诊断报告。同一范围缓存结果，force=True 时重新生成。"""
+    cache_key = f"{req.year}_{req.ep_start}_{req.ep_end}"
+
+    # 检查缓存
+    if not req.force and cache_key in _report_cache:
+        return _report_cache[cache_key]
+
+    # 1. 查询所选范围的 episodes
+    ep_stmt = (
+        select(Episode)
+        .where(extract("year", Episode.air_date) == req.year)
+        .where(Episode.episode_number >= req.ep_start)
+        .where(Episode.episode_number <= req.ep_end)
+        .where(Episode.audience_share.is_not(None))
+        .order_by(Episode.episode_number.asc())
+    )
+    episodes = session.exec(ep_stmt).all()
+
+    if not episodes:
+        return {"error": "所选范围内没有收视数据"}
+
+    # 2. 查年度目标
+    target_stmt = select(YearlyTarget).where(YearlyTarget.year == req.year)
+    target = session.exec(target_stmt).first()
+    if not target:
+        return {"error": f"{req.year}年没有设置年度目标"}
+
+    base_target = float(target.base_target)
+    stretch_target = float(target.stretch_target)
+
+    # 3. 计算统计数据
+    shares = [float(ep.audience_share) for ep in episodes]
+    avg_share = round(sum(shares) / len(shares), 4) if shares else 0
+    pass_count = sum(1 for s in shares if s >= base_target)
+    max_ep = max(episodes, key=lambda e: float(e.audience_share))
+    min_ep = min(episodes, key=lambda e: float(e.audience_share))
+
+    # 三档判定
+    if avg_share >= stretch_target:
+        tier = "excellent"
+    elif avg_share >= base_target:
+        tier = "on_target"
+    else:
+        tier = "danger"
+
+    # 4. 组装 user message
+    user_message = _build_user_message(episodes, base_target, stretch_target, avg_share, pass_count, max_ep, min_ep)
+
+    # 5. 读 system prompt
+    system_prompt = _PROMPT5_PATH.read_text(encoding="utf-8")
+
+    # 6. 调 DeepSeek
+    if not settings.DEEPSEEK_API_KEY:
+        return {"error": "DEEPSEEK_API_KEY 未配置"}
+
+    client = OpenAI(
+        api_key=settings.DEEPSEEK_API_KEY,
+        base_url="https://api.deepseek.com",
+    )
+    response = client.chat.completions.create(
+        model="deepseek-chat",
+        messages=[
+            {"role": "system", "content": system_prompt},
+            {"role": "user", "content": user_message},
+        ],
+        temperature=0.3,
+    )
+    report_markdown = response.choices[0].message.content
+
+    # 7. 组装返回
+    result = {
+        "tier": tier,
+        "avg_share": avg_share,
+        "episode_count": len(episodes),
+        "pass_count": pass_count,
+        "highest": {
+            "ep": max_ep.episode_number,
+            "name": max_ep.program_name,
+            "share": float(max_ep.audience_share),
+        },
+        "lowest": {
+            "ep": min_ep.episode_number,
+            "name": min_ep.program_name,
+            "share": float(min_ep.audience_share),
+        },
+        "report_markdown": report_markdown,
+        "generated_at": datetime.now().isoformat(),
+        "model": "deepseek-chat",
+        "disclaimer": "本报告基于已入库的收视数据、节目标签及内容摘要生成，未纳入同时段竞品、社会热点等外部因素。分析结论难免挂一漏万，仅供栏目内部讨论参考，不构成节目决策依据。",
+    }
+
+    # 8. 缓存
+    _report_cache[cache_key] = result
+    return result
@@ -20,6 +20,9 @@ _SESSION_MAX_AGE = int(os.environ.get("SESSION_MAX_AGE", "86400"))
 _MINIMAX_EMBED_API_KEY = os.environ.get("MINIMAX_EMBED_API_KEY", "")
 _MINIMAX_GROUP_ID = os.environ.get("MINIMAX_GROUP_ID", "")

+# DeepSeek API 凭证（诊断报告用）
+_DEEPSEEK_API_KEY = os.environ.get("DEEPSEEK_API_KEY", "")
+
 # 验证必需配置
 if not _DATABASE_URL:
    raise RuntimeError(f"[config] DATABASE_URL 未设置。请检查 {_env_path} 是否存在且内容正确。")
@@ -31,6 +34,7 @@ class Settings:
    SESSION_MAX_AGE: int = _SESSION_MAX_AGE
    MINIMAX_EMBED_API_KEY: str = _MINIMAX_EMBED_API_KEY
    MINIMAX_GROUP_ID: str = _MINIMAX_GROUP_ID
+    DEEPSEEK_API_KEY: str = _DEEPSEEK_API_KEY


 settings = Settings()
@@ -11,4 +11,5 @@ python-dotenv==1.0.1
 httpx==0.27.0
 python-frontmatter==1.1.0
 pandas>=2.0.0
-openpyxl>=3.1.0
+openpyxl>=3.1.0
+openai>=1.0.0,<1.55.0