fix: resolve 10 code review issues

High priority: - Fix concurrent race condition for view_count/like_count (atomic update) - Add route request ID tracking to prevent race conditions - Filter get_joke by status=approved (no pending content leak) - Add error feedback for like button Performance: - Optimize random joke query (avoid full table sort) - Limit page_size max to 100 (DoS prevention) Medium: - Add localStorage quota error handling - Handle empty AI response gracefully - Fix generate content title extraction Low: - Add rejected_jokes to stats API - Update dashboard to show rejected count
2026-06-02 20:35:08 +08:00
parent 0b43973236
commit ceed63fcb0
144 changed files with 191660 additions and 270 deletions
@@ -0,0 +1,341 @@
+"""笑话优化核心逻辑：质量检测 → AI 润色 → 评价分类。"""
+
+import hashlib
+import json
+import sys
+import time
+
+import httpx
+from openai import OpenAI
+
+from optimizer.prompts import (
+    QUALITY_CHECK_SYSTEM_PROMPT,
+    QUALITY_CHECK_USER_PROMPT,
+    POLISH_SYSTEM_PROMPT,
+    POLISH_USER_PROMPT,
+    EVALUATE_SYSTEM_PROMPT,
+    EVALUATE_USER_PROMPT,
+)
+
+
+class Optimizer:
+    def __init__(self, api_base: str, username: str, password: str):
+        self.api_base = api_base.rstrip("/")
+        self.username = username
+        self.password = password
+        self.token = None
+        self.ai_client = None
+        self.model_name = ""
+        self.types = []
+        self.crowds = []
+        # 统计
+        self.stats = {"checked": 0, "rejected": 0, "polished": 0, "evaluated": 0, "skipped": 0}
+
+    # === API 认证 ===
+    def _login(self) -> str:
+        resp = httpx.post(
+            f"{self.api_base}/api/auth/login",
+            json={"username": self.username, "password": self.password},
+            timeout=30,
+        )
+        resp.raise_for_status()
+        return resp.json()["access_token"]
+
+    def _get(self, path: str) -> dict | list:
+        resp = httpx.get(
+            f"{self.api_base}{path}",
+            headers={"Authorization": f"Bearer {self.token}"},
+            timeout=30,
+        )
+        resp.raise_for_status()
+        return resp.json()
+
+    def _put(self, path: str, data: dict) -> dict:
+        resp = httpx.put(
+            f"{self.api_base}{path}",
+            json=data,
+            headers={"Authorization": f"Bearer {self.token}"},
+            timeout=30,
+        )
+        resp.raise_for_status()
+        return resp.json()
+
+    # === 初始化 ===
+    def setup(self):
+        print("[*] 正在登录...")
+        self.token = self._login()
+        print("[*] 登录成功")
+
+        ai_config = self._get("/api/admin/settings/active")
+        self.ai_client = OpenAI(
+            base_url=ai_config["api_base"],
+            api_key=ai_config["api_key"],
+        )
+        self.model_name = ai_config["model_name"]
+        print(f"[*] AI 模型: {self.model_name}")
+
+        self.types = self._get("/api/categories/types")
+        self.crowds = self._get("/api/categories/crowds")
+        print(f"[*] 分类: {len(self.types)} 种类型, {len(self.crowds)} 种人群")
+
+    # === 读取笑话 ===
+    def get_jokes(self, status: str | None = None, limit: int | None = None,
+                  ids: list[int] | None = None) -> list[dict]:
+        """从 API 分页读取笑话"""
+        if ids:
+            jokes = []
+            for jid in ids:
+                try:
+                    j = self._get(f"/api/admin/jokes/{jid}")
+                    jokes.append(j)
+                except Exception as e:
+                    print(f"  [!] 获取笑话 #{jid} 失败: {e}")
+            return jokes
+
+        page = 1
+        page_size = 100
+        all_jokes = []
+
+        while True:
+            try:
+                path = f"/api/admin/jokes?page={page}&page_size={page_size}"
+                if status:
+                    path += f"&status={status}"
+                data = self._get(path)
+                items = data.get("items", [])
+                if not items:
+                    break
+                all_jokes.extend(items)
+                print(f"  [*] 已读取 {len(all_jokes)} 条...")
+                if limit and len(all_jokes) >= limit:
+                    all_jokes = all_jokes[:limit]
+                    break
+                page += 1
+            except Exception as e:
+                print(f"  [!] 分页读取失败 (page={page}): {e}")
+                break
+
+        return all_jokes
+
+    # === Stage 1: 质量检测 ===
+    def quality_check(self, content: str) -> dict:
+        """判断笑话是否有笑点，返回 {"has_punchline": bool, "reason": str}"""
+        resp = self.ai_client.chat.completions.create(
+            model=self.model_name,
+            messages=[
+                {"role": "system", "content": QUALITY_CHECK_SYSTEM_PROMPT},
+                {"role": "user", "content": QUALITY_CHECK_USER_PROMPT.format(content=content[:2000])},
+            ],
+            temperature=0.3,
+            max_tokens=200,
+        )
+        raw = resp.choices[0].message.content.strip()
+        return self._parse_json(raw, {"has_punchline": True, "reason": ""})
+
+    # === Stage 2: AI 润色 ===
+    def polish(self, content: str) -> str:
+        """润色笑话内容"""
+        resp = self.ai_client.chat.completions.create(
+            model=self.model_name,
+            messages=[
+                {"role": "system", "content": POLISH_SYSTEM_PROMPT},
+                {"role": "user", "content": POLISH_USER_PROMPT.format(content=content)},
+            ],
+            temperature=0.8,
+            max_tokens=1024,
+        )
+        return resp.choices[0].message.content.strip()
+
+    # === Stage 3: 评价分类 ===
+    def evaluate(self, content: str) -> dict:
+        """评价并分类，返回 {"types": [...], "crowds": [...], "score": int, "comment": str}"""
+        type_names = [t.get("name", "") for t in self.types]
+        crowd_names = [c.get("name", "") for c in self.crowds]
+
+        resp = self.ai_client.chat.completions.create(
+            model=self.model_name,
+            messages=[
+                {"role": "system", "content": EVALUATE_SYSTEM_PROMPT},
+                {"role": "user", "content": EVALUATE_USER_PROMPT.format(
+                    content=content[:2000],
+                    known_types=", ".join(type_names),
+                    known_crowds=", ".join(crowd_names),
+                )},
+            ],
+            temperature=0.3,
+            max_tokens=300,
+        )
+        raw = resp.choices[0].message.content.strip()
+        result = self._parse_json(raw, {"types": [], "crowds": [], "score": 5, "comment": ""})
+        # Backward compatibility: if LLM returns old single format, convert to array
+        if isinstance(result.get("types"), str):
+            result["types"] = [result["types"]] if result["types"] else []
+        if isinstance(result.get("crowds"), str):
+            result["crowds"] = [result["crowds"]] if result["crowds"] else []
+        return result
+
+    # === 辅助方法 ===
+    def _parse_json(self, raw: str, default: dict) -> dict:
+        """安全解析 LLM 返回的 JSON"""
+        try:
+            data = json.loads(raw)
+            if isinstance(data, dict):
+                return data
+            return default
+        except json.JSONDecodeError:
+            if "```json" in raw:
+                raw = raw.split("```json")[1].split("```")[0]
+            elif "```" in raw:
+                raw = raw.split("```")[1].split("```")[0]
+            try:
+                return json.loads(raw.strip())
+            except Exception:
+                return default
+
+    def _get_type_id(self, type_name: str) -> int | None:
+        for t in self.types:
+            if t.get("name") == type_name:
+                return t.get("id")
+        return None
+
+    def _get_crowd_id(self, crowd_name: str) -> int | None:
+        for c in self.crowds:
+            if c.get("name") == crowd_name:
+                return c.get("id")
+        return None
+
+    # === 单条笑话处理（3 阶段） ===
+    def process_joke(self, joke: dict) -> bool:
+        """处理单条笑话：质量检测 → 润色 → 评价分类，返回是否成功"""
+        joke_id = joke.get("id")
+        title = joke.get("title", "")
+        content = joke.get("content", "")
+
+        if not content:
+            print(f"  [!] #{joke_id} 内容为空，跳过")
+            self.stats["skipped"] += 1
+            return False
+
+        print(f"\n  {'='*40}")
+        print(f"  处理 #{joke_id}: {title[:30]}")
+        print(f"  {'='*40}")
+
+        # Stage 1: 质量检测
+        print(f"  [1/3] 质量检测...")
+        try:
+            check = self.quality_check(content)
+            if not check.get("has_punchline", True):
+                reason = check.get("reason", "无笑点")
+                print(f"  [!] 无笑点: {reason}")
+                # 标记为 rejected
+                self._put(f"/api/admin/jokes/{joke_id}", {"status": "rejected"})
+                self.stats["rejected"] += 1
+                self.stats["checked"] += 1
+                return True  # 处理完成（已拒绝）
+            print(f"  [OK] 有笑点: {check.get('reason', '')}")
+        except Exception as e:
+            print(f"  [!] 质量检测失败: {e}，跳过本条")
+            self.stats["skipped"] += 1
+            return False
+
+        self.stats["checked"] += 1
+
+        # Stage 2: AI 润色
+        print(f"  [2/3] AI 润色...")
+        try:
+            polished = self.polish(content)
+            if polished and polished != content:
+                print(f"  [OK] 润色完成 ({len(content)} -> {len(polished)} 字)")
+            else:
+                print(f"  [*] 润色后无变化")
+        except Exception as e:
+            print(f"  [!] 润色失败: {e}")
+            polished = content  # 润色失败时使用原文
+
+        # Stage 3: 评价分类
+        print(f"  [3/3] 评价分类...")
+        try:
+            eval_result = self.evaluate(polished)
+            type_names = eval_result.get("types", [])
+            crowd_names = eval_result.get("crowds", [])
+            score = eval_result.get("score", 5)
+            print(f"  [OK] 类型={type_names}, 人群={crowd_names}, 评分={score}/10")
+        except Exception as e:
+            print(f"  [!] 评价分类失败: {e}")
+            type_names = []
+            crowd_names = []
+            score = None
+
+        # 保存更新
+        try:
+            update = {
+                "polished_content": polished,
+                "status": "approved" if (score or 5) >= 4 else "pending",
+            }
+            if type_names:
+                update["type_ids"] = []
+                for n in type_names:
+                    tid = self._get_type_id(n)
+                    if tid:
+                        update["type_ids"].append(tid)
+                if not update["type_ids"]:
+                    del update["type_ids"]
+            if crowd_names:
+                update["crowd_ids"] = []
+                for n in crowd_names:
+                    cid = self._get_crowd_id(n)
+                    if cid:
+                        update["crowd_ids"].append(cid)
+                if not update["crowd_ids"]:
+                    del update["crowd_ids"]
+
+            self._put(f"/api/admin/jokes/{joke_id}", update)
+            self.stats["polished"] += 1
+            self.stats["evaluated"] += 1
+            print(f"  [OK] 更新成功")
+            return True
+        except Exception as e:
+            print(f"  [!] 更新失败: {e}")
+            return False
+
+    # === 主循环 ===
+    def run(self, status: str | None = None, limit: int | None = None,
+            ids: list[int] | None = None):
+        """主入口：读取笑话并逐个处理"""
+        print(f"\n>> 笑话优化模式启动")
+        print(f"    筛选状态: {status or '全部'}")
+        if limit:
+            print(f"    处理数量: {limit}")
+        if ids:
+            print(f"    指定 ID: {ids}")
+
+        self.setup()
+
+        jokes = self.get_jokes(status, limit, ids)
+        print(f"\n[*] 共读取 {len(jokes)} 条笑话，开始处理")
+
+        for idx, joke in enumerate(jokes):
+            print(f"\n  --- 进度 {idx+1}/{len(jokes)} ---")
+            try:
+                self.process_joke(joke)
+            except KeyboardInterrupt:
+                print("\n用户中断")
+                break
+            except Exception as e:
+                print(f"  [!] 处理异常: {e}")
+                self.stats["skipped"] += 1
+
+            # 每条间稍等，避免 API 限流
+            if idx < len(jokes) - 1:
+                time.sleep(1)
+
+        # 输出统计
+        print(f"\n{'='*40}")
+        print(f"  处理完成")
+        print(f"  {'='*40}")
+        print(f"  检查: {self.stats['checked']} 条")
+        print(f"  拒绝(无笑点): {self.stats['rejected']} 条")
+        print(f"  润色: {self.stats['polished']} 条")
+        print(f"  评价分类: {self.stats['evaluated']} 条")
+        print(f"  跳过(失败): {self.stats['skipped']} 条")
+        print(f"{'='*40}")