{ "metadata": { "id": "ch18", "title": "第18章：内容创作 Agent", "volume": "vol5", "volume_title": "专项篇", "word_count": 2326, "difficulty": "intermediate", "prerequisites": [ "ch05", "ch06" ], "key_concepts": [ "概述：当 AI 学会\"写作\"", "内容创作 Agent 的能力光谱", "内容创作 Agent 的核心挑战", "文章写作与编辑引擎", "文章结构化管理器", "内容风格控制器", "营销文案生成", "多渠道文案生成器", "多语言翻译与本地化", "翻译引擎", "SEO 优化模块", "智能关键词分析与布局", "内容审核与合规", "多层审核管道", "完整集成：内容创作 Agent" ], "learning_objectives": [], "estimated_tokens": 1396, "source_file": "vol5/ch18_内容创作Agent.md" }, "overview": "", "sections": [ { "id": "18.1", "title": "18.1 概述：当 AI 学会\"写作\"", "level": 2, "content": "内容创作是 Agent 技术在创意领域的深度应用。从营销文案到技术文档，从社交媒体帖子到长篇报告，内容创作 Agent 正在重新定义\"写作\"的边界。", "subsections": [ { "id": "18.1.1", "title": "18.1.1 内容创作 Agent 的能力光谱", "content": "- **文章写作与编辑**：长文生成、段落重组、风格调整、多版本迭代\n- **营销文案生成**：广告语、产品描述、社交媒体帖子、邮件营销\n- **多语言翻译与本地化**：不仅是翻译，更是文化适配\n- **SEO 优化**：关键词布局、标题优化、Meta 描述生成\n- **内容审核与合规**：敏感词检测、版权检查、品牌一致性" }, { "id": "18.1.2", "title": "18.1.2 内容创作 Agent 的核心挑战", "content": "| 挑战 | 说明 | 解决方案 |\n|------|------|---------|\n| 品牌一致性 | 输出内容需要符合品牌调性 | 品牌风格指南 + Few-Shot 示例 |\n| 事实准确性 | AI 可能产生幻觉 | RAG 检索增强 + 事实核查链 |\n| 多语言质量 | 不同语言的表达习惯差异大 | 本地化知识库 + 母语者审校 |\n| 合规风险 | 广告法、隐私法、平台规则 | 多层审核管道 + 敏感词库 |\n| 内容多样性 | 避免千篇一律的\"AI味\" | Temperature 控制 + 风格注入 |" } ] }, { "id": "18.2", "title": "18.2 文章写作与编辑引擎", "level": 2, "content": "", "subsections": [ { "id": "18.2.1", "title": "18.2.1 文章结构化管理器", "content": "高质量的文章需要清晰的结构。文章结构化管理器负责大纲生成、段落组织和逻辑流控制。\n\n{{\n \"title\": \"文章标题\",\n \"subtitle\": \"副标题（可选）\",\n \"target_audience\": \"目标读者\",\n \"sections\": [\n {{\n \"title\": \"章节标题\",\n \"points\": [\"要点1\", \"要点2\", \"要点3\"],\n \"estimated_words\": 400\n }}\n ]\n}}" }, { "id": "18.2.2", "title": "18.2.2 内容风格控制器", "content": "" } ] }, { "id": "18.3", "title": "18.3 营销文案生成", "level": 2, "content": "", "subsections": [ { "id": "18.3.1", "title": "18.3.1 多渠道文案生成器", "content": "营销文案需要针对不同渠道（微信、抖音、小红书、邮件等）调整格式和风格。\n\n{{\n \"version\": 1,\n \"headline\": \"标题/首句\",\n \"body\": \"正文内容\",\n \"cta\": \"行动号召语\",\n \"hashtags\": [\"标签1\", \"标签2\"],\n \"word_count\": 实际字数,\n \"tone\": \"该版本的语气描述\"\n}}" } ] }, { "id": "18.4", "title": "18.4 多语言翻译与本地化", "level": 2, "content": "", "subsections": [ { "id": "18.4.1", "title": "18.4.1 翻译引擎", "content": "{{\n \"translated_text\": \"翻译后的文本\",\n \"adaptations\": [\"所做的本地化调整说明\"],\n \"glossary_used\": [\"使用的术语表条目\"]\n}}" } ] }, { "id": "18.5", "title": "18.5 SEO 优化模块", "level": 2, "content": "", "subsections": [ { "id": "18.5.1", "title": "18.5.1 智能关键词分析与布局", "content": "{{\n \"optimized_title\": \"优化后的标题（含核心关键词，60字以内）\",\n \"meta_description\": \"Meta 描述（含核心关键词，150字以内）\",\n \"h1_suggestion\": \"H1 标签建议\",\n \"h2_suggestions\": [\"H2 子标题建议1\", \"H2 子标题建议2\"],\n \"content_improvements\": [\"内容改进建议1\", \"内容改进建议2\"],\n \"internal_link_suggestions\": [\"内链建议1\"],\n \"alt_text_suggestions\": {{\"image_1\": \"图片 alt 文本建议\"}},\n \"schema_suggestions\": [\"结构化数据建议\"]\n}}" } ] }, { "id": "18.6", "title": "18.6 内容审核与合规", "level": 2, "content": "", "subsections": [ { "id": "18.6.1", "title": "18.6.1 多层审核管道", "content": "" } ] }, { "id": "18.7", "title": "18.7 完整集成：内容创作 Agent", "level": 2, "content": "", "subsections": [ { "id": "18.7.1", "title": "18.7.1 统一工作流", "content": "" }, { "id": "18.7.2", "title": "18.7.2 端到端使用示例", "content": "" } ] }, { "id": "18.8", "title": "18.8 生产级考量", "level": 2, "content": "", "subsections": [ { "id": "18.8.1", "title": "18.8.1 内容质量保证", "content": "| 环节 | 策略 | 说明 |\n|------|------|------|\n| 事实核查 | RAG + 知识库检索 | 对关键声明进行事实验证 |\n| 原创性检测 | 指纹哈希 + 相似度计算 | 防止内容重复发布 |\n| A/B 测试 | 多版本生成 + 自动投放 | 数据驱动的内容优化 |\n| 反馈循环 | 用户评分 → Prompt 优化 | 持续提升生成质量 |" }, { "id": "18.8.2", "title": "18.8.2 多租户内容管理", "content": "" } ] }, { "id": "本章小结", "title": "本章小结", "level": 2, "content": "本章从文章写作引擎出发，构建了完整的内容创作 Agent 体系：\n\n1. **文章写作与编辑**：结构化大纲生成、逐节写作、多版本迭代修改\n2. **品牌风格控制**：品牌指南定义、风格合规检查、自动修复\n3. **多渠道营销文案**：渠道适配、多版本生成、统一活动管理\n4. **多语言翻译与本地化**：术语表驱动、文化适配、格式本地化\n5. **SEO 优化**：关键词分析、Meta 标签生成、内容优化建议\n6. **内容审核与合规**：多层审核管道、广告法检测、品牌一致性\n\n核心思想：**内容创作的核心不是\"生成文字\"，而是\"管理约束\"**——在品牌调性、法律法规、SEO 要求和用户偏好之间找到最佳平衡点。", "subsections": [] } ], "code_blocks": [ { "id": "code-1", "language": "text", "description": "内容创作是 Agent 技术在创意领域的深度应用。从营销文案到技术文档，从社交媒体帖子到长篇报告，内容创作 Agent 正在重新定义\"写作\"的边界。", "code": "补全 → 改写 → 摘要 → 翻译 → 生成 → 多语言适配 → SEO 优化 → 合规审核\n │ │ │ │ │ │ │ │\n续写 语气 提炼 语言 从零 本地化文化 关键词布局 敏感词检测\n 调整 核心 转换 创作 适配调整 搜索友好 政策合规", "section_ref": "18.1.1", "runnable": false, "dependencies": [] }, { "id": "code-2", "language": "python", "description": "高质量的文章需要清晰的结构。文章结构化管理器负责大纲生成、段落组织和逻辑流控制。", "code": "\"\"\"\n内容创作 Agent - 文章写作与编辑引擎\n\"\"\"\n\nfrom dataclasses import dataclass, field\nfrom enum import Enum\nfrom typing import Optional\nimport json\nimport re\nimport hashlib\nimport time\n\n\nclass ContentType(Enum):\n BLOG = \"blog\" # 博客文章\n TECH_DOC = \"tech_doc\" # 技术文档\n MARKETING = \"marketing\" # 营销文案\n SOCIAL = \"social\" # 社交媒体\n EMAIL = \"email\" # 邮件\n PRODUCT_DESC = \"product\" # 产品描述\n PRESS_RELEASE = \"press\" # 新闻稿\n WHITEPAPER = \"whitepaper\" # 白皮书\n\n\nclass ToneStyle(Enum):\n PROFESSIONAL = \"professional\" # 专业严谨\n CASUAL = \"casual\" # 轻松活泼\n HUMOROUS = \"humorous\" # 幽默风趣\n ACADEMIC = \"academic\" # 学术规范\n PERSUASIVE = \"persuasive\" # 说服性\n STORYTELLING = \"storytelling\" # 故事性\n\n\n@dataclass\nclass ArticleOutline:\n \"\"\"文章大纲\"\"\"\n title: str\n subtitle: str = \"\"\n sections: list[dict] = field(default_factory=list)\n target_audience: str = \"通用读者\"\n estimated_words: int = 0\n keywords: list[str] = field(default_factory=list)\n\n def to_prompt_context(self) -> str:\n \"\"\"转化为 Prompt 上下文\"\"\"\n lines = [f\"# 文章标题: {self.title}\"]\n if self.subtitle:\n lines.append(f\"副标题: {self.subtitle}\")\n lines.append(f\"目标读者: {self.target_audience}\")\n lines.append(f\"目标字数: {self.estimated_words}\")\n lines.append(f\"核心关键词: {', '.join(self.keywords)}\")\n lines.append(\"\\n## 大纲结构\\n\")\n for i, sec in enumerate(self.sections, 1):\n lines.append(f\"{i}. {sec['title']}\")\n lines.append(f\" - 核心要点: {', '.join(sec.get('points', []))}\")\n if sec.get(\"estimated_words\"):\n lines.append(f\" - 目标字数: {sec['estimated_words']}\")\n return \"\\n\".join(lines)\n\n\nclass ArticleGenerator:\n \"\"\"文章生成器\"\"\"\n\n def __init__(self, llm_client=None, brand_guide: Optional[dict] = None):\n self.llm_client = llm_client\n self.brand_guide = brand_guide or {}\n self.version_history: list[dict] = []\n\n def _call_llm(self, prompt: str, temperature: float = 0.7) -> str:\n \"\"\"调用 LLM\"\"\"\n if self.llm_client:\n return self.llm_client.chat(prompt, temperature=temperature)\n return self._mock_response(prompt)\n\n def _mock_response(self, prompt: str) -> str:\n \"\"\"模拟 LLM 返回\"\"\"\n return \"这是模拟生成的文章内容。实际使用时请替换为真实 LLM API 调用。\"\n\n def generate_outline(self, topic: str, content_type: ContentType,\n keywords: list[str], sections_count: int = 5,\n target_words: int = 2000) -> ArticleOutline:\n \"\"\"生成文章大纲\"\"\"\n prompt = f\"\"\"你是一位资深的{content_type.value}写作者。请为以下主题生成一份详细的文章大纲。\n\n主题: {topic}\n内容类型: {content_type.value}\n核心关键词: {', '.join(keywords)}\n目标字数: {target_words}\n章节数量: {sections_count}\n\n请以 JSON 格式返回:", "section_ref": "18.2.1", "runnable": true, "dependencies": [] }, { "id": "code-3", "language": "text", "description": "}}", "code": "\n要求:\n1. 标题要吸引眼球且包含核心关键词\n2. 章节之间有逻辑递进关系\n3. 每个章节的要点要具体、可展开\n4. 总字数分配要合理\"\"\"\n\n response = self._call_llm(prompt, temperature=0.8)\n\n try:\n json_match = re.search(r'```json\\s*(.*?)\\s*```', response, re.DOTALL)\n data = json.loads(json_match.group(1)) if json_match else json.loads(response)\n except (json.JSONDecodeError, AttributeError):\n data = {\n \"title\": topic,\n \"subtitle\": \"\",\n \"target_audience\": \"通用读者\",\n \"sections\": [\n {\"title\": f\"第{i}部分\", \"points\": [\"待展开\"],\n \"estimated_words\": target_words // sections_count}\n for i in range(1, sections_count + 1)\n ]\n }\n\n # 计算总预估字数\n total = sum(s.get(\"estimated_words\", 0) for s in data[\"sections\"])\n\n return ArticleOutline(\n title=data[\"title\"],\n subtitle=data.get(\"subtitle\", \"\"),\n sections=data[\"sections\"],\n target_audience=data.get(\"target_audience\", \"通用读者\"),\n estimated_words=total,\n keywords=keywords\n )\n\n def write_section(self, outline: ArticleOutline, section_index: int,\n context: str = \"\") -> str:\n \"\"\"根据大纲写某个章节\"\"\"\n if section_index >= len(outline.sections):\n return \"\"\n\n section = outline.sections[section_index]\n prev_summary = \"\"\n if section_index > 0 and self.version_history:\n prev = self.version_history[-1].get(\"content\", \"\")[:500]\n prev_summary = f\"\\n\\n上一节内容摘要: {prev}...\"\n\n brand_context = \"\"\n if self.brand_guide:\n brand_context = f\"\\n\\n品牌风格指南:\\n- 语气: {self.brand_guide.get('tone', '专业')}\\n\"\n brand_context += f\"- 禁用词: {', '.join(self.brand_guide.get('forbidden_words', []))}\\n\"\n brand_context += f\"- 品牌术语: {json.dumps(self.brand_guide.get('terminology', {}), ensure_ascii=False)}\"\n\n prompt = f\"\"\"你是一位专业写作者。请根据大纲撰写以下章节。\n\n## 全文大纲\n\n{outline.to_prompt_context()}\n\n## 当前要写的章节\n\n第 {section_index + 1} 节: {section['title']}\n核心要点: {', '.join(section.get('points', []))}\n目标字数: {section.get('estimated_words', 500)}\n{brand_context}\n{prev_summary}\n{f\"## 额外上下文\\n{context}\" if context else \"\"}\n\n要求:\n1. 严格围绕核心要点展开，不要跑题\n2. 段落之间要有自然的过渡\n3. 如果有品牌术语，请正确使用\n4. 保持全文的语气和风格一致\n5. 避免使用品牌禁用词\"\"\"\n\n content = self._call_llm(prompt, temperature=0.7)\n\n self.version_history.append({\n \"section_index\": section_index,\n \"section_title\": section[\"title\"],\n \"content\": content,\n \"timestamp\": time.time()\n })\n\n return content\n\n def write_full_article(self, outline: ArticleOutline,\n context_per_section: Optional[dict] = None) -> str:\n \"\"\"生成完整文章\"\"\"\n sections = []\n context_per_section = context_per_section or {}\n\n for i in range(len(outline.sections)):\n ctx = context_per_section.get(i, \"\")\n section_content = self.write_section(outline, i, ctx)\n sections.append(f\"## {outline.sections[i]['title']}\\n\\n{section_content}\")\n\n return f\"# {outline.title}\\n\\n\" + \"\\n\\n\".join(sections)\n\n def revise_section(self, section_index: int, instruction: str) -> str:\n \"\"\"根据指令修改某个章节\"\"\"\n if not self.version_history or section_index >= len(self.version_history):\n return \"\"\n\n original = self.version_history[section_index][\"content\"]\n\n prompt = f\"\"\"请根据修改指令，对以下文章段落进行修改。\n\n## 原始内容\n\n{original}\n\n## 修改指令\n\n{instruction}\n\n要求:\n1. 只修改与指令相关的部分，保持其余内容不变\n2. 保持原有的语气和风格\n3. 直接输出修改后的完整内容，不要解释修改了什么\"\"\"\n\n revised = self._call_llm(prompt, temperature=0.5)\n\n self.version_history[section_index][\"content\"] = revised\n self.version_history[section_index][\"revisions\"] = (\n self.version_history[section_index].get(\"revisions\", []) + 1\n )\n\n return revised", "section_ref": "18.2.1", "runnable": false, "dependencies": [] }, { "id": "code-4", "language": "python", "description": "", "code": "\"\"\"\n内容风格控制器 - 确保输出符合品牌调性\n\"\"\"\n\n\n@dataclass\nclass BrandGuide:\n \"\"\"品牌风格指南\"\"\"\n name: str\n tone: str # 语气描述\n vocabulary_level: str # 词汇难度: formal / casual / mixed\n forbidden_words: list[str] # 禁用词列表\n preferred_words: dict # 术语偏好: {\"AI\": \"人工智能\", \"app\": \"应用程序\"}\n sentence_style: str # 句式风格: short / mixed / complex\n emoji_usage: str # Emoji 使用: allowed / limited / forbidden\n perspective: str # 人称: first / third / second\n examples: list[str] # 风格示例文本\n\n\nclass StyleController:\n \"\"\"风格控制器\"\"\"\n\n def __init__(self, guide: BrandGuide):\n self.guide = guide\n\n def build_style_prompt(self) -> str:\n \"\"\"构建风格 Prompt\"\"\"\n g = self.guide\n lines = [\n \"## 写作风格规范\",\n f\"- 品牌名称: {g.name}\",\n f\"- 语气: {g.tone}\",\n f\"- 词汇级别: {g.vocabulary_level}\",\n f\"- 句式风格: {g.sentence_style}\",\n f\"- 人称视角: {g.perspective}\",\n f\"- Emoji: {g.emoji_usage}\",\n ]\n\n if g.forbidden_words:\n lines.append(f\"- 禁用词: {', '.join(g.forbidden_words)}\")\n if g.preferred_words:\n pairs = [f'\"{k}\" → \"{v}\"' for k, v in g.preferred_words.items()]\n lines.append(f\"- 术语偏好: {', '.join(pairs)}\")\n\n if g.examples:\n lines.append(\"\\n### 风格示例\\n\")\n for i, ex in enumerate(g.examples, 1):\n lines.append(f\"示例 {i}:\\n{ex}\\n\")\n\n return \"\\n\".join(lines)\n\n def check_compliance(self, text: str) -> dict:\n \"\"\"检查文本是否符合风格规范\"\"\"\n issues = []\n\n # 检查禁用词\n text_lower = text.lower()\n for word in self.guide.forbidden_words:\n if word.lower() in text_lower:\n # 找到上下文\n idx = text_lower.find(word.lower())\n start = max(0, idx - 20)\n end = min(len(text), idx + len(word) + 20)\n context = text[start:end]\n issues.append({\n \"type\": \"forbidden_word\",\n \"word\": word,\n \"context\": context,\n \"severity\": \"high\",\n \"suggestion\": f\"请替换'{word}'为更合适的表达\"\n })\n\n # 检查术语一致性\n for preferred, standard in self.guide.preferred_words.items():\n if preferred.lower() in text_lower:\n count = text_lower.count(preferred.lower())\n issues.append({\n \"type\": \"terminology\",\n \"word\": preferred,\n \"suggestion\": f\"应使用'{standard}'替代'{preferred}'，共 {count} 处\",\n \"severity\": \"medium\"\n })\n\n # 检查句子长度\n sentences = re.split(r'[。！？]', text)\n long_sentences = [s for s in sentences if len(s) > 80]\n if self.guide.sentence_style == \"short\" and long_sentences:\n issues.append({\n \"type\": \"sentence_length\",\n \"detail\": f\"发现 {len(long_sentences)} 个过长的句子（>80字），建议拆分\",\n \"severity\": \"low\"\n })\n\n # 检查 Emoji 使用\n emoji_pattern = re.compile(\n \"[\\U0001F600-\\U0001F64F\\U0001F300-\\U0001F5FF\"\n \"\\U0001F680-\\U0001F6FF\\U0001F700-\\U0001F77F\"\n \"\\U0001F780-\\U0001F7FF\\U0001F800-\\U0001F8FF\"\n \"\\U0001F900-\\U0001F9FF\\U0001FA00-\\U0001FA6F\"\n \"\\U0001FA70-\\U0001FAFF\\U00002702-\\U000027B0]\")\n emojis = emoji_pattern.findall(text)\n if self.guide.emoji_usage == \"forbidden\" and emojis:\n issues.append({\n \"type\": \"emoji\",\n \"detail\": f\"发现 {len(emojis)} 个 Emoji，当前风格不允许使用\",\n \"severity\": \"high\"\n })\n elif self.guide.emoji_usage == \"limited\" and len(emojis) > 3:\n issues.append({\n \"type\": \"emoji\",\n \"detail\": f\"Emoji 数量 {len(emojis)} 超过限制（最多3个）\",\n \"severity\": \"medium\"\n })\n\n return {\n \"compliant\": len([i for i in issues if i[\"severity\"] == \"high\"]) == 0,\n \"issues\": issues,\n \"score\": max(0, 100 - len(issues) * 10)\n }\n\n def auto_fix(self, text: str) -> str:\n \"\"\"自动修复简单的风格问题\"\"\"\n result = text\n for old, new in self.guide.preferred_words.items():\n result = result.replace(old, new)\n return result", "section_ref": "18.2.2", "runnable": true, "dependencies": [] }, { "id": "code-5", "language": "python", "description": "营销文案需要针对不同渠道（微信、抖音、小红书、邮件等）调整格式和风格。", "code": "\"\"\"\n营销文案生成器 - 多渠道适配\n\"\"\"\n\n@dataclass\nclass CampaignBrief:\n \"\"\"营销活动简报\"\"\"\n product_name: str\n product_description: str\n target_audience: str\n key_benefits: list[str]\n promotion_details: str\n cta: str # Call-to-Action\n channels: list[str] # 目标渠道\n constraints: list[str] = field(default_factory=list)\n\n\nclass MarketingCopyGenerator:\n \"\"\"多渠道营销文案生成器\"\"\"\n\n CHANNEL_TEMPLATES = {\n \"wechat\": {\n \"name\": \"微信公众号\",\n \"max_length\": 5000,\n \"style\": \"深度内容、情感共鸣\",\n \"format\": \"标题 + 导语 + 正文分段 + CTA\"\n },\n \"xiaohongshu\": {\n \"name\": \"小红书\",\n \"max_length\": 1000,\n \"style\": \"种草、真实体验、emoji 丰富\",\n \"format\": \"吸睛标题 + 痛点描述 + 使用体验 + Emoji + 标签\"\n },\n \"douyin\": {\n \"name\": \"抖音\",\n \"max_length\": 200,\n \"style\": \"口语化、节奏快、有悬念\",\n \"format\": \"3秒钩子 + 核心卖点 + CTA\"\n },\n \"email\": {\n \"name\": \"邮件营销\",\n \"max_length\": 3000,\n \"style\": \"专业、个性化、有紧迫感\",\n \"format\": \"主题行 + 个性化称呼 + 价值主张 + CTA\"\n },\n \"weibo\": {\n \"name\": \"微博\",\n \"max_length\": 500,\n \"style\": \"话题性强、可互动\",\n \"format\": \"话题标签 + 核心信息 + 互动引导\"\n }\n }\n\n def __init__(self, llm_client=None):\n self.llm_client = llm_client\n\n def _call_llm(self, prompt: str, temperature: float = 0.8) -> str:\n if self.llm_client:\n return self.llm_client.chat(prompt, temperature=temperature)\n return \"模拟生成的营销文案内容。\"\n\n def generate(self, brief: CampaignBrief, channel: str,\n variations: int = 3) -> list[dict]:\n \"\"\"为指定渠道生成多个版本的文案\"\"\"\n template = self.CHANNEL_TEMPLATES.get(channel)\n if not template:\n return [{\"error\": f\"不支持的渠道: {channel}\"}]\n\n prompt = f\"\"\"你是一位资深营销文案专家。请为以下产品生成 {variations} 个不同版本的{template['name']}文案。\n\n## 产品信息\n- 产品名称: {brief.product_name}\n- 产品描述: {brief.product_description}\n- 目标受众: {brief.target_audience}\n- 核心卖点: {', '.join(brief.key_benefits)}\n- 促销信息: {brief.promotion_details}\n- 行动号召: {brief.cta}\n\n## 渠道要求\n- 渠道: {template['name']}\n- 风格: {template['style']}\n- 格式: {template['format']}\n- 最大字数: {template['max_length']}\n\n## 限制条件\n{chr(10).join('- ' + c for c in brief.constraints) if brief.constraints else '- 无特殊限制'}\n\n## 输出要求\n\n请为每个版本生成以下 JSON 格式:", "section_ref": "18.3.1", "runnable": true, "dependencies": [] }, { "id": "code-6", "language": "text", "description": "}}", "code": "\n请生成 {variations} 个版本，每个版本风格略有不同（如理性/感性/幽默等）。\"\"\"\n\n response = self._call_llm(prompt)\n\n # 解析多个 JSON 对象\n versions = []\n try:\n json_matches = re.findall(r'```json\\s*(.*?)\\s*```', response, re.DOTALL)\n if not json_matches:\n # 尝试直接解析\n json_matches = [response]\n\n for match in json_matches:\n try:\n data = json.loads(match)\n if isinstance(data, list):\n versions.extend(data)\n else:\n versions.append(data)\n except json.JSONDecodeError:\n continue\n except Exception:\n pass\n\n # 如果解析失败，返回模拟结果\n if not versions:\n for i in range(variations):\n versions.append({\n \"version\": i + 1,\n \"headline\": f\"{brief.product_name} - 版本{i+1}\",\n \"body\": f\"模拟生成的{template['name']}文案版本{i+1}\",\n \"cta\": brief.cta,\n \"hashtags\": [brief.product_name],\n \"word_count\": 100,\n \"tone\": \"专业\"\n })\n\n return versions\n\n def generate_full_campaign(self, brief: CampaignBrief) -> dict:\n \"\"\"为所有目标渠道生成完整营销素材\"\"\"\n campaign = {\"brief\": brief, \"materials\": {}}\n\n for channel in brief.channels:\n versions = self.generate(brief, channel)\n campaign[\"materials\"][channel] = versions\n\n # 生成统一的品牌关键词\n campaign[\"brand_keywords\"] = (\n [brief.product_name] + brief.key_benefits[:3])\n\n return campaign", "section_ref": "18.3.1", "runnable": false, "dependencies": [] }, { "id": "code-7", "language": "python", "description": "", "code": "\"\"\"\n多语言翻译与本地化引擎\n支持术语表、风格适配和文化本地化\n\"\"\"\n\nfrom dataclasses import dataclass, field\n\n\n@dataclass\nclass Glossary:\n \"\"\"术语表\"\"\"\n language: str\n entries: dict[str, str] = field(default_factory=dict) # 源语言 -> 目标语言\n\n def to_prompt(self) -> str:\n lines = [f\"## {self.language} 术语表\\n\"]\n for src, tgt in self.entries.items():\n lines.append(f\"- {src} → {tgt}\")\n return \"\\n\".join(lines)\n\n\n@dataclass\nclass LocaleProfile:\n \"\"\"地区文化配置\"\"\"\n locale: str # 如 zh-CN, en-US, ja-JP\n date_format: str = \"YYYY-MM-DD\"\n number_format: str = \"1,234.56\"\n currency_symbol: str = \"¥\"\n reading_direction: str = \"ltr\" # ltr 或 rtl\n formality_level: str = \"formal\" # formal / casual\n cultural_notes: list[str] = field(default_factory=list)\n\n\nclass TranslationEngine:\n \"\"\"翻译与本地化引擎\"\"\"\n\n SUPPORTED_LOCALES = {\n \"zh-CN\": LocaleProfile(\"zh-CN\", \"YYYY年MM月DD日\", \"1,234.56\", \"¥\", \"ltr\", \"formal\",\n [\"使用敬语时注意等级\", \"避免直接否定，使用委婉表达\"]),\n \"en-US\": LocaleProfile(\"en-US\", \"MM/DD/YYYY\", \"1,234.56\", \"$\", \"ltr\", \"casual\",\n [\"直接表达，避免过于委婉\", \"使用美式英语拼写\"]),\n \"ja-JP\": LocaleProfile(\"ja-JP\", \"YYYY年MM月DD日\", \"1,234\", \"¥\", \"ltr\", \"very_formal\",\n [\"使用适当的敬语体系\", \"长句中保持主语明确\"]),\n \"ko-KR\": LocaleProfile(\"ko-KR\", \"YYYY.MM.DD\", \"1,234\", \"₩\", \"ltr\", \"formal\",\n [\"注意敬语和非敬语的使用场景\"]),\n \"de-DE\": LocaleProfile(\"de-DE\", \"DD.MM.YYYY\", \"1.234,56\", \"€\", \"ltr\", \"formal\",\n [\"德语名词首字母大写\", \"复合词使用正确\"]),\n \"ar-SA\": LocaleProfile(\"ar-SA\", \"DD/MM/YYYY\", \"١٬٢٣٤٫٥٦\", \"﷼\", \"rtl\", \"formal\",\n [\"从右到左书写\", \"避免使用猪相关的比喻\"]),\n }\n\n def __init__(self, llm_client=None, glossaries: Optional[dict[str, Glossary]] = None):\n self.llm_client = llm_client\n self.glossaries = glossaries or {}\n\n def _call_llm(self, prompt: str, temperature: float = 0.3) -> str:\n if self.llm_client:\n return self.llm_client.chat(prompt, temperature=temperature)\n return \"模拟翻译结果。\"\n\n def translate(self, text: str, source_locale: str,\n target_locale: str, context: str = \"\") -> dict:\n \"\"\"翻译文本\"\"\"\n source_profile = self.SUPPORTED_LOCALES.get(source_locale)\n target_profile = self.SUPPORTED_LOCALES.get(target_locale)\n\n if not target_profile:\n return {\"success\": False, \"error\": f\"不支持的目标语言: {target_locale}\"}\n\n glossary_text = \"\"\n if target_locale in self.glossaries:\n glossary_text = self.glossaries[target_locale].to_prompt()\n\n cultural_notes = \"\"\n if target_profile.cultural_notes:\n cultural_notes = \"\\n## 文化注意事项\\n\" + \"\\n\".join(\n f\"- {n}\" for n in target_profile.cultural_notes)\n\n prompt = f\"\"\"你是一位专业的翻译专家。请将以下文本从 {source_locale} 翻译为 {target_locale}。\n\n## 源文本\n\n{text}\n\n## 翻译要求\n\n1. 保持原文的语义和语气\n2. 使用 {target_profile.formality_level} 语体\n3. 确保表达符合目标语言的习惯用法\n4. 保留专业术语的准确性\n{f\"5. 额外上下文: {context}\" if context else \"\"}\n{glossary_text}\n{cultural_notes}\n\n请以 JSON 格式返回:", "section_ref": "18.4.1", "runnable": true, "dependencies": [] }, { "id": "code-8", "language": "\"\"\"", "description": "}}", "code": "\n response = self._call_llm(prompt, temperature=0.3)\n\n try:\n json_match = re.search(r'```json\\s*(.*?)\\s*```', response, re.DOTALL)\n result = json.loads(json_match.group(1)) if json_match else json.loads(response)\n result[\"success\"] = True\n return result\n except (json.JSONDecodeError, AttributeError):\n return {\"success\": False, \"translated_text\": response}\n\n def batch_translate(self, segments: list[dict], source_locale: str,\n target_locale: str) -> list[dict]:\n \"\"\"批量翻译（保持术语一致性）\"\"\"\n results = []\n for seg in segments:\n r = self.translate(\n seg[\"text\"], source_locale, target_locale,\n context=seg.get(\"context\", \"\"))\n results.append({\n \"source\": seg[\"text\"],\n **r,\n \"segment_id\": seg.get(\"id\")\n })\n return results\n\n def localize_content(self, content: dict, target_locale: str) -> dict:\n \"\"\"本地化完整内容（包括日期、数字、货币等格式）\"\"\"\n profile = self.SUPPORTED_LOCALES.get(target_locale)\n if not profile:\n return {\"success\": False, \"error\": f\"不支持: {target_locale}\"}\n\n localized = {\"locale\": target_locale}\n\n # 翻译文本字段\n for key, value in content.items():\n if isinstance(value, str) and len(value) > 10:\n result = self.translate(value, \"zh-CN\", target_locale)\n localized[key] = result.get(\"translated_text\", value)\n elif isinstance(value, (int, float)):\n # 数字格式化\n if isinstance(value, float):\n localized[key] = f\"{value:,.2f}\"\n else:\n localized[key] = f\"{value:,}\"\n else:\n localized[key] = value\n\n return localized", "section_ref": "18.4.1", "runnable": false, "dependencies": [] }, { "id": "code-9", "language": "python", "description": "", "code": "\"\"\"\nSEO 优化模块 - 关键词分析、内容优化和排名建议\n\"\"\"\n\n@dataclass\nclass SEOKeyword:\n \"\"\"SEO 关键词\"\"\"\n keyword: str\n search_volume: int # 月搜索量\n difficulty: float # 竞争难度 0-100\n current_rank: Optional[int] = None\n relevance: float = 0.0 # 与内容的关联度 0-1\n\n\nclass SEOOptimizer:\n \"\"\"SEO 优化器\"\"\"\n\n def __init__(self, llm_client=None):\n self.llm_client = llm_client\n\n def _call_llm(self, prompt: str, temperature: float = 0.5) -> str:\n if self.llm_client:\n return self.llm_client.chat(prompt, temperature=temperature)\n return '{\"keywords\": [], \"title\": \"\", \"meta_description\": \"\"}'\n\n def analyze_keywords(self, content: str,\n target_keywords: list[str]) -> dict:\n \"\"\"分析内容的关键词使用情况\"\"\"\n content_lower = content.lower()\n total_words = len(content)\n\n analysis = {}\n for kw in target_keywords:\n kw_lower = kw.lower()\n count = content_lower.count(kw_lower)\n density = (count * len(kw.split())) / max(total_words, 1) * 100\n\n # 检查关键词位置\n in_title = kw_lower in content_lower[:100]\n in_first_para = kw_lower in content_lower[:500]\n\n analysis[kw] = {\n \"count\": count,\n \"density\": round(density, 2),\n \"in_title\": in_title,\n \"in_first_paragraph\": in_first_para,\n \"status\": self._keyword_status(density, in_title)\n }\n\n return analysis\n\n def _keyword_status(self, density: float, in_title: bool) -> str:\n \"\"\"判断关键词使用状态\"\"\"\n issues = []\n if density < 0.5:\n issues.append(\"密度偏低\")\n elif density > 3.0:\n issues.append(\"密度过高（可能被视为关键词堆砌）\")\n if not in_title:\n issues.append(\"未出现在标题中\")\n return \"良好\" if not issues else \" | \".join(issues)\n\n def optimize_content(self, title: str, content: str,\n keywords: list[str]) -> dict:\n \"\"\"SEO 内容优化建议\"\"\"\n analysis = self.analyze_keywords(content, keywords)\n\n prompt = f\"\"\"你是一位 SEO 专家。请分析以下内容并提供优化建议。\n\n## 当前内容\n标题: {title}\n正文（前1000字）: {content[:1000]}\n\n## 目标关键词\n{', '.join(keywords)}\n\n## 关键词使用分析\n{json.dumps(analysis, ensure_ascii=False, indent=2)}\n\n请以 JSON 格式返回:", "section_ref": "18.5.1", "runnable": true, "dependencies": [] }, { "id": "code-10", "language": "\"\"\"", "description": "}}", "code": "\n response = self._call_llm(prompt)\n\n try:\n json_match = re.search(r'```json\\s*(.*?)\\s*```', response, re.DOTALL)\n return json.loads(json_match.group(1)) if json_match else json.loads(response)\n except (json.JSONDecodeError, AttributeError):\n return {\"optimized_title\": title, \"meta_description\": content[:150]}\n\n def generate_meta_tags(self, title: str, content: str,\n keywords: list[str]) -> dict:\n \"\"\"生成 SEO Meta 标签\"\"\"\n # 标题优化（含核心关键词，控制在60字内）\n primary_kw = keywords[0] if keywords else \"\"\n opt_title = f\"{title} | {primary_kw}\" if primary_kw and primary_kw not in title else title\n if len(opt_title) > 60:\n opt_title = opt_title[:57] + \"...\"\n\n # Meta 描述（含核心关键词，控制在150字内）\n desc = content[:300].replace(\"\\n\", \" \").strip()\n # 尝试在描述中自然插入关键词\n if primary_kw and primary_kw not in desc:\n desc = f\"{primary_kw}相关的{desc}\"\n meta_desc = desc[:147] + \"...\" if len(desc) > 150 else desc\n\n return {\n \"title\": opt_title,\n \"description\": meta_desc,\n \"keywords\": \", \".join(keywords),\n \"og_title\": opt_title,\n \"og_description\": meta_desc,\n \"twitter_card\": \"summary_large_image\",\n \"canonical_suggestion\": \"建议设置 canonical URL 防止重复内容\"\n }", "section_ref": "18.5.1", "runnable": false, "dependencies": [] }, { "id": "code-11", "language": "python", "description": "", "code": "\"\"\"\n内容审核与合规模块\n多层审核：敏感词 → 合规规则 → 品牌一致性\n\"\"\"\n\n@dataclass\nclass ComplianceRule:\n \"\"\"合规规则\"\"\"\n name: str\n description: str\n pattern: str # 正则表达式或关键词\n severity: str # \"error\" / \"warning\" / \"info\"\n suggestion: str\n\n\n@dataclass\nclass AuditResult:\n \"\"\"审核结果\"\"\"\n passed: bool\n score: float # 0-100\n issues: list[dict] = field(default_factory=list)\n summary: str = \"\"\n\n\nclass ContentAuditor:\n \"\"\"内容审核器\"\"\"\n\n # 中国广告法敏感词\n AD_LAW_FORBIDDEN = [\n \"最\", \"第一\", \"首个\", \"最好\", \"最大\", \"最高级\",\n \"极品\", \"顶级\", \"绝对\", \"独家\", \"首家\", \"首选\",\n \"国家级\", \"世界级\", \"全网最低价\", \"销量冠军\"\n ]\n\n # 通用敏感词模式\n SENSITIVE_PATTERNS = [\n {\"pattern\": r'包?(治|好|消|退)', \"severity\": \"warning\",\n \"suggestion\": \"避免使用保证性用语，建议改为'有助于'\"},\n {\"pattern\": r'[\\d]+%\\s*(保证|承诺)', \"severity\": \"warning\",\n \"suggestion\": \"避免使用百分比承诺\"},\n ]\n\n def __init__(self, brand_guide: Optional[BrandGuide] = None):\n self.brand_guide = brand_guide\n self.compliance_rules: list[ComplianceRule] = []\n\n def add_rule(self, rule: ComplianceRule):\n \"\"\"添加自定义合规规则\"\"\"\n self.compliance_rules.append(rule)\n\n def audit(self, text: str) -> AuditResult:\n \"\"\"执行完整的内容审核\"\"\"\n issues = []\n\n # 层1: 敏感词检测\n issues.extend(self._check_sensitive_words(text))\n\n # 层2: 正则模式检测\n issues.extend(self._check_patterns(text))\n\n # 层3: 自定义合规规则\n issues.extend(self._check_compliance_rules(text))\n\n # 层4: 品牌一致性检查\n if self.brand_guide:\n style_result = StyleController(self.brand_guide).check_compliance(text)\n for issue in style_result[\"issues\"]:\n issues.append({\n \"layer\": \"brand_consistency\",\n **issue\n })\n\n # 计算分数\n error_count = sum(1 for i in issues if i.get(\"severity\") == \"error\")\n warning_count = sum(1 for i in issues if i.get(\"severity\") == \"warning\")\n score = max(0, 100 - error_count * 20 - warning_count * 5)\n\n return AuditResult(\n passed=error_count == 0,\n score=score,\n issues=issues,\n summary=self._generate_summary(issues, score)\n )\n\n def _check_sensitive_words(self, text: str) -> list[dict]:\n \"\"\"检测敏感词\"\"\"\n issues = []\n text_lower = text.lower()\n\n for word in self.AD_LAW_FORBIDDEN:\n if word in text_lower:\n # 排除合理的上下文用法\n idx = text_lower.find(word)\n start = max(0, idx - 10)\n end = min(len(text), idx + len(word) + 10)\n context = text[start:end]\n\n issues.append({\n \"layer\": \"sensitive_word\",\n \"type\": \"ad_law_violation\",\n \"word\": word,\n \"context\": context,\n \"severity\": \"error\",\n \"suggestion\": f\"根据《广告法》第九条，禁止使用'{word}'等绝对化用语\"\n })\n\n return issues\n\n def _check_patterns(self, text: str) -> list[dict]:\n \"\"\"正则模式检测\"\"\"\n issues = []\n for p in self.SENSITIVE_PATTERNS:\n matches = re.findall(p[\"pattern\"], text)\n if matches:\n issues.append({\n \"layer\": \"pattern_match\",\n \"type\": \"risky_expression\",\n \"matches\": matches,\n \"severity\": p[\"severity\"],\n \"suggestion\": p[\"suggestion\"]\n })\n return issues\n\n def _check_compliance_rules(self, text: str) -> list[dict]:\n \"\"\"自定义合规规则检测\"\"\"\n issues = []\n for rule in self.compliance_rules:\n if re.search(rule.pattern, text):\n issues.append({\n \"layer\": \"custom_rule\",\n \"rule_name\": rule.name,\n \"severity\": rule.severity,\n \"suggestion\": rule.suggestion\n })\n return issues\n\n def _generate_summary(self, issues: list[dict], score: float) -> str:\n \"\"\"生成审核摘要\"\"\"\n errors = [i for i in issues if i.get(\"severity\") == \"error\"]\n warnings = [i for i in issues if i.get(\"severity\") == \"warning\"]\n\n summary = f\"审核完成。评分: {score:.0f}/100。\"\n if errors:\n summary += f\"发现 {len(errors)} 个严重问题（必须修复）。\"\n if warnings:\n summary += f\"发现 {len(warnings)} 个警告（建议修复）。\"\n if not issues:\n summary += \"内容合规，未发现问题。\"\n return summary", "section_ref": "18.6.1", "runnable": true, "dependencies": [] }, { "id": "code-12", "language": "python", "description": "", "code": "\"\"\"\n内容创作 Agent - 完整集成版\n\"\"\"\n\nclass ContentCreationAgent:\n \"\"\"\n 内容创作 Agent 核心入口\n\n 使用示例:\n agent = ContentCreationAgent(brand_guide=my_brand)\n article = agent.create_article(\"AI在教育领域的应用\", ContentType.BLOG, [\"AI\", \"教育\"])\n \"\"\"\n\n def __init__(self, llm_client=None, brand_guide: Optional[BrandGuide] = None):\n self.llm_client = llm_client\n self.brand_guide = brand_guide\n self.article_gen = ArticleGenerator(llm_client, brand_guide)\n self.style_ctrl = StyleController(brand_guide) if brand_guide else None\n self.seo = SEOOptimizer(llm_client)\n self.auditor = ContentAuditor(brand_guide)\n\n def create_article(self, topic: str, content_type: ContentType,\n keywords: list[str], target_words: int = 2000,\n tone: ToneStyle = ToneStyle.PROFESSIONAL,\n sections: int = 5) -> dict:\n \"\"\"\n 完整的文章创作工作流:\n 大纲生成 → 逐节写作 → SEO优化 → 风格检查 → 合规审核\n \"\"\"\n result = {\"topic\": topic, \"steps\": []}\n\n # Step 1: 生成大纲\n result[\"steps\"].append(\"生成文章大纲\")\n outline = self.article_gen.generate_outline(\n topic, content_type, keywords, sections, target_words)\n result[\"outline\"] = outline\n\n # Step 2: 逐节写作\n result[\"steps\"].append(\"逐节撰写内容\")\n full_text = self.article_gen.write_full_article(outline)\n result[\"draft\"] = full_text\n\n # Step 3: SEO 优化\n result[\"steps\"].append(\"SEO 优化分析\")\n seo_analysis = self.seo.analyze_keywords(full_text, keywords)\n meta_tags = self.seo.generate_meta_tags(outline.title, full_text, keywords)\n result[\"seo\"] = {\"keyword_analysis\": seo_analysis, \"meta_tags\": meta_tags}\n\n # Step 4: 风格检查\n if self.style_ctrl:\n result[\"steps\"].append(\"品牌风格检查\")\n style_result = self.style_ctrl.check_compliance(full_text)\n result[\"style\"] = style_result\n\n # Step 5: 合规审核\n result[\"steps\"].append(\"内容合规审核\")\n audit_result = self.auditor.audit(full_text)\n result[\"audit\"] = {\n \"passed\": audit_result.passed,\n \"score\": audit_result.score,\n \"issues_count\": len(audit_result.issues),\n \"summary\": audit_result.summary\n }\n\n # Step 6: 自动修复风格问题\n if self.style_ctrl:\n full_text = self.style_ctrl.auto_fix(full_text)\n\n result[\"final_content\"] = full_text\n result[\"word_count\"] = len(full_text)\n\n return result\n\n def create_marketing_campaign(self, brief: CampaignBrief) -> dict:\n \"\"\"创建多渠道营销活动素材\"\"\"\n generator = MarketingCopyGenerator(self.llm_client)\n campaign = generator.generate_full_campaign(brief)\n\n # 对每个渠道的每个版本进行审核\n for channel, versions in campaign[\"materials\"].items():\n for version in versions:\n body = version.get(\"body\", \"\")\n if body:\n audit = self.auditor.audit(body)\n version[\"audit\"] = {\n \"passed\": audit.passed,\n \"score\": audit.score,\n \"issue_count\": len(audit.issues)\n }\n\n return campaign\n\n def translate_and_localize(self, content: str, source: str,\n target: str) -> dict:\n \"\"\"翻译并本地化内容\"\"\"\n engine = TranslationEngine(self.llm_client)\n result = engine.translate(content, source, target)\n return result", "section_ref": "18.7.1", "runnable": true, "dependencies": [] }, { "id": "code-13", "language": "python", "description": "", "code": "\"\"\"\n端到端使用示例\n\"\"\"\n\nif __name__ == \"__main__\":\n # 创建品牌指南\n brand = BrandGuide(\n name=\"TechFlow\",\n tone=\"专业但不生硬，用简单语言解释复杂概念\",\n vocabulary_level=\"mixed\",\n forbidden_words=[\"颠覆性\", \"革命性\", \"一站式\", \"赋能\"],\n preferred_words={\"AI\": \"人工智能\", \"app\": \"应用\", \"bug\": \"缺陷\"},\n sentence_style=\"mixed\",\n emoji_usage=\"limited\",\n perspective=\"first_plural\", # \"我们\"\n examples=[\n \"我们相信技术的力量在于让复杂变简单。\",\n \"在这篇文章中，我们将一起探索...\"\n ]\n )\n\n # 创建 Agent\n agent = ContentCreationAgent(brand_guide=brand)\n\n # 示例 1: 创建博客文章\n print(\"=== 创建博客文章 ===\")\n result = agent.create_article(\n topic=\"人工智能如何改变教育行业\",\n content_type=ContentType.BLOG,\n keywords=[\"人工智能\", \"教育\", \"在线学习\", \"个性化教学\"],\n target_words=2000,\n tone=ToneStyle.PROFESSIONAL,\n sections=4\n )\n print(f\"大纲标题: {result['outline'].title}\")\n print(f\"字数: {result['word_count']}\")\n print(f\"审核评分: {result['audit']['score']}/100\")\n print(f\"审核通过: {'是' if result['audit']['passed'] else '否'}\")\n\n # 示例 2: 营销活动\n print(\"\\n=== 创建营销活动 ===\")\n brief = CampaignBrief(\n product_name=\"TechFlow 智能学习平台\",\n product_description=\"基于人工智能的个性化在线学习平台\",\n target_audience=\"25-40岁的职场人士\",\n key_benefits=[\"AI个性化推荐\", \"碎片化学习\", \"实时反馈\"],\n promotion_details=\"新用户注册即享7天免费体验\",\n cta=\"立即注册，开启智能学习之旅\",\n channels=[\"wechat\", \"xiaohongshu\", \"email\"]\n )\n campaign = agent.create_marketing_campaign(brief)\n for channel, materials in campaign[\"materials\"].items():\n print(f\"\\n{channel}: {len(materials)} 个版本\")\n\n # 示例 3: 合规审核\n print(\"\\n=== 内容合规审核 ===\")\n test_text = \"我们提供全网最好的AI教育产品，保证100%提升成绩，效果绝对显著！\"\n audit = agent.auditor.audit(test_text)\n print(f\"评分: {audit.score}/100\")\n print(f\"通过: {audit.passed}\")\n for issue in audit.issues:\n print(f\" [{issue['severity']}] {issue.get('word', issue.get('type', ''))}: \"\n f\"{issue.get('suggestion', '')}\")", "section_ref": "18.7.2", "runnable": true, "dependencies": [] }, { "id": "code-14", "language": "python", "description": "| 反馈循环 | 用户评分 → Prompt 优化 | 持续提升生成质量 |", "code": "\"\"\"\n多租户内容管理 - 品牌隔离与内容追踪\n\"\"\"\n\nclass TenantContentManager:\n \"\"\"租户内容管理器\"\"\"\n\n def __init__(self):\n self.tenants: dict[str, dict] = {}\n\n def register_tenant(self, tenant_id: str, brand_guide: BrandGuide):\n \"\"\"注册租户并绑定品牌指南\"\"\"\n self.tenants[tenant_id] = {\n \"brand_guide\": brand_guide,\n \"content_history\": [],\n \"usage_stats\": {\"total_words\": 0, \"total_articles\": 0}\n }\n\n def track_content(self, tenant_id: str, content_id: str,\n content: str, metadata: dict = None):\n \"\"\"记录内容产出\"\"\"\n if tenant_id not in self.tenants:\n return\n self.tenants[tenant_id][\"content_history\"].append({\n \"content_id\": content_id,\n \"word_count\": len(content),\n \"timestamp\": time.time(),\n \"metadata\": metadata or {}\n })\n self.tenants[tenant_id][\"usage_stats\"][\"total_words\"] += len(content)\n self.tenants[tenant_id][\"usage_stats\"][\"total_articles\"] += 1\n\n def get_usage_report(self, tenant_id: str) -> dict:\n \"\"\"获取租户使用报告\"\"\"\n if tenant_id not in self.tenants:\n return {\"error\": \"租户不存在\"}\n stats = self.tenants[tenant_id][\"usage_stats\"]\n history = self.tenants[tenant_id][\"content_history\"]\n return {\n \"total_articles\": stats[\"total_articles\"],\n \"total_words\": stats[\"total_words\"],\n \"average_words\": (stats[\"total_words\"] / max(stats[\"total_articles\"], 1)),\n \"recent_articles\": history[-5:] if history else []\n }", "section_ref": "18.8.2", "runnable": true, "dependencies": [] } ], "tables": [ { "headers": [ "挑战", "说明", "解决方案" ], "data": [ [ "品牌一致性", "输出内容需要符合品牌调性", "品牌风格指南 + Few-Shot 示例" ], [ "事实准确性", "AI 可能产生幻觉", "RAG 检索增强 + 事实核查链" ], [ "多语言质量", "不同语言的表达习惯差异大", "本地化知识库 + 母语者审校" ], [ "合规风险", "广告法、隐私法、平台规则", "多层审核管道 + 敏感词库" ], [ "内容多样性", "避免千篇一律的\"AI味\"", "Temperature 控制 + 风格注入" ] ] }, { "headers": [ "环节", "策略", "说明" ], "data": [ [ "事实核查", "RAG + 知识库检索", "对关键声明进行事实验证" ], [ "原创性检测", "指纹哈希 + 相似度计算", "防止内容重复发布" ], [ "A/B 测试", "多版本生成 + 自动投放", "数据驱动的内容优化" ], [ "反馈循环", "用户评分 → Prompt 优化", "持续提升生成质量" ] ] } ], "key_takeaways": [], "common_pitfalls": [], "related_chapters": [ "ch05", "ch06", "ch29" ] }