add Article.is_platform_knowledge_base
This commit is contained in:
Fam Zheng
parent
bd2d2eb358
commit
5d93ac5a32
@ -0,0 +1,28 @@
|
||||
# Generated by Django 4.2.23 on 2025-08-23 20:46
|
||||
|
||||
from django.db import migrations, models
|
||||
|
||||
|
||||
class Migration(migrations.Migration):
|
||||
|
||||
dependencies = [
|
||||
('products', '0105_chatsession_chatmessage'),
|
||||
]
|
||||
|
||||
operations = [
|
||||
migrations.AddField(
|
||||
model_name='article',
|
||||
name='is_platform_knowledge_base',
|
||||
field=models.BooleanField(default=False, verbose_name='是否为平台知识库'),
|
||||
),
|
||||
migrations.AlterField(
|
||||
model_name='article',
|
||||
name='options',
|
||||
field=models.TextField(blank=True, default=''),
|
||||
),
|
||||
migrations.AlterField(
|
||||
model_name='article',
|
||||
name='url',
|
||||
field=models.TextField(blank=True, null=True),
|
||||
),
|
||||
]
|
||||
@ -138,8 +138,9 @@ class Article(models.Model):
|
||||
title = models.CharField(max_length=128, null=True, blank=True, verbose_name="标题")
|
||||
body = models.TextField()
|
||||
tenant = models.ForeignKey(Tenant, related_name="articles", on_delete=models.CASCADE, null=True, blank=True)
|
||||
options = models.TextField(default='')
|
||||
url = models.TextField(null=True)
|
||||
options = models.TextField(default='', blank=True)
|
||||
url = models.TextField(null=True, blank=True)
|
||||
is_platform_knowledge_base = models.BooleanField(default=False, verbose_name="是否为平台知识库")
|
||||
|
||||
class AssetFile(models.Model):
|
||||
filename = models.TextField(verbose_name="文件名")
|
||||
|
||||
786
doc/ai-chat.md
786
doc/ai-chat.md
@ -47,36 +47,6 @@
|
||||
|
||||
### 4.1 AI模型集成
|
||||
|
||||
#### 4.1.1 模型选择
|
||||
- **主要模型**: Kimi K2 API ✅ **已实现**
|
||||
- **备选模型**: 智谱AI (ChatGLM)、百度文心一言
|
||||
- **理由**: Kimi K2在中文理解、多模态处理和联网搜索方面表现优秀，API稳定可靠，特别适合品牌客服场景
|
||||
|
||||
#### 4.1.2 API集成架构 ✅ **已实现**
|
||||
```python
|
||||
# api/products/aichat.py
|
||||
class AIChatService:
|
||||
def __init__(self, api_key: str = None, base_url: str = None):
|
||||
self.api_key = api_key or getattr(settings, 'KIMI_API_KEY', None)
|
||||
self.base_url = base_url or getattr(settings, 'KIMI_API_URL', None)
|
||||
self.client = OpenAI(api_key=self.api_key, base_url=self.base_url)
|
||||
|
||||
def chat(self, user_message: str) -> str:
|
||||
# 调用Moonshot API，支持工具调用
|
||||
response = self.client.chat.completions.create(
|
||||
model=self.model,
|
||||
messages=self.conversation_history,
|
||||
tools=self.tools,
|
||||
tool_choice="auto"
|
||||
)
|
||||
return self._process_response(response)
|
||||
```
|
||||
|
||||
#### 4.1.3 工具调用系统 ✅ **已实现**
|
||||
- **工具定义**: `start_qr_scan` 二维码扫描工具
|
||||
- **工具执行**: 模拟5秒扫描过程，返回验证结果
|
||||
- **工具集成**: 完整的工具调用流程，支持多轮对话
|
||||
|
||||
#### 4.1.4 智能能力调度系统
|
||||
- **架构选择**: 直接以Tool形式接入Kimi K2，无需复杂路由
|
||||
- **能力分类**: 知识库检索工具、二维码扫描工具、一般问答
|
||||
@ -91,371 +61,13 @@ class AIChatService:
|
||||
- **更新机制**: 支持实时更新和版本控制
|
||||
- **成本优化**: 本地处理，减少API调用
|
||||
|
||||
#### 4.1.6 RAG架构设计
|
||||
```python
|
||||
class AIChatService:
|
||||
def __init__(self):
|
||||
self.client = OpenAI(api_key=KIMI_API_KEY, base_url=KIMI_API_URL)
|
||||
self.knowledge_base = LocalKnowledgeBase() # 本地知识库
|
||||
self.tools = self._define_tools()
|
||||
|
||||
def chat(self, message: str) -> str:
|
||||
# 步骤1：从本地知识库检索相关信息
|
||||
relevant_docs = self.knowledge_base.search(message, k=3)
|
||||
|
||||
# 步骤2：构建增强的上下文
|
||||
enhanced_context = self._build_enhanced_context(message, relevant_docs)
|
||||
|
||||
# 步骤3：使用Kimi K2生成回答（基于检索到的信息）
|
||||
response = self.client.chat.completions.create(
|
||||
model=self.model,
|
||||
messages=[
|
||||
{"role": "system", "content": self.system_prompt + enhanced_context},
|
||||
{"role": "user", "content": message}
|
||||
],
|
||||
tools=self.tools,
|
||||
tool_choice="auto"
|
||||
)
|
||||
|
||||
# 步骤4：处理工具调用（如果需要）
|
||||
if response.choices[0].message.tool_calls:
|
||||
# 执行工具调用逻辑...
|
||||
pass
|
||||
|
||||
# 步骤5：将问答对保存到知识库（可选）
|
||||
self.knowledge_base.add_qa_pair(message, response.choices[0].message.content)
|
||||
|
||||
return response.choices[0].message.content
|
||||
|
||||
def _build_enhanced_context(self, query: str, docs: List[Document]) -> str:
|
||||
"""构建增强的上下文信息"""
|
||||
if not docs:
|
||||
return ""
|
||||
|
||||
context = "\n\n相关参考信息：\n"
|
||||
for i, doc in enumerate(docs, 1):
|
||||
context += f"{i}. {doc.page_content}\n"
|
||||
|
||||
context += f"\n请基于以上信息回答用户问题：{query}"
|
||||
return context
|
||||
```
|
||||
|
||||
**RAG流程**：
|
||||
- **检索(Retrieval)**：从本地知识库找到相关信息
|
||||
- **增强(Augmentation)**：将检索到的信息作为上下文
|
||||
- **生成(Generation)**：Kimi K2基于增强的上下文生成回答
|
||||
|
||||
#### 4.1.7 本地知识库实现
|
||||
```python
|
||||
class LocalKnowledgeBase:
|
||||
def __init__(self):
|
||||
# 使用轻量级向量数据库
|
||||
self.vector_store = Chroma(
|
||||
embedding_function=OpenAIEmbeddings(api_key=KIMI_API_KEY),
|
||||
persist_directory="./knowledge_base"
|
||||
)
|
||||
self.file_storage = FileStorage() # 文件存储
|
||||
|
||||
def search(self, query: str, k: int = 3, threshold: float = 0.5) -> List[Document]:
|
||||
"""搜索本地知识库，返回相关文档列表"""
|
||||
# 语义搜索
|
||||
results = self.vector_store.similarity_search_with_score(query, k=k)
|
||||
|
||||
# 过滤低置信度结果，但保留更多候选
|
||||
filtered_results = [r for r in results if r[1] > threshold]
|
||||
|
||||
# 转换为Document对象列表
|
||||
documents = []
|
||||
for doc, score in filtered_results:
|
||||
documents.append(Document(
|
||||
page_content=doc.page_content,
|
||||
metadata={
|
||||
**doc.metadata,
|
||||
'similarity_score': score
|
||||
}
|
||||
))
|
||||
|
||||
return documents
|
||||
|
||||
def add_content(self, content: str, content_type: str, metadata: dict):
|
||||
"""添加内容到知识库"""
|
||||
# 文本内容直接向量化
|
||||
if content_type == 'text':
|
||||
self.vector_store.add_texts([content], [metadata])
|
||||
|
||||
# 图片内容使用OCR提取文本
|
||||
elif content_type == 'image':
|
||||
text = self._extract_text_from_image(content)
|
||||
self.vector_store.add_texts([text], [metadata])
|
||||
|
||||
# 视频内容提取关键帧和音频
|
||||
elif content_type == 'video':
|
||||
frames = self._extract_key_frames(content)
|
||||
audio_text = self._extract_audio_text(content)
|
||||
self.vector_store.add_texts([audio_text], [metadata])
|
||||
|
||||
def add_qa_pair(self, question: str, answer: str):
|
||||
"""添加问答对到知识库"""
|
||||
qa_text = f"问题：{question}\n答案：{answer}"
|
||||
metadata = {
|
||||
'type': 'qa_pair',
|
||||
'question': question,
|
||||
'answer': answer,
|
||||
'created_at': datetime.now().isoformat()
|
||||
}
|
||||
self.vector_store.add_texts([qa_text], [metadata])
|
||||
```
|
||||
|
||||
**技术特点**：
|
||||
- **轻量级**：使用Chroma向量数据库，无需复杂部署
|
||||
- **多模态**：支持文字、图片、视频内容
|
||||
- **实时更新**：支持动态添加和更新内容
|
||||
- **成本优化**：本地处理，减少API调用
|
||||
|
||||
#### 4.1.8 完整RAG实现示例
|
||||
```python
|
||||
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
||||
from langchain.embeddings import OpenAIEmbeddings
|
||||
from langchain.vectorstores import Chroma
|
||||
from langchain.document_loaders import TextLoader, ImageLoader, VideoLoader
|
||||
|
||||
class RAGService:
|
||||
def __init__(self):
|
||||
self.embeddings = OpenAIEmbeddings(api_key=KIMI_API_KEY)
|
||||
self.text_splitter = RecursiveCharacterTextSplitter(
|
||||
chunk_size=1000,
|
||||
chunk_overlap=200
|
||||
)
|
||||
self.vector_store = Chroma(
|
||||
embedding_function=self.embeddings,
|
||||
persist_directory="./knowledge_base"
|
||||
)
|
||||
|
||||
def add_document(self, file_path: str, content_type: str):
|
||||
"""添加文档到知识库"""
|
||||
if content_type == 'text':
|
||||
loader = TextLoader(file_path)
|
||||
elif content_type == 'image':
|
||||
loader = ImageLoader(file_path)
|
||||
elif content_type == 'video':
|
||||
loader = VideoLoader(file_path)
|
||||
else:
|
||||
raise ValueError(f"不支持的内容类型: {content_type}")
|
||||
|
||||
documents = loader.load()
|
||||
# 分块处理
|
||||
chunks = self.text_splitter.split_documents(documents)
|
||||
# 添加到向量数据库
|
||||
self.vector_store.add_documents(chunks)
|
||||
|
||||
def retrieve_context(self, query: str, k: int = 3) -> str:
|
||||
"""检索相关上下文"""
|
||||
docs = self.vector_store.similarity_search(query, k=k)
|
||||
context = "\n\n".join([doc.page_content for doc in docs])
|
||||
return context
|
||||
|
||||
def generate_answer(self, query: str, context: str) -> str:
|
||||
"""使用Kimi K2生成回答"""
|
||||
enhanced_prompt = f"""
|
||||
基于以下参考信息回答用户问题：
|
||||
|
||||
参考信息：
|
||||
{context}
|
||||
|
||||
用户问题：{query}
|
||||
|
||||
请基于参考信息提供准确、详细的回答。如果参考信息不足以回答问题，请明确说明。
|
||||
"""
|
||||
|
||||
response = self.client.chat.completions.create(
|
||||
model="kimi-k2-0711-preview",
|
||||
messages=[
|
||||
{"role": "system", "content": "你是徵象防伪验证平台的AI助手，请基于提供的参考信息回答问题。"},
|
||||
{"role": "user", "content": enhanced_prompt}
|
||||
],
|
||||
temperature=0.7,
|
||||
max_tokens=2000
|
||||
)
|
||||
|
||||
return response.choices[0].message.content
|
||||
|
||||
# 使用示例
|
||||
rag_service = RAGService()
|
||||
|
||||
# 添加文档
|
||||
rag_service.add_document("产品说明.txt", "text")
|
||||
rag_service.add_document("防伪标识.jpg", "image")
|
||||
rag_service.add_document("操作演示.mp4", "video")
|
||||
|
||||
# 问答流程
|
||||
query = "如何验证产品真伪？"
|
||||
context = rag_service.retrieve_context(query)
|
||||
answer = rag_service.generate_answer(query, context)
|
||||
```
|
||||
|
||||
**RAG优势**：
|
||||
- **信息可控**：检索到的信息来自平台知识库
|
||||
- **回答质量**：Kimi K2基于准确信息生成高质量回答
|
||||
- **成本优化**：避免重复的API调用，知识库一次构建多次使用
|
||||
- **专业性强**：平台专业知识 + AI生成能力
|
||||
|
||||
#### 4.1.9 架构选择说明
|
||||
|
||||
**为什么选择直接以Tool形式接入Kimi K2？**
|
||||
|
||||
1. **符合Kimi K2设计理念**
|
||||
- Kimi K2本身就是为工具调用设计的
|
||||
- 原生支持Function Calling，无需额外路由层
|
||||
- AI能根据上下文智能选择合适的工具
|
||||
|
||||
2. **实现简单，维护成本低**
|
||||
- 代码逻辑清晰，易于理解和维护
|
||||
- 避免复杂的if-else路由规则
|
||||
- 减少系统复杂度，降低出错概率
|
||||
|
||||
3. **足够灵活，扩展性好**
|
||||
- 可以轻松添加新的工具
|
||||
- 工具优先级通过描述和系统提示词控制
|
||||
- 支持复杂的工具组合调用
|
||||
|
||||
4. **性能优势**
|
||||
- 减少额外的路由判断逻辑
|
||||
- 直接利用Kimi K2的智能判断能力
|
||||
- 避免不必要的中间层处理
|
||||
|
||||
**具体实现方式**：
|
||||
```python
|
||||
def _define_tools(self):
|
||||
return [
|
||||
{
|
||||
"type": "function",
|
||||
"function": {
|
||||
"name": "search_knowledge_base",
|
||||
"description": "搜索徵象平台知识库，获取产品信息、防伪验证方法、技术文档等相关内容。当用户询问平台相关问题时，优先使用此工具。",
|
||||
"parameters": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"query": {
|
||||
"type": "string",
|
||||
"description": "搜索查询，支持中文关键词"
|
||||
},
|
||||
"content_type": {
|
||||
"type": "string",
|
||||
"enum": ["all", "text", "image", "video"],
|
||||
"description": "内容类型过滤"
|
||||
}
|
||||
},
|
||||
"required": ["query"]
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"type": "function",
|
||||
"function": {
|
||||
"name": "start_qr_scan",
|
||||
"description": "启动二维码扫描功能，用于产品防伪验证。当用户需要验证产品真伪时使用。",
|
||||
"parameters": {
|
||||
"type": "object",
|
||||
"properties": {},
|
||||
"required": []
|
||||
}
|
||||
}
|
||||
}
|
||||
]
|
||||
```
|
||||
|
||||
**系统提示词优化**：
|
||||
```python
|
||||
system_prompt = """你是一个专业的徵象防伪验证平台AI客服助手。
|
||||
|
||||
使用策略：
|
||||
1. 当用户询问平台相关问题时，优先使用search_knowledge_base工具搜索知识库
|
||||
2. 当用户需要验证产品真伪时，使用start_qr_scan工具启动扫描
|
||||
3. 基于检索到的信息，结合你的知识，提供准确、详细的回答
|
||||
4. 如果知识库信息不足，可以结合你的通用知识回答，但要明确说明信息来源
|
||||
|
||||
工具优先级：知识库搜索 > 二维码扫描 > 一般问答
|
||||
"""
|
||||
```
|
||||
|
||||
### 4.2 数据模型扩展
|
||||
|
||||
#### 4.2.1 新增数据表
|
||||
```python
|
||||
# api/products/models.py
|
||||
|
||||
class AIConfig(models.Model):
|
||||
"""AI配置表"""
|
||||
tenant = models.ForeignKey(Tenant, on_delete=models.CASCADE)
|
||||
model_provider = models.CharField(max_length=50, default='kimi')
|
||||
api_key = models.CharField(max_length=255)
|
||||
model_name = models.CharField(max_length=100, default='kimi-k2')
|
||||
|
||||
class BrandKnowledge(models.Model):
|
||||
"""品牌知识库"""
|
||||
tenant = models.ForeignKey(Tenant, on_delete=models.CASCADE)
|
||||
title = models.CharField(max_length=200)
|
||||
content = models.TextField()
|
||||
content_type = models.CharField(max_length=20) # text, image, video, link
|
||||
file_url = models.URLField(null=True, blank=True)
|
||||
priority = models.IntegerField(default=0)
|
||||
created_at = models.DateTimeField(auto_now_add=True)
|
||||
|
||||
class PlatformKnowledge(models.Model):
|
||||
"""平台知识库"""
|
||||
title = models.CharField(max_length=200)
|
||||
content = models.TextField()
|
||||
content_type = models.CharField(max_length=20, choices=[
|
||||
('platform_intro', '平台介绍'),
|
||||
('service_guide', '服务指南'),
|
||||
('faq', '常见问题')
|
||||
])
|
||||
ai_summary = models.TextField(null=True, blank=True, verbose_name="AI摘要版本")
|
||||
ai_segments = models.JSONField(default=dict, verbose_name="AI分段内容")
|
||||
ai_keywords = models.JSONField(default=list, verbose_name="AI关键词")
|
||||
is_platform_content = models.BooleanField(default=True)
|
||||
ai_priority = models.IntegerField(default=0)
|
||||
created_at = models.DateTimeField(auto_now_add=True)
|
||||
|
||||
class ChatSession(models.Model):
|
||||
"""聊天会话"""
|
||||
session_id = models.CharField(max_length=100, unique=True)
|
||||
tenant = models.ForeignKey(Tenant, on_delete=models.CASCADE)
|
||||
user_openid = models.CharField(max_length=100, null=True)
|
||||
created_at = models.DateTimeField(auto_now_add=True)
|
||||
last_active = models.DateTimeField(auto_now=True)
|
||||
|
||||
class ChatMessage(models.Model):
|
||||
"""聊天消息"""
|
||||
session = models.ForeignKey(ChatSession, on_delete=models.CASCADE)
|
||||
role = models.CharField(max_length=20) # user, assistant, system
|
||||
content = models.TextField()
|
||||
content_type = models.CharField(max_length=20, default='text')
|
||||
metadata = models.JSONField(default=dict)
|
||||
created_at = models.DateTimeField(auto_now_add=True)
|
||||
```
|
||||
|
||||
#### 4.2.2 现有模型扩展
|
||||
```python
|
||||
# 扩展CodeBatch模型
|
||||
class CodeBatch(models.Model):
|
||||
# ... 现有字段 ...
|
||||
enable_ai_chat = models.BooleanField(default=False, verbose_name="启用AI聊天")
|
||||
ai_chat_config = models.JSONField(null=True, blank=True, verbose_name="AI聊天配置")
|
||||
|
||||
# 扩展Tenant模型
|
||||
class Tenant(models.Model):
|
||||
# ... 现有字段 ...
|
||||
ai_brand_name = models.CharField(max_length=100, null=True, blank=True, verbose_name="AI品牌名称")
|
||||
ai_welcome_message = models.TextField(null=True, blank=True, verbose_name="AI欢迎语")
|
||||
ai_theme_colors = models.JSONField(null=True, blank=True, verbose_name="AI主题色彩")
|
||||
|
||||
# TODO: 在MiniProgramContent模型中添加enable_ai_chat字段
|
||||
# class MiniProgramContent(models.Model):
|
||||
# # ... 现有字段 ...
|
||||
# enable_ai_chat = models.BooleanField(default=False, verbose_name="启用AI客服")
|
||||
```
|
||||
|
||||
### 4.3 后端API设计
|
||||
|
||||
#### 4.3.1 AI聊天API
|
||||
@ -507,397 +119,6 @@ class AIChatView(BaseView):
|
||||
'messages': [self._serialize_message(m) for m in messages]
|
||||
})
|
||||
|
||||
class BrandKnowledgeView(BaseView):
|
||||
name = 'brand-knowledge'
|
||||
auth_check = 'tenant'
|
||||
|
||||
def post(self, request):
|
||||
"""上传品牌知识"""
|
||||
title = request.data.get('title')
|
||||
content = request.data.get('content')
|
||||
content_type = request.data.get('content_type', 'text')
|
||||
file_url = request.data.get('file_url')
|
||||
|
||||
knowledge = BrandKnowledge.objects.create(
|
||||
tenant=request.tenant,
|
||||
title=title,
|
||||
content=content,
|
||||
content_type=content_type,
|
||||
file_url=file_url
|
||||
)
|
||||
|
||||
return JsonResponse({'id': knowledge.id})
|
||||
```
|
||||
|
||||
#### 4.3.2 配置管理API
|
||||
```python
|
||||
class AIConfigView(BaseView):
|
||||
name = 'ai-config'
|
||||
auth_check = 'tenant'
|
||||
|
||||
def get(self, request):
|
||||
"""获取AI配置"""
|
||||
config = AIConfig.objects.filter(tenant=request.tenant).first()
|
||||
if not config:
|
||||
config = AIConfig.objects.create(tenant=request.tenant)
|
||||
|
||||
return JsonResponse({
|
||||
'model_provider': config.model_provider,
|
||||
'model_name': config.model_name,
|
||||
'temperature': config.temperature,
|
||||
'max_tokens': config.max_tokens,
|
||||
'is_active': config.is_active
|
||||
})
|
||||
|
||||
def patch(self, request):
|
||||
"""更新AI配置"""
|
||||
config = AIConfig.objects.filter(tenant=request.tenant).first()
|
||||
if not config:
|
||||
config = AIConfig.objects.create(tenant=request.tenant)
|
||||
|
||||
for field in ['model_provider', 'model_name', 'temperature', 'max_tokens', 'is_active']:
|
||||
if field in request.data:
|
||||
setattr(config, field, request.data[field])
|
||||
|
||||
config.save()
|
||||
return JsonResponse({'status': 'success'})
|
||||
```
|
||||
|
||||
### 4.4 前端界面设计
|
||||
|
||||
#### 4.4.1 小程序AI聊天界面
|
||||
```javascript
|
||||
// scanner/pages/chat/chat.js
|
||||
Page({
|
||||
data: {
|
||||
messages: [],
|
||||
inputValue: '',
|
||||
sessionId: null,
|
||||
tenant: null,
|
||||
loading: false
|
||||
},
|
||||
|
||||
onLoad(options) {
|
||||
this.initChat(options);
|
||||
},
|
||||
|
||||
async initChat(options) {
|
||||
// 获取租户信息
|
||||
const tenant = await this.getTenantInfo(options.tenant);
|
||||
this.setData({ tenant });
|
||||
|
||||
// 创建或获取会话
|
||||
const sessionId = await this.createChatSession(tenant.id);
|
||||
this.setData({ sessionId });
|
||||
|
||||
// 显示欢迎消息
|
||||
this.showWelcomeMessage(tenant);
|
||||
|
||||
// 如果有验证结果，显示相关产品信息
|
||||
if (options.serial_code) {
|
||||
await this.showProductInfo(options.serial_code);
|
||||
}
|
||||
},
|
||||
|
||||
async sendMessage() {
|
||||
if (!this.data.inputValue.trim()) return;
|
||||
|
||||
const message = this.data.inputValue;
|
||||
this.addMessage('user', message);
|
||||
this.setData({ inputValue: '', loading: true });
|
||||
|
||||
try {
|
||||
const response = await this.callAIChat(message);
|
||||
this.addMessage('assistant', response);
|
||||
} catch (error) {
|
||||
this.addMessage('system', '抱歉，AI服务暂时不可用，请稍后再试。');
|
||||
} finally {
|
||||
this.setData({ loading: false });
|
||||
}
|
||||
},
|
||||
|
||||
async callAIChat(message) {
|
||||
const response = await wx.request({
|
||||
url: `${this.data.tenant.server_url}/api/v1/ai-chat/`,
|
||||
method: 'POST',
|
||||
data: {
|
||||
session_id: this.data.sessionId,
|
||||
message: message,
|
||||
context: this.getContext()
|
||||
},
|
||||
header: {
|
||||
'Authorization': `token ${this.data.tenant.token}`
|
||||
}
|
||||
});
|
||||
|
||||
return response.data.response;
|
||||
}
|
||||
});
|
||||
```
|
||||
|
||||
#### 4.4.2 Web管理端界面
|
||||
```vue
|
||||
<!-- web/src/views/ai-config.vue -->
|
||||
<template>
|
||||
<div class="ai-config">
|
||||
<CCard>
|
||||
<CCardHeader>
|
||||
<h4>AI聊天配置</h4>
|
||||
</CCardHeader>
|
||||
<CCardBody>
|
||||
<CForm>
|
||||
<CFormGroup label="AI模型提供商">
|
||||
<CSelect v-model="config.model_provider">
|
||||
<option value="kimi">Kimi K2</option>
|
||||
<option value="zhipu">智谱AI</option>
|
||||
<option value="baidu">百度文心一言</option>
|
||||
</CSelect>
|
||||
</CFormGroup>
|
||||
|
||||
<CFormGroup label="API密钥">
|
||||
<CInput v-model="config.api_key" type="password" />
|
||||
</CFormGroup>
|
||||
|
||||
<CFormGroup label="模型名称">
|
||||
<CInput v-model="config.model_name" />
|
||||
</CFormGroup>
|
||||
|
||||
<CFormGroup label="温度">
|
||||
<CInput v-model="config.temperature" type="number" step="0.1" min="0" max="2" />
|
||||
</CFormGroup>
|
||||
|
||||
<CFormGroup label="最大Token数">
|
||||
<CInput v-model="config.max_tokens" type="number" />
|
||||
</CFormGroup>
|
||||
|
||||
<CButton color="primary" @click="saveConfig">保存配置</CButton>
|
||||
</CForm>
|
||||
</CCardBody>
|
||||
</CCard>
|
||||
|
||||
<CCard class="mt-4">
|
||||
<CCardHeader>
|
||||
<h4>品牌知识库</h4>
|
||||
<CButton color="success" @click="showAddKnowledge">添加知识</CButton>
|
||||
</CCardHeader>
|
||||
<CCardBody>
|
||||
<CTable :items="knowledgeList" :fields="knowledgeFields">
|
||||
<template #actions="{ item }">
|
||||
<CButton color="danger" size="sm" @click="deleteKnowledge(item.id)">删除</CButton>
|
||||
</template>
|
||||
</CTable>
|
||||
</CCardBody>
|
||||
</CCard>
|
||||
</div>
|
||||
</template>
|
||||
```
|
||||
|
||||
### 4.5 内容管理系统
|
||||
|
||||
#### 4.5.1 知识库管理
|
||||
- 支持文本、图片、视频、链接等多种内容类型
|
||||
- 内容优先级设置
|
||||
- 内容分类标签
|
||||
- 版本控制
|
||||
|
||||
#### 4.5.2 品牌定制
|
||||
- 品牌色调配置
|
||||
- 欢迎语定制
|
||||
- 头像和图标上传
|
||||
- 回答风格调整
|
||||
|
||||
#### 4.5.3 首页AI客服开关
|
||||
- TODO: 在MiniProgramContent中添加enable_ai_chat字段
|
||||
- 后台管理：简单的on/off开关
|
||||
- 小程序端：判断bool值，如果为true则跳转到chat页面
|
||||
- 实现方式：几行代码即可完成
|
||||
|
||||
#### 4.5.3 平台知识库管理（简化版）
|
||||
- 平台介绍内容（4000字）的基础分块
|
||||
- 分层内容策略：摘要、完整两个层次
|
||||
- 基于关键词的简单内容选择
|
||||
- 基础内容压缩，避免信息过载
|
||||
|
||||
#### 4.5.4 智能内容处理（简化版）
|
||||
- 基于LangChain的文档基础预处理
|
||||
- 简单分块策略：按段落和句子分割
|
||||
- 基础内容压缩：摘要和完整内容
|
||||
- 固定上下文长度，简化Token管理
|
||||
|
||||
## 5. 技术实现细节
|
||||
|
||||
#### 5.1.3 智能能力路由（简化版）
|
||||
```python
|
||||
def route_capability(self, query, context):
|
||||
# 基于规则的快速路由
|
||||
capability = self._rule_based_routing(query)
|
||||
|
||||
# 根据能力类型构建相应的Prompt
|
||||
if capability == "rag_knowledge":
|
||||
return self._build_rag_prompt(query, context)
|
||||
elif capability == "platform_service":
|
||||
return self._build_platform_prompt(query, context)
|
||||
else:
|
||||
return self._build_general_prompt(query, context)
|
||||
```
|
||||
|
||||
### 5.2 多模态内容处理
|
||||
|
||||
#### 5.2.1 内容类型支持
|
||||
- **文本**: 直接返回
|
||||
- **图片**: 上传到OSS，返回URL
|
||||
- **视频**: 支持微信视频号链接
|
||||
- **链接**: 小程序码、商城链接等
|
||||
- **富文本**: Markdown格式支持
|
||||
|
||||
#### 5.2.2 内容渲染
|
||||
```javascript
|
||||
// 内容渲染组件
|
||||
function renderContent(content, type) {
|
||||
switch (type) {
|
||||
case 'text':
|
||||
return <Text>{content}</Text>;
|
||||
case 'image':
|
||||
return <Image src={content} mode="widthFix" />;
|
||||
case 'video':
|
||||
return <Video src={content} />;
|
||||
case 'link':
|
||||
return <Link href={content.url} text={content.text} />;
|
||||
case 'rich_text':
|
||||
return <RichText content={content} />;
|
||||
default:
|
||||
return <Text>{content}</Text>;
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
### 5.3 会话管理
|
||||
|
||||
#### 5.3.1 会话状态
|
||||
- 会话创建和销毁
|
||||
- 消息历史记录
|
||||
- 上下文保持
|
||||
- 会话超时处理
|
||||
|
||||
#### 5.3.2 用户身份识别
|
||||
- 微信OpenID绑定
|
||||
- 匿名会话支持
|
||||
- 用户偏好记录
|
||||
|
||||
### 5.4 智能调度系统（简化版）
|
||||
|
||||
#### 5.4.1 能力处理器架构
|
||||
- 基础能力处理器注册机制
|
||||
- 简单的能力降级策略
|
||||
- 固定路由规则
|
||||
|
||||
#### 5.4.2 调度策略
|
||||
- 基于关键词的能力选择
|
||||
- 基础异常处理和回退
|
||||
- 简单性能监控
|
||||
- 后续版本支持高级功能
|
||||
|
||||
### 5.5 上下文长度管理
|
||||
|
||||
#### 5.5.1 当前问题
|
||||
- **无限累积**：对话历史无限制增长
|
||||
- **Token超限**：可能超出API的token限制
|
||||
- **内存占用**：长时间对话占用大量内存
|
||||
- **API失败**：超长历史可能导致API调用失败
|
||||
|
||||
#### 5.5.2 优化策略
|
||||
```python
|
||||
class AIChatService:
|
||||
def __init__(self, api_key: str = None, base_url: str = None):
|
||||
# ... 现有代码 ...
|
||||
self.max_history_length = 20 # 最大历史记录数
|
||||
self.max_tokens_per_message = 1000 # 每条消息最大token数
|
||||
|
||||
def _truncate_history(self):
|
||||
"""截断对话历史，保持上下文长度"""
|
||||
if len(self.conversation_history) > self.max_history_length:
|
||||
# 保留系统提示词和最近的对话
|
||||
system_message = self.conversation_history[0] # 系统提示词
|
||||
recent_messages = self.conversation_history[-self.max_history_length+1:]
|
||||
self.conversation_history = [system_message] + recent_messages
|
||||
|
||||
def _estimate_tokens(self, text: str) -> int:
|
||||
"""估算文本的token数量（简化版）"""
|
||||
# 中文约1.5字符=1token，英文约4字符=1token
|
||||
chinese_chars = len([c for c in text if '\u4e00' <= c <= '\u9fff'])
|
||||
english_chars = len(text) - chinese_chars
|
||||
return int(chinese_chars / 1.5 + english_chars / 4)
|
||||
|
||||
def _smart_truncate(self):
|
||||
"""智能截断：基于token数量而非消息数量"""
|
||||
total_tokens = sum(self._estimate_tokens(msg['content']) for msg in self.conversation_history)
|
||||
max_total_tokens = 4000 # 预留1000token给AI回复
|
||||
|
||||
if total_tokens > max_total_tokens:
|
||||
# 保留系统提示词和最近的对话，直到token数合适
|
||||
system_message = self.conversation_history[0]
|
||||
truncated_history = [system_message]
|
||||
|
||||
for msg in reversed(self.conversation_history[1:]):
|
||||
truncated_history.insert(1, msg)
|
||||
current_tokens = sum(self._estimate_tokens(m['content']) for m in truncated_history)
|
||||
if current_tokens > max_total_tokens:
|
||||
truncated_history.pop(1) # 移除刚添加的消息
|
||||
break
|
||||
|
||||
self.conversation_history = truncated_history
|
||||
```
|
||||
|
||||
#### 5.5.3 分层历史策略
|
||||
```python
|
||||
def _build_context_aware_history(self, user_message: str) -> List[Dict]:
|
||||
"""构建上下文感知的对话历史"""
|
||||
# 策略1：保留最近的N轮对话
|
||||
recent_messages = self.conversation_history[-10:] # 最近10轮
|
||||
|
||||
# 策略2：保留关键信息（如工具调用结果）
|
||||
important_messages = [msg for msg in self.conversation_history
|
||||
if msg.get('role') == 'tool' or '重要' in msg.get('content', '')]
|
||||
|
||||
# 策略3：动态调整：根据用户问题复杂度决定保留多少历史
|
||||
if '刚才' in user_message or '之前' in user_message:
|
||||
# 用户引用之前内容，保留更多历史
|
||||
context_messages = self.conversation_history[-15:]
|
||||
else:
|
||||
# 新话题，保留较少历史
|
||||
context_messages = self.conversation_history[-5:]
|
||||
|
||||
return [self.conversation_history[0]] + context_messages # 系统提示词 + 上下文
|
||||
```
|
||||
|
||||
#### 5.5.4 错误处理和降级
|
||||
```python
|
||||
def chat(self, user_message: str) -> str:
|
||||
try:
|
||||
# 智能截断历史
|
||||
self._smart_truncate()
|
||||
|
||||
# 调用API
|
||||
response = self.client.chat.completions.create(
|
||||
model=self.model,
|
||||
messages=self.conversation_history,
|
||||
tools=self.tools,
|
||||
tool_choice="auto",
|
||||
temperature=0.7,
|
||||
max_tokens=2000
|
||||
)
|
||||
|
||||
# ... 处理响应 ...
|
||||
|
||||
except Exception as e:
|
||||
if "context_length_exceeded" in str(e) or "token_limit" in str(e):
|
||||
# Token超限，截断历史重试
|
||||
self._truncate_history()
|
||||
return self.chat(user_message) # 递归重试
|
||||
else:
|
||||
return f"AI服务调用失败: {str(e)}"
|
||||
```
|
||||
|
||||
## 7. 安全考虑
|
||||
|
||||
### 7.1 数据安全
|
||||
@ -938,9 +159,6 @@ def chat(self, user_message: str) -> str:
|
||||
- Week 3: 完成小程序AI聊天集成和内容管理
|
||||
- Week 4: 完成测试、优化和上线
|
||||
|
||||
### 10.3 Backup
|
||||
- 多AI服务商备选方案
|
||||
|
||||
### 11.2 开发周期
|
||||
- **总周期**: 4周（压缩版）
|
||||
- **核心功能**: AI聊天、RAG知识检索、工具调用、小程序集成、基础内容管理
|
||||
@ -967,3 +185,7 @@ def chat(self, user_message: str) -> str:
|
||||
- 降低维护成本
|
||||
- 提高系统稳定性
|
||||
- 保持架构简洁性
|
||||
|
||||
|
||||
TODO:
|
||||
- context length management
|
||||
Loading…
x
Reference in New Issue
Block a user