diff --git a/README-CN.md b/README-CN.md
index 0d029fc..420654a 100644
--- a/README-CN.md
+++ b/README-CN.md
@@ -42,6 +42,7 @@
- [MySQL](https://www.mysql.com/) 数据库
- [Conda](https://docs.conda.io/en/latest/)（可选，用于环境管理）
- 合法的微博账号（用于数据采集）
+- OpenAI API密钥或Anthropic（Claude）API密钥（用于AI分析功能）
### 安装步骤
@@ -68,7 +69,20 @@
- 运行 `createTables.sql` 创建所需的数据库表。
- 修改 `config.py` 中的数据库连接配置，确保与您的MySQL设置匹配。
-4. 启动Flask应用：
+4. 配置AI分析功能（可选）：
+
+ 设置AI分析功能所需的环境变量：
+ ```bash
+ # OpenAI API配置（使用GPT模型必需）
+ export OPENAI_API_KEY="你的openai密钥"
+
+ # Anthropic API配置（使用Claude模型必需）
+ export ANTHROPIC_API_KEY="你的anthropic密钥"
+ ```
+
+ 注意：至少需要配置一个API密钥才能使用AI分析功能。
+
+5. 启动Flask应用：
```bash
python app.py
@@ -90,6 +104,8 @@
- **[Matplotlib](https://matplotlib.org/)** - 数据可视化库。
- **[Scikit-learn](https://scikit-learn.org/)** - 机器学习库，用于模型训练和评估。
- **[TensorFlow](https://www.tensorflow.org/)** 或 **[PyTorch](https://pytorch.org/)** - 深度学习框架，用于高级模型开发。
+- **[OpenAI GPT](https://openai.com/)** - 先进的语言模型，用于文本分析。
+- **[Anthropic Claude](https://www.anthropic.com/)** - 智能AI模型，用于复杂文本分析。
## 🤝 贡献
diff --git a/README.md b/README.md
index 1a771dd..92fe2e3 100644
--- a/README.md
+++ b/README.md
@@ -40,6 +40,7 @@ Follow the steps below to run the project on your system.
- [MySQL](https://www.mysql.com/) Database
- [Conda](https://docs.conda.io/en/latest/) (optional, for environment management)
- A valid Weibo account (for data collection)
+- OpenAI API key or Anthropic (Claude) API key for AI analysis features
### Installation Steps
@@ -66,13 +67,26 @@ Follow the steps below to run the project on your system.
- Run `createTables.sql` to create the necessary database tables.
- Modify the database connection settings in `config.py` to match your MySQL configuration.
-5. Start the Flask application:
+5. Configure AI Analysis (Optional):
+
+ Set up environment variables for AI analysis features:
+ ```bash
+ # For OpenAI API (Required for GPT models)
+ export OPENAI_API_KEY="your-openai-key"
+
+ # For Anthropic API (Required for Claude models)
+ export ANTHROPIC_API_KEY="your-anthropic-key"
+ ```
+
+ Note: At least one API key must be configured to use AI analysis features.
+
+6. Start the Flask application:
```bash
python app.py
```
-6. Access the application: Open your browser and navigate to http://localhost:5000 to use the system.
+7. Access the application: Open your browser and navigate to http://localhost:5000 to use the system.
## 🛠️ Technology Stack
@@ -88,6 +102,8 @@ The Weibo Public Opinion Analysis and Prediction System employs a range of moder
- **[Matplotlib](https://matplotlib.org/)** - A data visualization library.
- **[Scikit-learn](https://scikit-learn.org/)** - A machine learning library used for model training and evaluation.
- **[TensorFlow](https://www.tensorflow.org/)** 或 **[PyTorch](https://pytorch.org/)** - Deep learning frameworks used for advanced model development.
+- **[OpenAI GPT](https://openai.com/)** - Advanced language models for text analysis.
+- **[Anthropic Claude](https://www.anthropic.com/)** - AI models for sophisticated text analysis.
## 🤝 Contribution
diff --git a/utils/ai_analyzer.py b/utils/ai_analyzer.py
index f5cb0ac..083209c 100644
--- a/utils/ai_analyzer.py
+++ b/utils/ai_analyzer.py
@@ -1,4 +1,5 @@
import openai
+import anthropic
import json
from typing import List, Dict
import os
@@ -8,11 +9,34 @@ from utils.logger import app_logger as logging
class AIAnalyzer:
def __init__(self):
# 从环境变量获取API密钥
- self.api_key = os.getenv('OPENAI_API_KEY')
- if not self.api_key:
- raise ValueError("请设置OPENAI_API_KEY环境变量")
+ self.openai_key = os.getenv('OPENAI_API_KEY')
+ self.claude_key = os.getenv('ANTHROPIC_API_KEY')
- openai.api_key = self.api_key
+ if not self.openai_key and not self.claude_key:
+ raise ValueError("请至少设置一个API密钥 (OPENAI_API_KEY 或 ANTHROPIC_API_KEY)")
+
+ if self.openai_key:
+ openai.api_key = self.openai_key
+ if self.claude_key:
+ self.claude_client = anthropic.Anthropic(api_key=self.claude_key)
+
+ # 支持的模型列表
+ self.supported_models = {
+ # OpenAI 模型
+ 'gpt-3.5-turbo': {'provider': 'openai', 'max_tokens': 2000, 'cost_per_1k': 0.0015},
+ 'gpt-3.5-turbo-16k': {'provider': 'openai', 'max_tokens': 16000, 'cost_per_1k': 0.003},
+ 'gpt-4': {'provider': 'openai', 'max_tokens': 8000, 'cost_per_1k': 0.03},
+ 'gpt-4-32k': {'provider': 'openai', 'max_tokens': 32000, 'cost_per_1k': 0.06},
+ 'gpt-4-turbo-preview': {'provider': 'openai', 'max_tokens': 128000, 'cost_per_1k': 0.01},
+
+ # Claude 模型
+ 'claude-3-opus-20240229': {'provider': 'anthropic', 'max_tokens': 4000, 'cost_per_1k': 0.015},
+ 'claude-3-sonnet-20240229': {'provider': 'anthropic', 'max_tokens': 3000, 'cost_per_1k': 0.003},
+ 'claude-3-haiku-20240307': {'provider': 'anthropic', 'max_tokens': 2000, 'cost_per_1k': 0.0025},
+ 'claude-2.1': {'provider': 'anthropic', 'max_tokens': 100000, 'cost_per_1k': 0.008},
+ 'claude-2.0': {'provider': 'anthropic', 'max_tokens': 100000, 'cost_per_1k': 0.008},
+ 'claude-instant-1.2': {'provider': 'anthropic', 'max_tokens': 100000, 'cost_per_1k': 0.0015}
+ }
# 不同深度的分析提示词
self.prompt_templates = {
@@ -73,48 +97,144 @@ class AIAnalyzer:
analysis_depth: str = "standard") -> List[Dict]:
"""分析一批消息并返回分析结果"""
try:
+ if model_type not in self.supported_models:
+ raise ValueError(f"不支持的模型类型: {model_type}")
+
+ model_info = self.supported_models[model_type]
+ provider = model_info['provider']
+ max_tokens = model_info['max_tokens']
+
+ # 根据模型类型调整批处理大小
+ adjusted_batch_size = min(batch_size, self._get_optimal_batch_size(model_type))
+ if adjusted_batch_size != batch_size:
+ logging.info(f"已将批处理大小从 {batch_size} 调整为 {adjusted_batch_size}")
+
all_results = []
+ total_cost = 0
# 分批处理消息
- for i in range(0, len(messages), batch_size):
- batch = messages[i:i + batch_size]
+ for i in range(0, len(messages), adjusted_batch_size):
+ batch = messages[i:i + adjusted_batch_size]
formatted_messages = []
for msg in batch:
formatted_messages.append(f"消息ID: {msg['id']}\n内容: {msg['content']}")
messages_text = "\n---\n".join(formatted_messages)
-
- # 获取对应深度的提示词
system_prompt = self.prompt_templates.get(analysis_depth, self.prompt_templates['standard'])
- # 调用OpenAI API
- response = await openai.ChatCompletion.acreate(
- model=model_type,
- messages=[
- {"role": "system", "content": system_prompt},
- {"role": "user", "content": f"请分析以下消息:\n{messages_text}"}
- ],
- temperature=0.3, # 降低随机性
- max_tokens=2000 if analysis_depth != 'deep' else 3000,
- n=1
- )
-
- try:
- result = json.loads(response.choices[0].message.content)
- if isinstance(result, dict) and 'analysis_results' in result:
- all_results.extend(result['analysis_results'])
- else:
- logging.error(f"API返回格式不正确: {response.choices[0].message.content}")
- except json.JSONDecodeError as e:
- logging.error(f"JSON解析失败: {e}")
- continue
+ if provider == 'openai':
+ result = await self._analyze_with_openai(
+ messages_text,
+ system_prompt,
+ model_type,
+ max_tokens
+ )
+ else: # anthropic
+ result = await self._analyze_with_claude(
+ messages_text,
+ system_prompt,
+ model_type,
+ max_tokens
+ )
+ if result:
+ all_results.extend(result)
+ # 计算本批次成本
+ batch_cost = self._calculate_cost(len(messages_text), model_type)
+ total_cost += batch_cost
+ logging.info(f"批次处理完成,成本: ${batch_cost:.4f}")
+
+ logging.info(f"分析完成,总成本: ${total_cost:.4f}")
return all_results
except Exception as e:
logging.error(f"AI分析过程出错: {e}")
return []
+ def _get_optimal_batch_size(self, model_type: str) -> int:
+ """根据模型类型获取最优批处理大小"""
+ model_info = self.supported_models[model_type]
+ max_tokens = model_info['max_tokens']
+
+ # 估算每条消息的平均token数(假设为200)
+ avg_tokens_per_message = 200
+
+ # 预留20%的token用于系统提示词和响应
+ available_tokens = int(max_tokens * 0.8)
+
+ # 计算最优批处理大小
+ optimal_batch_size = max(1, min(100, available_tokens // avg_tokens_per_message))
+
+ return optimal_batch_size
+
+ def _calculate_cost(self, input_length: int, model_type: str) -> float:
+ """计算API调用成本"""
+ model_info = self.supported_models[model_type]
+ cost_per_1k = model_info['cost_per_1k']
+
+ # 估算token数(假设每4个字符约等于1个token)
+ estimated_tokens = input_length // 4
+
+ # 计算成本(美元)
+ cost = (estimated_tokens / 1000) * cost_per_1k
+
+ return cost
+
+ async def _analyze_with_openai(self, messages_text: str, system_prompt: str,
+ model: str, max_tokens: int) -> List[Dict]:
+ """使用OpenAI API进行分析"""
+ try:
+ response = await openai.ChatCompletion.acreate(
+ model=model,
+ messages=[
+ {"role": "system", "content": system_prompt},
+ {"role": "user", "content": f"请分析以下消息:\n{messages_text}"}
+ ],
+ temperature=0.3,
+ max_tokens=max_tokens,
+ n=1,
+ response_format={"type": "json_object"} # 强制JSON响应格式
+ )
+
+ result = json.loads(response.choices[0].message.content)
+ if isinstance(result, dict) and 'analysis_results' in result:
+ return result['analysis_results']
+ else:
+ logging.error(f"OpenAI API返回格式不正确: {response.choices[0].message.content}")
+ return []
+
+ except Exception as e:
+ logging.error(f"OpenAI API调用失败: {e}")
+ return []
+
+ async def _analyze_with_claude(self, messages_text: str, system_prompt: str,
+ model: str, max_tokens: int) -> List[Dict]:
+ """使用Claude API进行分析"""
+ try:
+ response = await self.claude_client.messages.create(
+ model=model,
+ max_tokens=max_tokens,
+ temperature=0.3,
+ system=system_prompt,
+ messages=[
+ {
+ "role": "user",
+ "content": f"请分析以下消息:\n{messages_text}"
+ }
+ ]
+ )
+
+ result = json.loads(response.content[0].text)
+ if isinstance(result, dict) and 'analysis_results' in result:
+ return result['analysis_results']
+ else:
+ logging.error(f"Claude API返回格式不正确: {response.content[0].text}")
+ return []
+
+ except Exception as e:
+ logging.error(f"Claude API调用失败: {e}")
+ return []
+
def format_analysis_for_display(self, analysis: Dict) -> Dict:
"""将分析结果格式化为前端显示格式"""
base_result = {
diff --git a/views/page/templates/yuqingpredict.html b/views/page/templates/yuqingpredict.html
index 3241780..dd289d1 100644
--- a/views/page/templates/yuqingpredict.html
+++ b/views/page/templates/yuqingpredict.html
@@ -467,8 +467,21 @@