404 lines
14 KiB
Python
404 lines
14 KiB
Python
"""
|
||
代码意图识别模块
|
||
用于识别用户问题是否与代码相关，并进行详细的意图分类
|
||
支持多种代码问题类型的识别，为后续检索和Prompt生成提供指导
|
||
"""
|
||
import sys
|
||
import os
|
||
|
||
# 添加项目根目录到 Python 模块搜索路径
|
||
current_dir = os.path.dirname(os.path.abspath(__file__))
|
||
project_root = os.path.dirname(current_dir)
|
||
if project_root not in sys.path:
|
||
sys.path.insert(0, project_root)
|
||
|
||
import json
|
||
import asyncio
|
||
from typing import Dict, Any, Optional, List
|
||
from enum import Enum
|
||
from functools import lru_cache
|
||
from loguru import logger
|
||
from config import settings
|
||
from llama_index.llms.ollama import Ollama
|
||
from utils.prompt import INTENT_DETECTION_TEMPLATE
|
||
|
||
|
||
class CodeIntentCategory(Enum):
|
||
"""代码意图分类枚举 - 细粒度分类体系"""
|
||
|
||
# 代码解释与逻辑类 (Existing Code Focus)
|
||
LOGIC_EXPLANATION = "logic_explanation" # 解释既有代码的底层逻辑
|
||
ENTITY_INTRODUCTION = "entity_introduction" # 介绍具体的代码实体（函数定义、类属性、API参数）
|
||
CODE_STRUCTURE = "code_structure" # 询问项目组织
|
||
|
||
# 代码生成与实现类 (New Code Focus)
|
||
CODE_GENERATION = "code_generation" # 请求从零编写完整代码或功能块
|
||
BOILERPLATE_IMPLEMENTATION = "boilerplate_implementation" # 请求提供标准算法/模板
|
||
|
||
# 调试、优化与理论类
|
||
ERROR_DEBUGGING = "error_debugging" # 排查 Bug 或异常
|
||
CODE_OPTIMIZATION = "code_optimization" # 改进既有代码的性能或质量
|
||
ALGORITHM_THEORY = "algorithm_theory" # 算法原理或复杂度分析
|
||
|
||
# 非代码类
|
||
GENERAL_TECHNICAL = "general_technical" # 通用技术咨询
|
||
NON_TECHNICAL = "non_technical" # 非技术问题
|
||
UNKNOWN = "unknown" # 未知类型
|
||
|
||
|
||
class PromptTemplateType(Enum):
|
||
"""Prompt模板类型枚举"""
|
||
CODE_EXPLANATION = "code_explanation" # 代码解释模板（逻辑解释、实体介绍、代码结构）
|
||
CODE_GENERATION = "code_generation" # 代码生成模板（代码生成、模板实现）
|
||
CODE_DEBUGGING = "code_debugging" # 代码调试模板（错误调试）
|
||
CODE_OPTIMIZATION = "code_optimization" # 代码优化模板（代码优化）
|
||
ALGORITHM_EXPLANATION = "algorithm_explanation" # 算法解释模板（算法理论）
|
||
GENERAL_QA = "general_qa" # 通用问答模板（通用技术咨询、非技术问题）
|
||
|
||
|
||
class CodeIntentResult:
|
||
"""代码意图识别结果"""
|
||
|
||
def __init__(
|
||
self,
|
||
is_code_related: bool,
|
||
category: CodeIntentCategory,
|
||
confidence: float,
|
||
prompt_template_type: PromptTemplateType,
|
||
keywords: List[str],
|
||
reasoning: str,
|
||
requires_code_context: bool,
|
||
suggested_search_terms: List[str],
|
||
):
|
||
self.is_code_related = is_code_related # 是否与代码相关（True/False）
|
||
self.category = category # 代码意图分类
|
||
self.confidence = confidence # 置信度分数，范围0-1之间
|
||
self.prompt_template_type = prompt_template_type # Prompt模板类型
|
||
self.keywords = keywords # 相关关键词列表
|
||
self.reasoning = reasoning # 解释或理由
|
||
self.requires_code_context = requires_code_context # 是否需要代码上下文（True/False）
|
||
self.suggested_search_terms = suggested_search_terms # 建议搜索条款列表
|
||
|
||
def to_dict(self) -> Dict[str, Any]:
|
||
"""转换为字典格式"""
|
||
return {
|
||
"is_code_related": self.is_code_related,
|
||
"category": self.category.value,
|
||
"confidence": self.confidence,
|
||
"prompt_template_type": self.prompt_template_type.value,
|
||
"keywords": self.keywords,
|
||
"reasoning": self.reasoning,
|
||
"requires_code_context": self.requires_code_context,
|
||
"suggested_search_terms": self.suggested_search_terms,
|
||
}
|
||
|
||
def to_json(self) -> str:
|
||
"""转换为JSON格式"""
|
||
return json.dumps(self.to_dict(), ensure_ascii=False, indent=2)
|
||
|
||
|
||
class CodeIntentDetector:
|
||
"""代码意图检测器"""
|
||
|
||
def __init__(self, llm: Optional[Ollama] = None):
|
||
"""
|
||
初始化代码意图检测器
|
||
|
||
Args:
|
||
llm: LLM实例，如果为None则使用默认配置
|
||
"""
|
||
if llm is None:
|
||
self.llm = Ollama(
|
||
model=settings.OLLAMA_MODEL,
|
||
base_url=settings.OLLAMA_BASE_URL,
|
||
temperature=0.1, # 低温度确保输出稳定
|
||
request_timeout=1200.0
|
||
)
|
||
else:
|
||
self.llm = llm
|
||
|
||
logger.info("代码意图检测器初始化完成")
|
||
|
||
def _build_classification_prompt(self, query: str, history: Optional[str] = None) -> str:
|
||
"""
|
||
构建分类Prompt
|
||
|
||
Args:
|
||
query: 用户问题
|
||
history: 对话历史
|
||
|
||
Returns:
|
||
str: 分类Prompt
|
||
"""
|
||
history_str = "无" if not history else history
|
||
prompt = INTENT_DETECTION_TEMPLATE.format(
|
||
history_str=history_str,
|
||
query=query
|
||
)
|
||
return prompt
|
||
|
||
def _parse_llm_response(self, response_text: str) -> Optional[Dict[str, Any]]:
|
||
"""
|
||
解析LLM响应
|
||
|
||
Args:
|
||
response_text: LLM响应文本
|
||
|
||
Returns:
|
||
解析后的字典，如果解析失败返回None
|
||
"""
|
||
try:
|
||
response_text = response_text.strip()
|
||
|
||
# 尝试提取JSON部分
|
||
json_start = response_text.find('{')
|
||
json_end = response_text.rfind('}')
|
||
|
||
if json_start == -1 or json_end == -1:
|
||
logger.warning(f"未找到JSON格式响应: {response_text}")
|
||
return None
|
||
|
||
json_str = response_text[json_start:json_end + 1]
|
||
result = json.loads(json_str)
|
||
|
||
# 验证必需字段并使用默认值
|
||
required_fields = {
|
||
'is_code_related': False,
|
||
'category': 'unknown',
|
||
'confidence': 0.5,
|
||
'keywords': [],
|
||
'reasoning': '',
|
||
'requires_code_context': False,
|
||
'suggested_search_terms': []
|
||
}
|
||
|
||
for field, default_value in required_fields.items():
|
||
if field not in result:
|
||
logger.warning(f"缺少必需字段: {field}, 使用默认值: {default_value}")
|
||
result[field] = default_value
|
||
|
||
return result
|
||
|
||
except json.JSONDecodeError as e:
|
||
logger.error(f"JSON解析失败: {e}, 响应: {response_text}")
|
||
return None
|
||
except Exception as e:
|
||
logger.error(f"解析响应失败: {e}")
|
||
return None
|
||
|
||
def _map_to_enums(
|
||
self,
|
||
parsed_result: Dict[str, Any]
|
||
) -> tuple[CodeIntentCategory, PromptTemplateType]:
|
||
"""
|
||
将解析结果映射到枚举类型
|
||
|
||
Args:
|
||
parsed_result: 解析后的结果字典
|
||
|
||
Returns:
|
||
(CodeIntentCategory, PromptTemplateType)
|
||
"""
|
||
category_str = parsed_result.get('category', 'unknown')
|
||
|
||
try:
|
||
category = CodeIntentCategory(category_str)
|
||
except ValueError:
|
||
logger.warning(f"未知的分类: {category_str}, 使用默认值")
|
||
category = CodeIntentCategory.UNKNOWN
|
||
|
||
# 根据分类确定Prompt模板类型
|
||
if not parsed_result.get('is_code_related', False):
|
||
prompt_template_type = PromptTemplateType.GENERAL_QA
|
||
|
||
# 代码解释与逻辑类 -> CODE_EXPLANATION
|
||
elif category in [CodeIntentCategory.LOGIC_EXPLANATION, CodeIntentCategory.ENTITY_INTRODUCTION,
|
||
CodeIntentCategory.CODE_STRUCTURE]:
|
||
prompt_template_type = PromptTemplateType.CODE_EXPLANATION
|
||
|
||
# 代码生成与实现类 -> CODE_GENERATION
|
||
elif category in [CodeIntentCategory.CODE_GENERATION, CodeIntentCategory.BOILERPLATE_IMPLEMENTATION]:
|
||
prompt_template_type = PromptTemplateType.CODE_GENERATION
|
||
|
||
# 调试、优化与理论类
|
||
elif category == CodeIntentCategory.ERROR_DEBUGGING:
|
||
prompt_template_type = PromptTemplateType.CODE_DEBUGGING
|
||
elif category == CodeIntentCategory.CODE_OPTIMIZATION:
|
||
prompt_template_type = PromptTemplateType.CODE_OPTIMIZATION
|
||
elif category == CodeIntentCategory.ALGORITHM_THEORY:
|
||
prompt_template_type = PromptTemplateType.ALGORITHM_EXPLANATION
|
||
|
||
# 非代码问题 -> GENERAL_QA
|
||
elif category in [CodeIntentCategory.GENERAL_TECHNICAL, CodeIntentCategory.NON_TECHNICAL,
|
||
CodeIntentCategory.UNKNOWN]:
|
||
prompt_template_type = PromptTemplateType.GENERAL_QA
|
||
|
||
else:
|
||
prompt_template_type = PromptTemplateType.GENERAL_QA
|
||
|
||
return category, prompt_template_type
|
||
|
||
async def detect_intent_async(self, query: str, history: Optional[str] = None) -> CodeIntentResult:
|
||
"""
|
||
异步检测代码意图
|
||
|
||
Args:
|
||
query: 用户问题
|
||
history: 对话历史
|
||
|
||
Returns:
|
||
CodeIntentResult: 意图识别结果
|
||
"""
|
||
try:
|
||
# 构建Prompt
|
||
prompt = self._build_classification_prompt(query, history)
|
||
# 调用LLM
|
||
response = await self.llm.acomplete(prompt)
|
||
response_text = response.text
|
||
logger.debug(f"LLM响应: {response_text}")
|
||
|
||
# 解析响应
|
||
parsed_result = self._parse_llm_response(response_text)
|
||
|
||
if parsed_result is None:
|
||
logger.error("解析LLM响应失败，使用默认分类")
|
||
return self._get_default_result(query)
|
||
|
||
# 映射到枚举类型
|
||
category, prompt_template_type = self._map_to_enums(parsed_result)
|
||
|
||
# 构建结果对象
|
||
result = CodeIntentResult(
|
||
is_code_related=parsed_result.get('is_code_related', False),
|
||
category=category,
|
||
confidence=parsed_result.get('confidence', 0.5),
|
||
prompt_template_type=prompt_template_type,
|
||
keywords=parsed_result.get('keywords', []),
|
||
reasoning=parsed_result.get('reasoning', ''),
|
||
requires_code_context=parsed_result.get('requires_code_context', False),
|
||
suggested_search_terms=parsed_result.get('suggested_search_terms', [])
|
||
)
|
||
|
||
logger.info(f"代码意图检测完成: {result.to_json()}")
|
||
return result
|
||
|
||
except Exception as e:
|
||
logger.error(f"代码意图检测失败: {e}")
|
||
return self._get_default_result(query)
|
||
|
||
def detect_intent(self, query: str) -> CodeIntentResult:
|
||
"""
|
||
同步检测代码意图（包装异步方法）
|
||
|
||
Args:
|
||
query: 用户问题
|
||
|
||
Returns:
|
||
CodeIntentResult: 意图识别结果
|
||
"""
|
||
try:
|
||
loop = asyncio.get_event_loop()
|
||
except RuntimeError:
|
||
loop = asyncio.new_event_loop()
|
||
asyncio.set_event_loop(loop)
|
||
|
||
return loop.run_until_complete(self.detect_intent_async(query))
|
||
|
||
def _get_default_result(self, query: str) -> CodeIntentResult:
|
||
"""
|
||
获取默认结果（当检测失败时使用）
|
||
|
||
Args:
|
||
query: 用户问题
|
||
|
||
Returns:
|
||
CodeIntentResult: 默认结果
|
||
"""
|
||
logger.warning(f"使用默认意图识别结果: {query}")
|
||
|
||
return CodeIntentResult(
|
||
is_code_related=False,
|
||
category=CodeIntentCategory.UNKNOWN,
|
||
confidence=0.0,
|
||
prompt_template_type=PromptTemplateType.GENERAL_QA,
|
||
keywords=[],
|
||
reasoning="意图识别失败，使用默认分类",
|
||
requires_code_context=False,
|
||
suggested_search_terms=[query]
|
||
)
|
||
|
||
|
||
@lru_cache(maxsize=100)
|
||
def detect_code_intent_cached(query: str) -> CodeIntentResult:
|
||
"""
|
||
带缓存的代码意图检测（同步版本）
|
||
|
||
Args:
|
||
query: 用户问题
|
||
|
||
Returns:
|
||
CodeIntentResult: 意图识别结果
|
||
"""
|
||
detector = CodeIntentDetector()
|
||
return detector.detect_intent(query)
|
||
|
||
|
||
async def detect_code_intent_async_cached(query: str) -> CodeIntentResult:
|
||
"""
|
||
带缓存的代码意图检测（异步版本）
|
||
|
||
Args:
|
||
query: 用户问题
|
||
|
||
Returns:
|
||
CodeIntentResult: 意图识别结果
|
||
"""
|
||
detector = CodeIntentDetector()
|
||
return await detector.detect_intent_async(query)
|
||
|
||
|
||
def create_intent_detector(llm: Optional[Ollama] = None) -> CodeIntentDetector:
|
||
"""
|
||
创建代码意图检测器实例
|
||
|
||
Args:
|
||
llm: LLM实例，如果为None则使用默认配置
|
||
|
||
Returns:
|
||
CodeIntentDetector: 检测器实例
|
||
"""
|
||
return CodeIntentDetector(llm)
|
||
|
||
|
||
if __name__ == "__main__":
|
||
import asyncio
|
||
|
||
async def test_intent_detection():
|
||
"""测试代码意图检测"""
|
||
detector = CodeIntentDetector()
|
||
|
||
test_queries = [
|
||
"kth_number怎么实现的",
|
||
"kth_number的时间复杂度是多少",
|
||
"kth_number的测试用例有哪些",
|
||
"kth_number的代码质量如何",
|
||
"trieste实现的采集函数有哪些",
|
||
"这个函数是做什么的？",
|
||
"如何优化这段代码？",
|
||
"Python中如何实现异步？",
|
||
"今天天气怎么样？",
|
||
"为什么会报这个错误？",
|
||
"项目结构是怎样的？",
|
||
"如何调用这个API？"
|
||
]
|
||
|
||
for query in test_queries:
|
||
print(f"\n{'='*60}")
|
||
print(f"问题: {query}")
|
||
print('='*60)
|
||
|
||
result = await detector.detect_intent_async(query)
|
||
print(result.to_json())
|
||
|
||
asyncio.run(test_intent_detection())
|