012-kaopeilian/backend/app/services/ai/answer_judge_service.py

"""
答案判断服务 - Python 原生实现

功能：
- 判断填空题与问答题的答案是否正确
- 通过 AI 语义理解比对用户答案与标准答案

提供稳定可靠的答案判断能力。
"""

import logging
from dataclasses import dataclass
from typing import Any, Optional

from .ai_service import AIService, AIResponse
from .prompts.answer_judge_prompts import (
    SYSTEM_PROMPT,
    USER_PROMPT,
    CORRECT_KEYWORDS,
    INCORRECT_KEYWORDS,
)

logger = logging.getLogger(__name__)


@dataclass
class JudgeResult:
    """判断结果"""
    is_correct: bool
    raw_response: str
    ai_provider: str = ""
    ai_model: str = ""
    ai_tokens: int = 0
    ai_latency_ms: int = 0


class AnswerJudgeService:
    """
    答案判断服务
    
    使用 Python 原生实现。
    
    使用示例：
    ```python
    service = AnswerJudgeService()
    result = await service.judge(
        db=db_session,  # 传入 db_session 用于记录调用日志
        question="玻尿酸的主要作用是什么？",
        correct_answer="补水保湿、填充塑形",
        user_answer="保湿和塑形",
        analysis="玻尿酸具有补水保湿和填充塑形两大功能"
    )
    print(result.is_correct)  # True
    ```
    """
    
    MODULE_CODE = "answer_judge"
    
    async def judge(
        self,
        question: str,
        correct_answer: str,
        user_answer: str,
        analysis: str = "",
        db: Any = None  # 数据库会话，用于记录 AI 调用日志
    ) -> JudgeResult:
        """
        判断答案是否正确
        
        Args:
            question: 题目内容
            correct_answer: 标准答案
            user_answer: 用户答案
            analysis: 答案解析（可选）
            db: 数据库会话，用于记录调用日志（符合 AI 接入规范）
            
        Returns:
            JudgeResult 判断结果
        """
        try:
            logger.info(
                f"开始判断答案 - question: {question[:50]}..., "
                f"user_answer: {user_answer[:50]}..."
            )
            
            # 创建 AIService 实例（传入 db_session 用于记录调用日志）
            ai_service = AIService(module_code=self.MODULE_CODE, db_session=db)
            
            # 构建提示词
            user_prompt = USER_PROMPT.format(
                question=question,
                correct_answer=correct_answer,
                user_answer=user_answer,
                analysis=analysis or "无"
            )
            
            # 调用 AI
            messages = [
                {"role": "system", "content": SYSTEM_PROMPT},
                {"role": "user", "content": user_prompt}
            ]
            
            ai_response = await ai_service.chat(
                messages=messages,
                temperature=0.1,  # 低温度，确保输出稳定
                prompt_name="answer_judge"
            )
            
            logger.info(
                f"AI 判断完成 - provider: {ai_response.provider}, "
                f"response: {ai_response.content}, "
                f"latency: {ai_response.latency_ms}ms"
            )
            
            # 解析 AI 输出
            is_correct = self._parse_judge_result(ai_response.content)
            
            logger.info(f"答案判断结果: {is_correct}")
            
            return JudgeResult(
                is_correct=is_correct,
                raw_response=ai_response.content,
                ai_provider=ai_response.provider,
                ai_model=ai_response.model,
                ai_tokens=ai_response.total_tokens,
                ai_latency_ms=ai_response.latency_ms,
            )
            
        except Exception as e:
            logger.error(f"答案判断失败: {e}", exc_info=True)
            # 出错时默认返回错误，保守处理
            return JudgeResult(
                is_correct=False,
                raw_response=f"判断失败: {e}",
            )
    
    def _parse_judge_result(self, ai_output: str) -> bool:
        """
        解析 AI 输出的判断结果
        
        Args:
            ai_output: AI 返回的文本
            
        Returns:
            bool: True 表示正确，False 表示错误
        """
        # 清洗输出
        output = ai_output.strip().lower()
        
        # 检查是否包含正确关键词
        for keyword in CORRECT_KEYWORDS:
            if keyword.lower() in output:
                return True
        
        # 检查是否包含错误关键词
        for keyword in INCORRECT_KEYWORDS:
            if keyword.lower() in output:
                return False
        
        # 无法识别时，默认返回错误（保守处理）
        logger.warning(f"无法解析判断结果，默认返回错误: {ai_output}")
        return False


# ==================== 全局实例 ====================

answer_judge_service = AnswerJudgeService()


# ==================== 便捷函数 ====================

async def judge_answer(
    question: str,
    correct_answer: str,
    user_answer: str,
    analysis: str = ""
) -> bool:
    """
    便捷函数：判断答案是否正确
    
    Args:
        question: 题目内容
        correct_answer: 标准答案
        user_answer: 用户答案
        analysis: 答案解析
        
    Returns:
        bool: True 表示正确，False 表示错误
    """
    result = await answer_judge_service.judge(
        question=question,
        correct_answer=correct_answer,
        user_answer=user_answer,
        analysis=analysis
    )
    return result.is_correct