Function parse_evaluation_response

RealChart2Code_eval/evaluate_task2.py:606–658

(response_content: str, data_response_content: str)

def parse_evaluation_response(response_content: str, data_response_content: str) -> Dict:
    # 首先按照原有逻辑解析主要的视觉评估结果
    evaluation = None
    
    # 尝试从response_content解析JSON
    try:
        evaluation = json.loads(response_content)
    except json.JSONDecodeError:
        pass
    
    # 尝试从代码块中解析JSON
    if evaluation is None:
        json_match = re.search(r'```json\s*\n(.*?)\n```', response_content, re.DOTALL)
        if json_match:
            try:
                evaluation = json.loads(json_match.group(1))
            except json.JSONDecodeError:
                pass
    
    # 尝试从任意JSON对象中解析
    if evaluation is None:
        json_match = re.search(r'\{.*\}', response_content, re.DOTALL)
        if json_match:
            try:
                evaluation = json.loads(json_match.group(0))
            except json.JSONDecodeError:
                pass
    
    # 如果所有解析都失败，使用空的评估结构
    if evaluation is None:
        evaluation = create_empty_evaluation()
    
    # 确保evaluation是字典类型，不是字符串
    if isinstance(evaluation, str):
        evaluation = create_empty_evaluation()
    
    # 从data_response_content中提取<result>标签内容
    data_alignment = {'score': 0, 'reason': 'Data alignment not evaluated'}  # 默认值
    
    if data_response_content:
        # 尝试提取 <result> 标签内容
        result_match = re.search(r'<result>(YES|NO)</result>', data_response_content, re.IGNORECASE)
        if result_match:
            result = result_match.group(1).upper()
            if result == 'YES':
                data_alignment = {'score': 2, 'reason': 'Data alignment confirmed - identical data usage across all subplots'}
            else:
                data_alignment = {'score': 0, 'reason': 'Data alignment failed - different data usage detected'}
    
    # 将data_alignment添加到评估结果中
    evaluation['data_alignment'] = data_alignment
    
    return evaluation


def create_empty_evaluation() -> Dict: