LQAgentPlatform/utils_test/RAG_Pipeline_Test/print_review_results.py

"""
输出实际审查结果 — 将 RAG 管线的真实检索内容写入可读文件
"""
import sys, os, json

project_root = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
if project_root not in sys.path:
    sys.path.insert(0, project_root)

from utils_test.RAG_Pipeline_Test.test_data import TEST_SAMPLES
from utils_test.RAG_Pipeline_Test.rag_pipeline_runner import RAGPipelineRunner


def main():
    runner = RAGPipelineRunner()
    output_lines = []

    # 选取 3 个代表性样本深入分析
    target_ids = ["safety_Monitoring", "technology_Operations", "safety_Emergency"]
    samples = [s for s in TEST_SAMPLES if s["chunk_id"] in target_ids]

    for sample in samples:
        chunk_id = sample["chunk_id"]
        content = sample["content"]

        output_lines.append("=" * 80)
        output_lines.append(f"测试样本: {chunk_id}")
        output_lines.append(f"章节: {sample['section_label']}")
        output_lines.append("=" * 80)

        # 截取原文前 500 字
        output_lines.append(f"\n【原文摘录 (前500字)】")
        output_lines.append(content[:500])
        output_lines.append("...")

        # 执行管线
        result = runner.run_single(content, chunk_id=chunk_id)

        # 审查要点
        output_lines.append(f"\n{'─' * 60}")
        output_lines.append(f"【审查要点提取结果】共 {result.review_point_count} 个要点, 耗时 {result.extract_time:.1f}s")
        output_lines.append(f"{'─' * 60}")

        if not result.review_points:
            output_lines.append("  (提取失败)")
            continue

        for i, rp in enumerate(result.review_points):
            label = rp.get('label', rp.get('entity', ''))
            queries = rp.get('search_queries', rp.get('search_keywords', []))
            original = rp.get('original_text', rp.get('background', ''))
            parameter = rp.get('parameter', '')

            output_lines.append(f"\n  要点 [{i+1}] {label}")
            output_lines.append(f"  ├─ 检索语句:")
            for q in queries:
                output_lines.append(f"  │   • {q}")
            output_lines.append(f"  ├─ 原文摘录: {original}")
            output_lines.append(f"  └─ 技术参数: {parameter}")

        # 检索结果
        output_lines.append(f"\n{'─' * 60}")
        output_lines.append(f"【RAG 检索结果】{result.non_empty_pairs}/{result.review_point_count} 个查询对有结果, 耗时 {result.retrieval_time:.1f}s")
        output_lines.append(f"{'─' * 60}")

        if not result.retrieval_results:
            output_lines.append("  (检索失败)")
            continue

        for i, (rp, results) in enumerate(zip(result.review_points, result.retrieval_results)):
            label = rp.get('label', rp.get('entity', ''))

            output_lines.append(f"\n  ── 要点 [{i+1}] {label} 的检索结果 ──")

            if not results:
                output_lines.append("    (无结果)")
                continue

            for j, r in enumerate(results[:3]):  # 只显示 top-3
                rerank = r.get('rerank_score', 0)
                bfp = r.get('bfp_rerank_score', 0)
                text = r.get('text_content', '')[:200]
                meta = r.get('metadata', {})
                file_name = meta.get('file_name', '')
                parent_id = meta.get('parent_id', '')

                output_lines.append(f"\n    [{j+1}] rerank={rerank:.4f}, bfp_rerank={bfp:.4f}")
                if file_name:
                    output_lines.append(f"        来源: {file_name}")
                output_lines.append(f"        内容: {text}...")

        output_lines.append("\n\n")

    # 写入文件
    output_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), "reports", "review_results_detail.txt")
    os.makedirs(os.path.dirname(output_path), exist_ok=True)
    with open(output_path, "w", encoding="utf-8") as f:
        f.write("\n".join(output_lines))
    print(f"审查结果已输出: {output_path}")


if __name__ == "__main__":
    main()