Files
C.E.L_Slide_test2/tests/pipeline/results/v2_semantic_rerank_result.yaml
KyeongminandClaude Opus 4.8 b836e79ee1 wip: phase_z2 evidence 파이프라인 + matching 실험(phase2~26) + 프론트 trace 패널 진행분 스냅샷
- src: phase_z2 composition/mapper/pipeline/placement_planner/retry, ai_fallback(prompts/schema/validate), mdx_text_atoms 신규
- Front: PipelineTracePanel 신규, FramePanel/SlideCanvas/Home/designAgentApi 등 갱신 + 테스트 4종 추가
- templates/phase_z2: catalog(component_expansion_registry, node_slot_mapping 신규), frames, families, slide_base 갱신
- tests/matching: phase2~26 매칭 실험 스크립트·리포트·온톨로지 전체 (미커밋 진행분)
- tests: b4_v4 evidence, task5~28.5 시리즈, regression(imp95 baseline) 등 신규 테스트 대량 추가
- docs/reference: MDX 구조 인벤토리, MDX→Frame 구조 계약 문서
- scripts: mdx 계약/parity/coverage/viewport 체크, gitea comment, run sync 유틸
- .gitignore: tmp*.json, chromedriver, .orchestrator, *.pkl, Front_test* 등 임시/스냅샷 제외

미완성 작업의 보존용 스냅샷 커밋 (2026-07-02)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-02 17:03:42 +09:00

442 lines
11 KiB
YAML

meta:
pipeline_step: 8.v2
description: V1 Top-K 후보를 ko-sroberta cosine 으로 재정렬 (캐스케이드 rerank 1단)
model: jhgan/ko-sroberta-multitask
similarity: cosine
top_k: 5
mdx_summary_spec: title + 첫 일반 문단(≤120자) + slot label Top-5 join (detect_mdx.build_summary)
frame_source: analysis.md "내용" 섹션 (phase_common.load_32_frames)
v1_source: mdx_matching_result.yaml rank_by_matching_score
holdout_sections:
- 01-1
- 02-1
- 02-2.1
answer_map:
01-2: 18
02-2.2: 14
03-1: 13
03-2: 29
lock_snapshot:
timestamp: '2026-04-24T07:44:36'
model: jhgan/ko-sroberta-multitask
top_k: 5
files:
pipeline_08_v2_semantic_rerank.py: 87f0873dde540acaf4944f2e95a3d02dac7e6f6da5e7880bbe393120cce47190
embeddings.py: acc50f3f93935061fb146c3b15e30bc9734b4c96c53f82a603bc758944b50c9c
detect_mdx.py: aaf18b89aeb7bebb7c206348ef7a61d560999c4234ce28df070c24701c96227a
phase_common.py: b778840c8e246d0e5210359263e4a59ef6acfa6f3c9a8a1cc110f18506cc4f94
pipeline_01_extract_nodes.py: d31dec9c25bda9a893caa7121fe79b1f0f7e533f32d83743d04abb36f5e6a343
synonyms.yaml: 57bbc60d0ca8a95c445e3e0acf1af4e1cff101957ae8c1759311ec827c961893
mdx_matching_result.yaml: 2796504d30966f85492f3fad826f223969bba8fd5ee6e6ab21b56444527472c8
principle:
- V1 baseline 고정 — 재정렬 대상만 Top-K 로 한정
- 가중합 금지 — 순수 semantic rerank
- Holdout 성적을 설계 근거로 사용하지 않음
mdx_sections:
01-2:
section_type: target
answer_frame_number: 18
mdx_title: 2. 용어간 상호관계
mdx_summary: 2. 용어간 상호관계. ![DX와 핵심기술간 상호관계](/assets/images/DX1.png). DX / BIM
top_k: 5
v1_top_k:
- rank: 1
frame_id: '1171281195'
frame_number: 18
v1_score: 0.937
- rank: 2
frame_id: '1171281208'
frame_number: 27
v1_score: 0.392
- rank: 3
frame_id: '1171281182'
frame_number: 11
v1_score: 0.379
- rank: 4
frame_id: '1171281172'
frame_number: 1
v1_score: 0.362
- rank: 5
frame_id: '1171281180'
frame_number: 9
v1_score: 0.358
v2_rerank:
- frame_id: '1171281195'
frame_number: 18
v1_rank: 1
v1_score: 0.937
semantic_score: 0.7295
v2_rank: 1
- frame_id: '1171281208'
frame_number: 27
v1_rank: 2
v1_score: 0.392
semantic_score: 0.6442
v2_rank: 2
- frame_id: '1171281182'
frame_number: 11
v1_rank: 3
v1_score: 0.379
semantic_score: 0.5853
v2_rank: 3
- frame_id: '1171281172'
frame_number: 1
v1_rank: 4
v1_score: 0.362
semantic_score: 0.5741
v2_rank: 4
- frame_id: '1171281180'
frame_number: 9
v1_rank: 5
v1_score: 0.358
semantic_score: 0.5556
v2_rank: 5
02-2.2:
section_type: target
answer_frame_number: 14
mdx_title: 2.2 DX 시행 주체별 기대효과
mdx_summary: 2.2 DX 시행 주체별 기대효과. DX 시행 주체별 기대효과
top_k: 5
v1_top_k:
- rank: 1
frame_id: '1171281191'
frame_number: 14
v1_score: 0.84
- rank: 2
frame_id: '1171281180'
frame_number: 9
v1_score: 0.495
- rank: 3
frame_id: '1171281189'
frame_number: 12
v1_score: 0.459
- rank: 4
frame_id: '1171281174'
frame_number: 3
v1_score: 0.411
- rank: 5
frame_id: '1171281176'
frame_number: 5
v1_score: 0.404
v2_rerank:
- frame_id: '1171281191'
frame_number: 14
v1_rank: 1
v1_score: 0.84
semantic_score: 0.6416
v2_rank: 1
- frame_id: '1171281189'
frame_number: 12
v1_rank: 3
v1_score: 0.459
semantic_score: 0.4858
v2_rank: 2
- frame_id: '1171281174'
frame_number: 3
v1_rank: 4
v1_score: 0.411
semantic_score: 0.4471
v2_rank: 3
- frame_id: '1171281180'
frame_number: 9
v1_rank: 2
v1_score: 0.495
semantic_score: 0.3861
v2_rank: 4
- frame_id: '1171281176'
frame_number: 5
v1_rank: 5
v1_score: 0.404
semantic_score: 0.3762
v2_rank: 5
03-1:
section_type: target
answer_frame_number: 13
mdx_title: 1. DX 시행을 위한 필수 요건
mdx_summary: 1. DX 시행을 위한 필수 요건. 기술 / 사람 / 자연
top_k: 5
v1_top_k:
- rank: 1
frame_id: '1171281190'
frame_number: 13
v1_score: 0.598
- rank: 2
frame_id: '1171281192'
frame_number: 15
v1_score: 0.542
- rank: 3
frame_id: '1171281198'
frame_number: 20
v1_score: 0.388
- rank: 4
frame_id: '1171281208'
frame_number: 27
v1_score: 0.363
- rank: 5
frame_id: '1171281197'
frame_number: 19
v1_score: 0.269
v2_rerank:
- frame_id: '1171281190'
frame_number: 13
v1_rank: 1
v1_score: 0.598
semantic_score: 0.634
v2_rank: 1
- frame_id: '1171281192'
frame_number: 15
v1_rank: 2
v1_score: 0.542
semantic_score: 0.5959
v2_rank: 2
- frame_id: '1171281198'
frame_number: 20
v1_rank: 3
v1_score: 0.388
semantic_score: 0.5188
v2_rank: 3
- frame_id: '1171281208'
frame_number: 27
v1_rank: 4
v1_score: 0.363
semantic_score: 0.4577
v2_rank: 4
- frame_id: '1171281197'
frame_number: 19
v1_rank: 5
v1_score: 0.269
semantic_score: 0.4209
v2_rank: 5
03-2:
section_type: target
answer_frame_number: 29
mdx_title: 2. Process의 혁신과 Product의 변화
mdx_summary: 2. Process의 혁신과 Product의 변화. 과정 / 결과
top_k: 5
v1_top_k:
- rank: 1
frame_id: '1171281210'
frame_number: 29
v1_score: 0.705
- rank: 2
frame_id: '1171281189'
frame_number: 12
v1_score: 0.521
- rank: 3
frame_id: '1171281180'
frame_number: 9
v1_score: 0.426
- rank: 4
frame_id: '1171281182'
frame_number: 11
v1_score: 0.42
- rank: 5
frame_id: '1171281179'
frame_number: 8
v1_score: 0.383
v2_rerank:
- frame_id: '1171281210'
frame_number: 29
v1_rank: 1
v1_score: 0.705
semantic_score: 0.5991
v2_rank: 1
- frame_id: '1171281179'
frame_number: 8
v1_rank: 5
v1_score: 0.383
semantic_score: 0.4832
v2_rank: 2
- frame_id: '1171281182'
frame_number: 11
v1_rank: 4
v1_score: 0.42
semantic_score: 0.465
v2_rank: 3
- frame_id: '1171281189'
frame_number: 12
v1_rank: 2
v1_score: 0.521
semantic_score: 0.4395
v2_rank: 4
- frame_id: '1171281180'
frame_number: 9
v1_rank: 3
v1_score: 0.426
semantic_score: 0.3975
v2_rank: 5
01-1:
section_type: holdout
answer_frame_number: null
mdx_title: 1. 용어 정의
mdx_summary: '1. 용어 정의. fontSize: ''0.8rem'',. 건설산업 / BIM / DX'
top_k: 5
v1_top_k:
- rank: 1
frame_id: '1171281172'
frame_number: 1
v1_score: 0.551
- rank: 2
frame_id: '1171281208'
frame_number: 27
v1_score: 0.298
- rank: 3
frame_id: '1171281182'
frame_number: 11
v1_score: 0.249
- rank: 4
frame_id: '1171281189'
frame_number: 12
v1_score: 0.248
- rank: 5
frame_id: '1171281198'
frame_number: 20
v1_score: 0.238
v2_rerank:
- frame_id: '1171281208'
frame_number: 27
v1_rank: 2
v1_score: 0.298
semantic_score: 0.5881
v2_rank: 1
- frame_id: '1171281182'
frame_number: 11
v1_rank: 3
v1_score: 0.249
semantic_score: 0.5507
v2_rank: 2
- frame_id: '1171281189'
frame_number: 12
v1_rank: 4
v1_score: 0.248
semantic_score: 0.547
v2_rank: 3
- frame_id: '1171281172'
frame_number: 1
v1_rank: 1
v1_score: 0.551
semantic_score: 0.5368
v2_rank: 4
- frame_id: '1171281198'
frame_number: 20
v1_rank: 5
v1_score: 0.238
semantic_score: 0.5005
v2_rank: 5
02-1:
section_type: holdout
answer_frame_number: null
mdx_title: 1. DX의 궁극적 목표
mdx_summary: 1. DX의 궁극적 목표. ![DX의 궁극적 목표](/assets/images/궁극적목표.png). 안전과 품질 / 생산성 향상 / 소통과 신뢰
top_k: 5
v1_top_k:
- rank: 1
frame_id: '1171281189'
frame_number: 12
v1_score: 0.766
- rank: 2
frame_id: '1171281191'
frame_number: 14
v1_score: 0.382
- rank: 3
frame_id: '1171281208'
frame_number: 27
v1_score: 0.356
- rank: 4
frame_id: '1171281174'
frame_number: 3
v1_score: 0.302
- rank: 5
frame_id: '1171281198'
frame_number: 20
v1_score: 0.237
v2_rerank:
- frame_id: '1171281198'
frame_number: 20
v1_rank: 5
v1_score: 0.237
semantic_score: 0.5895
v2_rank: 1
- frame_id: '1171281191'
frame_number: 14
v1_rank: 2
v1_score: 0.382
semantic_score: 0.5779
v2_rank: 2
- frame_id: '1171281189'
frame_number: 12
v1_rank: 1
v1_score: 0.766
semantic_score: 0.57
v2_rank: 3
- frame_id: '1171281208'
frame_number: 27
v1_rank: 3
v1_score: 0.356
semantic_score: 0.5467
v2_rank: 4
- frame_id: '1171281174'
frame_number: 3
v1_rank: 4
v1_score: 0.302
semantic_score: 0.4707
v2_rank: 5
02-2.1:
section_type: holdout
answer_frame_number: null
mdx_title: 2.1 업무 수행 과정(Process)의 변화
mdx_summary: 2.1 업무 수행 과정(Process)의 변화. 업무 수행 과정
top_k: 5
v1_top_k:
- rank: 1
frame_id: '1171281182'
frame_number: 11
v1_score: 0.246
- rank: 2
frame_id: '1171281179'
frame_number: 8
v1_score: 0.193
- rank: 3
frame_id: '1171281174'
frame_number: 3
v1_score: 0.184
- rank: 4
frame_id: '1171281210'
frame_number: 29
v1_score: 0.176
- rank: 5
frame_id: '1171281180'
frame_number: 9
v1_score: 0.171
v2_rerank:
- frame_id: '1171281210'
frame_number: 29
v1_rank: 4
v1_score: 0.176
semantic_score: 0.5511
v2_rank: 1
- frame_id: '1171281179'
frame_number: 8
v1_rank: 2
v1_score: 0.193
semantic_score: 0.5021
v2_rank: 2
- frame_id: '1171281174'
frame_number: 3
v1_rank: 3
v1_score: 0.184
semantic_score: 0.4648
v2_rank: 3
- frame_id: '1171281182'
frame_number: 11
v1_rank: 1
v1_score: 0.246
semantic_score: 0.4311
v2_rank: 4
- frame_id: '1171281180'
frame_number: 9
v1_rank: 5
v1_score: 0.171
semantic_score: 0.4191
v2_rank: 5