#!/usr/bin/env python3 """Apply reviewed AAAI admissions while preserving existing records verbatim.""" import csv import hashlib import json import re from pathlib import Path from finance_taxonomy_rules import classify ROOT = Path(__file__).resolve().parents[1] DEST = ROOT / 'checks/aaai_expansion' def hash_record(record): return hashlib.sha256(json.dumps(record, ensure_ascii=False, sort_keys=True).encode()).hexdigest() def sanitize(value): """Remove PDF extraction control characters that cannot be stored in Excel.""" if isinstance(value, str): return re.sub(r'[\x00-\x08\x0b\x0c\x0e-\x1f]', '', value) if isinstance(value, list): return [sanitize(item) for item in value] if isinstance(value, dict): return {key: sanitize(item) for key, item in value.items()} return value def main(): if not (ROOT / 'finance_catalog.csv').exists(): raise SystemExit('Historical import tool: use scripts/build_collection.py for the separated long-paper collection.') with (ROOT / 'finance_catalog.csv').open(encoding='utf-8-sig', newline='') as stream: reader = csv.DictReader(stream) columns = reader.fieldnames rows = {row['id']: row for row in reader} records = {r['id']: r for r in map(json.loads, (ROOT / 'finance_catalog.jsonl').read_text().splitlines())} assignments_path = ROOT / 'checks/finance_taxonomy/assignments.json' assignments = json.loads(assignments_path.read_text()) overrides_path = ROOT / 'checks/finance_taxonomy/manual_overrides.json' overrides = json.loads(overrides_path.read_text()) baseline = DEST / 'baseline_record_hashes.json' if not baseline.exists(): baseline.write_text(json.dumps({ 'source_commit': 'e7abe45cccd7eaeee41692249e34f180d5fa3236', 'papers': {pid: {'csv_record': hash_record(row), 'jsonl_record': hash_record(records[pid]), 'assignment': hash_record(assignments[pid])} for pid, row in rows.items()}, }, indent=2) + '\n') schema = json.loads((ROOT / 'checks/finance_taxonomy/schema.json').read_text()) definitions = {kind: {item['id']: item['label'] for item in schema[kind]} for kind in ('topics', 'subtopics', 'methods')} candidates = {p['id']: p for p in json.loads((DEST / 'candidates.json').read_text())} decisions = sanitize(json.loads((DEST / 'review_decisions.json').read_text())) (DEST / 'review_decisions.json').write_text(json.dumps(decisions, ensure_ascii=False, indent=2) + '\n') pdfs = {p['id']: p for p in json.loads((DEST / 'pdf_results.json').read_text())} for pid, decision in decisions.items(): paper = candidates[pid] download = pdfs.get(pid, {'id': pid, 'status': 'not_downloaded', 'local_path': ''}) assignment = classify(paper, decision, decision.get('taxonomy_override')) if decision.get('taxonomy_override'): overrides[pid] = decision['taxonomy_override'] assignments[pid] = assignment finance = { **decision, 'relevance_zh': ('以金融任务或金融数据为核心研究对象。' if decision['scope'] == 'core' else '通用方法包含已核验的金融数据、实验或场景建模。'), 'primary_topic': definitions['topics'][assignment['primary_topic_id']], 'topic_ids': assignment['topic_ids'], 'topics': [definitions['topics'][i] for i in assignment['topic_ids']], 'subtopic_ids': assignment['subtopic_ids'], 'subtopics': [definitions['subtopics'][i] for i in assignment['subtopic_ids']], 'method_ids': assignment['method_ids'], 'methods': [definitions['methods'][i] for i in assignment['method_ids']], 'taxonomy_version': schema['version'], } code_urls = sorted(set(re.findall(r'https?://(?:github\.com|gitlab\.com)/[^\s<>\\)]+', paper['abstract']))) record = {key: paper[key] for key in ['id', 'title', 'authors', 'venue', 'year', 'abstract', 'paper_url', 'pdf_url', 'doi', 'volume', 'issue', 'proceedings_pages']} record.update(track='main', abstract_source=paper['metadata_sources'][0], booktitle='Proceedings of the AAAI Conference on Artificial Intelligence', metadata_sources=paper['metadata_sources'] + [decision['main_track_evidence']], awards=[], award_evidence=[], oral=False, oral_evidence=[], presentation_original=[], code_urls=code_urls, finance=finance, download=download, research={'primary_topic_id': 'finance' if finance['scope'] == 'core' else None, 'topic_ids': ['finance'], 'subtopic_ids': [], 'review_status': 'finance_only_classification'}) records[pid] = record row = {key: '' for key in columns} row.update(id=pid, title=paper['title'], authors='; '.join(paper['authors']), venue='AAAI', year=str(paper['year']), track='main', research_primary_topic='AI for Finance / 金融人工智能' if finance['scope'] == 'core' else '', research_topics='AI for Finance / 金融人工智能', research_topic_ids='finance', research_review_status='finance_only_classification', is_best_or_outstanding='False', is_oral='False', oral_candidate_unverified='False', finance_scope=finance['scope'], finance_primary_topic=finance['primary_topic'], finance_topics='; '.join(finance['topics']), finance_subtopics='; '.join(finance['subtopics']), finance_methods='; '.join(finance['methods']), finance_topic_ids='; '.join(finance['topic_ids']), finance_relevance_zh=finance['relevance_zh'], finance_evidence_type=finance['evidence_type'], finance_evidence_excerpt=finance['evidence_excerpt'], paper_url=paper['paper_url'], pdf_url=paper['pdf_url'], download_status=download['status'], code_urls='; '.join(code_urls), metadata_sources='; '.join(record['metadata_sources']), abstract=paper['abstract']) if download['status'] == 'downloaded': row.update({key: str(download[key]) for key in ['local_path', 'pdf_version', 'pages', 'bytes', 'sha256']}) rows[pid] = row note = [f"# {paper['title']}", '', f"AAAI {paper['year']} · main · {finance['scope']}", '', f"- Official paper: {paper['paper_url']}", f"- DOI: https://doi.org/{paper['doi']}", f"- Metadata source: {paper['metadata_sources'][0]}", f"- Main-track evidence: {decision['main_track_evidence']} (volume {paper['volume']}, issue {paper['issue']})", f"- Financial evidence: `{finance['evidence_type']}`", f"- Primary topic: {finance['primary_topic']}", f"- Topics: {'; '.join(finance['topics'])}", f"- Subtopics: {'; '.join(finance['subtopics'])}", f"- Methods: {'; '.join(finance['methods'])}", '- [Classification definitions](../finance_taxonomy.md) · ' '[Classification evidence](../../checks/finance_taxonomy/assignments.json)'] if row['local_path']: note.append(f"- Local PDF: [open](../../{row['local_path']}); version: `{row['pdf_version']}`") else: note.append(f"- Online PDF: {paper['pdf_url']}") if code_urls: note.append('- Code links from the abstract: ' + '; '.join(code_urls)) note.extend(['', '## Financial relevance evidence', '', finance['evidence_excerpt'], '', '## Abstract', '', paper['abstract'], '']) (ROOT / f'indexes/finance_notes/{pid}.md').write_text( '\n'.join(line.rstrip() for line in note).rstrip() + '\n') ordered = sorted(rows.values(), key=lambda r: (r['venue'], int(r['year']), r['title'].casefold())) with (ROOT / 'finance_catalog.csv').open('w', encoding='utf-8-sig', newline='') as stream: writer = csv.DictWriter(stream, fieldnames=columns) writer.writeheader() writer.writerows(ordered) (ROOT / 'finance_catalog.jsonl').write_text(''.join(json.dumps(records[row['id']], ensure_ascii=False) + '\n' for row in ordered)) assignments_path.write_text(json.dumps(assignments, ensure_ascii=False, indent=2) + '\n') overrides_path.write_text(json.dumps(overrides, ensure_ascii=False, indent=2) + '\n') updates = [{'date': '2026-11-10', 'venue': 'AAAI', 'track': 'main', 'years': [2017, 2026], 'added_papers': len(decisions), 'audit': 'checks/aaai_expansion/README.md'}] (ROOT / 'checks/collection_updates.json').write_text(json.dumps(updates, indent=2) + '\n') print(f'Applied {len(decisions)} AAAI records. Collection total: {len(rows)}.') if __name__ == '__main__': main()