Zayden-Jung/paper-atlas
1
0
Fork 0
Code Issues Pull Requests 4 Actions Packages Projects Releases Wiki Activity

按官方证据识别早期 ICML 与 IROS 主系列 #37

Merged
Zayden-Jung merged 1 commits from codex/historical-primary-series into main 2026-11-09 05:19:50 +00:00
Conversation 0 Commits 1 Files Changed 4
+177
4 changed files with 177 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+2
View File
@@ -60,6 +60,8 @@ paper-atlas collect colm --cache /data/paper-index/raw/http-cache \
--out /data/paper-index/collected
```
早期主系列的Workshop名称按[ICML/IROS精确届次证据](docs/coverage/historical-primary-series-2026-11-09.md)识别，保留原名与范围证据；代码支持不代表已补齐生产数据。
`paper-atlas collect --help` 查看来源；`paper-atlas import --help` 查看导入方式。完整批次步骤见[使用指南](docs/usage.md)。
共享实例名为 `foundry-postgres`，数据位于 `foundry_data` 库的 `research` schema；其他项目可按权限复用同一组表。库名通过 libpq service 配置、schema 通过 search_path 或 `--pg-schema` 指定，不写死在查询中。PostgreSQL 凭据放在服务器权限为 `0600` 的 libpq service 文件中；命令行只使用服务名称。`paper-atlas maintain postgres --help` 提供迁移入口：先完整核验 21 张业务表，再发布目标 schema。SQLite FTS5 内部表由 PostgreSQL GIN 索引替代。切换后 SQLite 为只读备份；`import records`、`import arxiv` 和 `maintain backfill` 使用 `--pg-service foundry-research` 写入主库，整个写入会话只允许一个协调写入者，会议文件失败时整体回滚并保留失败批次记录。
docs/coverage/historical-primary-series-2026-11-09.md
+33
View File
@@ -0,0 +1,33 @@
# ICML 与 IROS 早期主系列的来源范围
本轮修正采集阶段的届次识别，不修改任何现有生产投稿轨道或历史。正式主系列曾以Workshop命名，不等于附属workshop；也不能反过来把所有旧workshop升为主会。
## 已核实的精确范围
| 会议 | 论文集键 | 依据与行为 |
| --- | --- | --- |
| ICML | `conf/icml/1988`、`1989`、`1990`、`1991`、`1992` | [官方历史](https://icml.cc/Conferences/2008/past_icmls.shtml.html)将其列为第5–9届，年份与举办地对应固定目录中的完整标题。1989/1991/1992的Workshop名称不再导致主系列被默认排除；1988/1990保留main并增加来源范围证据 |
| IROS | `conf/iros/1988`、`1989`、`1990` | [IEEE RAS官方历届目录](https://www.ieee-ras.org/conferences-workshops/financially-co-sponsored/iros/iros-past-and-future-venues/)列出三届与论文集链接，地点及日期对应固定目录标题；这些是IROS主系列 |
规则同时要求family、booktitle基础别名（ML或IROS）、空轨道后缀、完整精确论文集键、匹配的事件年份及固定完整标题。仍执行原有书目年份核查：与事件年不同且标题无明确事件年时拒绝推断。ML91原始标题没有四位年份，使用精确1991卷键、完整标题、逐篇书目年和官方第8届记录，不从两位标题代码猜世纪。
`1991w`/`1988w`等卷键、booktitle中的Workshops后缀、不同届次/地点/完整标题、未列入的年份均不会获得上述主系列豁免。AISTATS/COLT既有精确规则保留；真正的附属workshop仍受默认不采集workshop的边界约束。
## 原始字段、决定与后续导入
采集器仍保存原始proceedings标题（包括Workshop）、书目年、原始作者名、完整来源书目字段及稳定标识；新增 `track_scope_note` 与官方 `track_scope_evidence_url`。接受状态仍只源自该论文的inproceedings书目证据；不从历史目录生成占位论文或拒稿，也不重新导入已经存在的记录。
新规则的输出必须重新做原始来源/标题/作者/稳定标识与现有库冲突预检，再按增量导入。此前被隔离的ICML1988/1990记录不能仅凭代码规则而直接解除全部冲突。当前正在运行的历史批次使用原固定版本和文件，不能在运行中替换其源码或输入。
## 验证与证据
新增回归覆盖：ICML三届Workshop名称、IROS最初三届、ICML原Conference名称的两届、错误卷键/后缀/标题/未知年份，以及归一化记录经过原生文件导入后保留轨道证据。构建与测试仅在已授权的ces-control远程runner执行，本机仅做语法和文本/JSON一致性检查；结果以相应commit的CI回执为准。
官方HTML在工作证据目录 `paper-index/historical-primary-series-evidence-v1/`，附响应状态、字节数、时间与SHA256；仅取元数据网页，没有下载论文全文。
| 来源 | UTC采样时间 | HTTP | 字节数 | SHA256 |
| --- | --- | ---: | ---: | --- |
| [icml-history](https://icml.cc/Conferences/2008/past_icmls.shtml.html) | 2026-11-08T18:36:43.043135+00:00 | 200 | 11993 | `80cce886fe985551339a52b13c4077acf827e8fcc01555f854dc24077bdebdf0` |
| [iros-history](https://www.ieee-ras.org/conferences-workshops/financially-co-sponsored/iros/iros-past-and-future-venues/) | 2026-11-08T18:36:45.687171+00:00 | 200 | 466059 | `fdb4594621e68567e1099f5c88aae53de835b7b8146025148c3b1968df990697` |
本次不关闭历史年度缺口，也不替代全部导入后的全库去重、来源/年度/track覆盖、摘要和实际查询验收。
paper_atlas/collectors/dblp_scope.py
+27
View File
@@ -54,7 +54,34 @@ HISTORICAL_WORKSHOP_TITLES = {
}
# Exact pinned volume titles corroborated by official series histories. The
# original title (including Workshop) remains in source metadata. In particular
# ML91 has no four-digit year in its title; its exact key/title and bibliographic
# year are checked independently, rather than guessing a title year.
HISTORICAL_PRIMARY_EVIDENCE = {
'icml': ('ML', 'https://icml.cc/Conferences/2008/past_icmls.shtml.html'),
'iros': ('IROS', 'https://www.ieee-ras.org/conferences-workshops/financially-co-sponsored/iros/iros-past-and-future-venues/'),
}
HISTORICAL_PRIMARY_TITLES = {
('icml', 1988): 'Machine Learning, Proceedings of the Fifth International Conference on Machine Learning, Ann Arbor, Michigan, USA, June 12-14, 1988',
('icml', 1989): 'Proceedings of the Sixth International Workshop on Machine Learning (ML 1989), Cornell University, Ithaca, New York, USA, June 26-27, 1989',
('icml', 1990): 'Machine Learning, Proceedings of the Seventh International Conference on Machine Learning, Austin, Texas, USA, June 21-23, 1990',
('icml', 1991): 'Proceedings of the Eighth International Workshop (ML91), Northwestern University, Evanston, Illinois, USA',
('icml', 1992): 'Proceedings of the Ninth International Workshop on Machine Learning (ML 1992), Aberdeen, Scotland, UK, July 1-3, 1992',
('iros', 1988): "IEEE International Workshop on Intelligent Robots and Systems '88, Proceedings. IROS 1988, Tokyo, Japan, October 31 - November 2, 1988",
('iros', 1989): "IEEE International Workshop on Intelligent Robots and Systems '89, Proceedings. IROS 1989, September 4-6, 1989, Tsukuba, Japan",
('iros', 1990): "IEEE International Workshop on Intelligent Robots and Systems '90, Towards a New Frontier of Applications, Proceedings. IROS 1990, July 3-6, 1990, Ibaraki, Japan",
}
def historical_track(family, base, suffix, volume_key, year, title):
primary = HISTORICAL_PRIMARY_EVIDENCE.get(family)
if (primary and not suffix and base.casefold() == primary[0].casefold()
and volume_key == f'conf/{family}/{year}'
and title == HISTORICAL_PRIMARY_TITLES.get((family, year))):
return ('main',
'Historical primary series; exact proceedings key, complete title and bibliographic/event year corroborated by official series history. Workshop in this original title is not an affiliated track.',
primary[1])
if suffix or base.casefold() != family or volume_key != f'conf/{family}/{year}':
return None
prefix = HISTORICAL_WORKSHOP_TITLES.get((family, year))
tests/test_historical_primary_series.py
+115
View File
@@ -0,0 +1,115 @@
"""Regression coverage for primary series once named Workshop, with strict scope boundaries."""
from contextlib import closing
import hashlib
import json
from pathlib import Path
import sqlite3
import tempfile
import unittest
from paper_atlas.collectors.dblp_records import normalize_row
from paper_atlas.collectors.dblp_scope import scope
from paper_atlas.ingestion.records import ingest_file
ICML_HISTORY = 'https://icml.cc/Conferences/2008/past_icmls.shtml.html'
IROS_HISTORY = 'https://www.ieee-ras.org/conferences-workshops/financially-co-sponsored/iros/iros-past-and-future-venues/'
class HistoricalPrimarySeriesTests(unittest.TestCase):
def example(self, family, year, title, book=None, key=None):
key = key or f'conf/{family}/{year}'
book = book or ('ML' if family == 'icml' else 'IROS')
row = {'record_type':'inproceedings', 'key':f'conf/{family}/Paper{year}', 'publtype':None,
'fields':{'title':['An independently identified historical research paper'],
'author':['Jane Doe'], 'booktitle':[book], 'year':[str(year)],
'crossref':[key], 'ee':[f'https://doi.org/10.1234/{family}.{year}']}}
volume = {'record_type':'proceedings', 'key':key,
'fields':{'title':[title], 'year':[str(year)]}}
return row, volume
def test_icml_workshop_primary_editions_are_not_silently_filtered(self):
examples = {
1989:'Proceedings of the Sixth International Workshop on Machine Learning (ML 1989), Cornell University, Ithaca, New York, USA, June 26-27, 1989',
1991:'Proceedings of the Eighth International Workshop (ML91), Northwestern University, Evanston, Illinois, USA',
1992:'Proceedings of the Ninth International Workshop on Machine Learning (ML 1992), Aberdeen, Scotland, UK, July 1-3, 1992',
}
for year, title in examples.items():
with self.subTest(year=year):
row, volume = self.example('icml', year, title)
record, reason = normalize_row(row, {volume['key']:volume}, {'ICML'})
self.assertIsNone(reason)
self.assertEqual((record['venue'],record['year'],record['track'],record['status']), ('ICML',year,'main','accepted'))
self.assertEqual(record['metadata']['proceedings_title'], title)
self.assertEqual(record['metadata']['track_scope_evidence_url'], ICML_HISTORY)
self.assertEqual(record['metadata']['bibliographic_year'], year)
def test_iros_first_three_editions_are_primary_series(self):
examples = {
1988:"IEEE International Workshop on Intelligent Robots and Systems '88, Proceedings. IROS 1988, Tokyo, Japan, October 31 - November 2, 1988",
1989:"IEEE International Workshop on Intelligent Robots and Systems '89, Proceedings. IROS 1989, September 4-6, 1989, Tsukuba, Japan",
1990:"IEEE International Workshop on Intelligent Robots and Systems '90, Towards a New Frontier of Applications, Proceedings. IROS 1990, July 3-6, 1990, Ibaraki, Japan",
}
for year, title in examples.items():
with self.subTest(year=year):
row, volume = self.example('iros', year, title)
record, reason = normalize_row(row, {volume['key']:volume}, {'IROS'})
self.assertIsNone(reason)
self.assertEqual((record['venue'],record['year'],record['track']), ('IROS',year,'main'))
self.assertEqual(record['metadata']['proceedings_title'], title)
self.assertEqual(record['metadata']['track_scope_evidence_url'], IROS_HISTORY)
def test_main_named_icml_predecessors_also_carry_primary_scope_provenance(self):
examples = {
1988:'Machine Learning, Proceedings of the Fifth International Conference on Machine Learning, Ann Arbor, Michigan, USA, June 12-14, 1988',
1990:'Machine Learning, Proceedings of the Seventh International Conference on Machine Learning, Austin, Texas, USA, June 21-23, 1990',
}
for year, title in examples.items():
row, volume = self.example('icml', year, title)
edition, reason = scope(row, volume)
self.assertIsNone(reason)
self.assertEqual(edition['track'], 'main')
self.assertEqual(edition['track_scope_evidence_url'], ICML_HISTORY)
def test_wrong_keys_titles_suffixes_and_unknown_years_keep_workshop_boundary(self):
title = 'Proceedings of the Eighth International Workshop (ML91), Northwestern University, Evanston, Illinois, USA'
for year, changed, book, key in [
(1991,title,'ML','conf/icml/1991w'),
(1991,title,'ML (Workshops)','conf/icml/1991'),
(1991,title.replace('Eighth','Ninth'),'ML','conf/icml/1991'),
(1991,title + ', Affiliated Workshop','ML','conf/icml/1991'),
(1987,'Machine Learning Workshop 1987, Irvine','ML','conf/icml/1987'),
]:
with self.subTest(year=year, book=book, key=key, title=changed):
row, volume = self.example('icml',year,changed,book,key)
record, reason = normalize_row(row,{volume['key']:volume},{'ICML'})
self.assertIsNone(record)
self.assertEqual(reason,'separate_workshop')
self.assertNotIn('track_scope_evidence_url',scope(row,volume)[0])
title = "IEEE International Workshop on Intelligent Robots and Systems '88, Proceedings. IROS 1988, Tokyo, Japan, October 31 - November 2, 1988"
for changed, book, key in [
(title,'IROS','conf/iros/1988w'),
(title,'IROS (Workshops)','conf/iros/1988'),
(title.replace('Tokyo','Another city'),'IROS','conf/iros/1988'),
]:
row, volume = self.example('iros',1988,changed,book,key)
self.assertEqual(normalize_row(row,{volume['key']:volume},{'IROS'})[1],'separate_workshop')
def test_primary_scope_proof_survives_native_file_ingestion(self):
title = 'Proceedings of the Eighth International Workshop (ML91), Northwestern University, Evanston, Illinois, USA'
row, volume = self.example('icml',1991,title)
record, reason = normalize_row(row,{volume['key']:volume},{'ICML'})
self.assertIsNone(reason)
with tempfile.TemporaryDirectory() as temporary:
root=Path(temporary);path=root/'records.jsonl';path.write_text(json.dumps(record)+'\n')
path.with_suffix('.manifest.json').write_text(json.dumps({'source':'dblp','count':1,'list_url':ICML_HISTORY,'file_sha256':hashlib.sha256(path.read_bytes()).hexdigest()}))
ingest_file(root/'test.sqlite',path)
with closing(sqlite3.connect(root/'test.sqlite')) as connection:
self.assertEqual(connection.execute('SELECT track FROM editions').fetchone()[0],'main')
self.assertEqual(connection.execute('SELECT status FROM decisions').fetchone()[0],'accepted')
metadata=json.loads(connection.execute('SELECT metadata_json FROM source_records').fetchone()[0])
self.assertEqual(metadata['proceedings_title'],title)
self.assertEqual(metadata['track_scope_evidence_url'],ICML_HISTORY)
if __name__ == '__main__':
unittest.main()