Handle MLSys year-labelled proceedings and bridge retained publisher evidence

docs/data-sources.md

@@ -52,7 +52,7 @@ paper-atlas collect mlsys --cache /data/paper-index/raw/http-cache \
 
 每年清单核对官方声明数量，缺摘要标为 partial；未知现代轨道、空目录或读取失败不能记为成功。`SysML 2019` 归于同一 MLSys 系列并保留别名。2018 年官方说明接受的投稿为 non-archival，论文集从 2019 年开始；不拼造 2018 年归档卷，也不把未出现在论文集中的投稿判为拒稿。参见 [2018 年官方说明](https://mlsys.org/Conferences/2018/)。
 
-MLSys 老式 `/paper/<year>/hash/…` 与新式 `/paper_files/paper/<year>/hash/…` 链接规范为同一个 `publication_mlsys` 标识；DBLP 和官方摘要源共用它，避免同一发表记录成为两个论文实体或两次投稿。只从自身出版社链接提取，不从摘要参考文献提取，也不把发表标识用于合并某次拒稿尝试。共享投稿身份入口要求 MLSys 源具有有效的同年出版社身份，缺失或错年的文件不能生成投稿事件。
+MLSys 老式 `/paper/<year>/hash/…` 与新式 `/paper_files/paper/<year>/hash/…` 链接规范为同一个 `publication_mlsys` 标识，也支持 2023 年实际使用的 `-Abstract-mlsys2023.html` 及对应年份标签；标签年份必须与目录年份一致。DBLP 和官方摘要源共用它，避免同一发表记录成为两个论文实体或两次投稿。只从自身出版社链接提取，不从摘要参考文献提取，也不把发表标识用于合并某次拒稿尝试。共享投稿身份入口要求 MLSys 源具有有效的同年出版社身份，缺失或错年的文件不能生成投稿事件；旧版 DBLP 整理文件可利用已保留的唯一自身出版社链接补得此身份，不需要丢弃来源或依靠标题猜测。
 
 ACM MM 使用固定 DBLP 原始快照和已核验的 proceedings crossref 盘点历史，主会 `ACM Multimedia`、历史编号分卷与同系列 workshop 分开。主会卷也包含 Doctoral Symposium、Reproducibility Companion Papers、Tutorials 等章节（参见 [2025 年目录](https://dblp.org/db/conf/mm/mm2025.html)），而原始 XML 不含逐篇章节，所以这类记录暂标 `proceedings_unspecified`，不能整卷当作 regular 主轨道；后续需逐篇目录证据补齐。旧版候选 JSONL 没有 `mm` / `mlsys` 家族，不能用它假装覆盖新增会议；应从已经下载并校验的原始 XML 重新筛选这两个会议，保留完整扫描数量、卷证据与遗漏原因。DBLP 没有摘要，后续须通过 DOI、真实 forum 等独立标识关联官方或公开元数据，不能将书目导入成功表述为摘要已齐全。
 

paper_atlas/collectors/proceedings.py

@@ -84,7 +84,8 @@ def parse_paper(html,url,source,year,index_url):
         identifiers['publication_mlsys'] = own
         if track_text:
             explicit = {'conference': 'main', 'main track': 'main', 'research track': 'main',
-                        'industry track': 'industry', 'industrial track': 'industry'}
+                        'industry track': 'industry', 'industrial track': 'industry',
+                        'mlsys' + str(year): 'main'}
             if track_text.casefold() not in explicit:
                 raise ValueError('Unclassified official MLSys track: ' + track_text)
             track = explicit[track_text.casefold()]

paper_atlas/db/submissions.py

@@ -1,6 +1,7 @@
 """Canonical submission events with independent source and decision evidence."""
 from datetime import datetime, timezone
 import re
+from paper_atlas.io.publication_ids import mlsys_id
 
 
 PUBLICATIONS={'pmlr','acl_anthology','colm','neurips','mlsys','cvf','ecva','dblp'}
@@ -25,8 +26,16 @@ def identities_for_row(row):
     if row['source'] in PUBLICATIONS and row.get('status')=='accepted' and own.get('doi'):
         doi=re.sub(r'^https?://(?:dx\.)?doi\.org/','',str(own['doi']),flags=re.I).strip().lower()
         if re.fullmatch(r'10\.\d{4,9}/\S+',doi):identities.append({'scheme':'publication_doi','value':doi})
-    if row['source'] in PUBLICATIONS and row.get('status')=='accepted' and own.get('publication_mlsys'):
-        identities.append({'scheme':'publication_mlsys','value':own['publication_mlsys']})
+    publication_mlsys = own.get('publication_mlsys')
+    if not publication_mlsys and row['source']=='dblp' and row['venue']=='MLSys':
+        # Older normalized DBLP files retain their own publisher links even
+        # when a newly supported historical URL label was not yet parsed.
+        links = (row.get('metadata') or {}).get('bibliography', {}).get('ee', [])
+        candidates = {identity for link in links if (identity := mlsys_id(link))}
+        if len(candidates)==1:
+            publication_mlsys = candidates.pop()
+    if row['source'] in PUBLICATIONS and row.get('status')=='accepted' and publication_mlsys:
+        identities.append({'scheme':'publication_mlsys','value':publication_mlsys})
     meta=row.get('metadata') or {}
     if row['source']=='iclr_program' and row['venue']=='ICLR' and meta.get('program_identity_type')=='submission' and meta.get('program_submission_number_verified') is True:
         identities.append({'scheme':'iclr_number','value':str(meta['program_source_id'])})

paper_atlas/io/publication_ids.py

@@ -7,5 +7,7 @@ def mlsys_id(url):
     parsed = urlsplit(url)
     if parsed.scheme not in ('https', 'http') or parsed.netloc != 'proceedings.mlsys.org':
         return None
-    match = re.fullmatch(r'/paper(?:_files/paper)?/((?:19|20)\d{2})/hash/([a-f0-9]{32})-Abstract(?:-Conference)?\.html', parsed.path)
-    return match[1] + '/' + match[2] if match else None
+    match = re.fullmatch(r'/paper(?:_files/paper)?/((?:19|20)\d{2})/hash/([a-f0-9]{32})-Abstract(?:-(?:Conference|mlsys(20\d{2})))?\.html', parsed.path)
+    if not match or match[3] and match[3] != match[1]:
+        return None
+    return match[1] + '/' + match[2]

tests/test_mlsys_multimedia.py

@@ -55,6 +55,28 @@ class MlsysMultimediaTests(unittest.TestCase):
         with self.assertRaisesRegex(ValueError, 'non-archival'):
             parse_paper(html, page(2018), 'mlsys', 2018, ROOT)
 
+    def test_year_label_and_retained_bibliography_links_bridge_the_2023_format(self):
+        url = page(2023).replace('-Conference.html', '-mlsys2023.html')
+        self.assertEqual(mlsys_id(url), '2023/' + HASH)
+        self.assertIsNone(mlsys_id(url.replace('mlsys2023.html', 'mlsys2022.html')))
+        official = parse_paper('<h1>Updated title</h1><span class="paper-track">mlsys2023</span><p class="paper-abstract">Official abstract</p>', url, 'mlsys', 2023, ROOT)
+        self.assertEqual(official['track'], 'main')
+        legacy = {'source': 'dblp', 'external_id': 'conf/mlsys/paper23', 'url': 'https://dblp.org/rec/conf/mlsys/paper23',
+                  'venue': 'MLSys', 'year': 2023, 'track': 'main', 'status': 'accepted', 'title': 'Bibliographic title',
+                  'abstract': None, 'authors': [], 'identifiers': {}, 'metadata': {'bibliography': {'ee': [url]}}}
+        self.assertEqual(identities_for_row(legacy), [('publication_mlsys', '2023/' + HASH)])
+        self.assertEqual(identities_for_row({**legacy, 'status': 'unknown'}), [])
+        with tempfile.TemporaryDirectory() as tmp:
+            root = Path(tmp); db = root / 'papers.sqlite'
+            for row in [legacy, official]:
+                path = root / (row['source'] + '.jsonl'); path.write_text(json.dumps(row) + '\n')
+                path.with_suffix('.manifest.json').write_text(json.dumps({'source': row['source'], 'parser_version': 2, 'list_url': ROOT, 'count': 1}))
+                ingest_file(db, path)
+            with closing(sqlite3.connect(db)) as c:
+                self.assertEqual(c.execute('SELECT count(*) FROM papers').fetchone()[0], 1)
+                self.assertEqual(c.execute('SELECT count(*) FROM submissions').fetchone()[0], 1)
+                self.assertEqual(c.execute('SELECT count(*) FROM submission_sources').fetchone()[0], 2)
+
     def test_legacy_mlsys_and_sysml_alias_preserve_same_publication_identity(self):
         self.assertEqual(mlsys_id(page(2019, False)), mlsys_id(page(2019)))
         self.assertIsNone(mlsys_id(page().replace('proceedings.mlsys.org', 'example.org')))