FL-paper-update-tracker

May 20, 2026 · View on GitHub

This project is a part of Awesome-FL. Stars

An automated paper tracking bot for Federated Learning research. It periodically queries the DBLP API for new publications across 40+ top-tier conferences and journals, deduplicates entries by their electronic edition (ee) field and by title, and automatically opens GitHub Issues to notify subscribers.

Features

  • Broad Coverage: Monitors 40+ leading venues in AI, ML, CV, NLP, Systems, Security, and more.
  • Smart Deduplication: Uses the ee (electronic edition) field and the title field to eliminate duplicate records caused by minor author-name variations or multiple URLs for the same paper in DBLP.
  • Year-Based Filtering: Only tracks papers published within the last 3 years and the next 1 year (e.g., 2023–2027 when running in 2026).
  • Auto-Notification: Creates nicely formatted GitHub Issues daily via GitHub Actions.
  • Automatic Abstract Fetching: Newly discovered papers automatically retrieve abstracts from Crossref, Semantic Scholar, arXiv, and OpenAlex (in that order) to enrich the cache.
  • Automatic Chinese Translation: Successfully retrieved abstracts are automatically translated into Chinese via Qwen-MT-plus and stored as abstract_cn.
  • Automatic Related Code Extraction: Scans fetched abstracts for GitHub repository links and stores the first match as related_code.
  • Automatic DOI Backfill: A standalone script is provided to backfill missing DOIs via DBLP, Crossref, and Semantic Scholar.

Extending to Other Research Domains

This tracker is not limited to Federated Learning. The domain keywords are controlled by the keywords list in config.yaml:

dblp:
  keywords:
    - federate                # primary keyword
    - grad inversion               # secondary keyword (optional)
  queries:
    - "venue:IJCAI:"
    - "venue:ICML:"
    # ... adjust venue list as needed

To switch to a new domain (e.g., diffusion models, LLM, reinforcement learning):

  1. Change the keywords — edit config.yaml → dblp.keywords. The first keyword is the primary keyword; during automatic runs, secondary keywords are only scanned on venues where the primary keyword discovered new papers, reducing API load.
  2. Adjust the venue list — keep, add, or remove entries under dblp.queries.
  3. Update category mappings — edit scripts/convert_cache_to_md.py so that venue names and categories match your new domain.
  4. Reset the cache — delete or rename cached/dblp.yaml so the next run treats every paper as new.

No other code changes are required.

Supported Conferences & Journals

AI & Machine Learning — IJCAI, AAAI, AISTATS, ALT, AI, NeurIPS, ICML, ICLR, COLT, UAI, Machine Learning, JMLR, IEEE TPAMI

Data Mining, Information Retrieval & Web — KDD, WSDM, SIGIR, WWW

Computer Vision & Multimedia — CVPR, ICCV, ECCV, ACM MM, IJCV

Natural Language Processing — ACL, NAACL-HLT, EMNLP, COLING

Security & Privacy — IEEE S&P, CCS, USENIX Security, NDSS

Systems, Architecture & Databases — OSDI, SOSP, ISCA, MLSys, EuroSys, SIGMOD, ICDE, VLDB, ACM TOCS, ACM TOS

Networking — SIGCOMM, INFOCOM, MobiCom, NSDI

Other Related Venues — DAC, IEEE TPDS, IEEE TCAD, IEEE TC, ICSE, FOCS, STOC

The venue list can be customized. For details, see TECHNICAL.md.

How to Get Notifications

Watch the Main Repository 🚀

Click Watch on Awesome-FL to receive updates on new papers and major changes.

Track Individual Issues (Optional)

Watch this repository if you want real-time alerts on paper submissions.

Note: Issues will be closed once merged into the main repo.

Technical Details

For installation, configuration, workflow mechanics, and local development, please refer to TECHNICAL.md.

Thanks

This repository is based on dblp-watcher. We use DBLP API to search papers and construct a paper update tracker for federated learning.

FL-paper-update-tracker（中文版）

本项目是 Awesome-FL 的配套子项目。

一个面向 联邦学习 研究的自动化论文追踪机器人。它定期通过 DBLP API 查询 40 余个顶级会议与期刊的新发表论文，通过 ee（电子版链接）和 title（标题）字段智能去重，并自动创建 GitHub Issue 通知订阅者。

功能特性

  • 覆盖广泛：持续监控人工智能、机器学习、计算机视觉、自然语言处理、系统、安全等领域的 40+ 顶级学术会议与期刊。
  • 智能去重：利用 ee（电子版链接）和 title（标题）字段消除 DBLP 中因作者名称微差异或同一论文多版本链接导致的重复记录。
  • 年份过滤：仅追踪近三年及未来一年内发表的论文（例如 2026 年运行时，保留 2023–2027 年的论文）。
  • 自动通知：通过 GitHub Actions 每日自动生成格式化的 GitHub Issue。
  • 自动获取摘要：新发现的论文会自动从 Crossref、Semantic Scholar、arXiv 和 OpenAlex 依次查询并补充摘要，以丰富缓存数据。
  • 自动中文翻译：成功获取的英文摘要会自动通过 Qwen-MT-plus 翻译为中文，存储为 abstract_cn 字段。
  • 自动提取代码链接：从获取的摘要中扫描 GitHub 仓库链接，并将首个匹配结果存储为 related_code。
  • 自动 DOI 回填：提供独立脚本，通过 DBLP、Crossref 和 Semantic Scholar 回填缺失的 DOI。

扩展到其他研究领域

本追踪器不限于联邦学习。领域关键词由 config.yaml 中的列表控制：

dblp:
  keywords:
    - federate                # 主关键词
    - grad inversion              # 次关键词（可选）
  queries:
    - "venue:IJCAI:"
    - "venue:ICML:"
    # ... 按需增删会议/期刊

切换到新领域（例如 扩散模型、大语言模型、强化学习）的步骤：

  1. 修改关键词 — 编辑 config.yaml → dblp.keywords。第一个关键词为主关键词；在自动运行时，次关键词仅会在主关键词发现新论文的会议/期刊上进行扫描，以减少 API 调用。
  2. 调整会议/期刊列表 — 在 dblp.queries 中保留、添加或删除条目。
  3. 更新类别映射 — 编辑 scripts/convert_cache_to_md.py，使其 venue 名称与类别匹配新领域。
  4. 重置缓存 — 删除或重命名 cached/dblp.yaml，下次运行将重新抓取全部论文。

无需修改其他代码。

支持的会议与期刊

人工智能与机器学习 — IJCAI、AAAI、AISTATS、ALT、AI、NeurIPS、ICML、ICLR、COLT、UAI、Machine Learning、JMLR、IEEE TPAMI

数据挖掘、信息检索与网络 — KDD、WSDM、SIGIR、WWW

计算机视觉与多媒体 — CVPR、ICCV、ECCV、ACM MM、IJCV

自然语言处理 — ACL、NAACL-HLT、EMNLP、COLING

安全与隐私 — IEEE S&P、CCS、USENIX Security、NDSS

系统、体系结构与数据库 — OSDI、SOSP、ISCA、MLSys、EuroSys、SIGMOD、ICDE、VLDB、ACM TOCS、ACM TOS

计算机网络 — SIGCOMM、INFOCOM、MobiCom、NSDI

其他相关领域 — DAC、IEEE TPDS、IEEE TCAD、IEEE TC、ICSE、FOCS、STOC

追踪列表支持自定义，详情请见 TECHNICAL.md。

如何获取通知

关注主仓库 🚀

点击 Awesome-FL 的 Watch 按钮，以接收新论文和重要更新的通知。

关注本仓库的 Issue（可选）

如果你希望获得论文提交的实时提醒，可以 Watch 本仓库。

注意：Issue 会在论文合并到主仓库后关闭。

技术细节

有关安装部署、配置文件、工作流程机制及本地开发说明，请参阅 TECHNICAL.md。

致谢

本仓库基于 dblp-watcher 构建。我们使用 DBLP API 检索论文，并构建了面向联邦学习的论文更新追踪器。