Skip to content

ROADMAP.md

Latest commit

 

History

History
132 lines (103 loc) · 4.17 KB

ROADMAP.md

File metadata and controls

132 lines (103 loc) · 4.17 KB
Raw

ResearchScope — Roadmap

v0.1 — MVP (current)

Data sources:

  • arXiv preprints (via arxiv package + Atom API fallback)
  • ACL Anthology (search API + venue JSON fallback)

Intelligence pipeline:

  • Normalised Paper/Author/Lab/University schema
  • Near-duplicate detection (Jaccard title similarity)
  • Keyword-based topic tagging (25+ topics)
  • Paper type detection (survey, benchmark, dataset, methods, …)
  • L1–L4 difficulty classification with plain-English reason
  • 4-score system: paper_score, read_first_score, content_potential, momentum
  • Score breakdowns + transparent reasoning
  • Author aggregation with momentum scoring
  • Lab / industry aggregation
  • University aggregation
  • 3-layer research gap engine (explicit, pattern, starter)
  • Creator content generation (tweet, LinkedIn, newsletter, video script)
  • Daily editorial queue

Frontend:

  • Homepage with stats, top papers, editorial digest
  • Papers page (filterable, sortable, score breakdown)
  • Topics page (reading paths, starter/frontier packs)
  • Research Gaps page (3 gap types, starter ideas)
  • Authors page (momentum, topics, profile)
  • Labs & Universities page (output, momentum)
  • Conferences page
  • Dark mode

Infrastructure:

  • GitHub Actions daily pipeline
  • GitHub Pages static deployment
  • Config-driven weights and venues

v0.2 — Conference Expansion

  • OpenReview connector — ICLR, NeurIPS, ICML, COLM with oral/spotlight/poster acceptance-tier signals (feeds quality_hint; oral/spotlight surfaced as paper badges)
  • PMLR connector — ICML proceedings
  • Per-conference explorer pages with theme detection
  • Best paper award tracking
  • Acceptance rate context per venue
  • Conference comparison view

v0.3 — Semantic Intelligence

  • Sentence-transformer embeddings for semantic dedup and clustering
    • Replace Jaccard with cosine similarity
    • Better topic clustering beyond keywords
  • Citation network via Semantic Scholar API
    • Real citation counts
    • Influential citations
    • Paper ancestry / descendant graphs
  • Paper lifecycle tracking
    • Link arXiv preprint → accepted conference version
    • "First seen / accepted at" timeline
    • Maturity stage auto-classification

v0.4 — LLM Enrichment (optional)

  • Optional LLM-powered enrichment stage (requires API key)
    • Better summaries from full PDF text
    • Richer "why it matters" explanations
    • Automatic limitations extraction from PDFs
  • PDF parsing via pymupdf4llm or similar
  • Preserve rule-based fallback when no API key is set

v0.5 — Learning System

  • Curated reading paths per topic (hand-edited + auto-generated)
  • Prerequisites graph visualisation
  • "Start here" vs "Go deeper" track separation
  • Paper difficulty quiz / self-assessment
  • User reading history (localStorage)

v0.6 — Creator Tools

  • Creator dashboard — per-paper content package download
  • Email newsletter template export
  • Slides outline generator
  • "Explain to non-expert" generator

v0.7 — Community & Personalisation

  • Topic subscriptions (RSS feed per topic)
  • Custom watchlists (stored in localStorage)
  • "Papers like this one" recommendations
  • Author follow + alert

Future Conference Connectors

Conference Area Status
ICLR ML Planned (v0.2)
NeurIPS ML Planned (v0.2)
ICML ML Planned (v0.2)
COLM LM Planned (v0.2)
CVPR CV Planned
ICCV CV Planned
ECCV CV Planned
CHI HCI Planned
SIGIR IR Planned
AAAI AI Planned
IJCAI AI Planned
WWW Web Planned

Known Limitations (MVP)

  • Affiliation extraction is heuristic (keyword matching) — no ground-truth author-institution database
  • Deduplication uses title Jaccard similarity — may miss paraphrased titles
  • All tagging and difficulty are rule-based — LLM enrichment planned for v0.4
  • ACL search API availability is not guaranteed — falls back gracefully to empty
  • No persistent state between pipeline runs (each run is a full refresh)