Paper Discovery

Quick Start

Use the bundled script first when the user asks for paper discovery:

python .\scripts\discover_papers.py --query "short video multimodal user engagement health emergency" --years 2020:2026 --max-results 30 --format md --output papers.md

For journal-specific searches, pass the journal ISSN and name. For Information Processing & Management:

python .\scripts\discover_papers.py --query "short video multimodal comments user perception engagement" --years 2020:2026 --issn 0306-4573 --journal "Information Processing & Management" --max-results 25 --format md --output ipm-papers.md

If shell network access is blocked, use available browsing/fetch tools to query the same sources manually, then follow the same curation criteria below.

Workflow

1. Translate the user request into 2-4 concrete query variants.

   • Include domain terms: e.g. short video, multimodal, comments, health information, emergency, engagement.
   • Include method terms only when relevant: e.g. LLM, NLP, sentiment analysis, computer vision.
   • For a target journal, use ISSN over journal-name text matching.

2. Run scripts/discover_papers.py.

   • Use crossref for DOI-backed journal metadata and journal/ISSN filters.
   • Add openalex or semantic_scholar for broader discovery when the user did not restrict to a specific journal.
   • Use --years START:END for date limits.
   • Use --min-score only after checking whether the first pass is too broad.

3. Review and curate.

   • Prefer papers with a DOI, clear journal metadata, and a close conceptual match.
   • Deduplicate by DOI first, then normalized title.
   • Keep papers that help the user's argument, not just keyword matches.
   • Flag uncertain metadata, online-first records, or source conflicts.

4. Report results.

   • Include APA-style citation, DOI link, relevance reason, and how the paper can be used.
   • If the user needs manuscript integration, group papers by section: Introduction, Related Work, Methods, Discussion.
   • If the user asks for files, write a Markdown bibliography and optionally JSON metadata.

Script Reference

scripts/discover_papers.py supports:

• --query: search query string.
• --years: year range such as 2020:2026, 2024, or :2026.
• --sources: any of crossref, openalex, semantic_scholar; defaults to all three unless an ISSN is supplied, where Crossref is usually best.
• --journal: journal name filter or preference.
• --issn: ISSN filter, recommended for journal-specific discovery.
• --max-results: maximum curated results after deduplication.
• --min-score: minimum local relevance score.
• --format: md or json.
• --output: output file path.

Read references/sources.md only when you need source-specific details or troubleshooting.

Quality Rules

• Do not present search results as verified if metadata was not checked.
• For final bibliographies, prefer DOI URLs in https://doi.org/... form.
• Use publication year from the DOI metadata when source records disagree, and mention uncertainty for online-first records.
• When the target is a specific journal, verify every retained result has that journal in metadata.
• Avoid padding the list. If fewer than requested papers are truly relevant, say so and explain the search coverage.