beita6969 Scientific Retrieval & Recommendation
Purpose
Retrieve relevant documents, datasets, and resources from large scientific and domain-specific archives.
Key Datasets
- Financial Reports SEC (JanosAudran/financial-reports-sec): SEC 10-K filings with 20 sections and sentiment labels
- Historical Newswire (dell-research-harvard/newswire): Historical news article corpus for digital humanities research
Protocol
- Query analysis — Parse information need, identify key concepts and constraints
- Source selection — Choose appropriate databases and archives
- Search execution — Multi-strategy search (keyword, semantic, citation-based)
- Relevance ranking — Score and rank results by relevance, authority, recency
- Result synthesis — Organize and present findings with metadata
Retrieval Domains
- Financial documents: SEC filings (10-K, 10-Q, 8-K), earnings calls, analyst reports
- Historical archives: Newspapers, government records, digitized manuscripts
- Scientific literature: Journal articles, preprints, conference proceedings
- Patent databases: USPTO, EPO, WIPO patent documents
Recommendation Types
- Similar documents: Find related papers/reports based on content similarity
- Citation chain: Forward/backward citation tracking
- Cross-domain: Find analogous work in different disciplines
- Temporal: Track how a topic evolves over time
Rules
- Always report search coverage and potential gaps
- Rank by relevance, not just recency
- Include document metadata (date, source, section, author)
- For financial documents, note the filing period and any restatements