beita6969 Knowledge Discovery & Graphs
Purpose
Discover hidden patterns, build knowledge graphs, and extract novel insights from structured and unstructured data.
Key Datasets
- WALS (wals.info): World Atlas of Language Structures — 192 linguistic features across 2,679 languages in CLDF format (CC-BY 4.0)
- HistWords (nlp.stanford.edu/projects/histwords): Historical word embeddings tracking semantic change across 4 languages over centuries (.npy/.pkl format)
Protocol
- Data exploration — Profile data, identify patterns, check distributions
- Feature engineering — Create derived features, temporal features, cross-references
- Pattern detection — Apply clustering, association rules, anomaly detection
- Knowledge graph construction — Build entity-relation graphs from discovered patterns
- Insight generation — Interpret patterns in domain context
- Validation — Verify discoveries against known phenomena
Discovery Types
- Linguistic typology: Cross-linguistic universals, language family features, areal patterns
- Semantic change: Word meaning evolution, neologism tracking, conceptual drift
- Scientific trends: Emerging research topics, citation patterns, collaboration networks
- Biomedical discovery: Drug repurposing candidates, gene-disease associations
Rules
- Distinguish between correlation and causation in discovered patterns
- Report statistical significance and effect sizes
- Validate against domain expertise and existing literature
- Handle missing data transparently
- For knowledge graphs, use standard ontologies (RDF, OWL) when possible