Knowledge Discovery, Search, and Semantic Technologies
This section covers Knowledge Discovery, Search, and Semantic Technologies. The exam tests search relevance, indexing, metadata filters, semantic models, knowledge graphs, and evaluation.
Core rules
- Search quality depends on content, indexing, query understanding, ranking, and interface. A better algorithm cannot fix missing or inconsistent metadata.
- Lexical search matches terms and handles exact identifiers well. Vector or semantic search handles paraphrase and synonymy but can miss precise codes. Hybrid retrieval often balances both.
- Query understanding includes spelling, acronyms, entity recognition, and intent. Facets support narrowing by type, date, owner, classification, or status; they are navigation controls, not ranking.
- Semantic technologies represent meaning with triples, ontologies, and controlled vocabularies. RDF expresses subject-predicate-object statements; SPARQL queries RDF; OWL adds richer logical constraints.
- Evaluate with a relevance set: queries, expected results, precision, recall, and task success. Monitor zero-result queries and click-through as signals, not proof.
Worked example
Engineers search for pump failure high temperature across maintenance reports, procedures, and incident tickets. A keyword-only search misses reports that say overheating. A hybrid design expands synonyms, applies equipment and date facets, and ranks validated procedures above raw tickets. Evaluation uses known incidents and checks whether the correct procedure appears in the top five.
Traps
Do not equate search with navigation. Do not build an ontology at unnecessary granularity. Do not ignore query logs and failed searches. Do not treat semantic search as automatically accurate.