Scholar Gateway Search
semanticSearchFull Description
Semantic search across peer-reviewed literature. Returns text passages with citations and provenance metadata. Use complete natural language queries — do NOT reduce to keywords unless the user explicitly provided them. For best results, include enough context to distinguish your topic from related fields — for example, 'cold virus treatment' rather than 'common cold', or 'plant drought stress' rather than 'stress in plants'. Expand acronyms and disambiguate short or polysemous terms — for example, 'ALS motor neuron disease' rather than 'ALS', or 'major depressive disorder treatment outcomes' rather than 'depression treatment'. When calling this tool you must: 1. Generate a UUID for interaction_id. Reuse the same UUID for all searches stemming from the same user prompt. 2. Write a brief free-text inferred_intent describing the user's underlying information need — not the query itself, but the goal behind it (e.g. 'comparing treatment efficacy for metabolic syndrome interventions').
Parameters (1 required, 6 optional)
querystringNatural language research question. Preserve intent, scope, and structure. Rewrite for semantic clarity but do not compress to keywords. Expand acronyms and add field context for ambiguous or polysemous terms.
end_yearnumberInclusive upper bound for publication year filter. Omit to include publications up to the present.
includeRetractedContentbooleanWhether to include retracted publications in results. Default false; set true only when retraction history is itself the subject of inquiry.
Falseinferred_intentstringFree-text description of the user's underlying information need — why they need this information, not what they searched for. Examples: 'trying to understand whether a treatment works', 'looking for evidence to support or challenge a position', 'seeking background on an unfamiliar concept', 'checking whether a claim is well-supported in the literature'.
interaction_idstringUUID grouping all searches from a single user prompt. Generate once per user prompt; reuse for all parallel or follow-up searches within that episode.
start_yearnumberInclusive lower bound for publication year filter. Omit to search across all available years.
topNnumberNumber of passages to return. Higher values improve recall for broad or multi-faceted queries; lower values are appropriate for narrow or well-defined topics.
15