Semantic Search
Search that retrieves results based on meaning rather than exact keyword matching — how tools like Elicit and Semantic Scholar find relevant papers even when they use different terminology than your query.
What it means
Semantic search retrieves documents or passages based on their meaning rather than on the presence of specific keywords. A traditional keyword search for “myocardial infarction treatment” would miss a paper that discusses “heart attack therapy” without using those exact words. Semantic search, by contrast, understands that these phrases mean the same thing and would return both.
The mechanism: documents and queries are both converted into numerical vectors (see Embedding) by a language model. Documents with similar meanings have vectors that are close together in the high-dimensional space, even when the vocabulary differs. Search becomes a matter of finding the nearest neighbors to a query vector.
Why it matters for researchers
This is how Elicit, Semantic Scholar, and similar tools find relevant papers. When you type a natural language research question — “how does sleep deprivation affect cortisol regulation?” — the tool converts your question to a vector and retrieves papers whose meaning is closest, not just papers containing those exact words. This is why you can describe a concept rather than remember the exact terminology a field uses.
Advantages over keyword search:
- Finds relevant work across disciplines that use different terminology for the same concept
- Handles paraphrase and synonymy automatically
- Useful for interdisciplinary researchers who don’t yet know the vocabulary of a new field
Limitations:
- Recall is not exhaustive — relevant papers with unusual terminology may still be missed
- The quality of results depends entirely on the quality of the embedding model; smaller or older models may miss nuanced semantic relationships
- Does not replace controlled vocabulary search (MeSH terms in PubMed, thesaurus terms in PsycINFO) for formal systematic reviews where reproducibility of the search strategy is required