Understanding lexical & semantic search

When searching for information there are two main approaches: lexical search and semantic search. While both methods aim to retrieve relevant documents, they use different techniques to do so.

Lexical search - keyword search

• Matches exact words or phrases that appear in a query with those in the documents.
• Relatively simple and fast.
• May not be able to handle misspellings, synonyms, or polysemy (when a word has multiple meanings).
• Doesn't take into account the context or meaning of the words, which can lead to irrelevant results.

Semantic search - vector search

• Uses natural language processing (NLP) techniques to analyze the meaning of words and their relationships.
• Represents words as vectors, where the distance between vectors indicates their semantic similarity.
• Can handle misspellings, synonyms, and polysemy (when a word has multiple meanings).
• Can produce more accurate and relevant results.
• Can be computationally expensive and time-consuming.

Pinecone Docs "Differences between Lexical and Semantic Search regarding relevancy"

AI-aided search tools using keyword search

There are several AI-aided search tools offering keyword search. Some of them also serves as the back end for other AI-aided search tools, with other features than traditional keyword search. None of the sources are as comprehensive as Google Scholar, which is estimated to have almost 400 million records. You can find information about coverage and more through Search Smart.

Search Smart

Search Smart gives you information on different databases, like coverage (prevalence of certain record types in a database), subject coverage, functionalities (the capabilities a search system offers users to search, filter, retrieve and manipulate search results), search provider and more.

Search Smart

Semantic Scholar

More information on Semantic Scholar from Search Smart

Open Alex

More information on Open Alex from Search Smart

Lens

More information on Lens from Search Smart

Scite

More information on Scite from Search Smart

AI-aided search tools using vector search

Undermind

Scopus AI

Elicit

Scite Assistant

When to use which kind of AI-aided search tool?

Aaron Tay, librarian Singapore Management University LIbrary: "Google Scholar vs other AI search tools (Underdmind, Eilcit, SciSpace) - how and when to use each"

Limitations with AI-aided search tools

• Access to citation/abstract metadata and the full-text of open access articles still omits a vast amount of scholarly research contained in full-text paywalled articles.
• When crafting answers and summaries, tools without access to full-text will base their answers on abstracts.
• These tools can suffer from poor metadata in the index of scholarly literature they search, which can lead to less than comprehensive search results

From University of California Irvine "AI in research"

Trust in AI-aided search tools

• AI tools like Scite Assistant, Elicit, Consensus, and Scopus AI can generate summaries and references quickly, but the accuracy and completeness of outputs vary.
• Generating accurate citations doesn’t ensure their alignment with the arguments in the summary.
• These tools should be used as a starting point for research and not as a replacement for a thorough literature review. The information they generate should always be verified.

Aster Zhao, librarian Lee Shau Kee Library "Trust in AI-aided search tools: Evaluating Scite, Elicit, Consensus, and Scopus AI for Generating Literature Reviews"

Critique AI-aided search tools using vectors

We see two fundamental problems: first in these semantic search systems it is opaque algorithms that decide about the "right" information that is shown (either absolutely or by order). We currently have neither insight nor control over these decisions. This is particularly problematic for systematic searching, where our study has shown that all semantic search systems in our sample fail to meet requirements. Second, we must stay alert as these efficient-slick systems aim at transforming "inefficient" exploratory searching into "efficient" lookup searching (eg. through presentation of preselected cues). This means exploratory searching (and thus learning) might be more and more crippled toward quick, unconsciously biased lookup searching (cherry picking) that users more and more expect when engaging with online systems.

Gusenbauer & Haddaway. (2021). What every researcher should know about searching–clarified concepts, search advice, and an agenda to improve finding in academia. Research Synthesis Methods, 12(2), 136-147. https://doi.org/10.1002/jrsm.1457