Glossary

Research and AI terms, in plain words

What the words around AI reading tools and literature reviews mean, and where each one shows up in practice.

AI hallucination

An AI hallucination is a confident statement a model made up, like a fake citation. Why it happens and how to check answers about papers.

Bibliographic coupling

Two papers are bibliographically coupled when they cite the same sources. Why it finds recent related work that co-citation misses.

Chunking

Chunking splits a long document into passages small enough to embed, search and send to a model. Why chunk size matters for AI answers.

Citation graph

A citation graph is a network where papers are nodes and citations are links. How to read one to find foundations, clusters and new directions.

Co-citation

Two papers are co-cited when a third paper cites both. How co-citation reveals related work, and how it differs from bibliographic coupling.

Context window

A context window is how much text a language model can take in at once, measured in tokens. Why long papers need retrieval, not just a bigger window.

DOI (Digital Object Identifier)

A DOI is a permanent identifier for a paper, like 10.1000/xyz123. What it is, how to resolve it, and why it is the best way to find a paper.

Embeddings

Embeddings turn text into lists of numbers so that texts with similar meaning sit close together. The basis of semantic search and RAG.

Grey literature

Grey literature is research published outside journals and books: reports, theses, preprints, conference papers. Why reviews should search it.

Local LLM

A local LLM is a language model that runs on your own computer, through Ollama or LM Studio, so your text never leaves it. What you need and why.

Meta-analysis

A meta-analysis statistically combines the results of comparable studies into one estimate. When it applies, and the data it needs from each paper.

Model Context Protocol (MCP)

MCP is an open protocol that lets AI apps like Claude Desktop call tools on your machine through MCP servers. What it is and how it is used.

Open access

Open access means research anyone can read for free. Gold and green open access explained, and how to find the free version of a paper.

Peer review

Peer review is the evaluation of a paper by independent experts before publication. What it checks, what it misses, and why preprints skip it.

PICO

PICO frames a focused research question: Population, Intervention, Comparison, Outcome. How to use it to plan a literature search.

Preprint

A preprint is a paper shared publicly before peer review, on servers like arXiv or bioRxiv. How preprints differ from published papers.

PRISMA

PRISMA 2020 is the reporting guideline for systematic reviews: a checklist and a flow diagram of records found, screened, excluded and included.

Reference manager

A reference manager stores the papers you cite and formats citations and bibliographies. How it differs from a reading tool, and when you need both.

Retraction

A retraction withdraws a published paper because of serious errors or misconduct. How to check whether a paper you cite was retracted.

Retrieval-augmented generation (RAG)

RAG explained simply: find the relevant passages in real documents first, then ask a language model to answer from them, and cite them.

Scoping review

A scoping review maps the research on a broad topic to show what exists and where the gaps are, with a systematic search but lighter appraisal.

Semantic search

Semantic search finds text by meaning instead of exact words, by comparing embeddings. How it differs from keyword search, and when to use each.

Snowballing

Snowballing finds studies by following the references of included papers backward and their citations forward. How to do it systematically.

Systematic review

A systematic review answers a focused question with a predefined, reproducible method for finding, selecting and appraising all relevant studies.

Vector database

A vector database stores embeddings and finds the nearest ones fast. What it is, why RAG needs one, and how pgvector adds it to Postgres.