Gå till huvudinnehållet

Using AI to identify and review research

Understanding academic searching

According to Michael Gusenbauer and Neal R. Haddaway (2021) researchers have taken their searching for information for granted, and in order to improve academic searching they propose that we need to understand the different goals that define the way searching needs to be conducted.

We need to understand that where and how we search greatly impacts what we find and miss, what we conclude [...]. Improving academic searching helps to improve the quality of science [...]. Decisions concerning the design of search strategies profoundly affect the resultant evidence that researchers identify, what they (often unknowingly) fail to identify, and what conclusions they draw based on the emergent evidence.

Different search types

As researchers, we must start understanding the basic types of searching we engage in and how the objectives behind each search type (why we search) should determine the search methods—that is, system choice (where we search) and search heuristics (how we search).

Lookup search

Lookup searches - also called "known item searches" or "navigational searches" - are conducted with a clear goal in mind and "yield precise results with a minimal need for result set examination and item comparison".

Exploratory search

Exploratory searching is a process characterized by learning where users aim to be exposed to a multitude of different, sometimes contradicting knowledge sources to build their mental models on a topic.

Systematic search

Systematic reviews (including meta-analyses) and systematic maps, has introduced many disciplines to the concept of systematic searches, with the goal to (a) identify all relevant records (within the resource constraints) in a (b) transparent and (c) reproducible manner.

Search types: Goals, uses cases, heuristics & search system requirements

Search types

Goals

Use cases

Heuristics (how)

Search system requirements

Lookup

Clear goal with targeted search, to find one or a small number of articles meeting a narrow set of criteria. Retrieve specific facts, question answering, verification or re-finding. Straightforward searches. Most relevant records first. Simple, straightforward search. High coverage for high recall/sensitivity. Interpretation of the system of what the user "means".

Exploratory

Learn about a concept or research area. Initially unclear goal but gets clearer during the iterative search process. Research discovery, learning, evaluation, keeping up-to-date. Learning with little prior knowledge. Most relevant records first. Snowballing (associations between papers), post-query filtering (limits). Efficient navigation,various navigational options (querying, browsing, filtering), offering cues to the user as how to assess the search result.

Systematic

Clear search goal to identify all records on a topic. Unbiased, transparent & reproducible search strategy, constructed to balance recall/sensitivity and precision. Multiple bibliographic databases are searched. Systematic reviews, meta-analyses, scoping reviews. Search blocks using boolean operators (AND/OR), snowballing (associations between papers), handsearching (manual screening of journals), "succesive fraction" (inclusion/exclusion). Comprehensive, transparent, reproducible, unbiased search with efficient retrieval. High precision searching multiple databases. Download entire dataset (search results).

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

AI-aided search tools and search types

AI-aided search tools  are tuned for high precision rather than high recall searches, which while good for explorary searches, isn't ideal when you want to do a deep thorough literature review much less an evidence synthesis where the aim is to find as many relevant articles as possible.

From Aaron Tay "A conceptual view of information retrieval - Can we do better with AI?"

Precision

What fraction of the returned results are relevant to the information need?

Recall

What fraction of the relevant documents in the collection were returned by the system?

Manning, C.D., Raghavan, P. & Schütze, H. (2008). Introduction to information retrieval. Cambridge: Cambridge University Press.