Computational Literature Reviews: Method, Algorithms, and Roadmap

Abstract

The substantial volume, continued growth, and resulting complexity of the scientific literature not only increases the need for systematic, replicable, and rigorous literature reviews, but also highlights the natural limits of human researchers’ information processing capabilities. In search of a solution to this dilemma, computational techniques are beginning to support human researchers in synthesizing large bodies of literature. However, actionable methodological guidance on how to design, conduct, and document such computationally augmented literature reviews is lacking to date. We respond by introducing and defining computational literature reviews (CLRs) as a new review method and put forward a six-step roadmap, covering the CLR process from identifying the review objectives to selecting algorithms and reporting findings. We make the CLR method accessible to novice and expert users alike by identifying critical design decisions and typical challenges for each step and provide practical guidelines for tailoring the CLR method to four conceptual review goals. As such, we present CLRs as a literature review method where the choice, design, and implementation of a CLR are guided by specific review objectives, methodological capabilities, and resource constraints of the human researcher.

Artificial intelligence and the conduct of literature reviews

“…The expanding computational resources allow ever greater numbers of research papers to be extracted and analyzed with some projects reporting tens (Larsen et al, 2019) and others reporting hundreds of thousands (Dang et al, 2009) of papers processed. In such cases, the technical quality of paper documents would vary (e.g., regarding Optical character recognition and the inclusion of additional, non-content-related text), and the scale of work necessitates novel methods and metrics for preprocessing and establishing the quality of inputs into AILRs (Antons et al, 2021). Considering the scale, an important consideration in undertaking this work is the need to establish and, if necessary and possible, improve quality of inputs automatically, with little human intervention.…”

Major and recent trends in creativity research: An overview of the field with the aid of computational methods

“…To overcome the information overload, researchers have opted for computer‐assisted methods that could enable them to perform comprehensive navigation of the field (Antons et al, 2021). Still, these assisted overviews are not abundant and are usually focused on narrow sets of information, from where general trends are inferred.…”

A Network-Graph Based IT Artifact Aiding the Theory Building Process

“…The use of computational techniques to conduct systematic and rigorous literature reviews also informed the development of this IT artifact. Antons et al (2021) provide a six-step roadmap outlining the computational literature review (CLR) process, which details roles of both human and machine with issues for consideration at each step [2]. Mortenson and Vidgen (2016) offer a CLR approach to compliment human researchers while investigating the technology acceptance model [22].…”

“…Literature reviews are essential in establishing a foundation of understanding for a given phenomenon, as well as situating the phenomenon within existing theoretical frameworks. In developing this foundation there is a critical need for greater replicability and repeatability in the literature review process, which is complicated by the substantial volume and continued grown of scientific literature [2,3]. The documents selected for review have a strong effect on the theories and relationships identified, making the need to start with a broad set of data critical [4].…”