Text analysis tools for identification of emerging topics and research gaps in conservation science

Abstract

Keeping track of conceptual and methodological developments is a critical skill for research scientists, but this task is increasingly difficult due to the high rate of academic publication. As a crisis discipline, conservation science is particularly in need of tools that facilitate rapid yet insightful synthesis. We show how a common text-mining method (latent Dirichlet allocation, or topic modeling) and statistical tests familiar to ecologists (cluster analysis, regression, and network analysis) can be used to investigate trends and identify potential research gaps in the scientific literature. We tested these methods on the literature on ecological surrogates and indicators. Analysis of topic popularity within this corpus showed a strong emphasis on monitoring and management of fragmented ecosystems, while analysis of research gaps suggested a greater role for genetic surrogates and indicators. Our results show that automated text analysis methods need to be used with care, but can provide information that is complementary to that given by systematic reviews and meta-analyses, increasing scientists' capacity for research synthesis.

75 years of dryland science: Trends and gaps in arid ecology literature

“…LDA identifies sets of words that occur together with unusual frequency, thus allowing topics to be defined [6]. Multiple topics can occur in any article, and LDA weights topics within an article and the words that define a topic [11].…”

“…Multiple topics can occur in any article, and LDA weights topics within an article and the words that define a topic [11]. We set the number of topics to 25, which adequately represented variation in the data, captured the complexity of topics in the literature, but was not so high as to hamper interpretation of patterns [6]. We named each topic based on assessment of the 20 highest weighted words for that topic and expert knowledge of the literature (S1 Table).…”

“…(6). Topics with positive random intercepts can be interpreted as having higher than average numbers of articles written about them and topics with positive random slopes as having higher than average growth in publications in the time period (1940-2015) [6, 7]. GLMM was run using the lme4 1.1–10 package [16] in the R Statistical program [15].…”

“…For example, bibliometric reviews have explored research trends in specific fields, such as how research in urban ecology aligns with professional society statements of research goals [5], how the ecological surrogate and indicator literature informs conservation science [6] and topics studied within specific journals [7]. Bibliometric reviews have also been used to track topics in the field of ecology as a whole [4, 8, 9].…”

Distilling knowledge from the flood

“…Just keeping abreast of a field of study is now a very significant task in its own right. It may be that technology has some of the answers to managing this task and software tools are likely to play a growing role in making sense of the literature 2,3 . Aiming to go further and synthesize the available evidence requires a systematic review (see the Comment by Neal Haddaway and Biljana Macura), which typically intends to satisfy one of two goals: to assess the efficacy of specific interventions -such as payment for carbon storage -or to reach some broader generalizations across studiesfor example, changing species distributions in response to warming.…”