No 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.

Knowledge of the number and distribution of species is fundamental to biodiversity conservation efforts, but this information is lacking for the majority of species on earth. Consequently, subsets of taxa are often used as proxies for biodiversity; but this assumes that different taxa display congruent distribution patterns. Here we use a global meta-analysis to show that studies of cross-taxon congruence rarely give consistent results. Instead, species richness congruence is highest at extreme spatial scales and close to the equator, while congruence in species composition is highest at large extents and grain sizes. Studies display highest variance in cross-taxon congruence when conducted in areas with dissimilar areal extents (for species richness) or latitudes (for species composition). These results undermine the assumption that a subset of taxa can be representative of biodiversity. Therefore, researchers whose goal is to prioritize locations or actions for conservation should use data from a range of taxa.

120 max) 25 We propose, and formalize, a new framework for research synthesis of both 26 evidence and influence, named 'research weaving'. It summarizes and visualizes 27 information content, history, and networks among a collection of diverse publication 28 types on any given topic. Research weaving achieves this feat by combining the 29 power of two methodologies: systematic mapping and bibliometrics. Systematic 30 mapping provides a snapshot of the current state of knowledge, identifying areas 31 needing more research attention and those ready for full synthesis (e.g., using meta-32 analysis). Bibliometrics enables researchers to see how pieces of evidence are 33 connected, revealing the structure and the evolution of a field. We explain how to 34 become a 'research weaver', and discuss how research weaving may change the 35 landscape of research synthesis. 36 37 38 Keywords: meta-research, quantitative synthesis, systematic review, Big Data, data 39 visualization, evidence synthesis (max 6) 40 41 2 Influence 43 Research fields are flooded with torrents of publications, and researchers require 44 informative reviews to stay afloat. For many years, researchers sought expert 45 opinions from narrative reviews (see Glossary) to obtain and update their 46 knowledge of a research topic or question [1]. These reviews were valuable not just 47 for summarizing 'facts' about a particular research field, but also for giving broader 48 insights, such as identifying the origin and development of key theoretical concepts, 49 or drawing attention to ideas that deserved greater research focus. More 50 sophisticated syntheses are now commonly used -systematic review and meta-51 analysis [2-8] -which incorporate systematic and often quantitative methods to 52 extract factual information from the literature in a reliable manner. However, both 53 these syntheses have their limitations. They are not practical for broad fields 54 encompassing thousands of publications, and cannot handle a highly heterogeneous 55 literature. A new technique has emerged to deal with these limitations: mapping. 56Currently, scientists' 'map' research evidence using two complementary 57 methodologies of different origins: systematic mapping and bibliometrics. 58Systematic mapping (sometimes called 'evidence mapping') is a method derived 59 from systematic reviews, with the goal of classifying the types of research on a broad 60 topic [9][10][11][12][13][14]. Systematic mapping is still a nascent methodology, with the first 61 systematic maps appearing only in the last decade [9, 10]. In addition to providing a 62 written report, a systematic map typically involves the production of a database of 63 studies and their attributes, which can be provided to users as a searchable 64 database or a series of visualisations [10][11][12]. In contrast, bibliometrics (more 65