Using machine learning to disentangle homonyms in large text corpora

Abstract

Systematic reviews are an increasingly popular decision-making tool that provides an unbiased summary of evidence to support conservation action. These reviews bridge the gap between researchers and managers by presenting a comprehensive overview of all studies relating to a particular topic and identify specifically where and under which conditions an effect is present. However, several technical challenges can severely hinder the feasibility and applicability of systematic reviews, for example, homonyms (terms that share spelling but differ in meaning). Homonyms add noise to search results and cannot be easily identified or removed. We developed a semiautomated approach that can aid in the classification of homonyms among narratives. We used a combination of automated content analysis and artificial neural networks to quickly and accurately sift through large corpora of academic texts and classify them to distinct topics. As an example, we explored the use of the word reintroduction in academic texts. Reintroduction is used within the conservation context to indicate the release of organisms to their former native habitat; however, a Web of Science search for this word returned thousands of publications in which the term has other meanings and contexts. Using our method, we automatically classified a sample of 3000 of these publications with over 99% accuracy, relative to a manual classification. Our approach can be used easily with other homonyms and can greatly facilitate systematic reviews or similar work in which homonyms hinder the harnessing of large text corpora. Beyond homonyms we see great promise in combining automated content analysis and machine-learning methods to handle and screen big data for relevant information in conservation science.

On the overlap between scientific and societal taxonomic attentions — Insights for conservation

“…Scientific names represent a reliable proxy and preferable alternative to vernacular names, due to a strong and culturally independent association between their representation in digital corpora (Jarić et al, 2016;Correia et al, 2017Correia et al, , 2018. At the same time, search based on scientific names avoids numerous problems related to vernacular language, such as frequent vernacular synonyms and homonyms (Roll et al, 2018), differing names among languages, as well as lack of vernacular names for some species (Jarić et al, 2016). Accounting for taxonomic synonyms is also critical, as they can strongly affect the accuracy of species data retrieval .…”

Online tools supporting the conduct and reporting of systematic reviews and systematic maps: a case study on CADIMA and review of existing tools

“…This is an area that we believe will be of increasing interest to users, particularly for updating existing reviews (algorithms can be trained to identify relevant studies based on similarity to previously included studies) [31] and dealing with very large bodies of literature.…”

A systematic map of evidence on the contribution of forests to poverty alleviation

“…Third, the topic of this systematic map straddles multiple disciplines and sectors (conservation, development, natural resource management), thus substantial semantic diversity is more than likely to exist. Unlike medical fields, these fields lack a standardized ontology [45,46] and are characterized by rapid radiation of terms in use over the past few decades. While our search strategy attempted to capture this diversity through piloting and testing with an interdisciplinary and multisector review team, we recognize that not including specific terms in our search may have resulted in literature areas missed.…”