Introduction

The frequency of biological invasions, driven by globalisation, is increasing and shows no sign of reaching saturation (Seebens et al. 2018; Essl et al. 2020). As invasions increase, there will be further disruption to ecosystems and negative consequences for native species (Ricciardi et al. 2013) and human livelihoods (Shackleton et al. 2019). Preventing a conservation crisis requires effective management of invasive non-native species (INNS). However, previous studies have found a disconnect between the research generated and its policy and management applications (Esler et al. 2010). This is reflected in concern over the limited achievement of international targets related to INNS. The assessment of global actions to meet the Convention on Biological Diversity’s (CBDs) Aichi Biodiversity Targets highlight that the efforts to combat species invasions have been outpaced by globalisation, and in particular the impact of massively expanded trade (CBD 2020). It is therefore important to consider if and how research is supporting global conservation goals of preventing and managing the impacts of INNS.

The body of literature related to INNS is large and continues to proliferate (Richardson and Pyšek 2008; Esler et al. 2010). However, its accessibility can be hindered by differences in the approach and terminology adopted by each discipline within invasion science, be that theoretical, practical, policy or management focussed, for example (Pyšek et al. 2004; Blackburn et al. 2011; Enders et al. 2020; Robertson et al. 2020). Extracting key messages from such a complex and wide-ranging body of literature is therefore challenging, especially when attempting to navigate the full extent of the publishing landscape. Topic modelling provides a semi-automated statistical tool to assess the content of articles in a corpus (a large body of literature; Blei and Lafferty 2009). Information is collected from article abstracts, with the patterns of word co-occurrence then analysed to identify common ideas or topics (Griffiths and Steyvers 2004). This provides a more quantitative alternative to standard literature reviews, summarising trends without researcher bias and allowing information from different thematic, spatial, and temporal scales to be consolidated (Westgate et al. 2015). The approach has already been applied to conservation science and conservation planning research (Westgate et al. 2015; Mair et al. 2018).

We used topic modelling to assess how scientific research contributes to developments in the field of invasion science, using terms commonly utilised by separate disciplines to capture the diversity in the literature. For the purposes of this study, invasive non-native species are a species, subspecies, or lower taxon; including any part, gametes, seeds, eggs, or propagules of such species that might survive and subsequently reproduce, whose introduction and/or spread outside their natural past or present distribution threatens biological diversity (CBD 2002). Adopting a broad inclusive definition of INNS allowed us to determine potential gaps and understudied aspects within the literature.

Our aim was to identify the key themes and knowledge gaps in invasion science and how these have changed over the period 1985–2019. Our objectives were to (1) identify the topics most popular in invasion science research; (2) identify how topics have changed in prevalence over time; and (3) assess how different topics interact (co-occur) within articles, as a means of identifying more or less well-connected topics in the literature. We specifically focus on how well research topics on invasive species are connected to management and policy, which is key to achieving scientific impact. Highlighting such absences can help pinpoint knowledge gaps needing to be bridged, areas where a lack of connectedness between topics/disciplines could help resolve stubborn problems, or in prioritising future research agendas (Esler et al. 2010).

Methods

Literature search

Documents included in the literature search were classed as articles (peer reviewed publications) according to the search engine and refined to only include articles written in English. The article inclusion criteria for this study required that each article clearly focused on invasive non-native species, as opposed to non-native/alien or naturalised species that were not considered invasive. A study by Golebie et al. (2022) found the use of terminology to broadly align with the stage of invasion, in that “invasive” was most commonly used except when the research was conducted at early stages of invasion, when “non-native” was most commonly used. The results of this study may therefore be biased towards the biodiversity aspects of invasion biology, as other areas of research such as animal and plant health interact with INNS but may use different terminology, such as “pest” (see Roda et al. 2011 as an example).

We searched Web of Science and Scopus on April 17th, 2020, using the search terms: “invasive” AND [“non-native” OR “alien”] AND “species”. These terms were considered to capture the most relevant publications specific to “invasive non-native species”. A range of other terms are used in some literature (e.g., non-indigenous, feral) however, we found that additional search terms returned a larger volume of irrelevant articles (for a comparison of results obtained from an alternative search string, see Supplementary Information S1.1). We considered publications up to and including 2019. The citations and abstracts for each article were then downloaded into the reference management software EndNote X9 (Clarivate, 2020) for review. For a full repository of all included articles see Supplementary Information 2.

Data cleaning

Duplicate articles and those without abstracts were excluded, reducing the number of papers from 17,558 to 10,945. Titles and abstracts of the 10,945 articles were screened and then manually reviewed to ensure that they met the article inclusion criteria for this study. To be included, articles needed to focus specifically and primarily on INNS, so articles were excluded if they discussed irrelevant topics such as medicine, or only mentioned INNS in passing as an example or as part of a list of issues affecting biodiversity. This was done by two people (Stevenson and Witts), due to the large volume of articles returned. A subset of 600 articles were cross-checked to refine inclusion criteria and ensure consistency between reviewers, with disparities discussed in a larger group to solidify the inclusion criteria prior to reviewing the rest of the articles. Examples of article abstracts that were excluded, along with the reason for exclusion, are provided in Supplementary Information S1.2. This left 9882 articles for processing.

Following the methodology used by Westgate et al. (2015) abstracts were converted into a corpus and processed using the R package tm (Meyer et al. 2008; R Core Team, 2021). To optimise the output of the model, a range of commonly occurring or generic words were removed. This included all numbers, written as words or digits (Grün and Hornik 2011), and the search terms as these would be present in all abstracts. Words that did not provide information on the topic being discussed, such as “like'', “may” and “either”, were removed using a combination of the list of “stop words” provided as part of the tm package (Meyer et al. 2008) and terms that we chose to remove following a preliminary run of the model, as they did not provide meaning to the results, such as “can” and “will” (Full list in Supplementary Information S1.3). Hyphens and forward slashes were changed to spaces, capital letters were converted to lowercase and all other punctuation was removed (Grün and Hornik 2011). The remaining words were then stemmed to their common root and all spelling converted to the American style to prevent duplicates. Finally, words that appeared in five or fewer articles were removed to speed up processing and because such infrequent words contribute little to topic generation (Griffiths and Steyvers 2004).

Topic modelling

Topic modelling identifies the main ideas being discussed in a corpus using sets of words that co-occur with unusual frequency, which are then grouped into topics (Griffiths and Steyvers 2004). The model generates the weight that each word contributes to a topic, allowing the focus of each topic to be determined from the most highly weighted words. Individual articles will often discuss multiple topics and the relative weighting of each topic will vary across articles. This allows the main topic and diversity of topics within an article to be identified.

The number of topics fitted to a particular corpus is determined a priori (Mair et al. 2018). We ascertained the most suitable number of topics by using ten-fold block cross-validation to run a series of models with varying numbers of topics (n = 10, 20, 30, 40, 50, 60, 100, 200). For each number of topics, models were fitted to 90% of the corpus. Model fit was tested by calculating perplexity on the with-held 10% of the corpus; lower perplexity indicates better model fit (Grün and Hornik 2011). This process was repeated for each candidate number of topics. The block-cross validation showed that as the number of topics increased, model perplexity decreased, suggesting there were many topics discussed in the corpus (Supplementary Information S1.4).

It was necessary to balance the need to investigate the complexity of the corpus with our aim of simplifying the overarching patterns in the literature, for easier consolidation by all stakeholders (sensu Westgate et al. 2015; Mair et al. 2018). We therefore fitted 50 topics using a Latent Dirichlet Allocation (LDA) model with Gibbs sampling, using the R package topicmodels (Grün and Hornik 2011). Model perplexity also had declined substantially by 50 topics, indicating better model fit (Supplementary Information S1.4).

We used the 20 highest weighted words in each topic, with particular focus on the top five words (Supplementary Information S1.5), to name the topics generated by the model. Alongside this, a workshop was held with all authors who brought wide and varying perspectives on INNS, to agree on topic names based on the data and their expertise. Where words associated with a topic appeared vague, articles were extracted for those topics and reviewed to gain further insight into the weighted words describing a topic. Topics were named to simplify the information being presented. Thereafter, topics were grouped into broad themes, based on an analysis of topic similarity, to allow for easier identification of patterns in the data (Supplementary Information S1.6). Topic similarity was calculated using the weight with which each unique word was assigned to each topic, following the methods described in Westgate et al. (2015). We acknowledge that individual topics may reasonably be placed in more than one theme, however the combination of inspection of similarity and expert input provided us with a means to organise topics into themes in the most appropriate way to facilitate communication of results.

To assess topic frequency within the corpus, we assigned articles to their single highest weighted topic (for the distribution of weights of highest weighted topic per article versus the weights of all other topics per article see Supplementary Information S1.7). There are a range of approaches that can be used to assign articles topics, including applying a topic weight threshold, such that an article is assigned to one or more topics that have a weight greater than the selected threshold, and using a cumulative weight threshold, such that an article is assigned to topics with weights that cumulatively sum to a selected threshold. We investigated the effect of using a cumulative threshold to assign articles to topics on our results, and present these analyses in Supplementary Information S1.8.

Topic generality and specificity

Within the literature, some topics are “general”. General topics reflect broad ideas that are discussed across many articles, often in association with other topics. Meanwhile, other topics are more “specific”, and these topics tend to be the sole or primary focus of an article, contributing a large weight to the article. To determine whether individual topics tended to be more general or specific, the distribution of topic weights within articles was used. For each article, the topic with the highest weight was selected. We calculated the mean weight of a topic when it was selected, and the mean weight of a topic when it was not selected. These values were then plotted against each other, which allowed us to observe the generality or specificity of each topic (Westgate et al. 2015). To further investigate the effect of applying a cumulative weight threshold to article assignment to topics, we calculated topic generality and specificity across a range of cumulative weight thresholds, see Supplementary Information S1.9.

Topic prevalence

To assess how topic prevalence changed over time, we generated a matrix of the weights with which each article was assigned to each of the 50 topics. Article publication date was then used to calculate the mean topic weight per decade, to give the relative prevalence of each topic within each decade. This allowed interpretation of variations in relative topic prevalence over time. For this analysis, articles from the 1980s and 1990s were combined due to the low volume of articles in the corpus from the 1980s.

Co-occurrence of topics within articles

To assess the frequency with which any two topics co-occurred within any article in the corpus, we used the distribution of topic weights within articles. The matrix of the topic weight per article was log10 transformed and the Euclidean distance between each topic pairing was then calculated and scaled from zero to one. A distance of one showed that a pair of topics frequently co-occurred within the same article, while a distance of zero showed that a pair of topics rarely co-occurred (Westgate et al. 2015).

Results

The database search yielded 8411 articles in Web of Science and 9147 articles in Scopus, giving a combined total of 17,558 papers, on April 17th, 2020. The EndNote software identified 6613 duplicated articles, leaving a total of 10,945 unique papers. Ensuring inclusion criteria were met, 9882 papers were deemed suitable for inclusion. Following abstract cleaning the vocabulary of the corpus was reduced from 34,027 unique words to 7427.

The number of articles published in the field has increased dramatically over the last 35 years. The oldest articles in the corpus were published in 1985 (the search returned no articles before 1985). There were fewer than a dozen publications per year until 1998, after which the frequency of publications increased rapidly, surpassing 100 articles a year by 2004 and 500 articles a year by 2010. By 2019 the number of articles discussing INNS had doubled again, surpassing 1000 publications a year (Fig. 1).

Fig. 1
figure 1

The number of articles discussing invasive non-native species, published per year from 1985 (the earliest year in which articles in our corpus were published) to 2019

Full size image

Topics and themes

Fifty topics were identified to describe the content of the corpus (Table 1). Topics were assigned to seven broad themes, (i) Ecosystems: topics which discussed a specific region, biome, or focused on a particular species strongly associated with one ecosystem type (e.g. Topic 6: Crayfish); (ii) Monitoring: topics regarding all aspects of monitoring INNS, including detection, identification, and distributional mapping; (iii) Management and decision-making: topics discussing the management and socio-political aspects of invasion science, such as prevention, control, and policy; (iv) Interactions: topics discussing the interactions of INNS with native species, or the effects of those interactions; (v) Assessing change: topics focused on studying and analysing temporal and ecological change; (vi) Traits: topics that explored the characteristics of INNS; and (vii) Invasion mechanisms: topics discussing INNS dispersal pathways and drivers of spread.

Table 1 Fifty topics generated by the topic model, with the five highest weighted words (reduced to the word stem in all cases) for each topic, topic name, and theme to which each topic was assigned. Each article was assigned to its highest weighted topic, generating the number of articles per topic
Full size table

The largest theme was Ecosystems, which represented 26.3% of articles in the corpus and had the greatest number of topics (Fig. 2). The second largest theme was Monitoring which contained 21.8% of articles, followed by Management and decision-making, which contained 15.8%. The fourth largest theme, Interactions, contained 12.2% of articles across seven topics, while the theme Assessing change contained nine topics, but only 9.8% of articles. The smallest themes were each made up of four topics, with Traits containing 8.3% of articles, followed by Invasion mechanisms with 5.7%.

Fig. 2
figure 2

The number of articles per topic. Articles were assigned to the topic with the highest weighting. The topic name and its respective topic number are given on the y-axis, alongside the theme each topic was assigned to, and the total number of articles within each theme

Full size image

Some geographic regions and taxonomic groups had sufficient presence in the corpus to be represented as distinct topics. Two countries / regions had associated words weighted highly enough to be represented as a topic: Americas (Topic 11) and South African invasions (Topic 9) (Table 1). Europe and North America also featured as highly weighted words (Topic 18). Words associated with plants featured throughout the corpus and several topics were plant specific such as Aquatic plants (Topic 8); Plant taxonomy (Topic 15) and Leaf litter (Topic 31). For animals, the only topic specific to an organism was Crayfish (Topic 6). The top 20 words associated with each topic can be found in the Supplementary Information S1.5.

Topics associated with the largest number of articles, in decreasing order, were Population genetics (Topic 12), Marine systems (Topic 1), Policy (Topic 21), First records (Topic 13) and Insect biocontrol (Topic 22), which all came from the three largest themes (which were Ecosystems, Monitoring, and Management and decision-making). Topics associated with the smallest number of articles included Non-native introductions (Topic 50) and Plant herbivory (Topic 33).

Topic generality or specificity

The two topics with the largest number of articles, Population genetics and Marine systems, were highly specific (Topics 12 and 1 respectively, Fig. 3), meaning they were more likely to be the sole focus of an article. In contrast, the most specific topic in the corpus, Pollination (Topic 29), was associated with a relatively small number of articles (Fig. 2). Many topics within the theme Assessing change were more general (such as Temporal change and Community composition; Topics 39 and 35 respectively). These topics were therefore broad and likely to be discussed in conjunction with other topics. Topics from the theme Invasion mechanisms were consistently general, whilst topics from the theme Traits showed high specificity, except for Life history (Topic 44). The theme Ecosystems predominantly contained specific topics, with South African invasions (Topic 9) being the most general. Within the Management and decision-making theme Management strategies (Topic 24) and Insect biocontrol (Topic 22) were the most and least general topics respectively, although Policy (Topic 21) and Risk Assessment (Topic 26) both had surprisingly high specificity.

Fig. 3
figure 3

Topic generality versus specificity. The topics in the bottom right corner were general (broad topics that tended to occur with other topics in an article), while topics in the top left corner were specific (more likely to be the only topic discussed in an article). Numbers represent Topics (see Table 1) and colours represent Theme

Full size image

Changes in topic prevalence over time

Within the corpus, the prevalence of topics changed across decades (Fig. 4; Supplementary Information S1.10). Many of the topics associated with the largest number of articles overall (Fig. 2) had lower prevalence in earlier decades. The relative prevalence of the topic with the largest number of articles overall, Population genetics (Topic 12, Fig. 2), increased dramatically between the 1980/90 s and 2010s. Risk assessment (Topic 26) also showed a dramatic increase in relative prevalence since the 1990s. First records (Topic 13) showed a similar pattern; First records exponentially increased in prevalence and by the 2010s ranked 2nd, rising from position 31 in the 1980s and 1990s, and position 29 in the 2000s (Supplementary Information S1.10). Policy (Topic 21) was consistently prevalent across decades (1980s and 1990s position 9; 2000s position 3) and in the 2010s, was the most prevalent topic within articles of the corpus.

Fig. 4
figure 4

Change in topic prevalence over time, calculated as mean topic weight per decade. Topics are grouped by theme

Full size image

Some topics experienced both increases and decreases in relative prevalence; the prevalence of Management methods (Topic 23) and Marine systems (Topic 1) peaked in the 2000s, before dropping in the 2010s (Fig. 4). Several topics dropped in ranking in the 2000s but rose again in the 2010s, including Island threats and impacts (Topic 4), Climate change (Topic 34), Economic impacts (Topic 25) and Detection (Topic 16). South African Invasions (Topic 9) consistently decreased relative to other topics, from first position in the 1980s and 1990s, to 30th in the 2000s, to 36th in the 2010 (Supplementary Information S1.10). Seasonality (Topic 38) and Plant herbivory (Topic 33) also showed a decline in prevalence in the 2010s. While Crayfish (Topic 6) and Freshwater fauna (Topic 2) increased sharply in prevalence in recent decades, other freshwater related topics such as Aquatic plants (Topic 8) and River systems (Topic 5) remained consistently low ranked over time.

Co-occurrence of topics

Generally, topics were most likely to co-occur with topics from within the same theme; the most frequently co-occurring pairs of topics were River systems (Topic 5) and Freshwater fauna (Topic 2), followed closely by River systems (Topic 5) and Aquatic plants (Topic 8) (Fig. 5). The topic Grasslands (Topic 7) commonly occurred with Management methods (Topic 23), while Leaf litter (Topic 31) was very likely to co-occur with Soil nutrients (Topic 28) and Chemical properties (Topic 46).

Fig. 5
figure 5

Correlation matrix of topic co-occurrence within articles. Topics which co-occurred least often are scored zero, while the topics which co-occurred the most frequently scored one. The grids mark out theme groups for comparison

Full size image

The analysis identified gaps in topic co-occurrence, mostly between topics from the theme Management and decision making and other themes. Policy (Topic 21) had consistently low co-occurrence scores, and rarely co-occurred with any topic outside of the Management and decision-making theme. Topics such as Marine systems (Topic 1) had low interactions with Economic impacts (Topic 25) and Management strategies (Topic 24). Meanwhile, Population genetics (Topic 12) had very low co-occurrence values for Ecological impacts, Interactive effects, and Management methods (Topics 37, 32 and 23 respectively).

Beyond consideration of topic pairs, the average co-occurrence value for all pairs associated with a topic provides an overall measure of how well connected the topic is within the corpus (Fig. 6). The five topics with the highest average co-occurrence values were Sampling (Topic 41), Coastal systems (Topic 10), Chemical properties (Topic 46), Field studies (Topic 36) and Colonisation pressure (Topic 49). The five with the lowest average co-occurrence values were Policy (Topic 21), Marine systems (Topic 1), Population genetics (Topic 12), Woodland structure (Topic 3), and Management strategies (Topic 24).

Fig. 6
figure 6

Mean co-occurrence value for each topic across all articles. A lower co-occurrence rank indicates that the topic does not appear in articles with other topics as often as topics with a higher co-occurrence rank. The mean co-occurrence value across all topics is indicated by the solid vertical line. The topic name and its respective topic number are given on the y-axis, alongside the theme each topic was assigned to, and the total number of articles within each theme

Full size image

Discussion

The publication landscape of invasion science is broad and covers a very wide range of topics, reflecting the multidisciplinary nature of the field. However, there was a focus on certain taxonomic groups, with several topics exclusively focused on plants, and articles on crayfish being popular enough to generate a unique topic. Topics related to freshwater systems were also represented in the literature, although the prevalence of Aquatic plants and River systems has remained consistently low over the last 35 years. The prevalence of topics changed dramatically over time, with articles on Management methods declining. Conversely, the advent of novel technologies has led to increases in the prevalence of articles on Distribution modelling and Population genetics, visibly impacting the publishing landscape. Assessment of topic co-occurrences revealed a distinct lack of interaction between articles discussing management or policy with other topics, showing a lack of integration of scientific research into policy and management on INNS.

Taxonomic and geographic patterns

There is an uneven global distribution of scientific publishing and readership in ecology (Nuñez et al. 2019), and we found that only two regions, Americas and South African invasions, were sufficiently represented in our corpus to emerge as distinct topics. There was wider representation within topics, with several countries mentioned by name including China, Canada, New Zealand, and Australia. However, the most dominant country was the USA, which was the only country to have a state, Florida, named within a topic (Supplementary Information Table S1.5).

While various taxa were listed in the top five highest weighted words for several topics, the only topic to represent a specific organism group was Crayfish. This topic exclusively focused on studies of this family, with “crayfish”, “red” and “clarkii” appearing in the top 20 words, in reference to the red swamp crayfish, Procambarus clarkii (Supplementary Information Table S1.5). Analysis showed that the topic has grown in prevalence over the duration of the corpus, with the total number of articles in which it was the highest weighted topic being only fractionally smaller than the much broader topic River systems. Crayfish may provide an ideal model for studying many aspects of invasion science (Twardochleb et al. 2013; Manfrin et al. 2019), but a large focus on one taxon might also increase the risk that other potentially more damaging species or impacts are overlooked.

Taxonomic bias is a recognised issue in the literature, with plants commonly being the most studied taxa in invasion science (Pyšek et al. 2008), confirmed by our results given the number of topics focused exclusively on plants. This bias can lead to over- and under-representation of biota within the literature that prevents novel synthesis of information in invasion science (Matzek et al. 2015). The species most popular in invasion science have changed over time (Richardson and Pyšek 2008), and it is likely that other species will become more prevalent in the literature as the drivers of invasion by non-native species change (Seebens et al. 2018). To avoid lack of attention to novel/unrecognised threats, researchers need to be more aware of understudied species which can be determined through conversations with stakeholders and horizon scanning (Ricciardi et al 2017). Future research could quantify if the popularity of species or groups of organisms in the literature is reflective of the relative threat they pose.

Emerging trends in the literature

Overall, Population genetics and First records increased fastest in relative prevalence. The prevalence of articles discussing First records, as well as Pathways and Distribution modelling, is likely a result of the emphasis on pathways, spread, risk assessment and management in the Convention on Biological Diversity’s Aichi Biodiversity Targets (part of the CBD’s Strategic Plan for Biodiversity 2011–2020; CBD 2010) and the increasing availability of dedicated journals, and large collated databases based on first records (Seebens et al. 2018), in some cases boosted by growing enthusiasm for community science and recording platforms. This trend in prevalence is likely to continue as awareness of INNS and the rate of invasions increase, and new invaders and pathways emerge (Ricciardi et al. 2017).

Population genetics had the largest numbers of articles in the corpus and appears set to increase. The growth in prevalence of genetics is likely a result of novel technology becoming more widely accessible. Developments in environmental DNA sampling and sequencing allow for the identification or improved detection of INNS, which enables monitoring in systems that were previously difficult to survey (Thomas et al. 2020). A more controversial development, that may support the resurgence in prevalence of the topic Insect biocontrol, is the advent of genome modification which could have widespread applications for management of INNS (Thresher et al. 2014; Champer et al. 2016). Discussion of genetic research and biocontrol is predicted to increase in the literature as these technologies develop (Ricciardi et al. 2017), yet there is currently a lack of interaction between Population genetics and the wider literature. Greater interdisciplinary research is required to achieve uptake of these technologies and realise their potential benefits for INNS monitoring and management efforts.

Management methods was one of the most prevalent topics in the 2000s yet declined in relative prevalence over time. Prevalence in the 2000s may be associated with efforts to achieve the CBD’s 2010 Biodiversity Target but, as the 2010 target was superseded by the Aichi Biodiversity Targets, this does not explain the subsequent drop in prevalence (UNEP 2002; CBD 2010). Decline in prevalence may be because managing INNS is perceived as a “wicked problem” (Woodford et al. 2016), although improved methods to prioritise management actions based on cost and feasibility can improve effectiveness (Booy et al. 2017). Given ambitious global targets to halt the loss of biodiversity caused by INNS, there is a clear need for more evidence designed to support effective management and problem solving (IUCN 2019).

The policy-management–science interface

Policy has been a consistently popular topic across decades, and the most prevalent topic of the 2010s. However, we found that Policy was a highly specific topic, therefore most likely being the only focus of an article. Policy also consistently showed low co-occurrence with other topics, particularly topics within other themes, such as Ecosystems. The lack of co-occurrence between policy and Ecosystems topics was particularly noticeable for Marine systems, despite the results showing that Marine systems were discussed in many articles. This may be due to the greater challenges of identifying, monitoring and managing INNS in the marine environment (Giakoumi et al. 2019).

The overall co-occurrence between management topics and the wider literature was similarly low. While there was a topic dedicated to the economic impacts of INNS, there was no evidence that the terms relating to the cost of management had high weight in either Management strategy or Management methods. In the same vein, there was no evidence of ‘success’ being a highly weighted term in any topic of the management theme. This may be because the number of successful control or eradication campaigns is relatively small, especially outside of island ecosystems (Gardener et al. 2010; Glen et al. 2013; Robertson et al. 2017). It is also difficult to guarantee whether any eradication scheme has been 100% effective, hence the cautionary use of ‘success’. This leads to ongoing management, which in turn makes calculation of costs challenging (Rout et al. 2009). These issues are recognised in the literature, leading to calls for clearer definitions of management objectives and success criteria, and establishment of cost estimation frameworks (Jardine and Sanchirico 2018; Robertson et al 2020). Considering the limited budgets many management schemes operate under, facilitating widespread discussion of management outcomes and the costs of management is crucial for success (Larson et al. 2011).

Invasion science casts light on a range of disciplines, ranging from evolutionary theory, island biogeography to management. However, management is currently the focus of international targets to reduce the impacts of INNS and is worthy of separate consideration. Management-based research in invasion science has been found to receive fewer citations than more fundamental research (Pyšek et al. 2006; Esler et al. 2010). This result reinforces previous studies that found research outputs in this and other disciplines are often poorly translated into policy or management action (Esler et al. 2010; McGeoch et al. 2010; Matzek et al. 2015). Scientific articles are often not the primary mechanism for disseminating outcomes of management and policy actions, and often these documents can be difficult to access, both for practitioners and researchers (Catalano et al. 2019; Kadykalo et al. 2021). While there is a need for both broad and focused research on INNS, improved communication and collaboration between disciplines, and between researchers and stakeholders, would be mutually beneficial and will increase the prospect of meeting future global targets for INNS management.

Knowledge gaps in invasion research

When considering the topics generated and the words that represented them, there were some notable absences. The social dimensions of INNS were not well represented within the corpus. Terms such as socio-economic, livelihood and well-being for example, were not present amongst highly weighted words. The role and impact of INNS in these and related contexts is a complex and important topic that requires further research (Shackleton et al. 2019). Alongside this is a need to better understand human perceptions towards INNS to better inform communication, management and policy (Shackleton et al. 2018). Terms referring to invasions in the Arctic were also not present amongst highly weighted words. While this region has been viewed as having few currently established INNS and a low invasion risk, loss of sea ice and increased shipping activities are expected to drive an increase in invasions (Chan et al. 2019). Recent horizon scanning has suggested this issue will soon feature widely in the literature (Ricciardi et al. 2017), but our findings suggest that a greater scale of invasions may be required first, before the scientific community responds.

While the application of topic modelling provides a quantitative approach to literature analysis, there was still bias in this work. For example, using abstracts written in English will cause an uneven geographical representation. Within the corpus, the abstracts contained a high number of synonyms for INNS, such as “alien”, “exotic”, “introduced” and “non-indigenous”. While the aim was to encompass the majority of relevant literature with the search terms applied, papers solely using these alternative terms are likely to have been excluded from the corpus which might have contributed to the prevalence of particular disciplines and uneven geographical spread linked to usage of these terms. However, we found that using expanded search terms accounting for synonyms yielded a higher proportion of irrelevant articles, including from e.g., the medical sciences, and that the same topics and key trends were returned. We conclude that our narrower search string captures the diversity of the scientific literature, while excluding a greater proportion of irrelevant articles. Based on the results of this study, future investigations could be conducted using a similar methodology on a subsection of invasion science, such as Marine systems or Management methods. As these were such broad topics it would be beneficial to determine the more specific topics and trends within these areas.

Conclusion

This study provides an insight into the publishing landscape of invasion science by quantitatively assessing the research trends within the literature. The topics addressed varied widely, reflecting the multidisciplinary nature of the field, and new technological developments visibly impacted the publishing landscape. However, our results revealed a prevalence of specific taxonomic groups and geographical regions within the literature, and a paucity of interdisciplinary collaboration in the critical areas of policy and management. The risk is that research outcomes have low relevance or perceived transferability for practical management and policy, or that relevant research fails to reach the right audiences, and that some stubborn problems will persist for lack of more creative or joined-up thinking. To achieve targets for INNS management there will need to be improvements in how research is targeted, communicated and used. While emerging topics such as genetic identification and control offer new capabilities, we found a general lack of cross-sectoral interdisciplinarity, and we suggest that addressing this gap, particularly in relation to policy and management, would strengthen and steer the policy-management-science interface.