A Metascience Study of the Low-Code Scientific Field

Low-Code promotes the development of software in a simplified manner, commonly involving a graphical user interface while requiring minimal coding skills from the user. Because of such simplification, Low-code is trending and replacing other similar model-based practices, such as Model-Driven Architecture (MDA), Model-Driven Development (MDD), Model-Driven Engineering (MDE),… At least in terms of the terminology used by many tools.

Despite the potential technical differences between the two concepts, both Low-Code and traditional modeling share commonalities and the goal to accelerate the creation of software systems. But beyond the technical aspects, we were interested in studying the community/adoption aspects of low-code, in complement (or competition!) with other model-driven approaches. In particular , we were wondering about questions such as:

• What are the characteristics of the Low-Code community and how is this community evolving?
• How does the Low-Code community compare to and influence the traditional Modeling community?

Such analysis could benefit both communities, especially to help them better understand their relationship and, therefore, their potential synergies and growth opportunities.

To perform the analysis described above, we focus on answering the following two questions, with 5 sub-questions each:

1. What is the composition and growth of the Low-Code research field?
   1. How has the number of publications related to Low-Code evolved over time?
   2. What are the most common publication types and venues for Low-Code research?
   3. How do Low-code publications contribute to the field of low-code?
   4. What is the typical author profile in Low-Code publications?
   5. How many Low-Code platforms and tools are available as open-source repositories, and which are the most popular?
2. What is the impact of Low-Code on the traditional Modeling research community?
   1. Are Low-Code publications affecting the number of traditional Modeling publications?
   2. How many Modeling conferences and workshops aim to explicitly attract Low-Code publications?
   3. Are Low-Code publications present in Modeling conferences and workshops targeting them?
   4. How are authors distributed across Low-Code and traditional Modeling publications?
   5. How many Low-Code open-source platforms and tools explicitly relate to traditional Modeling techniques?

Metascience analysis of low-code publications

To answer questions 1.1, 1.2, 1.4, 2.1, and 2.4, we used Lens.org to collect articles containing “Low-Code” or “Lowcode” in their title, abstract, or keywords; and articles containing either “Model-Driven Architecture”, “Model-Driven Development”, “Model-Driven Engineering”, “Model-Based Architecture”, “Model-Driven Software Engineering”, or “Model-Based Engineering” in the same fields. As a result, we collected 1,309 articles related to Low-Code; and 14,105 articles related to traditional Modeling. We manually cleaned the Low-Code set, removing 465 irrelevant, inaccessible, or duplicate articles. Due to scale, the traditional Modeling set was not manually cleaned. We then conducted various analyses, detailed below.

To answer question 1.3, we manually classified the previously identified Low-Code articles into one of the following categories: (1) Low-Code users, for papers using Low-Code in their solutions; (2) Low-Code solutions, for papers proposing frameworks, techniques, or solutions to improve the Low-Code area; (3) Low-Code platforms, for papers introducing new platforms for Low-Code users; (4) Low-Code evolution, for papers analyzing existing Low-Code publications; (5) Low-Code learning/teaching, for papers focused on teaching or using Low-Code in education; and (6) Others, for papers not fitting into these categories.

To answer questions 1.5 and 2.5, we used the Dashboard of open source low-code tools to identify open-source Low-Code repositories on GitHub.

For question 2.2, we reviewed 64 conferences and 84 workshops on modeling, extracting relevant text from their websites (e.g., call for papers, topics of interest, keynotes) and searched for Low-Code-related terms.

Finally, for question 2.3, we analyzed all abstracts from MODELS, MODELS-C, and ECMFA published in 2023, extracting keyphrases to identify the main topics covered in those conferences.

1.1. Number of publications related to Low-Code

In Figure 1, we see that the first Low-Code publications appeared in 2017 and its presence in the literature has been increasing since then with a steep rise. Additionally, we note that in 2023, the number of Low-Code papers surpassed the number of “traditional” Modeling papers.

1.2. Most common publication types and venues for Low-Code research

In Figure 2, we observe that Low-Code publications are mostly published in conference proceedings, accounting for 46% of the total articles. In Figure 3, we note that The ACM/IEEE International Conference on Model-Driven Engineering Languages and Systems Companion (MODELS-C) is the most popular venue for Low-Code articles — as expected as it hosts the International Workshop on Modeling in Low-Code Development Platforms (LowCode). Additionally, we note that 56% of Low-Code papers were published in conferences with no other Low-Code publications, suggesting that authors may lack an established venue to publish their work.

 

1.3. Contribution of Low-Code publications to the field

In Figure 4, we observe that most Low-Code publications are aimed at expanding the current state of the practice by developing Low-Code solutions for specific problems. Furthermore, publications presenting new Low-Code platforms are lower than those using them. Interestingly, Low-Code papers often use the paradigm as a tool in disciplines outside Computer Science, such as Medicine, Meteorology, and Business.

1.4. Low-Code author profile

In Figure 5, we note that the majority of Low-Code authors are relatively new to the field, with only 4% of them having three or more publications related to Low-Code. Interestingly, in Figure 6 we see that authors with a larger number of publications tend to be part of larger co-authorship groups, having collaborated with a broader network of researchers. Still, in Figure 6b we observe that collaboration among research groups is limited.

 

1.5. Open-source Low-Code tools and platforms on GitHub

As can be observed in the Dashboard of open source low-code tools, there is a large number of Low-Code repositories on GITHUB. In Table 1 we list the ones that have garnered more attention from the community, achieving tens of thousands of stars. This demonstrates that the Low-Code topic is being actively explored by developers.

2.1. Impact of Low-Code in Modeling publications

From Figures 1 and 7 we observe that Low-Code publications began to increase about 6 years after traditional Modeling publications started to decline. Today, Low-Code publications receive comparable attention from researchers as to that of traditional Modeling paradigms.

2.2. Modeling conferences Targeting Low-Code

From Table 2 we see that few Modeling venues are specifically targeting Low-Code. Despite the increasing attention Low-Code is receiving from researchers and developers, the number of Modeling events encouraging submissions on this topic remains limited.

2.3. Presence of Low-Code in Modeling Conferences

In Table 3 we note that Low-Code does not appear as one of the most prominent topics in any of the proceedings of the Modeling venues we analyzed. This suggests that the Low-Code topic has not yet fully integrated into the traditional Modeling community despite the openness of such conferences.

2.4. Authorship Distribution between Low-Code and Modeling

In Figure 8a, it is shown that the overlap between authors who have published in both traditional Modeling and Low-Code is small, around 15% of the analyzed authors. Such overlap increases (cf. Figure 8b) to 28% when analyzing the most influential authors of each field. Furthermore, in Figure 9 we see that 10 of the most influential and productive authors in Low-Code also have a significant presence in the traditional Modeling community. On the other hand, some authors are highly productive in Low-Code but have no publications in traditional Modeling.

2.5. Low-Code tools explicitly related to Modeling

In Table 4, we show that only nine Low-Code platforms and tools available on GitHub are explicitly related to traditional Modeling. These tools primarily focus on accelerating the development of AI-based software or automating CRUD-related operations. Among them, BESSER is the only research-oriented platform identified.

 

Insights and recommendations to the Low-Code and the traditional Modeling communities:

• Decline in the modeling community: Low-Code publications now surpass any single Modeling paradigm and are approaching the total volume of all Modeling-related research. This is due to both rising interest in Low-Code and declining engagement in traditional Modeling, particularly outside core conferences.
  • Recommendation: Researchers should recognize the overlap between Low-Code and Modeling. Modeling authors unfamiliar with Low-Code risk missing broader audiences and reducing their impact. Adopting a Low-Code perspective can enhance visibility and application. Conversely, Low-Code can benefit from decades of Modeling research to avoid repeating past challenges.

• Do not reinvent the wheel: A 28% author overlap suggests senior Modeling researchers are influencing Low-Code. However, younger researchers may be unaware of existing Modeling work.
  • Recommendation: Senior researchers should guide newcomers in understanding foundational Modeling research. Repackaging existing techniques with Low-Code terminology alone is not acceptable.

• Workshops and arXiv entry points for Low-Code: Low-Code is entering the Modeling community mainly through MODELS workshops, while the high presence of publications on arXiv may demonstrate new trends in publishing research work.
  • Recommendation: The Modeling community should ensure Low-Code research is not sidelined to workshops and is fairly considered for core conference tracks.

• Low-Code papers outside of Modeling venues: Despite leveraging Modeling concepts, most Low-Code papers appear in unrelated venues. Additionally, 56% of venues with Low-Code publications have published a single Low-Code paper, indicating the lack of an established venue in the area.
  • Recommendation: Modeling venues should encourage Low-Code submissions via application tracks, tool demos, or special workshops to promote cross-field collaboration. This could initially focus on topics that bridge the two fields, such as applying traditional modeling techniques to address Low-Code challenges or presenting novel Low-Code solutions inspired by modeling principles.

• Low industry association between Low-Code and Modeling: Only 6% of Low-Code tools on GitHub explicitly reference Modeling, suggesting industry does not perceive Low-Code as part of Modeling.
  • Recommendation: Increase industry awareness by showcasing Low-Code solutions in Modeling venues—industry tracks, demos, and forums—to foster collaboration. Such measure could encourage the exchange between both communities and additionally highlight the importance of traditional Modeling solutions to Low-Code developers.

• Research on Low-Code Improvements: During the analysis of Low-Code publications, we observed that most Low-Code papers focus on domain-specific frameworks rather than advancing the broader paradigm.
  • Recommendation: We see a clear need for research that advances the Low-Code paradigm itself. Thus, more research should address fundamental Low-Code challenges, including usability, best practices, testing, and integration with Modeling solutions, to drive long-term progress.

You can read an expanded version of these analysis and reflections in our arxiv paper. This work will also be part of the upcoming ECMFA conference.

Let us know whether you agree with our reflections and/or what other analysis you’d like to see!