When Optimization Meets Clickbait: Metaphor-Based Metaheuristics

Özgür Oskay
Eindhoven University of Technology

Not everything that glitters is gold, and not every algorithm inspired by a dancing octopus is novel.

“Ten tricks to change your life instantly.” “This one simple method scientists don’t want you to know.” Phrases like these have become a staple of digital culture. They represent clickbait: content whose main purpose is to attract attention and encourage visitors to click, often through eye-catching headlines or titles. Unsurprisingly, they are now almost unavoidable, shaping the way we encounter news, entertainment, and even everyday information. What is less obvious, however, is that a similar phenomenon can be found in academia. In the field of optimization, certain metaphor-based metaheuristics (MBMs) can be classified as academic clickbait

Rise of Metaphor-Based Meteheuristics

The term “metaphor-based metaheuristics” may sound complicated at first, but the underlying ideas are straightforward. Heuristics, from the Greek word heuriskein meaning “to find” or “to discover,” are problem-solving strategies designed to quickly produce good or approximate solutions. In optimization, heuristics are not guaranteed to deliver the perfect answer, but they are valued for their efficiency in tackling complex problems where exact solutions would take too long to compute. Metaheuristics extend this idea further.  They are higher-level strategies that provide a general framework for guiding the search process across a wide range of problems. A good example is tabu search, proposed by Glover (1986) in the same work where the term metaheuristic was formally introduced. Tabu search keeps a memory of recently visited solutions (with a tabu list) and temporarily excludes them from consideration, preventing the search from circling back to the same points and pushing it toward unexplored areas. This simple principle illustrates the idea of metaheuristics: adaptable strategies that can be applied to many different problems.

From the 1970s onward, researchers began to design metaheuristics inspired by natural or physical processes. One of the earliest and most influential was simulated annealing, introduced by Kirkpatrick et al. (1983). The method borrowed its metaphor from metallurgy, where a material is slowly cooled in order to settle atoms into a stable, low-energy structure. In simulated annealing, this notion of slow cooling is implemented as a gradual decrease in the probability of accepting worse solutions while exploring the search space. Accepting worse solutions allows the algorithm to broader exploration and increases the chances of finding a global optimum. The algorithm has proven useful for many types of problems and remains influential in the design of metaheuristics. Over the years, many other important MBMs were proposed, ranging from evolutionary algorithms and genetic algorithms to ant colony optimization and particle swarm optimization. Together, these approaches established a diverse set of strategies that combined inspiration from nature with genuine algorithmic contributions. For a comprehensive overview of this development, see Martí et al. (2025).

Figure 1: Dancing Octopus

With the success of early nature-inspired algorithms, the number of MBMs began to rise rapidly in the late 1990s and early 2000s Sörensen (2015). What started as occasional creative analogies soon turned into a flood of publications proposing methods inspired by increasingly unusual sources. Alongside natural metaphors like birds or bees, researchers began drawing from themes as diverse as reincarnation, zombies, interior design, and even sports tournaments! Thanks to the humorous catalog named the Evolutionary Computation Bestiary¹, we have a record of this phenomenon. The bestiary tries to collect all known metaphor-based algorithms, ranging from the influential to the outright bizarre. Importantly, neither the website nor this article makes claims about the quality of the individual papers listed in the bestiary. Classic, well-validated contributions appear side by side with more suspicious ones. As the authors put it:

I encourage readers to explore the Bestiary themselves; its bizarre entries capture both the creativity and the excesses of this trend. In the early 2000s, only a handful of new MBMs appeared each year. By the 2010s, however, the field was producing them at a rate of one or two per month (Campelo and Aranha (2023)). As the numbers grew, however, so did the problems.

Problems Started Arising

One of the central problems in MBMs was the lack of genuine novelty. Many MBMs presented as “novel” were in fact just repackaged versions of existing methods Sörensen (2015). An example of this was shown by Weyland (2010), who carefully analyzed the Harmony Search (HS) algorithm - an algorithm inspired by the improvisation process of jazz musicians - and demonstrated that it was nothing more than a special case of another algorithm, Evolutionary Strategies (ES). This meant that the HS algorithm couldn’t outperform the best ES. Even with this fact, the study identified several other papers which claimed successful results using the HS algorithm. This was possible because of another recurring issue: poor experimental validation.

Several concerns in MBM papers relate directly to experimental validation procedures (Campelo and Aranha (2023)). To begin, many studies focused almost exclusively on competitive testing rather than on the underlying principles of the algorithms. Moreover, reported results often relied on overfitting to specific benchmark instances, with little evidence that performance generalizes more broadly. Futhermore, comparisons were frequently made only against similar metaphor-based approaches, instead of against state-of-the-art methods, and sometimes even under conditions of unbalanced tuning. Altogether, these practices created a misleading impression of novelty.

In addition to these concerns about novelty and empirical rigor, another recurring issue was the usage of unnecessary metaphorical terminology. Instead of describing algorithms with standard optimization concepts, many MBM papers relied on the language of their chosen metaphor. So even if there exists a novelty, such terminology hindered clear analysis, complicated comparison with existing methods, and ultimately reduced the accessibility of the research.

Unfortunately, the impact of such MBM papers extends beyond their own limitations. They add noise to the literature, making it harder for genuinely novel contributions to be identified. They also consume the time and attention of researchers. For students and researchers outside the field, these papers can be particularly misleading, creating the impression of progress where little exists. The overall effect is “disappointment and detrimental” to the field as a whole Sörensen (2015).

Figure 2: Clickbait

References: 

Campelo, F., Aranha, C., 2023. Lessons from the evolutionary computation bestiary. Artificial Life 29, 421–432.

Glover, F., 1986. Future paths for integer programming and links to artificial intelligence. Computers & operations research 13, 533–549.

Glover, F., Laguna, M., 1997. Tabu search. Springer. doi:https://doi.org/10.1007/978-1-4615-6089-0. Kirkpatrick, S., Gelatt Jr, C.D., Vecchi, M.P., 1983. Optimization by simulated annealing. Science 220, 671–680.

Martí, R., Sevaux, M., Sörensen, K., 2025. Fifty years of metaheuristics. European Journal of Operational Research 321, 345–362. doi:https://doi.org/10.1016/j.ejor.2024.04.004.

Sörensen, K., 2015. Metaheuristics—the metaphor exposed. International Transactions in Operational Research 22, 3–18. doi:https://doi.org/10.1111/itor.12001.

Weyland, D., 2010. A rigorous analysis of the harmony search algorithm: How the research community can be misled by a "novel" methodology. Int. J. Appl. Metaheuristic Comput. 1, 50–60. doi:10.4018/jamc.2010040104.

Acknowledgments: We would like to thank Marziehsadat Rezaei for taking time to review this article. Illustration credit goes to Vera Lal Erdoğdu. Photo credit goes to Anne Nygård for header photo.

Disclaimer: The author of this article does not intend to make or imply any claims regarding the integrity, competence, scientific quality or conduct of any individual researcher, paper, or journal. The discussion is limited to general trends in the field and references to previously published critiques, and should be understood as commentary on systemic academic practices.