It’s possible that I shall make an ass of myself. But in that case one can always get out of it with a little dialectic. I have, of course, so worded my proposition as to be right either way (K.Marx, Letter to F.Engels on the Indian Mutiny)
Tuesday, January 06, 2026
The book only gets 3 stars, but is considered great literature
A new study from the Center for Humanities Computing and the Center for Contemporary Cultures of Text at Aarhus University shows that star ratings of books are not always accurate.
You may have tried it yourself: to deselect a book because it "only" has about three stars on Goodreads. But according to a new study from the Center for Humanities Computing (CHC) and the Center for Contemporary Cultures of Text (TEXT), these books may well contain great literary value.
Goodreads is an international platform where millions of readers rate books between one and five stars. The average is often used as a quick indicator of quality – also by publishers, authors and researchers. But when a book ends up in the middle of the scale, the number says far less than you might think.
The researchers from CHC and TEXT have analyzed about 9,000 American novels published between 1880 and 2000. They have particularly focused on just over 2,000 books with average Goodreads ratings in the middle field. By comparing the readers' stars with other measures of literary quality, the researchers have investigated what is hidden behind the seemingly mediocre figures.
The results show that about 30 percent of these 2,000 "mediocre" books are rated as literary important or of high quality according to other criteria – for example, whether they are considered classics, are part of education or have had great cultural significance.
According to the researchers, the mediocre ratings are often not due to the fact that the books are boring. On the contrary.
A key finding of the study A key finding of the study is that disagreement among readers does not arise by chance:
"For books that are considered to be of literary significance, we see that the more readers who rate them, the greater the disagreement between readers. Some give top marks, others are critical – and it is precisely this spread that characterises books that engage," says PhD student Pascale Feldkamp, who is behind the study together with colleagues from the Center for Humanities Computing and TEXT.
For books that are generally assessed as less important, the same correlation is not seen. Here, several assessments do not lead to major disagreement. This indicates that split ratings are not just an expression of random noise, but are linked to books that actually mean something to readers.
When disagreement grows as more people read along, it is not a sign of indifference – but of importance." The study thus challenges the notion that a book's value can be read directly in its average star rating. An average rating can cover very different situations. Sometimes it is an expression of a broad but lukewarm agreement. Other times, it hides strong and opposing reading experiences that cancel each other out on average," says Pascale Feldkamp.
An average may look neutral, but can in reality be the sum of strong opinions that point in different directions.
The study's main conclusion is therefore that an average Goodreads rating does not automatically mean that a book is unimportant. On the contrary, it can point to works that are controversial, polarizing – or later recognized as literarily important.
According to the researchers, if reader data from platforms like Goodreads is to be used to say something meaningful about literary success or value, it requires a more nuanced approach. It is not enough to look at one number. You also have to look at how many people are assessing and how much they disagree.
Behind the research results Study type: Computational study Authors: Pascale Feldkamp Moreira, Yuri Bizzoni, Mia Jacobsen, Mads Rosendahl Thomsen and Kristoffer L. Nielbo Link to the scientific publication "The Goodreads' ›Mediocre‹: Assessing a Grey Area of Literary Judgements" in Zeitschrift für digitale Geisteswissenschaften External funding: Velux Fonden
fact box
Highly rated books with a polarizing character
Highly rated, acclaimed books can get a medium rating because they evoke strong, opposing reactions in readers – either because of style, theme or point of view. Some examples:
James Joyce: Ulysses (stylistically experimental)
Vladimir Nabokov: Lolita(provocative theme)
William Faulkner: The Sound and the Fury (fragmented narrator's voice)
Malcolm Lowry: Under the Volcano(complex style)
Ayn Rand: The Fountainhead(politically controversial)
Tim LaHaye & Jerry B. Jenkins: Left Behind (Ideologically Polarizing)
The average rating can thus hide both fascination and frustration.
The house sparrow is about 15 centimeters long and weighs about 25–35 grams. It has brown and black striped upperparts and gray undersides. The male has a black throat patch and gray skullcap, while the female has a brown skullcap.
Researchers are trying to understand why some wild species do better than others over time, as the environment changes.
Researcher Kenneth Aase's research focuses on a new mathematical approach that could shed light on this question, which in turn could move us closer to understanding the loss of biological diversity. Aase is a statistician and a PhD research fellow at the Norwegian University of Science and Technology (NTNU's) Department of Mathematical Sciences. He is associated with the GPWILD project, funded by a European Research Council Consolidator Grant. The project involves using biology and mathematics to understand more about a species' adaptive evolutionary potential, and relies on genetic and body data from tens of thousands of house sparrows who live in the northern Norwegian district of Helgeland.
Why house sparrows?
House sparrows turn out to be the perfect critter for research like Aase’s.
"Because our island populations are small and delimited, they are exceptionally well suited for research. Biologists can record and follow almost all individual sparrows from birth until they die,” Aase said.
NTNU researchers at the Department of Biology and the Gjærevoll Centre have been studying these house sparrows for more than 30 years, and have an enormous database of information, he said.
"They can investigate what affects their survival, and how many young they have. We have been collecting such data for over 30 years, and have produced long-term datasets that are both unusual and completely invaluable. They help us understand the consequences of changes in the environment, as well as genetic and ecological development over many generations,” he said.
And what’s more, “what we learn is transferable to many other species,” he said.
Genomic prediction as a tool
Aase’s work focuses on a technique called genomic prediction, or GP.
This is a statistical method for finding out how an individual's genes affect a trait in an individual or a human being. The trait can be anything that can be measured, such as height, illness, or body weight. The method can be used to predict how much yield a grain plant can yield, or whether a person is genetically predisposed to certain diseases, Aase said.
"The method can also tell us whether the genes of a given house sparrow will give it a higher or lower body weight. This is important for the sparrow's ability to survive. GP is widely used in plant and animal breeding, but so far it has not been used much in research on populations of wild animals and plants,” he said.
As the researchers have gained more and more access to genetic material from wild populations, Aase and his colleagues, led by Professor Stefanie Muff, will investigate how useful the method can be in ecology, evolution and conservation biology.
From a training group to an individual
All of this genetic information enables the researchers to use hundreds to millions of genetic markers spread across the genetic material, and link them to measurements of the trait from a “training group”.
The statistical model works even if the trait they’re interested in hasn’t been measured in the given individual – it is enough to have measured it in the training group.
"As long as we have information from the same genetic markers in both the training group and the individual, we can calculate how the genes affect the trait. The accuracy depends in part on how many individuals are in the training group, the number of genetic markers, and the heritability of the trait,” Aase said.
Tested in different populations
The researchers wanted to know how accurate their statistical model would be if their training group consists of sparrows from a different population than the individuals they were interested in.
“This is important in order to be able to investigate crucial natural processes in new and more efficient ways. For example, we can save a lot of fieldwork, because the researchers do not need to obtain measurements of traits from each individual population they are interested in,” Aase said.
Here’s where the house sparrow data from Helgeland was perfect for the researchers to test their model on.
"In the new study, we used measurements of body traits from wild house sparrow populations from islands along the Helgeland coast. Because the islands are more or less clearly separated, we could answer the research question by making predictions across different islands and archipelagos,” Aase said.
Aase and his colleagues found that making predictions with a training group from one sparrow population for other, different sparrow populations in the islands didn’t work as well as when they made predictions with a training group from the exact same sparrow population.
"We found that predicting across different populations works less well than within populations. This was expected, based on previous studies in breeding and medical research. However, we were the first to demonstrate this in wild populations. We also provided new insights that can be useful for improving GP across populations,” he said.
Wild populations challenging
The technique of using GP was developed for use with domestic animals, where researchers have access to all the genetic information they need. That’s not always the case with wild populations, Aase said.
“For us statisticians, perhaps the biggest challenge is that field datasets are often incomplete. We do not always get genetic data or measurements of all individuals. In addition, we usually do not have data from controlled trials, for example because environmental conditions change over time and space,” he said.
"Studies of wild animals are often exploratory rather than confirmatory. There are few such thorough and long-term studies of wild populations in the wild, but the house sparrow data made this new study possible,” he added.
That’s where the Helgeland house sparrow data offer clear advantages, because it is almost complete, he said.
"As a statistician, I am fortunate to have the data served on a silver platter by the biologists I collaborate with at the Gjærevoll Centre and the Department of Biology at NTNU. They have been working on collecting this unique dataset for more than three decades. So there is a lot of field and laboratory work involved before I get to play with the end result,” Aase said.
"In addition to data from the house sparrow populations, I also use computer simulations where you can test model assumptions. My everyday life is spent programming statistical analyses. For the most challenging computations, I use NTNU’s supercomputer IDUN,” he said.
Facing the “sixth mass extinction”
Natural resource managers and conservation biologists need to know how changes in the environment, either from a warming planet or from loss of habitat – or both – will affect wild populations.
That’s where using GP as a tool can make a difference, Aase said.
"Climate change and increased land use mean that many populations of wild animals and plants are exposed to increased external pressure and faster environmental changes. Understanding both the genetic and ecological consequences of this is necessary for nature managers and conservation biologists to be able to prioritize measures, such as which populations need protection and how,” he said.
"GPs can tell us about how viable individuals are under given environmental conditions. Thus, it can be used to reintroduce or strengthen populations. This knowledge also helps to increase our basic understanding of natural processes, and how evolution actually plays out in nature,” he added.
In this way, “studies of house sparrows in populations along the Norwegian coast can help us preserve populations of other species that are threatened with extinction due to the changes we humans make in nature,” he said.
The world is facing a biodiversity crisis, where human activity is causing what is called the “sixth mass extinction”.
“If we want to stop this development through targeted measures, we need both good analytical tools and basic knowledge about how evolution in nature works. I would also argue that there is an intrinsic value in such a basic understanding,” Aase said.
From house sparrows to Svalbard reindeer
As he proceeds with his PhD research, Aase will continue to investigate how genomic prediction can be used in wild populations.
“In GPWILD, we're going to put these questions into a broader framework. The project will work with several other animal species, such as Svalbard reindeer, deer from Scotland, arctic foxes, and several bird species,” he said.
But he isn’t quite done with house sparrows just yet,
"Currently, I am working on an applied study where GP is used to investigate how certain genetic processes affect the fitness of the house sparrow,” he said.
Aase's academic article was as an "Editor's Choice" article for December in the academic journal Evolution.
The house sparrow feeds on grains, seeds and insects, but also eats leftovers, buds, flowers and berries. It often visits bird feeders in winter.
Credit
Photo: Thor Harald Ringsby, NTNU
Helgeland has many islands - with their own limited populations of house sparrows. This makes the region particularly suitable for research. Biologists can record and follow the life cycle of almost all individual sparrows.
(A) Object transport demonstration. (i) The full process of lifting the target object. After the tethered UAV autonomously performs knotting on top of a tall building, the winch reels in the tether to lift the object. (ii) Autonomous knotting procedure. The UAV first performs object enclosing, then enters the tether search state, and finally executes tether binding to complete knotting. (B) Radar chart comparing the proposed system with other state-of-the-art aerial/cable-driven transport robots, including the commercial high-payload multicopter (DJI FC 30), commercial high-payload helicopter (Blowfish A2G), and a portable wire-driven parallel robot (Cubix). PWR (payload-to-weight ratio) is the metric describing the ratio of the maximum payload capacity to the weight of the robot. The outer ring of the radar chart is color-matched to the best-performing system for each corresponding metric.
Credit: Lihua Xie, Nanyang Technological University, School of Electrical and Electronic Engineering
“Cable-driven systems excel at heavy-load transport but are limited by fixed anchoring points in unstructured environments,” explained study corresponding author Lihua Xie from Nanyang Technological University. The core innovations include (a) a human-in-the-loop knot planner integrating enclosing plane extraction, frontier-based path search, and knotting trajectory generation; (b) three key optimization metrics (enclosing planarity, tether visibility, tether clearance) ensuring task reliability; and (c) seamless integration of UAV mobility and winch load-bearing capability. “This system enables rapid deployment of transport tasks without pre-designed anchors, expanding robotic logistics to complex scenes.”
The system leverages key technical advancements: The knot planner interprets user sketches to extract enclosing planes, searches for paths around target structures via frontier clusters, and generates optimized trajectories. The UAV is equipped with LiDAR, RGB-D cameras, and an onboard computer for real-time perception, mapping, and tether detection. “The three metrics work synergistically—enclosing planarity prevents tether slippage, tether visibility maintains real-time monitoring, and tether clearance avoids collisions,” said co-first author Rui Jin.
The study authors validated the system through real-world and simulation experiments: In an urban outdoor environment, the system autonomously completed knotting on a linkway roof and lifted a 15.3-kg payload to 3.5 m in 42.1 s. Simulation tests confirmed shape-agnostic performance, with success rates exceeding 90% on four distinct structures (pipeline, archway, billboard, bridge) across 30 trials per target. Ablation experiments verified the necessity of the three metrics—removing enclosing planarity reduced success rate to 8%, while omitting tether visibility or clearance impaired binding robustness.
“While the system shows strong performance, it faces limitations: reliance on clear visual access to the tether, sensitivity to environmental disturbances, and the need for mechanical optimization of the winch-tether mechanism,” said co-first author Xinhang Xu. Future work will focus on advanced control algorithms to compensate for tether-induced disturbances, multimodal sensing for reliable tether detection, and explicit modeling of tether-environment interactions. Overall, this autonomous knotting system offers a novel solution for rapid-deployable heavy-load transport, unlocking new capabilities in robotic logistics for unstructured environments.
Authors of the paper include Rui Jin, Xinhang Xu, Yizhuo Yang, Jianping Li, Muqing Cao, and Lihua Xie.
This research was partially supported by the Ministry of Education, Singapore, under AcRF TIER 1 Grant RG64/23, and the National Research Foundation Medium-Sized Centre for Advanced Robotics Technology Innovation.
The paper, “Tethered UAV Autonomous Knotting on Environmental Structures for Transport” was published in the journal Cyborg and Bionic Systems on Dec. 26, 2025, at DOI: 10.34133/cbsystems.0450.
