When we rank things—whether they are power grid projects, university teachers, or nodes in a social network—we rarely rely on a single metric. Instead, we look at a constellation of factors: cost, efficiency, impact, and risk. The central challenge in these Multi-Criteria Decision-Making (MCDM) problems is not just gathering the data, but deciding how much each factor should "count."
Traditionally, this has been solved by assigning weights to each criterion. However, as Juan-juan Peng and Chao Tian point out, relying on experts to manually assign these weights introduces significant subjective bias. If two experts disagree on the importance of environmental impact versus economic return, the final ranking reflects their personal preferences rather than the intrinsic properties of the data.
To solve this, researchers are moving toward objective weighting mechanisms—mathematical frameworks that derive the importance of a factor directly from the data itself.
The Entropy of Information
One of the most prominent tools for removing human bias is the Entropy Weighting Method (EWM). The intuition is rooted in information theory: if a specific criterion (like "years of experience") is nearly identical across all candidates, it provides very little information for ranking them. Conversely, if a criterion shows wide variation, it is highly "informative."
Xiaojuan Tang demonstrates this by applying entropy to rank the competitiveness of power enterprises, avoiding the "ranking reversals" common in subjective models. Similarly, Zhenlin Dai and colleagues used EWM to balance satellite imagery and socio-economic data for light pollution indexing, ensuring that regional comparisons remain robust across diverse urban landscapes. By letting the statistical dispersion of the data dictate the weights, researchers like Hongwu Qin can create more reliable models for high-stakes environments like sepsis diagnosis.
Ranking Without Weights: Pareto Dominance
A more radical approach to objectivity is to avoid assigning numerical weights altogether. Instead of trying to collapse multiple scores into a single weighted sum, researchers use Pareto dominance (or "equivalence classes").
In this framework, an item is considered "better" only if it is at least as good as another in all categories and strictly better in at least one. Bojin Zheng and Jianmin Wang 0001 applied this to complex networks, identifying important nodes without needing to decide if "degree" is more important than "betweenness." In academia, Dan Song and Sha Fu used Pareto non-dominated sorting to evaluate teachers, treating research and teaching as distinct objectives to identify "non-inferior" candidates rather than forcing a biased compromise.
Extracting Structure with Rough Sets
When data is "fuzzy" or incomplete, researchers often turn to Rough Set Theory. This method identifies the "core" attributes of a dataset—the minimum set of factors needed to maintain the same classification power as the full set.
Yuheng Sha and Yanlai Zhang utilized this to evaluate power grid projects, using the theory to objectively weight performance indicators based on their actual influence on project success. Similarly, Xu Zheng applied Rough Sets to power network planning to remove the "arbitrariness" of expert opinion, ensuring that long-term infrastructure decisions are grounded in data-driven coordination degrees.
Learning Weights through Neural Networks
Perhaps the most sophisticated evolution of this field is the use of machine learning to "learn" the optimal weighting function. Traditional linear models assume that if you double a score, you double the quality. But real-world performance is often non-linear.
Xianggen Gan and Wu Ren replaced subjective linear models with Backpropagation (BP) Neural Networks for teaching evaluation. By training the network on historical data, the model discovers the complex, non-linear mapping between performance indexes and outcomes. In network science, Tingshuai Jiang and Yirun Ruan developed EMCNN, a deep learning framework that uses entropy to weight structural metrics before passing them through a convolutional neural network to predict a node's true influence.
Hybridizing Subjectivity and Objectivity
While the goal is to minimize bias, some researchers argue that expert intuition still holds value if properly constrained. This has led to hybrid models like CRITIC-TOPSIS and Fuzzy AHP.
Jingdan Liu and Xun Bao used the CRITIC method (Criteria Importance Through Intercriteria Correlation) to evaluate AI in education. CRITIC calculates weights by looking at both the standard deviation of a criterion and its correlation with others—mathematically capturing the "conflict" between factors. Meanwhile, Tanmayee Tele and Murat Kar used Fuzzy Analytical Hierarchy Process (FAHP) to model the "fuzziness" of human judgment, transforming qualitative expert intuition into a rigorous, data-driven ranking for construction contractors.
By moving from "expert-assigned" to "data-derived" weights, these researchers are building evaluation systems that are not only more accurate but also more justifiable and transparent across fields as diverse as genomics, maritime logistics, and stock selection.