Divisiveness-Consistent Label Distribution Learning

Label Distribution Learning (LDL) is an effective learning paradigm for predicting entire conditional label distributions, improving the trustworthiness of predictions in risk-sensitive tasks. Although previous LDL methods achieve satisfactory performance on conventional evaluation metrics, they generally overlook the divisiveness within label distributions, i.e., the propensity of label distribution to exhibit dissension between semantically opposing labels, which is an essential indicator of the practical decision risk. Therefore, we propose a divisiveness‑consistent label distribution learning framework to quantify and preserve the divisiveness information. First, we formalize a divisiveness measure that satisfies the axiomatic property of polarity monotonicity to quantify the divisiveness information. Second, we theoretically demonstrate the inconsistency between conventional loss functions and divisiveness error. Besides, in order to address the adversarial gradient problem arising from directly minimizing the divisiveness error, we propose a pairwise divisiveness loss as an unbiased estimator of the original divisiveness error. Experiments confirm the effectiveness of the proposed method.

This paper studies how machine learning models can better capture divisiveness, i.e., situations where opinions are strongly split between opposing labels, such as highly polarized movie ratings or public reactions. Existing label distribution learning methods mainly focus on matching overall prediction accuracy, but often fail to preserve whether opinions are actually polarized or consensual. To address this, the paper introduces a new framework called divisiveness-consistent label distribution learning. We first define a principled way to measure divisiveness and show theoretically that common learning objectives cannot reliably preserve this information. They then design a new training loss that helps models maintain divisiveness while still learning accurate label distributions. Experiments on multiple real-world datasets show that the proposed method more accurately captures polarization patterns than previous approaches, while remaining competitive on standard evaluation metrics. These findings may help improve machine learning systems used in risk-sensitive applications such as recommendation systems, emotion analysis, and decision-making under uncertainty.