Listwise reranking with Large Language Models (LLMs) has emerged as the state-of-the-art approach, consistently establishing new performance benchmarks in passage reranking. However, their practical application faces two critical hurdles: the prohibitive computational overhead and high latency of processing long token sequences, and the performance degradation caused by phenomena like "lost in the middle" in long contexts. To address these challenges, we introduce Compress-then-Rank (C2R), an efficient framework that performs listwise reranking not on original passages, but on their compact multi-vector surrogates. These surrogates can be pre-computed and cached for all passages in the corpus. The effectiveness of C2R hinges on three key innovations. First, the compressor model is pre-trained on a combination of text restoration and continuation objectives, enabling high-fidelity compressed vector sequences that mitigate the semantic loss common in single-vector methods. Second, a novel input scheme prepends embeddings of each ordinal index (e.g., [1]:) to its corresponding compressed vector sequence, which both delineates passage boundaries and guides the reranker LLM to generate a ranked list. Finally, the compressor and reranker are jointly optimized, making the compression explicitly ranking-aware for the ranking objective. Extensive experiments on major reranking benchmarks demonstrate that C2R provides substantial speedups while achieving competitive and even superior ranking performance compared to full-text reranking methods.

Label distribution learning (LDL) is a framework to solve label ambiguity. To improve the performance of LDL, some existing algorithms exploit global and local label correlations. In reality, local correlations are more reasonable than global ones because few correlations can be globally adapted to all samples. The existing algorithms exploiting local correlations assume that the smaller the Euclidean distance between samples, the more likely the samples are to share the same label correlations. Specifically, they all use K-means to divide samples into several clusters, and then mine local correlations in each cluster based on this assumption. However, this assumption is incorrect in some cases and may lead to inappropriate clustering results and biased correlations. In this paper, we propose a novel LDL algorithm which mines local label correlations in self-regulating clusters independent of sample distance. In particular, we introduce clustering with learnable parameters into the model to realize that the clustering can be optimized jointly with the objective function instead of depending on the distance between samples. In this way, our proposed algorithm can mine more accurate local label correlations in these more appropriate clusters. Experimental results on 15 real-world datasets demonstrate the effectiveness of the proposed algorithm.