Cluster-R1: Large Reasoning Models Are Instruction-following Clustering Agents
This work addresses the limitations of existing embedding models in capturing user-specified text characteristics and inferring latent corpus structures, offering a novel approach for tasks like daily dialogue, legal cases, and financial reports.
The paper tackles the problem of instruction-following clustering by reframing it as a generative task and training large reasoning models as autonomous clustering agents, resulting in consistent outperformance over embedding-based methods and LRM baselines across diverse datasets and clustering scenarios.
General-purpose embedding models excel at recognizing semantic similarities but fail to capture the characteristics of texts specified by user instructions. In contrast, instruction-tuned embedders can align embeddings with textual instructions yet cannot autonomously infer latent corpus structures, such as determining the optimal number of clusters. To address both limitations, we reframe instruction-following clustering as a generative task and train large reasoning models (LRMs) as autonomous clustering agents. Our reasoning-driven training pipeline enables LRMs to interpret high-level clustering instructions and then infer the corresponding latent groupings. To evaluate this paradigm, we introduce ReasonCluster, a comprehensive benchmark comprising 28 diverse tasks spanning daily dialogue, legal cases, and financial reports. Experiments across diverse datasets and clustering scenarios show that our approach consistently outperforms strong embedding-based methods and LRM baselines, demonstrating that explicit reasoning fosters more faithful and interpretable instruction-based clustering.