Jun 2026· Informatica· Vol 50· 0 citations· 28 references
TL;DR
A dynamic Chain-of-Thought prompting method based on problem difficulty assessment that effectively resolve the two major limitations of traditional exemplar-based methods, enabling LLMs to obtain appropriately tailored exemplars for both multi-step mathematical reasoning and interdisciplinary question answering.
Abstract
Large language models (LLMs) have demonstrated exceptional capabilities across multiple domains and emerged as the core driving force in natural language processing. Their reasoning however can be associated with logical flaws and lack of stability when attempting to solve difficult problems that require multi-step deduction, cross-domain knowledge or implicit constraints, with redundant or insufficient exemplars in conventional prompts and poor fit with exemplar and target problems. To address these issues, we propose a dynamic Chain-of-Thought (CoT) prompting method based on problem difficulty assessment: first, the model performs zero-shot self-evaluation of the required solution steps to dynamically determine the number of exemplars; then, it integrates BM25 retrieval to select the most similar high-quality question-answer pairs, constructing precise Few-shot prompts. Experiments conducted on multiple datasets effectively resolve the two major limitations of traditional exemplar-based methods, enabling LLMs to obtain appropriately tailored exemplars for both multi-step mathematical reasoning and interdisciplinary question answering. Consequently, the accuracy of complex reasoning is improved to varying degrees across tasks.
A scale-aware comparative study of reasoning enhancement for SLMs across three major families of methods: prompting-based reasoning, retrieval-based augmentation, and knowledge graph guided scaffolding shows that reasoning-enhancement strategies are not universally transferable across model scales under the evaluated settings.
Zhen-Zhen Gu, Jie Liu, Xian Liu· Journal of King Saud Univers...· 0 citations
Document-level relation extraction (DocRE) aims to extract relations among multiple entities across extended contexts while maintaining consistency across predicted triples. Although large language models (LLMs) show remarkable reasoning capabilities in information extraction, their predictions are typically generated independently for each candidate triple and may violate fundamental relational constraints such as transitivity, symmetry, and functional uniqueness, leading to contradictory and unreliable outputs. We propose CONSISTRE, a unified consistency-aware framework for DocRE that addresses this limitation through two complementary tracks. The first operates at inference time for black-box LLMs, combining constraint-aware prompting, constraint-based verification, and iterative self-reflection to refine predictions without task-specific fine-tuning. The second injects consistency knowledge into smaller open-source models via a knowledge distillation and reinforcement learning pipeline: reasoning traces from a powerful teacher are distilled into a student via supervised fine-tuning, followed by GRPO alignment using a composite reward that jointly optimizes extraction performance and relational consistency. Together, the two tracks cover both API-accessible and locally deployable scenarios under a unified consistency formulation. Experiments on DocRED show that both tracks outperform their baselines, with the inference-time track achieving competitive F1 using off-the-shelf black-box LLMs and the training-time track substantially narrowing the gap between 7--8B open-source models and state-of-the-art proprietary LLMs at a fraction of their inference cost. Ablation studies confirm that explicit consistency modeling mitigates relational contradictions and enhances the reliability of LLM-based DocRE across both deployment paradigms.
Large Language Models (LLMs) demonstrate significant potential in sequential recommendation, and leveraging their Chain-of-Thought (CoT) reasoning capabilities can further unlock profound user preference understanding. However, deploying explicit CoT reasoning in real-world systems faces prohibitive challenges: (i) the conflict between the large model scale required for high-fidelity reasoning and the resource constraints of online services, and (ii) the excessive latency introduced by auto-regressive rationale generation. To address these issues, we propose I Reasoning via Multi-Teacher Distillation (IRMD), a novel framework that 'compiles' the reasoning abilities of large teacher LLMs into a lightweight student Small Language Model (SLM). IRMD first employs a Multi-Teacher CoT Synthesis with Dual-Constraint Rejection Sampling module to generate a high-quality, diverse set of reasoning paths. Subsequently, our Annealing-Scheduled Reasoning Distillation strategy progressively trains the student to internalize this logic, transitioning from mimicking explicit CoT to performing purely implicit reasoning. Extensive experiments on multiple benchmark datasets demonstrate that IRMD significantly outperforms state-of-the-art baselines in both recommendation accuracy and inference efficiency. Our code is accessible at https://github.com/Cxx-0/IRMD.