Conventional feedback, even when accompanied by brief explanations, rarely uncovers the hidden contradictions that trigger a learner's mistake. We bridge this gap with counterfactual question generation (CFQG): given a learner's answer, generate a follow-up question that deliberately contradicts it, compelling the learner to confront the underlying conflict. CFQG thus transforms assessment from passive scoring into an interactive and contradiction-centered dialogue that supports knowledge repair. To automate CFQG, we propose GapProbe, which probes the knowledge gap between a learner’s belief and curated facts through a knowledge graph (KG), then designs counterfactual questions (CFQs) that negate the belief. Identifying contradiction-aware triples, and more importantly, selecting those most likely to confuse the learner, are highly challenging in large-scale KGs. GapProbe tackles these challenges with an iterative ProConB cycle coupled with a schema-aware KGMap. By caching one- and multi-hop schema patterns of the KG, KGMap provides ``roadmap'' to guide LLMs jump to deep and contradiction-aware triples, beyond traditional step-wise graph traversal. We present the CFQG benchmark and corresponding metrics for evaluating how generated CFQs trigger, focus, and deepen learner reflection through explicit contradictions. Experiments on multiple datasets and LLMs show that GapProbe boosts LLM reasoning over KGs and generates follow-up questions that consistently promote deeper and more focused learner reflection.

Yirui QI, Xiaoming Zhang, Ruilin Zeng, Mengyao Liu, Ziyi Zhou, Dezhuang Miao, Bingyu Yan, Zhenyu Guan. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Knowledge-based questions are typically employed to evaluate LLM's knowledge boundaries; meanwhile, numerous studies focus on question generation as a means to enhance the capabilities of both models and individuals. However, there is a lack of in-depth exploration about what constitutes a good question from the perspective of knowledge cognition. This paper proposes aligning the complete knowledge underlying questions with educational criteria effectively employed in physics courses, thereby developing novel knowledge-intensive metrics of question quality. To this end, we propose Meta-Fact Checking (MFC), which transforms questions into knowledge graph (KG) triples utilizing LLMs through few-shot prompting, thereby quantifying question quality based on the patterns observed within these triples. MFC introduces a novel interaction mechanism for KGs that communicates meta-facts, illustrating the types of knowledge that KGs can offer to the LLM for reasoning questions, rather than relying solely on the original triples. This strategy ensures that MFC remains unaffected by unexplored triples that LLM has not yet encountered within KGs compared to the retrieve-while-reasoning routine. Experiments across multiple datasets and LLMs demonstrate that MFC significantly improves the accuracy and efficiency of both question answering and assessing. This research marks a pioneering effort to automate the evaluation of question quality based on cognitive capabilities.

Large language models (LLMs) exhibit strong language understanding capabilities, but encounter challenges when integrating structured knowledge from knowledge graphs (KGs) for complex reasoning tasks such as knowledge graph question answering (KGQA). Existing methods often rely on prompt engineering or fixed templates, which obscure the relational structure and limit generalization. To address these limitations, this paper introduces the Reverse Chain-of-Thought (R-CoT) and Causal Path Verification Plugin, a modular framework that reconstructs retrieved KG triples into reverse chains of sub-questions. Each reasoning step is aligned with a supporting triple, forming interpretable multi-hop paths. In particular, Semantic Causal Scoring (SCS) module is further incorporated to evaluate the causal alignment between each reverse sub-question and the original question through dynamic semantic vector matching. The SCS design avoids frequent interactions with LLMs and effectively filters irrelevant or unsupported reasoning steps. Based on the scoring results, a template-free, model-agnostic R-CoT input format is constructed as a semi-structured sequence. This design preserves the KG structure in natural language form and enables seamless integration with standard LLMs without fine-tuning. Experimental results demonstrate that the R-CoT Plugin consistently improves factual alignment, enhances reasoning stability, and outperforms conventional prompt-based methods in both accuracy and coherence.

Traffic forecasting on the entire road network is challenging due to the non-linear temporal dynamics and complex spatial correlations. Multi-step traffic forecasting further increases the difficulty because of the accumulated prediction errors. Existing forecasting models attempt to extract both spatial and temporal features of all locations on the road network for prediction, but often overlook the interaction between the two types of features, which has led to sub-optimal performance. In this work, we tackle this problem by proposing InterNet, which applies the multi-head attention mechanism on the extracted spatio-temporal features and enables the interaction of the spatial (temporal) features of one location with the temporal (spatial) features of all locations. Moreover, we extract the features of all locations using a graph convolutional layer and a bidirectional LSTM layer, before feeding them into the multi-head attention layer. The three layers are seamlessly integrated and thereby enable end-to-end learning. Experimental results show that the InterNet model outperforms the state-of-the-art models in terms of the prediction accuracy, which demonstrates the potential of such interactions.

Music recommendation is a prominent application of recommender systems, which has been attracting more and more attentions. There are two research streams of music recommender systems: one is static recommendation based on learning user’s preference according to historical data, and the other is dynamic recommendation considering user’s feedback. But the individual music preference for a certain moment is closely related to personal experience of the music and music literacy, as well as temporal scenario with diversity. Thus, it’s necessary to design a new music recommendation framework by integrating static recommendation and dynamic recommendation. Therefore, we propose a novel approach for music recommendation HRRS (Humming-Query and Reinforcement-Learning based Recommender Systems) by integrating prior two research streams. This novel recommendation framework HRRS based on humming query and reinforcement learning is learning and adapting to user’s current preference continually by collecting interactive data in real time. This preliminary recommendation framework captures song characters, personal dynamic preferences, and yields a better listening experience with proper interaction.