Junyu Lu

Knowledge Tracing (KT), a pivotal component of intelligent tutoring systems, models the evolution of student knowledge states to predict future performance. While KT fundamentally relies on the premise that performance on similar questions is highly correlated, existing approaches often depend on generalized question representations, neglecting the rich, multi-faceted nature of question attributes. To address this limitation, we propose the Hierarchical Question Attribute-Fused KT (HQAF-KT) model, a novel architecture that deconstructs question similarity through three hierarchical dimensions: inherent, dynamic, and statistical. HQAF-KT first enriches foundational representations by integrating inherent question attributes. It then deploys a Dynamic Computing module that leverages student-specific dynamic attributes to personalize similarity assessments based on individual cognitive contexts. Furthermore, a Statistic Correction module refines generalized statistical attributes to account for unique student abilities. This hierarchical fusion enables a nuanced, individualized modeling of question relationships. Extensive experiments on three large-scale, real-world datasets demonstrate that HQAF-KT significantly outperforms state-of-the-art baselines by effectively capturing multi-level question similarity.

Multimodal intent recognition (MIR) seeks to accurately interpret user intentions by integrating verbal and non-verbal information across video, audio and text modalities. While existing approaches prioritize text analysis, they often overlook the rich semantic content embedded in non-verbal cues. This paper presents a novel Wavelet-Driven Multimodal Intent Recognition (WDMIR) framework that enhances intent understanding through frequency-domain analysis of non-verbal information. To be more specific, we propose: (1) a wavelet-driven fusion module that performs synchronized decomposition and integration of video-audio features in the frequency domain, enabling fine-grained analysis of temporal dynamics; (2) a cross-modal interaction mechanism that facilitates progressive feature enhancement from bimodal to trimodal integration, effectively bridging the semantic gap between verbal and non-verbal information. Extensive experiments on MIntRec demonstrate that our approach achieves state-of-the-art performance, surpassing previous methods by 1.13% on accuracy. Ablation studies further verify that the wavelet-driven fusion module significantly improves the extraction of semantic information from non-verbal sources, with a 0.41% increase in recognition accuracy when analyzing subtle emotional cues.

The understanding and representation of songs is a crucial issue in music platforms, as it can facilitate numerous applications in the music field. Songs are a common multimodal art form within the music domain, achieving rich musical connotations and strong expressiveness. However, the data of songs exhibit obvious multimodal and heterogeneous characteristics, presenting significant challenges to the understanding and representation of songs. Regrettably, the current methods do not respond to these challenges effectively. To this end, in this study, a unified multimodal representation framework for songs, namely SUMMR, is proposed. Specifically, first, a two-layer framework is put forward. In embedding layer, the features of different modalities data are embedded into a unified space. In content layer, a novel cross-modal attention mechanism is designed, which effectively capture the cross-modal semantic associations and deep music features, thereby obtaining a unified representation of songs. Then, a two-level hierarchical pre-training algorithm is proposed, which can effectively lower the training cost. Finally, experiments are conducted on two typical music tasks with public datasets of songs, where the experimental results demonstrate the effectiveness of SUMMR for understanding and representation of songs, and also show that SUMMR has good capability of being fine-tuned in many song-based tasks.

Adaptive learning, also known as adaptive teaching, relies on learning path recommendations that sequentially suggest personalized learning items (such as lectures and exercises) to meet the unique needs of each learner. Despite the extensive research in this field, previous approaches have primarily modeled the interaction sequences between learners and items using simple indexing, leading to three issues: (1) The utilization of information from both learners and items is not sufficient. For instance, these models are unable to leverage the semantic information contained within the textual content of the items. (2) Models need to be retrained on different datasets separately, which makes it difficult to adapt to the continuously expanding item pool in online educational scenarios. (3) The existing recommendation paradigm based on trained reinforcement learning frameworks, suffers from unstable recommendation performance in sparse learning logs. To address these challenges, we propose a generalized Generative Agent for Adaptive Learning (GenAL), which integrates educational tools with LLMs' semantic understanding to enable effective and generalizable learning path recommendations across diverse data distributions. Specifically, our framework consists of two components: the Global Thinking Agent, which updates the learner profile and reflects on recommendation outcomes based on the learner's historical learning records. The other is the Local Teaching Agent, which recommends items using educational prior knowledge. Leveraging the LLM's robust semantic understanding, our framework does not rely on item indexing but instead extracts relevant information from the textual content. We evaluated our approach on three real-world datasets, and the experimental results demonstrate that our GenAL not only consistently outperforms all baselines but also exhibits strong generalization ability.

Instruction-tuned Code Large Language Models (Code LLMs) have excelled in diverse code-related tasks, such as program synthesis, automatic program repair, and code explanation. To collect training datasets for instruction-tuning, a popular method involves having models autonomously generate instructions and corresponding responses. However, the direct generation of responses does not ensure functional correctness, a crucial requirement for generating responses to code instructions. To overcome this, we present Verification-Based Self-Play (VERSE), aiming to enhance model proficiency in generating correct responses. VERSE establishes a robust verification framework that covers various code instructions. Employing VERSE, Code LLMs engage in self-play to generate instructions and corresponding verifications. They evaluate execution results and self-consistency as verification outcomes, using them as scores to rank generated data for self-training. Experiments show that VERSE improves multiple base Code LLMs (average 7.6%) across various languages and tasks on many benchmarks, affirming its effectiveness.