Zixiang Lu

Multimodal recommender systems have become essential to modern content platforms, leveraging user–item interactions to continually improve recommendation quality. However, many existing approaches still rely on coarse, globally shared fusion schemes and struggle to remain robust when multimodal content is incomplete. Consequently, we propose IPRec, a multimodal recommendation model, to tackle two persistent challenges: the lack of item-aware fusion strategies and the limited robustness of multimodal representations. IPRec mainly consists of two key components, i.e., Adaptive Item-specific Feature Learning (AIFL) and Progressive Knowledge Distillation Enhancement (PKDE). The AIFL module learns item-specific weight distributions across refined visual and textual representations from a large language model, yielding content-aware multimodal fusion weights that adapts to each item’s characteristics. The PKDE module enhances model robustness by combining masked auto-encoders with teacher–student distillation, allowing the student to effectively handle missing content while aligning with the teacher’s cross-modal representations. Extensive experiments on real-world datasets demonstrate that IPRec consistently outperforms the state-of-the-art methods, highlighting the importance of two components for enhancing mul-timodal recommenders.

Large Language Models (LLMs) often generate inaccurate responses (hallucinations) when faced with questions beyond their knowledge scope. Retrieval-Augmented Generation (RAG) addresses this by leveraging external knowledge, but a critical challenge remains: determining whether retrieved contexts effectively enhance the model`s ability to answer specific queries. This challenge underscores the importance of knowledge boundary awareness, which current methods-relying on discrete labels or limited signals-fail to address adequately, as they overlook the rich information in LLMs` continuous internal hidden states. To tackle this, we propose a novel post-retrieval knowledge filtering approach. First, we construct a confidence detection model based on LLMs` internal hidden states to quantify how retrieved contexts enhance the model`s confidence. Using this model, we build a preference dataset (NQ_Rerank) to fine-tune a reranker, enabling it to prioritize contexts preferred by the downstream LLM during reranking. Additionally, we introduce Confidence-Based Dynamic Retrieval (CBDR), which adaptively triggers retrieval based on the LLM`s initial confidence in the original question, reducing knowledge conflicts and improving efficiency. Experimental results demonstrate significant improvements in accuracy for context screening and end-to-end RAG performance, along with a notable reduction in retrieval costs while maintaining competitive accuracy.

Degradation prediction for aerospace electronic systems plays a crucial role in maintenance work. This paper proposes a concise and efficient framework for multivariate time series forecasting that is capable of handling diverse sequence representations through a Channel-Independent (CI) strategy. This framework integrates a decomposition-aware layer to effectively separate and fuse global trends and local variations and a temporal attention module to capture temporal dependencies dynamically. This design enables the model to process multiple distinct sequences independently while maintaining the flexibility to learn shared patterns across channels. Additionally, the framework incorporates probabilistic distribution forecasting using likelihood functions, addressing the dynamic variations and uncertainty in time series data. The experimental results on multiple real-world datasets validate the framework’s effectiveness, demonstrating its robustness and adaptability in handling diverse sequences across various application scenarios.