Knowledge Base Question Answering (KBQA) aims to answer natural language questions by reasoning over large-scale structured Knowledge Bases (KBs). Among existing approaches, semantic parsing-based methods have emerged as a mainstream solution, where Large Language Models (LLMs) are employed to translate questions into structured graph queries such as Logical Forms (LFs). However, this paradigm faces two critical challenges: (1) The complex semantic mapping and graph retrieval operations render direct one-shot LF generation difficult; (2) LLMs suffer from inherent hallucination issues, generating semantically plausible-seeming but factually incorrect or invalid LFs, which are non-executable. To address these challenges, this article proposes HP-Corr , a novel framework that integrates H op-wise P lanning with iterative explainable self- Corr ection for faithful knowledge reasoning. Specifically, the system utilizes a fine-tuned open source LLM for query planning and explainable self-correction. The query planner generates reasoning paths hop-by-hop, while an explainable self-correction provides hop-wise feedback, enabling interpretable path editing based on existing reasoning paths and retrieved KB knowledge. By introducing the dual-module cooperative architecture, our system performs iterative plan-then-correct to refine query paths progressively, ensuring answer reliability and LFs executability. Experimental results demonstrate significant improvements, with our approach achieving higher accuracy while substantially reducing the search space, particularly in complex multi-hop KBQA scenarios.

Among various disasters, fire poses one of the most widespread threats to public safety. The early stage of fire, marked by small and slow-spreading fire objects, is the ideal time for firefighting intervention. Therefore, early fire detection is crucial to prevent potential hazards and reduce loss of life and property. However, due to the variable shapes and small size, existing methods struggle to precisely locate smoke and flame in the initial stage. Compared to rigid objects, flame has unique local morphology, such as sharp tip, irregular edge and cavity. These prior knowledge could guide the model to focus on the local morphological features of the flame, thereby improving early fire detection ability. To address the current challenges, the paper proposes a Local Morphological Knowledge Matching based Early Fire Detector(LMKMFD), which accurately detects early fires and locates the multi-scale fires by exploring local flame morphology in fire images. Firstly, the local morphological features are extracted by matching the input fire image with the knowledge templates of the designed local morphological knowledge base. Secondly, fire multi-scale semantic features at four scales are mined by a Transformer-based backbone. Finally, fire local morphological features and multi-scale semantic features are aggregated by single-level depth prediction module to achieve region-level localization of fire objects. Experimental results on private and public datasets show that LMKMFD exhibits high detection precision for fires of different scales, particularly small early fires. LMKMFD outperforms baseline models, with mean Average Precision(mAP) of 88.38% and 83.12% on two datasets. Notably, due to the significance of local morphology of flame, the method performs better in detecting flame than smoke.

While significant progress has been made in recognizing entities from plain text, the exploration of entity recognition from multimodal data remains limited due to disparities in semantic representation. In light of this challenge, given the supportive nature of visual and text data, we propose a novel entity recognition model called Heterogeneous Graph Reasoning(HGR), leveraging the synergistic nature of visual and textual data. This is achieved through the utilization of the Vision Refine and Graph Cross Inference modules. In the Vision Refine module, semantically relevant objects hidden in the image are selected to aid in the text entity extraction. In the Graph Cross Inference module, cross-association inference between visual regions and textual entities is constructed through graph construction, heterogeneous graph fusion, visual region refinement and cross inference. Extensive experiments on four multimodal datasets are demonstrate the superiority of our model, when compared to the second-best state-of-the-art model.