International Conference on Cognitive Computing [Services Society]

With the rapid advancement of multimodal large models, there is an increasing demand for intelligent question-answering systems capable of comprehending and leveraging complex, heterogeneous, and multi-source knowledge. However, existing approaches often face challenges in semantic alignment, efficient retrieval, and coherent generation, particularly when processing unstructured multimodal data and multi-turn dialogues. To address these limitations, this study presents a systematically designed multimodal retrieval-augmented generation (RAG) system that integrates advanced knowledge segmentation, hybrid retrieval, and generation techniques. In the offline stage, the system performs fine-grained knowledge segmentation and structured annotation through multimodal understanding, while enhancing global retrieval performance via multi-dimensional knowledge augmentation. In the online stage, it establishes an end-to-end, closed-loop answering pipeline that incorporates multi-turn dialogue rewriting, hybrid retrieval, reranking, and large model generation. Furthermore, a comprehensive automatic evaluation framework is developed to assess the system across multiple dimensions, including retrieval accuracy, factual consistency, generation quality, and structural clarity. Experimental results show that the proposed system substantially improves retrieval hit rates and generation performance over baseline methods. Overall, this work demonstrates the effectiveness of integrating RAG frameworks with multimodal large models and provides a solid foundation for deployment in high-demand domains such as healthcare, finance, and education.

The integration of artificial intelligence (AI) assisted diagnosis and treatment has ushered in a new era of healthcare, but it also presents significant challenges to traditional theories of causation in medical malpractice liability. This paper examines the limitations of conventional causation determination methods in the context of AI-assisted diagnosis and proposes a shift toward probabilistic causation theory. We construct a rule system centered on “probability gain” supported by apparent cause identification, joint causation identification, spurious cause analysis, and causal chain reconstruction. This approach assists courts in determining causation when evidentiary proof is limited and causation is uncertain. We demonstrate how this framework can clarify the roles of medical personnel and AI entities in mixed causation cases and determine the attributable causal power and liability share in joint causation scenarios. Our findings suggest that probabilistic causation theory offers a promising avenue for unraveling the complex causation puzzles in AI-assisted medical malpractice cases, potentially reshaping the legal landscape for healthcare liability in the age of artificial intelligence.

As AI becomes a core driver of enterprise transformation, traditional organizations face challenges such as rigid structures, data fragmentation, and outdated evaluation models. Existing digital transformation maturity models focus primarily on IT infrastructure and industrial-era processes, lacking a comprehensive framework for evaluating AI effectiveness. To address this gap, this paper proposes the IMPRESSIVE Enterprise Framework, which defines ten critical capabilities for AI-powered organizations: Innovation, Machine Learning, Personalization, Reliability, Enhanced Experience, Scalability, Security, Integrated Ecosystem, Value-Driven Culture, and Efficiency. Derived from this framework, the IMPRESSIVE Maturity Assessment Model (IMAM) is developed as a quantitative tool to measure AI maturity across these dimensions. Validation through case studies demonstrates IMAM’s applicability in identifying strengths, gaps, and strategic pathways for enterprises transitioning to AI-native operations. This research contributes a holistic framework to guide organizations in systematically navigating AI transformation, bridging the gap between theoretical insights and actionable practice.

With the proposal of the Belt and Road Initiative, research in the field of Silk Road trade has once again become a hotspot. This paper focuses on Named Entity Recognition (NER) for Silk Road Trade Texts Based on Deep Learning for Silk Road trade texts to support the knowledge graph development process. To address the absence of standardized datasets and noise interference in Silk Road trade research, we construct an annotated corpus by aggregating multi-source raw data, deploying DeepSeek-R1’s domain adaptation for noise filtering, followed by LangChain-based paragraph-level semantic segmentation to support NER training. Subsequently, it proposes an innovative deep learning model named BERT-BiGRU-Attention-CRF for entity recognition in Silk Road trade texts. This model synergistically combines BERT’s deep semantic understanding, BiGRU’s capability for modeling long-range dependencies, the attention mechanism’s focus on critical information, and CRF’s optimization of label sequence constraints. Results from comparative experiments with other entity recognition models illustrate that our proposed model exhibits better performance and higher accuracy. It effectively addresses the key challenge of entity recognition posed by long-range dependencies and complex contextual semantics inherent in Silk Road trade texts, thereby laying a solid foundation for constructing high-quality Silk Road trade knowledge graphs.

Named Entity Recognition (NER) in texts about Chinese traditional musical instruments faces challenges due to composite structures that blur entity boundaries and difficulties like dialect transliterations. To address these issues, this paper proposes a novel NER model that integrates adversarial training and a cross-attention mechanism. During the fine-tuning stage of Bidirectional Encoder Representations from Transformers (BERT) word embeddings, the model introduces adversarial training. It enhances the model’s ability to learn robust features by applying perturbations. Next, the model constructs a parallel feature extraction network comprising Iterated Dilated Convolutional Neural Network (IDCNN) and Bidirectional Gated Recurrent Unit (BiGRU). It also designs a cross-attention fusion module. This module enables the dynamic weighted fusion of context-aware features. Finally, the model employs a Conditional Random Field (CRF) decoder to constrain label transition logic. Experiments conducted on our self-built Chinese musical instrument text dataset show that the proposed model achieves an F1 score of 83.27