The rapid development of large language models has greatly advanced human-computer dialogue research. However, applying these models to specialized fields like maternity and infant care often leads to subpar performance due to a lack of domain-specific datasets. To address this problem, we have created MicDialogue, a Chinese dialogue dataset for maternity and infant care. MicDialogue involves a wide range of specialized topics, including gynecological health, pediatric care, pregnancy preparation, emotional counseling and other related topics. This dataset is curated from two types of Chinese social media: short videos and blog posts. Short videos capture real-time interactions and pragmatic dialogue patterns, while blog posts offer comprehensive coverage of various topics within the domain. We have also included detailed annotations for topics, diseases, symptoms, and causes, enabling in-depth research. Additionally, we developed a knowledge-driven benchmark model using LLM-based prompt learning and multiple knowledge graphs to address diverse dialogue topics. Experiments validate MicDialogue's usability, providing benchmarks for future research and essential data for fine-tuning language models in maternity and infant care.

Generative conversational agents, driven by large language models, have gained widespread popularity. However, a significant drawback lies in their tendency to produce uncontrollable and unpredictable contents, thereby increasing the risk of generating malevolent responses that potentially exacerbate users’ mental health issues. Although existing research on malevolence detection in dialogues addressed the modeling of interaction patterns in dialogues, the implicitly expressed emotion and user-specific context are often neglected. Addressing this gap, we propose a hypergraph-enhanced context modeling approach for detecting malevolence in mental health counseling dialogues. Our approach harnesses the emotion reasoning capabilities of large language models to generate implicit emotional prompts. Employing hypergraph neural networks, our approach effectively integrates emotional context, user-specific context, and interactive context, fusing them into high-order semantic representations using hypergraph convolution. Experimental results on two benchmark datasets, MDRDC and Dialogue Safety, demonstrate the superiority of our model over state-of-the-art baseline models, particularly in complex contextual scenarios.

Malevolence detection in dialogues aims to identify harmful or inappropriate utterances, significantly impacting dialogue quality and user satisfaction. Although existing studies have shown promising performance by modeling interaction patterns from dialogue history, various malevolence-invoking factors, such as fine-grained emotions, evolving topics and user profiles, are often overlooked. To comprehensively consider these factors, we propose a hypergraph fusion model by employing multi-view LLM-driven prompts for malevolence detection in dialogues. Our model integrates emotion context, topic context, user profile context and interaction context, utilizing hypergraphs to establish high-order contextual relationships from multi views for deducing malevolence-invoking semantics. Experimental results on two benchmark datasets demonstrate that our model achieves the state-of-the-art performance.