Despite the rapid digitization of intangible cultural heritage (ICH), the complex mechanisms governing how users interact and co-create knowledge in digital spaces remain underexplored. Understanding the internal dynamics and engagement logic of these interactive environments is therefore essential to developing sustainable heritage knowledge ecosystems. Conceptualizing the Zhihu community as such an ecosystem, this study investigates ICH thematic structures, knowledge demands, and user participation. By employing an LLM-refined BERTopic framework, this study identified 36 core topics and mapped them onto a four-layer architecture (Cultural Resource Layer, Action Subject Layer, Social Support Layer, and External Interaction Layer) and five knowledge demand dimensions (Basic Knowledge, Cultural Experience, Professional Development, Protection and Inheritance, and Modern Application) through weighted semantic similarity and Spearman correlation analysis. The results reveal a structural configuration dominated by the External Interaction Layer. A dual-track demand mechanism was identified, comprising a professionalized ability-oriented pathway and an affective experience-driven mode. Furthermore, deep engagement was primarily catalyzed by topics that integrate technology, action, and narrative, rather than structural prominence alone. The ICH knowledge ecosystem was characterized by an outward-looking and emotion-driven orientation. This research study contributes an ecosystem framework to heritage information while providing insights for practitioners to optimize digital ICH information services through multi-dimensional semantic integration and public co-creation.

Audio classification plays an essential role in sentiment analysis and emotion recognition, especially for analyzing customer attitudes in marketing phone calls. Efficiently categorizing customer purchasing propensity from large volumes of audio data remains challenging. In this work, we propose a novel Multi-Segment Multi-Task Fusion Network (MSMT-FN) that is uniquely designed for addressing this business demand. Evaluations conducted on our proprietary MarketCalls dataset, as well as established benchmarks (CMU-MOSI, CMU-MOSEI, and MELD), show MSMT-FN consistently outperforms or matches state-of-the-art methods. Additionally, our newly curated MarketCalls dataset will be available upon request, and the code base is made accessible at GitHub Repository MSMT-FN, to facilitate further research and advancements in audio classification domain.