Research on dialogue mode recognition in meeting scenes

Research on dialogue mode focuses on understanding the ways people engage in conversation and their learning processes, with exploratory dialogue serving as an external indicator of learning occurrences. With the continuous expansion of dialogue data in different learning scenarios, automatic recognition of dialogue modes becomes increasingly crucial. Building upon previous research on dialogue mode encoding schemes, this paper identified six categories of dialogue modes in meeting scenes. This paper collected and cleaned 11K utterances, annotated them with DM labels, and constructed a dataset called MRDM (Meeting Record Dialogue Mode). Additionally, this paper proposed the CAHDMR (Context-Attention Hierarchical Dialogue Mode Recognition) model, which integrates different levels of context into utterance representations using encoders and contextual attention layers, to generate more robust representations of dialogue. The experimental results indicate that the CAHDMR model demonstrates excellent performance on the constructed MRDM dataset.