A Hierarchical Intent Classification Framework for Mobile Focused E Commerce Conversational Systems

Conversational systems in e-commerce rely heavily on accurate intent classification to provide relevant and timely responses to user queries. However, real-world user utterances are often ambiguous, noisy, and may contain multiple intents, making flat intent classification insufficient for complex product domains. This study proposes a hierarchical intent classification framework designed specifically for mobile-focused e-commerce conversational systems. The framework introduces a multi-level intent taxonomy comprising an uber intent layer, a common functional intent layer, and a fine-grained product specification intent hierarchy. To evaluate the proposed framework, a synthetically generated dataset simulating realistic user queries was constructed, focusing on mobile product specifications and transactional interactions. The dataset captures multi-intent queries, factual and subjective intent distinctions, and varying levels of intent granularity. Experimental analysis highlights the advantages of hierarchical intent decomposition in improving intent clarity, handling multi-intent queries, and enabling structured response generation. The findings suggest that the proposed framework provides a scalable and domainadaptable approach for designing intent classification systems in e-commerce conversational agents.