Talks

The ecommerce chatbot requires precise intent recognition to streamline order processes and reduce the burden on customer service staff. However, several challenges persist:

• Customer messages are often brief and lack sufficient information, reducing intent recognition accuracy.
• Increasing service types lead to confusing intent definitions, with over 200 intents, impacting response appropriateness.
• Maintaining conversational coherence and context is challenging.
• Diverse ecommerce business rules require ongoing improvements to dialogue systems to ensure they remain flexible, easy to maintain, and consistently reliable.

This talk will discuss how we use the Rasa Framework’s flexible pipeline architecture, efficient corpus cleaning, and specific mechanism to refine the vector space for business terms, creating a cost-effective and practical chat system.