Data-driven Intent Models for Visual Analysis Tools and Chatbot Platforms


Natural language interaction in visual analysis tools supports expressive ways for users to interact with their data. Chatbot systems have recently garnered interest as conversational interfaces for a variety of tasks. Crafting the business logic for handling user intent in natural language input using a pre-defined grammar can be precise, but often covers a small set of intent models for a specific platform. More recently, machine learning approaches have shown to be promising for supporting complex responses based on the current conversational state of the interface. Such techniques could be employed for bootstrapping a range of chatbot interfaces for visual analysis. One approach is to use a labeled dataset of natural language interactions that capture user intent distribution, co-occurrence, and flow patterns. Another approach is employing deep learning techniques that approximate the heuristics and conversational cues for continuous learning in a chatbot interface. This paper explores the implications for these data-driven approaches in broadening the scope for visual analysis workflows across various chatbot experiences.

Publication Date: 
Saturday, April 25, 2020
Publication Information: 
Workshop on Artificial Intelligence for HCI: A Modern Approach at CHI (April 25, 2020)