Inferring a user’s propensity for elaborative thinking based on natural language

Abstract

Natural language-based aids (e.g., intelligent cognitive assistants) that assist humans with various tasks and decisions, often need to recognize the user’s propensity (low-high) to elaborate on the task or decision, to ensure that the information provided matches the user’s thinking level. We conducted two qualitative studies of natural language usage in customers’ written product reviews (Study 1) and conversational transcripts of customer-store associate interactions (Study 2) to generate (Study 1) and validate (Study 2) four rules that can be employed to infer a user’s propensity for elaborative thinking. These include: consideration of multiple (2+) attributes/alternatives; detailed description (word count) about a single attribute/alternative; demonstration of specific knowledge (use of specific terms) about an attribute/alternative; and consideration of pros and cons about an attribute/alternative. Implications for natural language-based, intelligent cognitive assistants emerge as a result of this work.