Text as Data

APPENDIX BA Design Approach to Characterizing Users Based on Audio Interactions on a Conversational AI Platform

Online, interactive, intelligent personal agents (referred to as botsⁱ) are now ubiquitous and accomplish a wide range of tasks that enhance the interface between human users and cloud-based computer services. Increasingly, these agents are built with artificial intelligence (AI) capabilities, and this expands their utility. The utility of these agents depends on their ability to correctly infer intentions and circumstances that characterize the user. In this use case, we outline a collection of processes and techniques to extract and quantify aspects of the sound and audio signal from audio interactions between user and the bot agent, to establish characteristics that lead to the attribution of end-user persona attributes such as gender, age, accent, and so on. Additionally, the audio signals are converted to a textual format and pretrained text analytics models are brought into the process to establish feature weights to infer latent intent from the interaction. This adds information to the interaction and thereby increases the accuracy and utility of the user-bot interaction.

AUDIO-BASED USER INTERACTION INFERENCE

A wide range of organizations is moving toward automation in conversational user interaction through the development of increasingly capable AI based audio user interaction bots and assistants. Historically, these AI assistants were text based; now ...