Using customers' online reviews to identify and classify human robot interaction failures in domestic robots

Abstract

Little information is available regarding which types of failures robots experience in domestic settings. To further the community's knowledge, we manually classified 3062 customer reviews of robotic vacuum cleaners on Amazon.com. The resulting database was analyzed and used to create a Natural Language Processing (NLP) model capable of predicting whether a review contains a description of a failure or not. The current work describes the initial analysis and model development process as well as preliminary findings.