Eliciting User Food Preferences in terms of Taste and Texture in Spoken Dialogue Systems

Abstract

Food preference varies from person to person and is not easy to verbalize. This study proposes a dialogue system that elicits the user's food preference through human-robot interaction. First, as the default knowledge of the dialogue system, we determined the ingredients of each dish from a large-scale recipe database, and collected the taste and texture of each dish and its ingredients by analyzing a large number of Twitter messages. Subsequently, the dialogue system asks questions to elicit the user's preferred taste/texture of the food by using the default knowledge base, while employing frame-based dialogue management. Finally, we created a food vector space that represents the relationship between the dish names, ingredients, and taste/texture expressions. We also discuss the possibility of using this vector space in dish recommendation.