Exploring Personal Knowledge Extraction from Conversations with CHARM

Abstract

Incorporating users' personal facts enhances the quality of many downstream services. Automated extraction of such personal knowledge has recently received considerable attention. However, often the operation of extraction models is not exposed to the user, making predictions inexplicable. In this work we present a web demonstration platform showcasing a recent personal knowledge extraction model, CHARM, which provides information on how the prediction was made and which data was decisive for it. Our demonstration explores two potential sources of input data: conversational transcripts and social media submissions.