Emotion Recognition in Written Conversations Using CASTLE
Filip Karabeleski, Nick Sancovic Persson · Lund University Publications Student Papers (Lund University) · 2021
Conversational emotion recognition is a notoriously challenging task for computers as conversations often contain subtle nuances that are di cult to identify.In this work, we investigated the recognition of conversational emotions in a set of dialogue corpora.We started from an existing baseline and, using empirical observation; we combined it with a previously untested set of architectures.We created a new framework, CASTLE, that builds upon the previous state of the art and complements it with more e ective recurrent devices.As a result of using this framework, we could improve the F -score of previously published results on two of the three corpora we evaluated.