We present Graph2Bots, a tool for assisting conversational agent designers. It extracts a graph representation from human-human conversations by using unsupervised learning. The generated graph contains the main stages of the dialogue and their inner transitions. The graphical user interface (GUI) then allows graph editing.

In a human-to-human conversation between a user and his interlocutor in an assistance center, we suppose a context where the conclusion of the dialog can characterize a notion of success or failure, explicitly annotated or deduced. The study involves different approaches expected to have an influence on predictive classification model of failures. On the one hand, we will aim at taking into account the asymmetry of the speakers’ roles in the modelling of the lexical distribution. On the other hand, we will determine whether the part of the lexicon most closely relating to the domain of customer assistance studied here, modifies the quality of the prediction. We will eventually assess the perspectives of generalization to morphologically comparable corpora.