Abstract
Recently, intelligent dialogue systems have shown promise in terms of reducing the load of human customer care agents and decreasing user wait times. In some cases, these systems still cannot understand user intent which leads to the generation of inappropriate responses. Therefore, their inability to handle inappropriate responses has limited their utility in the real world. In this work, we propose a character deep learning model for the detection of chatbot quality of services to handle inappropriate responses by intelligently transferring the dialogue to a human agent. The proposed model has two goals: detect CQoS based on the sentiment score of the utterance using a deep learning model and transferring the user to a live agent when the utterance is inappropriate. The proposed model's effectiveness is evaluated on the dialogue breakdown detection task. The results of the experiment show that our proposed model is effective in achieving the desired goals.