Gunrock: Building a human-like social bot by leveraging large scale real user data

Gunrock is a social bot designed to engage users in open domain conversations. We improved our bot iteratively using large scale user interaction data to be more capable and human-like. Our system engaged in over 40, 000 conversations during the semi-finals period of the 2018 Alexa Prize. We developed a context-aware hierarchical dialog manager to handle a wide variety of user behaviors, such as topic switching and question answering. In addition, we designed a robust threestep natural language understanding module, which includes techniques such as sentence segmentation and automatic speech recognition (ASR) error correction. Furthermore, we improve the human-likeness of the system by adding prosodic speech synthesis. As a result of our many contributions and large scale user interactions analysis, we achieved an average score of 3.62 on a 1− 5 Likert scale on Oct 14th. Additionally, we achieved an average of 22.14 number of turns and a 5.22 minutes conversation duration.