2021/11/24 by Jordan J. Bird, Bird, Jordan J.
Computer Science · #AI in Service Interactions #Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Human-Computer Interaction (cs.HC) #Sentiment Analysis and Opinion Mining #Topic Modeling
paper · pdf · doi:10.48550/arxiv.2111.14621
openalex publication_date 2021/11/24 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
With growing societal acceptance and increasing cost efficiency due to mass\nproduction, service robots are beginning to cross from the industrial to the\nsocial domain. Currently, customer service robots tend to be digital and\nemulate social interactions through on-screen text, but state-of-the-art\nresearch points towards physical robots soon providing customer service in\nperson. This article explores two possibilities. Firstly, whether transfer\nlearning can aid in the improvement of customer service chatbots between\nbusiness domains. Second, the implementation of a framework for physical robots\nfor in-person interaction. Modelled on social interaction with Twitter customer\nsupport accounts, transformer-based chatbot models are initially assigned to\nlearn one domain from an initial random weight distribution. Given shared\nvocabulary, each model is then tasked with learning another domain by\ntransferring knowledge from the previous. Following studies on 19 different\nbusinesses, results show that the majority of models are improved when\ntransferring weights from at least one other domain, in particular those that\nare more data-scarce than others. General language transfer learning occurs, as\nwell as higher-level transfer of similar domain knowledge, in several cases.\nThe chatbots are finally implemented on Temi and Pepper robots, with\nfeasibility issues encountered and solutions are proposed to overcome them.\n