PI: Dejing Dou

Automatic customer service is a computer system that can interact with human through nature language conversation. It is considered to be a new generation of apps after web and mobile apps, due to its efficiency: compared with most companies’ apps, it takes at least half a dozen clicks to get what you want. The objectives of this project involve developing a customer service ontology represented as knowledge graph and building an automatic customer service system backed by the knowledge graph. To achieve those objectives, the first phase of the project is to investigate the novel ways to jointly extract typed entities and relations from various text corpora on mobile phone customer service, such as standard questions-answer pairs, multi-rounds conversation logs between users and customer service agents. We first fine-tune a contextual-based language model by our corpora, which is able to capture semantic relations between entities. Then, we formulate a joint optimization problem to train the extraction model with labeled data, which consist of human-labeled ones and non-human-labeled ones heuristically derived from existing knowledge base. The research will not only facilitate the development of ontology on domains that lack structured knowledge, but also build the automatic customer service based on the developed ontology.