A Real-World Human-Machine Interaction Platform in Insurance Industry

All authors contributed equally to this manuscript. A Real-World Human-Machine Interaction Platform in Insurance Industry Wei Tan, Chia-Hao Chang†, Yang Mo, Lian-Xin Jiang, Gen Li, Xiao-Long Hou Chu Chen, Yu-Sheng Huang, Meng-Yuan Huang, Jian-Ping Shen AI Department Ping An Life Insurance of China, Ltd. {tanwei818,zhangjiahao206†,moyang853,jianglianxin769,ligen947,houxiaolong430, chenchu870,huangyusheng112,huangmengyuan334,shenjianping324}@pingan.com.cn Correspondence†:strategist922@gmail.com Abstract In the insurance industry, lots of effort is putting into helping the customer to solve their problems that occurred during and after purchasing cycle and helping telemarketers to practice selling skills. Chat bots and assistant bots are widely used in these business scenarios, but building a bot application from scratch is expensive. In this paper, a human-machine interaction platform specially designed for intelligent bot applications in insurance industry that combined the technologies of Question Answering (QA), task-oriented dialogue and chit-chat was proposed and we demonstrate the architecture design of this platform, key technologies and the scenario of applications in real-world insurance industry. It has been supporting many intelligent bot applications of insurance industry already, such as Intelligent Coach Bot (ICB) which helps telemarketers to practice their selling skills, Intelligent Customer Service Bot (ICSB) which provides after-sales services and Insurance Advisor Bot (IAB) which helps customer to purchase the most suitable insurance product. Currently, these bot applications serve millions of users per day and are able to solve 80% of the online problems. KEYWORDS Human-Machine Interaction Schema, Dialogue Tree, Text Similarity, Text Matching, Chit Chat Engine, Anaphora Resolution, Intention/Slot Driven, Attention Mechanism 1 INTRODUCTION The 2019 Conference on Computational Linguistics and Speech Processing ROCLING 2019, pp. 82-91 ©The Association for Computational Linguistics and Chinese Language Processing 82 . With the development of deep learning technology, speech recognition and natural language understanding (NLU) have developed rapidly and the way of human-computer interaction has also changed from GUI to CUI (Conversation User Interface). The traditional GUI depends on the keyboard, mouse, or touch screen. But CUI carries out input and output through dialogue, breaking through the limitation of contact and getting closer to the way of communication between people. CUI is a tremendous change in the way of interaction with a variety of AI techniques. We have implemented a human-computer interaction platform that uses CUI as a starting point to construct a conversational input and output interface and a smart brain. Based on the insurance scenario, our platform has been widely and solidly applied in the fields of salesman training, intelligent customer service, and recommendation of products. In salesman training scenario, we rely on this platform to build intelligent coach robot that can simulate customers to do human-machine dialogue with the telesales man. It helps them learn sales and handle customer objections, improving their sales skills. In the intelligent customer service scenario, we built a customer service robot under the technical support of the platform which can answer users' questions at anytime and anywhere and guide users to apply the request for business, such as the policy loan. In the scenario of insurance products recommendation, we implemented the insurance advisor bot that could find customers' …

Content truncated to 500 words