'Could You Describe the Reason for the Transfer?': A Reinforcement Learning Based Voice-Enabled Bot Protecting Customers from Financial Frauds

With the booming of the Internet finance and e-payment business, telecom and online fraud has become a serious problem which grows rapidly. In China, 351 billion RMB (approximately 0.3% of China's GDP) was lost in 2018 due to telecommunication and online fraud, influencing tens of millions of individual customers. Anti-fraud algorithms have been widely adopted by major Internet finance companies to detect and block transactions induced by scam. However, due to limited contextual information, most systems would probably mistakenly block the normal transactions, leading to poor user experience. On the other hand, if the transactions induced by scam are detected yet not fully explained to the users, the users will continue to pay, suffering from direct financial losses. To address these problems, we design a voice-enabled bot that interacts with the customers who are involved with potential telecommunication and online frauds decided by the back-end system. The bot seeks additional information from the customers through natural conversations to confirm whether the customers are scammed and identify the actual fraud types. The details about the frauds are then provided to convince the customers that they are on the edge of being scammed. Our bot adopts offline reinforcement learning (RL) to learn dialogue policies from real-world human-human chat logs. During the conversations, our bot also identifies fraud types every turn based on the dialogue state. The bot proposed outperforms baseline dialogue strategies by 2.8% in terms of task success rate, and 5% in terms of dialogue accuracy in offline evaluations. Furthermore, in the 8 months of real-world deployment, our bot lowers the dissatisfaction rate by 25% and increases the fraud prevention rate by 135% relatively, indicating a significant improvement in user experience as well as anti-fraud effectiveness. More importantly, we help prevent millions of users from being deceived, and avoid trillions of financial losses.