IFDDS: An Anti-fraud Outbound Robot

With the rapid growth of internet finance and e-payment, payment fraud has attracted increasing attention. To prevent customers from being cheated, systems often block risky payments depending on a risk factor. However, this may also inadvertently block cases which are not actually risky. To solve this problem, we present IFDDS, a system that proactively chats with customers through intelligent speech interaction to precisely determine the actual payment risk. Our system adopts imitation learning to learn dialogue policies. In addition, it encompasses a dialogue risk detection module which identifies fraud probability every turn based on the dialogue state. We create a web-based user interface which simulates a practical voice-based dialogue system. Introduction In tandem with the development of digital finance and epayment, financial fraud has been emerging. To prevent customers from being deceived, anti-fraud systems typically block payments with high risk probability. If some payments are not actually risky, however, they may also be stopped. This is not desirable, as it is time-consuming and laborious for customers to ask customer service to unblock legitimate payments. To effectively tackle this problem, we build IFDDS (Interactive Fraud Detection Dialogue System), an anti-fraud outbound robot. In its operation, if payment is blocked by the anti-fraud system, an outbound call is triggered to converse with the customer to refine the payment risk. The robot will persuade them to stop the payment if any risk is detected during the conversation, otherwise it will lift payment restrictions and let users continue to pay. The most common way to establish an outbound robot is the flow-based method, e.g., Dialogflow1. Flow-based robots adopt an intent-based model by combining multiple utterances in a state machine which imitates a conversation flow. The flow-based robot gradually guides a user towards the right response. At each turn, the robot analyzes the user’s intention based on the text received. The state tracking module maintains the dialogue state consisting of user intentions and other dialogue states. Conversation flow outputs possible responses and moves to the next state at each turn based Copyright c © 2021, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. https://cloud.google.com/dialogflow Customer ASR SLU