DialCSP: A Two-Stage Attention-Based Model for Customer Satisfaction Prediction in E-commerce Customer Service.

Nowadays, customer satisfaction prediction (CSP) on ecommerce platforms has become a hot research topic for intelligent customer service. CSP aims to discover customer satisfaction according to the dialogue content of customer and intelligent customer service, for the purpose of improving service quality and customer experience. Previous works have made some progress in many aspects, but they mostly ignore the huge expressional differences between customer questions and customer service answers, and fail to adequately consider the internal relations of these two kinds of personalized expressions. In this paper, we propose a two-stage dialogue-level classification model containing an intra-stage and an inter-stage, to emphasize the importance of modeling customer part (content of customer questions) and service part (content of customer service answers) separately. In the intra-stage, we model customer part and service part separately by using attention mechanism combined with personalized context to obtain a customer state and a service state. Then we interact these two states with each other in the inter-stage to obtain the final satisfaction representation of the whole dialogue. Experiment results demonstrate that our model achieves better performance than several competitive baselines on our in-house dataset and four public datasets.