Hierarchical LSTMs with Joint Learning for Estimating Customer Satisfaction from Contact Center Calls

This paper presents a joint modeling of both turn-level and call-level customer satisfaction in contact center dialogue. Our key idea is to directly apply turn-level estimation results to call-level estimation and optimize them jointly; previous work treated both estimations as being independent. Proposed joint modeling is achieved by stacking two types of long short-term memory recurrent neural networks (LSTM-RNNs). The lower layer employs LSTM-RNN for sequential labeling of turn-level customer satisfaction in which each label is estimated from context information extracted from not only the target turn but also the surrounding turns. The upper layer uses another LSTM-RNN to estimate call-level customer satisfaction labels from all information of estimated turn-level customer satisfaction. These two networks can be efficiently optimized by joint learning of both types of labels. Experiments show that the proposed method outperforms a conventional support vector machine based method in terms of both turn-level and call-level customer satisfaction with relative error reductions of over 20%.