RNSC: A Hierarchical Deep Learning Model for Net Promoter Scoring Understanding by Combining Review and Note Through Semantic Consistency

Abstract

The Net Promoter Score (NPS) is widely used metric to measure customer loyalty and is an essential element in customer service analysis. It has garnered considerable attention in both academia and industry. Customergenerated content, such as customer reviews, is commonly utilized to understand and analyze NPS. However, the information provided in customergenerated content may sometimes be inadequate. In this study, we aim to enhance NPS understanding by incorporating a valuable employee-generated content known as employee notes, which provide insights into the internal operations of the enterprise's customer service. This paper proposes a hierarchical deep learning approach, called RNSC (Review, Note and Semantic Consistency), to effectively capture the important semantic consistency relationship and further enhance the understanding of NPS, consequently helping to achieve satisfactory learning performance. Specifically, this approach employs a bidirectional LSTM (BiLSTM) and a soft attention mechanism to encode and identify the local relationship between customer reviews and employee notes. Building upon this, a semantic consistency recognition module is designed. Finally, these semantic relationship representations are aggregated using meanpool and a multi-layer perception (MLP) to evaluate the NPS. Additionally, based on the trained model, the attention mechanism can detect and output the keywords that significantly contribute to the NPS through semantic consistency, showing its potential for supporting diagnostic analysis and further service optimization. Extensive data experiments, such as comparison with baseline methods, ablation experiments, and sparse data experiments, verify the superiority of the proposed RNSC approach.

Keywords: NPS understanding, customer review, employee note, semantic consistency, Deep learning