Predicting the evolution of service value features from user reviews for continuous service improvement

Abstract

Facing with a highly competitive service market where customers have more choices on services to fulfill their demands, service providers have to improve their services continuously to make them adapt to constantly-changing value expectations of customers. An enormous quantity of reviews published by customers who have experienced services is an essential basis for service providers to understand which fine-grained features are cared more by customers and what others are less. In this paper, we present a method (VFAMine) for extracting Service Value Features (VF) from review texts by text mining and measuring customers’ attention degrees on VFs by sentiment analysis. As a result, a Time-series Service Value Feature Distribution model (TSVFD) is constructed to delineate the evolution history of attention degrees on various VFs. To help providers identify VFs which are to be extensively concerned by customers and improve them in advance, we give a convolutional sliding window and random forest based algorithm (CSRF) for predicting the future trend of the attention degree on one VF, either for a single service or for services belonging to the same region/domain. In terms of Maximum Information Coefficient (MIC) based correlation analysis, we find that there are latent correlations between the evolution history of different VFs, and such correlation would help service providers improve multiple correlated VFs together. Experiments are conducted on a Yelp dataset and the results demonstrate the effectiveness of our approach.