Identifying individual expectations in service recovery through natural language processing and machine learning

摘要

Identifying customers' expectations for recovery service plays an important role in firms' service recovery. We proposed a customer expectation detection method, which includes data acquisition, expectation classification, experimental preparation and expectation detection, to identify individual service recovery expectations from her/his complaints. Before describing the method, to ensure detection accuracy, we first presented a new expectation classification. In the method, we acquired data from a website called automobile complaint network and sorted the expectations into new categories. We extracted the features of complaints using Word2vec and examined the validity of classifiers by k-fold cross validation. Meanwhile, we also examined the feature dimensions and n-gram approaches, finding that 100-dimension features obtained using 1-gram approach can be used as the independent variables in our method. Then, using the new expectation classifications as dependent variables, we programmed different machine learning classifiers and selected the suitable classifier for each type of expectation detection. Finally, the advantage of new classification was validated and Word2vec, GloVe and TF-IDF had been compared. (C) 2019 Elsevier Ltd. All rights reserved.