Start Date
12-16-2013
Description
The emergence of social media has brought up plenty of platforms where dissatisfied customers can share their service encounter experiences. Those customers’ feedbacks have been widely recognized as valuable information sources for improving service quality. Due to the sparse distribution of customer complaints and diversity of topics related to non-complaints in social media, manually identifying complaints is time-consuming and inefficient. In this study, a supervised learning approach including samples enlargement and classifiers construct was proposed. Applying small labeled samples as training samples, reliable complaints samples and non-complaints samples were identified from the unlabeled dataset during the sample enlargement process. Combining the enlarged samples and the labeled samples, SVM and KNN algorithms were employed to construct the classifier. Empirical results show that the proposed approach can efficiently distinguish complaints from non-complaints in social media, especially when the number of labeled samples is very small.
Recommended Citation
Jin, Jiahua; Yan, Xiangbin; Yu, You; and Li, Yijun, "Service Failure Complaints Identification in Social Media: A Text Classification Approach" (2013). ICIS 2013 Proceedings. 82.
https://aisel.aisnet.org/icis2013/proceedings/ResearchInProgress/82
Service Failure Complaints Identification in Social Media: A Text Classification Approach
The emergence of social media has brought up plenty of platforms where dissatisfied customers can share their service encounter experiences. Those customers’ feedbacks have been widely recognized as valuable information sources for improving service quality. Due to the sparse distribution of customer complaints and diversity of topics related to non-complaints in social media, manually identifying complaints is time-consuming and inefficient. In this study, a supervised learning approach including samples enlargement and classifiers construct was proposed. Applying small labeled samples as training samples, reliable complaints samples and non-complaints samples were identified from the unlabeled dataset during the sample enlargement process. Combining the enlarged samples and the labeled samples, SVM and KNN algorithms were employed to construct the classifier. Empirical results show that the proposed approach can efficiently distinguish complaints from non-complaints in social media, especially when the number of labeled samples is very small.