TOWARDS EFFICIENT ONLINE STORE MANAGEMENT: AN NLP-BASED SYSTEM FOR CLIENTCOMPLAINT CLASSIFICATION

Abstract

Efficient complaint management in e-commerce is essential, as the growing trend toward e-commerce transactions generates an overwhelming number of customer complaints and associated managementchallenges, including unreliable, inefficient, time-consuming, and ineffective handling of customercomplaints. Conventional manual methods for handling customer complaints are unreliable because of inconsistencies in complainants' language,noisy text, and semantic similarities among complaints,making them difficult to manage effectively. Earlier research efforts toward customer complaint classification focused more on polarity detection or two-class problems, with little consideration for decision-making purposes in dynamic online retailing business environments. The objective of this research effort is to propose a multiclass, natural language processing-based technique for classifying customer complaint problems to automate the categorization of online retailer customer complaints into relevant business domain services. Text pre-processing methods used in this study include tokenization, stop-word removal, lemmatization, and noise elimination, with Term Frequency - Inverse Document Frequency (TF-IDF) feature extraction and a Multinomial Naïve Bayes (MNB) classifier. Our dataset consists of 10,000 customer complaints that have been manually annotated, and the generated features are used to train MNB classifiers to assign customer complaints to predefined classes such as Delivery Delay, Product Quality, and Payment Issue. The web-based framework achieved an accuracy of approximately 91%, demonstrating its effectiveness in identifying complaint categories andtransforming raw customer feedback into structured information for business intelligence. This system will improve customer satisfaction, enhance operational efficiency, and reduce the workload associated with manual complaint management.