Automating Customer Complaint Classification in Telecom using Pretrained Natural Language Processing Models
Azhaguvelan Thayumanavan · 2025
Customer complaint management is a key operational challenge for telecommunication service providers, where manual classification operations utilize significant human resources and present classification inconsistencies. This study introduces an extensive examination of automated complaint classification systems using pretrained Natural Language Processing (NLP) models that attain 92.5% accuracy using transformer-based architecture. The research compares various techniques such as standard machine learning techniques, bidirectional LSTM models, and BERT-based transformer models on a 6,000-strong telecom customer complaint dataset with six categories. Experiments prove the improvement of results rapidly compared to standard methods, with transformer models outperforming regular Naive Bayes classifiers with 5.2% improvement and retaining production-level inference times of 12 milliseconds. The study fills significant gaps in domain-specific NLP applications for telecommunications by offering an extensible framework for real-time complaint handling and customer care automation.