Application of Natural Language Processing to Analyse Customer Complaints in Nigerian Telecommunications

Analysing large volumes of unstructured customer complaint text provides telecommunications companies with actionable intelligence for service improvement and prioritisation. This study applied natural language processing techniques to a corpus of 35,000 customer complaint records from a major Nigerian telecommunications operator with branches across the South West and North Central zones. The corpus was collected from the operator's customer relationship management system under a non-disclosure agreement. Text preprocessing included language detection, removal of non-standard characters and Pidgin expressions, tokenisation, and TF-IDF vectorisation. Topic modelling using Latent Dirichlet Allocation identified seven distinct complaint themes: network downtime, data billing discrepancies, SIM registration problems, call drops, customer service delays, mobile money failures, and internet speed dissatisfaction. Network downtime complaints accounted for 28.6 percent of the corpus and peaked consistently on weekday mornings. Geographic mapping of complaint metadata showed that data billing issues were disproportionately concentrated among users in North Central states. Sentiment polarity analysis confirmed that unresolved complaints escalated in negativity after three interactions. The LDA coherence score of 0.61 indicated acceptable topic interpretability. Named entity recognition identified specific base station identifiers frequently mentioned in network downtime complaints, enabling targeted infrastructure diagnostics. Recommendations include an automated complaint routing system based on LDA topic classification, proactive billing anomaly detection, and enhanced self-service resolution workflows to reduce escalation rates in the identified high-frequency complaint categories.

Keywords: natural language processing, customer complaints, telecommunications, LDA topic modelling, Nigeria