An Approach to Predict Users' Portability in the Peruvian Communications Sector Using Vector Representation Models and Community Detection Algorithm

Maguina R.; Pozo R.; Wong L. · 2026 9th International Conference on Artificial Intelligence and Big Data, ICAIBD 2026 (2026) · DOI: 10.1109/icaibd69640.2026.11637208

Resumen

Number portability and client migration represent one of the main challenges in the Peruvian telecommunications sector, due to their impact on revenues, retention and operational sustainability. To address this issue, a hybrid approach that combines semantic representation techniques and community detection applied to large volumes of customer interactions is proposed. The study is conducted in five steps: (i) dataset identification and curation, (ii) data preprocessing, (iii) vector representation models Word2Vec and Node2Vec, (iv) community detection algorithms Louvain and Leiden, (v) community-based prediction of collective churn, and (vi) performance evaluation. The embeddings made it possible to retain recurring semantic patterns in customer service reasons, while the communities revealed structured groups of behavior prior to churn. The results show that Word2Vec and the Louvain Method are superior due to their effectiveness and balance across the various metrics and tests performed.

Instituciones: Universidad Peruana de Ciencias Aplicadas