Complaint Process Management in an Electric Power Company

Abstract

The complaint process is a mechanismfor citizen participation that provides the means tosubmit petitions, complaints, and claims to companiesproviding goods or services. These appeals arrivein large quantities, must be answered in the timesestablished by law, and are costly to process manually.In this article, we propose a computational method toprocess the complaints written in natural language inSpanish arriving at the Pereira Electric Power Companyin Colombia and then classify the complaints that belongto the area of energy solutions to respond in a faster andmore effective way. Natural Language Processing andMachine Learning techniques are used to classify thetext to construct the method. It starts with the receptionof documents for prediction, performs a preprocessingphase, texts are vectorized, a Recurrent Neural Networkis configured and trained, and finally, the prediction ofeach text is presented. The results show that the methodprocesses and classifies the complaints correspondingto the area of electric power solutions and achieves anaccuracy of 94.35%, a precision of 95%, a recall of 94%,an F-measure of 94.49% and 93.77% according to theROC curve metric. The system was tested preliminarilyand then with a more formal test in a real environment.Compared to the evaluation criteria of other approaches,the method shows promising results. It was developedunder a Service Oriented Software Architecture (SOA)which allowed deployment on a web server and whichhelps the company to process real complaints efficiently.