Customer opinion mining in electricity distribution company using twitter topic modeling and logistic regression

With the continuous increase in twitter datasets, effective opinion mining techniques are needed for mining these datasets. Subsequent application of results obtained using opinion mining can be applied in the electricity industry to improve its service delivery. This research develops a hybrid model for opinion mining of Abuja Electricity Distribution Company (AEDC) customers’ tweets using topic modeling and classification techniques. The Latent Dirichlet Allocation (LDA) topic modeling was used to generate dominant topics from Twitter API tweets. The topics were grouped into four categories: positive, negative, neutral, or vandalism. To further verify these topics generated by LDA, the opinion retrieved was split into train, test, and validate data using the k-fold cross-validation technique. This step is further classified into four categories using the logistic regression classification technique with a prediction accuracy of 94.8%. However, three more different classifiers: Naïve Bayes, k-nearest neighbors (K-NN), and Support Vector Machine with the resulting accuracies of 93.5%, 92.7%, and 61.6% respectively, are also used. Customer tweets dataset can provide great insight for various companies to understand the opinion of their customers. Electricity companies can apply this knowledge to enhance their service delivery.