Improved Cuckoo Search Optimization and Hybrid Firefly Artificial Neural Network Algorithm for Cyberbullying Discovery on Twitter Dataset

Abstract

The exponential growth of SM (Social Media) and SNSs (Social Networking Sites) has seen escalations in Cyberbullying
or bullying electronically making it imperative to focus researches in this area. Cyberbullying has grown into a pervasive and
significant problem, impacting Internet users. TMTs (Text Mining Techniques) can help identify Cyberbullying while addressing
many such similar concerns. However, proposed algorithms have certain issues on the Twitter dataset while identifying
Cyberbullying texts or mining such texts. Hence this study proposes the use of ECSOs (Enhanced Cuckoo Search Optimizations)
along with HFANNs (Hybrid Firefly Artificial Neural Networks) to overcome aforesaid hurdles. This work follows three steps
namely pre-processing, feature subset selections, and classifications in order. K-Means clustering is a pre-processing step in this
work and is used on the Twitter dataset to reduce processing record counts where k-means centroid values and min max
normalisations manage missing and redundant features. K-Means used in this work improves categorization accuracy. The
dataset features are pre-processed for obtaining more information and then used in the feature selection process where ECSOs
compute a feature’s importance based on fitness values and given by an objective function. The proposed HFANN subsequently
classifies the selected features by training on the features to learn and use this learning to predict in tests. The best firefly is
used to classify features accurately. This work’s experimental result demonstrates the effectiveness of the proposed method.
ECSO+HFANN algorithm provides better classification performance in terms of lower time complexity, higher precision, recall, fmeasure and accuracy than the existing algorithms.