Corrections to “A YouTube Spam Comments Detection Scheme Using Cascaded Ensemble Machine Learning Model”

Abstract

In the above article [1], the author would like to correct the Acknowledgment as follows: This work was supported by the Ministry of Education and National Research Foundation of Korea through the ``Convergence and Open Sharing System'' Project. The authors would like to thank Min Chul Jeong and Jihyeon Lee for their great contributions as Ajou University undergraduate students. This study will be extended by applying a deeplearning model to the existing implementation. The Conclusion should also be updated as follows: In this article, we proposed a technique to detect spam comments on YouTube, which has recently seen tremendous growth using a Cascaded Ensemble Machine Learning Model both the Federated Learning(w/o) and the Federated Learning(w/). It examined related studies on YouTube spam comment screening and conducted classi_cation experiments with six different machine learning techniques (decision tree, logistic regression, Bernoulli Na ve Bayes, random forest, support vector machine with linear kernel, support vector machine with Gaussian kernel) and two ensemble models (ensemble with hard voting, ensemble with soft voting) combining these techniques in the comment data. The experimental results showed that the ESM-S model proposed in this paper had the best performance in four of ve evaluation measures both the Federated Learning(w/o) and the Federated Learning(w/). We proposed a new model, combining various techniques, that improved the performance results unlike, previous studies that used one model for detection. We also applied the ensemble model to videos in various categories. It showed that the ESM-S model performed the best in Acc, F1-score, and MCC in both datasets, the ANN model in SC, and the NB-B model in BH. In addition, the data set with 1,000 spam comments and 1,000 normal comments performed better than the data set with 5,000 comments of the increase in outliers and missing values. In future research, it is expected that the performance would be better if a TF-IDF or deep-learning technique is added. This study will be extended by applying a deep-learning model to the existing implementation.