Enhancing Malicious Phishing Video Detection and Classification Using Sparrow Search Algorithm with Deep Learning Model

The growing significance of short videos on video-sharing sites is accompanied by an escalation in misleading content, causing adverse social concerns. Therefore, it is important to develop and utilize advanced methods for automatically detecting the unreliable videos. These methods must include techniques namely user engagement analysis, metadata, and content, artificial intelligence (AI), fact-checking integration, and community reporting, integrated with efforts for promoting media skills and potentially robust regulatory measures, to mitigate the risky effects of misleading video content. Recent developments of deep learning (DL) models have been found useful in developing malicious video detection models. Consequently, this research presents a sparrow search technique with deep learning-assisted malicious video detection (SSADL-MVD) technique. The goal of the SSADL-MVD technique is to discriminate the input videos into real and malicious ones. To do so, the SSADL-MVD technique initially employs the DenseNet121 model for deriving features from the input videos. This study has employed three types of features for malicious video detection namely features of content features, uploader, and environment. Content features result from video content. Before feature extraction, each audio data changed into text. Additionally, four sorts of content features exist namely counts of personal pronouns, number of modal particles, sentiment polarity, and text length. Besides, the features are provided in long short-term memory (LSTM) system for detection of malicious videos. Finally, the SSA can be applied for optimal hyperparameter selection of LSTM system, which in turn boosts the overall classifier outcomes. The simulation results of the SSADL-MVD approach on malicious video datasets demonstrated its superiority over other DL models.