AUTOMATED INAPPROPRIATE CONTENT DETECTION IN YOUTUBE VIDEOS USING DEEP LEARNING TECHNIQUES
Murali Mohan Reddy .M, PARSHAVENI UMA MAHESHWARI · International journal of engineering science and advanced technology. · 2024
YouTube's video content has grown exponentially, drawing billions of viewers, most of whom are in the younger age range. Additionally, malicious uploaders use this site as a means of disseminating disturbing visual information. For example, they use animated cartoon videos to disseminate stuff that is improper for children. Therefore, it is strongly advised that social media networks have an automated real-time video content screening method. This paper proposes a new architecture based on deep learning for the identification and categorisation of objectionable information in films. A pretrained convolutional neural network (CNN) model called EfficientNet-B7 is used in the proposed framework to extract video descriptors, which are then fed into a bidirectional long short-term memory (BiLSTM) network to enable multiclass video classification and the learning of effective video representations. In order to apply the attention probability distribution in the network, an attention mechanism is also included after the BiLSTM. A carefully annotated dataset including 111,156 cartoon clips gathered from YouTube videos is used to assess these algorithms. EfficientNetBiLSTM (accuracy D 95.66%) outperforms the attention mechanism-based EfficientNetBiLSTM (accuracy D 95.30%) framework, according to experimental data. Second, deep learning classifiers outperform typical machine learning classifiers in terms of performance. All things considered, the EfficientNet and BiLSTM design with 128 hidden units produced state-of-the-art results (f1 score D 0.9267). Additionally, the performance comparison against current cutting-edge methods confirmed that BiLSTM on top of CNN captures better contextual information of video descriptors in network architecture, leading to better results in the detection and classification of inappropriate video content for children. An essential element of the expanding discipline of data science is machine learning. Several types of algorithms are taught to create predictions or classifications and to unearth important insights in this project by use of statistical methodologies. Subsequently, these insights inform business and application decisions, which ideally influence important growth metrics. Without being specifically taught to do so, machine learning algorithms create a model based on this project data, sometimes referred to as training data, in order to generate predictions or judgements. In many different datasets, machine learning algorithms are used when it is impractical or impossible to create traditional algorithms to carry out the necessary tasks.