Enhanced Video Summarization Using BiLSTM Encoder–Decoder With Attention Mechanism and Particle Swarm Optimization

ABSTRACT Due to the explosion of video content on different platforms such as YouTube, TikTok, and streaming applications, as well as the increase in mobile accessibility and remoteness, the information overload has become overwhelming. This has introduced the need to come up with effective ways of video summarization to make it more accessible and user‐friendly. Nonetheless, current solutions tend to be semantically shallow, contextually lacking, and storytelling. To address these shortcomings, we also suggest a hybrid architecture that combines bidirectional long short‐term memory (BiLSTM) with two different attention mechanisms (Bahdanau attention, Luong attention) under the identical settings, as well as particle swarm optimization (PSO). BiLSTM considers both forward and backward temporal dependence, whereas the attention mechanism is spatial and temporal in nature in keeping the relevance and continuity. PSO also improves the determination of the keyframes by eliminating redundancy and maximizing the presentation of features. The model under consideration was tested against benchmark datasets such as SumMe and TVSum, and measured in terms of such metrics as precision, recall, F1‐score, and accuracy. Experimental results demonstrate that our framework outperforms several state‐of‐the‐art methods, offering improved summarization accuracy and adaptability across various video genres and application scenarios.