<p>Analyzing all the available resources, namely audio, text metadata, and visual content in the video, is essential for accurate Video Categorization (VC). However, none of the prevailing works focused on extracting text from video for filtering relevant information. Also, text derived from audio was not utilized for VC based on extracted textual features, affecting the accuracy of VC. Therefore, the Distribution of Pearson Correlation Coefficient-based Latent Dirichlet Allocation (DPCC-LDA)-based Video Recognition (VR) is proposed, which utilizes the features extracted from video frames and audio content to effectively filter and analyze the most relevant information. Primarily, the video is converted into frames, and key frames are extracted. Next, frames undergo contrast enhancement using Contrast Limited Adaptive Histogram Equalization (CLAHE). Further, the enhanced frames are checked to verify whether the video contains text. If text is identified, then the frames undergo binarization, followed by text extraction using Modified Region Proposal Network–based Faster Regional-Convolutional Neural Network (MRPN-FRCNN) and relevant information filtering. If text is not detected, then textual information is extracted from the corresponding audio file. Further, the extracted texts are pre-processed, and the VR is performed using DPCC-LDA. Then, keywords are extracted from the VR output and converted into word embeddings using Central Limit Theorem-based Bidirectional Encoder Representations from Transformers (CLT-BERT). Finally, the different categories of respective classified videos are identified using Stochastic Weight Averaging-based Long Short-Term Memory (SWA-LSTM). Thus, the proposed approach performed VC with an average accuracy of 98.34% and average precision of 98.46%, showing better performance than prevailing works.</p>