Movie trailers can be classified from an internet-based program through a combination of multimedia analysis techniques that use visual, textual, and audio data extracted from the trailer. Common approaches include:

• Extracting visual features from trailers using pretrained convolutional neural networks (CNNs) or other deep learning image analysis models.
• Analyzing the subtitles or dialogue from the trailer using natural language processing techniques such as TF-IDF vectors or automatic speech recognition.
• Using audio-based models to classify music and sound effects that are typical of specific movie genres.
• Fusing features from these different modalities (video frames, audio, text) using transformer models or classification trees.
• Training classifiers (e.g., classification and regression trees, Naïve Bayes, or deep neural networks) on these extracted features to predict the genre of the movie trailer.
• Segmenting the trailer into scenes and ranking key scenes based on genre-specific models to enhance classification accuracy.

Popular methods perform genre identification first and then use genre-specific models to further analyze scenes or produce trailers. This multimodal approach often results in improved classification accuracy compared to single-modal methods. In brief, classifying movie trailers from an internet-based program involves automatic extraction of multimodal features (visual, textual, audio) followed by machine learning classification models trained to recognize movie genres and key scenes from these features.