Etiquetado semántico de vídeos basado en aprendizaje profundo y minería de texto

Abstract

El vídeo online es responsable del 40% del tráfico de Internet, con una tendencia al alza debido al crecimiento de las redes sociales y las plataformas destinadas a este fin, tales como YouTube, Vimeo, Viddler, etc. Para estas plataformas, debido al gran volumen de vídeos que manejan, se presenta de forma natural la problemática referente a la clasificación de los mismos en tópicos o clases. Este es, precisamente, el objetivo desarrollado en este TFM: una metodología que permite realizar clasificación de vídeo desde un enfoque basado en minería de texto. Aunque los resultados de clasificación obtenidos han sido dispares, la metodología desarrollada es perfectamente válida y funcional, siempre en función del dataset que se este utilizando

Online video is responsible for 40% of Internet traffic, with an upward trend due to the growth of social networks and platforms for this purpose, such as YouTube, Vimeo, Viddler, etc. For these platforms, due to the large volume of videos they have, the problem of classifying them into topics or classes is presented naturally. This is precisely the objective developed in this TFM: a methodology that allows video sorting from a text-mining-based approach. Although the classification results obtained have been uneven, the methodology developed is perfectly valid and functional, always depending on the dataset used.

Online video is responsible for 40% of Internet traffic, with an upward trend due to the growth of social networks and platforms for this purpose, such as YouTube, Vimeo, Viddler, etc. For these platforms, due to the large volume of videos they have, the problem of classifying them into topics or classes is presented naturally. This is precisely the objective developed in this TFM: a methodology that allows video sorting from a text-mining-based approach. Although the classification results obtained have been uneven, the methodology developed is perfectly valid and functional, always depending on the dataset used.

Online video is responsible for 40% of Internet traffic, with an upward trend due to the growth of social networks and platforms for this purpose, such as YouTube, Vimeo, Viddler, etc. For these platforms, due to the large volume of videos they have, the problem of classifying them into topics or classes is presented naturally. This is precisely the objective developed in this TFM: a methodology that allows video sorting from a text-mining-based approach. Although the classification results obtained have been uneven, the methodology developed is perfectly valid and functional, always depending on the dataset used.