Abstract:
With the explosion of user-generated video material on sites like Facebook and YouTube and
e-learning platforms like Coursera, Udacity, and Udemy, new approaches for categorizing,
annotating, and retrieving digital information are needed to make it more valuable in our daily
lives. As a result, cognitive measure prediction, such as memorability, offers a wide range of
possible uses. The attribute or state of being easy to recall is referred to as memorability.
Memorability, like other important video cues like aesthetics or interestingness, can be used to
assist people in choosing between otherwise similar videos. Therefore, a wide range of
applications, such as Knowledge and training, content retrieval, summarization of content, story
narration, online advertising, content promotion and screening, would benefit from systems able
to rank videos according to their memory. In the proposed method, visual and semantic features
have been used to train different Machine Learning and Deep learning models on the dataset of
10,000 soundless videos provided by the MediaEval Benchmarking Initiative to predict short and
long term video memorability scores. According to the findings, short-term memorability is more
predictable than long-term memorability since all models scored higher in short-term memorability
than long-term memorability. The best results have been achieved with RNN using video captions
and embedding.
