An annotated video dataset for computing video memorability

Area of Science:

• Cognitive Psychology
• Computer Vision
• Human-Computer Interaction

Background:

• Understanding video memorability is crucial for content creation and recommendation systems.
• Short-form videos are increasingly prevalent, necessitating research into their cognitive impact.
• Existing datasets often lack comprehensive annotations for memorability prediction.

Purpose of the Study:

• To create and analyze a dataset of short-form videos annotated for memorability.
• To investigate factors influencing short-term and long-term video recall.
• To facilitate research in computational video memorability prediction.

Main Methods:

• Collected 1275 user annotations on video memorability (short-term and long-term recall).
• Utilized an online memory game format for data collection, measuring recognition accuracy and reaction times.
• Extracted video features including text captions and image-level features from key frames.

Main Results:

• The dataset provides a rich resource for studying human memory for video content.
• Analysis of annotations can reveal patterns correlating with video memorability.
• The dataset was successfully used in the Video Memorability task at MediaEval 2020.

Conclusions:

• The developed dataset enables further research into predicting video memorability.
• Understanding memorability factors can improve video content design and delivery.
• This work contributes to the field of affective computing and multimedia analysis.