Textual Analysis for Video Memorability Prediction

Camille Guinaudeau (1, 2, 3, 4) , Andreu Girbau Xalabarder (5)

1 JFLI - Japanese French Laboratory for Informatics
2 Université Paris-Saclay
3 LISN - Laboratoire Interdisciplinaire des Sciences du Numérique
4 STL - Sciences et Technologies des Langues - LISN
5 NII - National Institute of Informatics

This article presents the analysis carried out by the Japanese French Laboratory for Informatics (JFLI) and the National Institute of Informatics (NII) to understand what makes a video memorable. To do so, we first propose an analysis of the results obtained by two sequential models applied on visual and textual representations. We then study the manual descriptions and automatic captions in order to identify specificities in the textual representations of videos associated with a high memorability score. We show that they are described by longer and more precise texts (manual and automatic) than the videos associated with lower memorability scores, opening the way to research on the correlation between textual vagueness and video memorability.

Mots clés: Video Memorability; Textual Analysis

Domaines: Informatique et langage [cs.CL]

Camille Guinaudeau, Andreu Girbau Xalabarder. Textual Analysis for Video Memorability Prediction. the 13th MediaEval Multimedia Benchmark Workshop, Jan 2023, Bergen, Norway. ⟨hal-04091024⟩