Λογότυπο αποθετηρίου

Πλοήγηση ανά Συγγραφέα "Samoili, Varvara"

A B C D E F G H I J K L M N O P Q R S T U V W X Y Z
Α Β Γ Δ Ε Ζ Η Θ Ι Κ Λ Μ Ν Ξ Ο Π Ρ Σ Τ Υ Φ Χ Ψ Ω
Φιλτράρετε τα αποτελέσματα πληκτρολογώντας τα πρώτα γράμματα
Τώρα δείχνει 1 - 1 από 1
  • Μικρογραφία εικόνας
    Τεκμήριο
    Emotion-aware content representation and retrieval for movie dialogues
    (2018-12-21) Samoili, Varvara; Σαμοΐλη, Βαρβάρα; Athens University of Economics and Business, Department of Informatics; Giannakopoulos, Theodoros; Androutsopoulos, Ion
    The vast variety of information available on today’s web has created a need for state of-the-art Recommender Systems. Apart from collaborative methods, which are based on modeling the similarities between the preferences of different users, content-based retrieval applications for Recommender Systems and User Profiling, particularly in the area of movie recommendations, are also important. Their contribution become seven more valuable when, apart from static metadata, they also use underlying information related to the content consumed by the user. Furthermore, during the last two decades, Emotion Recognition has peaked the interest of researchers involved in Speech and Text Analytics. Meanwhile, emotion and the way it is conveyed is particularly important in films, as it undoubtedly plays a major role in the final aesthetic result. This leads us to believe that speech emotion in movie dialogues can act as an extra ‘dimension’ in content-based movie retrieval and recommendation, resulting in emotion-aware content-based movie retrieval.In this work, we show how specific high-level attributes, which derive from speech emotion estimates in movie dialogues, can constitute a discriminative factor when separating movie content. This is demonstrated through the use of an open and widely used dialogue benchmark, which first undergoes appropriate preprocessing. Experiments show that while, on one hand, emotion-based information alone is not a reliable enough factor for movie retrieval, there is, nonetheless, a statistically significant correlation between the ‘emotion-aware’ features and high-level movie attributes. Further research should be conducted in order to explore fusion methods of this emotion-based information along with metadata, as well as other types of content-based information(music, vision, etc.) towards the improvement of recommender systems.