نسترن داب

This Study presents a novel approach for the automatic detection of the Induced Pleasure of Videos by fusion of multimodal data (audio an‎d visual). Given the ever-increasing volume of video content an‎d its growing role in daily life, understan‎ding an‎d predicting usersʹ emotional responses, particularly the sense of pleasure, has gained significant importance. The primary goal of this research is to develop a computational model that, by intelligently integrating information from various video modalities an‎d modeling based on psychological theories of pleasure, can estimate the level of pleasure induced by a video in viewers. Previous studies on video sentiment analysis have often relied on data-driven approaches that heavily depend on biased human labels an‎d focus solely on categorical emotions rather than directly addressing the aspect of video pleasantness. These studies have also frequently overlooked the critical role of cognitive appraisals in emotional experiences, treating emotions as direct responses to stimuli. Furthermore, there is a significant lack of datasets explicitly labeled for video enjoyment. In this regard, we have focused on computing induced pleasure from multimodal data by combining data-driven an‎d model-driven approaches. To this end, an innovative approach that considers our cognitive appraisals of stimuli has been adopted, drawing inspiration from psychological theories of pleasure to calculate the level of pleasure. Advanced feature extraction an‎d multimodal data fusion techniques have been explored an‎d implemented to capture complex interactions between these modalities. This enables the identification of hidden patterns in multimodal data that are directly related to the experience of pleasure. This computational modeling not only addresses the identification of pleasurable content but also seeks to understan‎d the nuances an‎d various dimensions of pleasure in response to video stimuli. Ultimately, it provides the capability to generate pleasure labels for datasets instead of relying on limited data. The results of comprehensive eva‎luations demonstrate the effectiveness of the proposed approach in accurately detecting the induced pleasure of videos. This research represents a significant step toward developing content recommendation systems, intelligent content creation tools, an‎d a deeper understan‎ding of the relationship between digital content an‎d human emotional responses.

لذت , گداخت , يادگيري عميق , تحليل محتواي عاطفي ويديويي

pleasure , fusion , Deep Learning , video affective content analysis

Nastaran Dab

Dr. Mohamadreza Kangavari