Sala 1.1/2 IMDEA Networks Institute, Avda. del Mar Mediterráneo 22, 28918 Leganés – Madrid
The use of deceptive techniques in user-generated video portals is ubiquitous. Unscrupulous uploaders deliberately mislabel video descriptors, such as headline, thumbnail etc., in order to increase their views and subsequently their ad revenue. However, such deceptive practices come at the expense of the end-user’s experience as the content does not meet the viewer’s expectations, which are formed by the misleading headline and thumbnail.
This problem is widely known as "clickbait". In this work, we perform measurements that reveal the great extent of this problem on user-generated video portals, in particular YouTube. To address the clickbait problem, we leverage recent advances in the field of deep learning. Specifically, we devise an innovative formulation of variational autoencoders that contemplates the diverse modalities of data pertaining to videos. The proposed model relies on a limited amount of manually labeled data to classify a large corpus of unlabeled data. Our evaluation indicates that the proposed model offers satisfactory performance. We envision the adoption of such model by YouTube and other user-generated-video portals for effectively mitigating the clickbait problem.
About Savvas Zannettou
Savvas Zannettou is a PhD student from Cyprus University of Technology. Currently, he is working as a Research Intern at Telefonica I+D. In 2014 and 2016 he received respectively the BSc and MSc degrees in Computer Engineering from Cyprus University of Technology. During 2014, he was a Research Intern at NEC Labs Europe for 6 months where he worked on Software-Defined Networks. His research interests include social networks analysis, deep learning and software-defined networks. He also has experience with working on EU-funded projects.
Este evento se impartirá en inglés