Learning to Rank for Joy


Orellana-Rodriguez C., Nejdl W., Diaz-Aviles E., ALTINGÖVDE İ. S.

23rd International Conference on World Wide Web (WWW), Seoul, South Korea, 7 - 11 April 2014, pp.569-570 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1145/2567948.2576961
  • City: Seoul
  • Country: South Korea
  • Page Numbers: pp.569-570

Abstract

User-generated content is a growing source of valuable information and its analysis can lead to a better understanding of the users needs and trends. In this paper, we leverage user feedback about YouTube videos for the task of affective video ranking. To this end, we follow a learning to rank approach, which allows us to compare the performance of different sets of features when the ranking task goes beyond mere relevance and requires an affective understanding of the videos. Our results show that, while basic video features, such as title and tags, lead to effective rankings in an affective-less setup, they do not perform as good when dealing with an affective ranking task.