8. Sentiment Feature Extraction
• Nowadays the Internet, as a major platform for
communication and information exchange,
provides a rich repository of people’s opinion and
sentiment about a vast spectrum of topics
• The analysis of such information, such as
comments, tags, browsing actions, as well as
shared media objects, plays an important role in
behavior sciences, which aims to understand and
predict human decision making
Borth, Damian, et al. "Large-scale visual sentiment ontology and detectors using adjective noun
pairs." Proceedings of the 21st ACM international conference on Multimedia. ACM, 2013.
9. Sentiment Feature Extraction
• The computational analysis of sentiment
mostly concentrates on the textual content.
Limited efforts have been conducted to
analyze sentiments from visual content
• Visual content conveying valuable sentiment
information primarily visually
19. Learning with Label Proportions(LLP)
Yu, F. X., et al. "SVM for learning with label proportions." Proceedings of the 30rd
International Conference on Machine learning. 2013.
Negative
Positive
33% positive
75% positive
20. Video Detection by LLP
Lai, Kuan-Ting, et al. "Video event detection by inferring temporal instance labels." Computer
Vision and Pattern Recognition (CVPR), 2014 IEEE Conference on. IEEE, 2014.
21. Experiment Setting
• Dataset
– This dataset contains 2 events
– 5816 people
– Every person has 100 images
– Split 2/3 of people as training set and 1/3 as test set
• Features
– Frame features are quantized into 1 and -1 to represent
positive and negative emotion respectively
• Label
– Bags are labeled positive when having 50% more positive
images and negative if less than 50%
22. Experiment Result
• Precision of Event1(Positive)
– 99.92%
• Precision of Event2(Negative)
– 99.28%
• Average Precision
– 99.6%