Reducing noisy labels in weakly labeled data for visual sentiment analysis

Wu, L, Liu, S, Jian, M, Luo, J, Zhang, X and Qi, M 2017, 'Reducing noisy labels in weakly labeled data for visual sentiment analysis', in Proceedings of the IEEE International Conference on Image Processing (ICIP 2017), Beijing, China, 17-20 September 2017, pp. 1322-1326.


Document type: Conference Paper
Collection: Conference Papers

Title Reducing noisy labels in weakly labeled data for visual sentiment analysis
Author(s) Wu, L
Liu, S
Jian, M
Luo, J
Zhang, X
Qi, M
Year 2017
Conference name ICIP 2017
Conference location Beijing, China
Conference dates 17-20 September 2017
Proceedings title Proceedings of the IEEE International Conference on Image Processing (ICIP 2017)
Publisher IEEE
Place of publication United States
Start page 1322
End page 1326
Total pages 5
Abstract Deep learning-based visual sentiment analysis requires a large dataset for training. Dataset from social networks is popular but noisy because some images collected in this manner are mislabeled. Therefore, it is necessary to refine the dataset. Based on observations to such datasets, we propose a refinement algorithm based on the sentiments of adjective-noun pairs (ANPs) and tags. We first determine the unreliably labeled images through the sentiment contradiction between the ANPs and tags. These images are removed if the numbers of tags with positive and negative sentiments are equal. The remaining images are labeled again based on the majority vote of the tags' sentiments. Furthermore, we improve the traditional deep learning model by combining the softmax and Euclidean loss functions. Additionally, the improved model is trained using the refined dataset. Experiments demonstrate that both the dataset refinement algorithm and improved deep learning model are beneficial. The proposed algorithms outperform the benchmark results.
Subjects Pattern Recognition and Data Mining
Keyword(s) visual sentiment analysis
mislabeled images
deep learning
sentiment conflict
DOI - identifier 10.1109/ICIP.2017.8296496
Copyright notice © 2017 IEEE
ISBN 9781509021758
Versions
Version Filter Type
Citation counts: TR Web of Science Citation Count  Cited 2 times in Thomson Reuters Web of Science Article | Citations
Scopus Citation Count Cited 0 times in Scopus Article
Altmetric details:
Access Statistics: 10 Abstract Views  -  Detailed Statistics
Created: Thu, 31 Jan 2019, 11:26:00 EST by Catalyst Administrator
© 2014 RMIT Research Repository • Powered by Fez SoftwareContact us