Title :
Evaluating visual and textual features for predicting user ‘likes’
Author :
Guntuku, Sharath Chandra ; Roy, Sujoy ; Weisi, Lin
Author_Institution :
Sch. of Comput. Eng., Nanyang Technol. Univ., Singapore, Singapore
fDate :
June 29 2015-July 3 2015
Abstract :
Computationally modeling users `liking´ for image(s) requires understanding how to effectively represent the image so that different factors influencing user `likes´ are considered. In this work, an evaluation of the state-of-the-art visual features in multimedia understanding at the task of predicting user `likes´ is presented, based on a collection of images crawled from Flickr. Secondly, a probabilistic approach for modeling `likes´ based only on tags is proposed. The approach of using both visual and text-based features is shown to improve the state-of-the-art performance by 12%. Analysis of the results indicate that more human-interpretable and semantic representations are important for the task of predicting very subtle response of `likes´.
Keywords :
image representation; recommender systems; Flickr; feature representation; likes; semantic representations; text-based features; visual-based features; Computational modeling; Feature extraction; Probabilistic logic; Semantics; Sparse matrices; Training; Visualization; Feature Representation; Image; Likes; Recommendation; Tags;
Conference_Titel :
Multimedia and Expo (ICME), 2015 IEEE International Conference on
Conference_Location :
Turin
DOI :
10.1109/ICME.2015.7177381