Learning to tag Export

@inproceedings{citeulike:4387938, abstract = {Social tagging provides valuable and crucial information for large-scale web image retrieval. It is ontology-free and easy to obtain; however, irrelevant tags frequently appear, and users typically will not tag all semantic objects in the image, which is also called semantic loss. To avoid noises and compensate for the semantic loss, tag recommendation is proposed in literature. However, current recommendation simply ranks the related tags based on the single modality of tag co-occurrence on the whole dataset, which ignores other modalities, such as visual correlation. This paper proposes a multi-modality recommendation based on both tag and visual correlation, and formulates the tag recommendation as a learning problem. Each modality is used to generate a ranking feature, and Rankboost algorithm is applied to learn an optimal combination of these ranking features from different modalities. Experiments on Flickr data demonstrate the effectiveness of this learning-based multi-modality recommendation strategy.}, address = {New York, NY, USA}, author = {Wu, Lei and Yang, Linjun and Yu, Nenghai and Hua, Xian S.}, booktitle = {WWW '09: Proceedings of the 18th international conference on World wide web}, citeulike-article-id = {4387938}, citeulike-linkout-0 = {http://portal.acm.org/citation.cfm?id=1526709.1526758}, citeulike-linkout-1 = {http://dx.doi.org/10.1145/1526709.1526758}, comment = {[Talk] Std. image search: query -> annotations (text/content/tags) -> images -> results. Flickr tag distribution (more frequent to less frequent): uninformative, popular, specific, noise. Three types of features for tag similarity: tag coocurrence + tag content correlation (among the text of tags, e.g. synonyms -- handled from PicNet distance ACM MM'09) + tag visual correlation (among images) }, doi = {10.1145/1526709.1526758}, isbn = {978-1-60558-487-4}, keywords = {image, search, tagging, text}, location = {Madrid, Spain}, pages = {361--370}, posted-at = {2009-04-23 17:01:03}, priority = {0}, publisher = {ACM}, title = {Learning to tag}, url = {http://dx.doi.org/10.1145/1526709.1526758}, year = {2009} }

TY - CONF ID - citeulike:4387938 L3 - citeulike-article-id:4387938 N2 - Social tagging provides valuable and crucial information for large-scale web image retrieval. It is ontology-free and easy to obtain; however, irrelevant tags frequently appear, and users typically will not tag all semantic objects in the image, which is also called semantic loss. To avoid noises and compensate for the semantic loss, tag recommendation is proposed in literature. However, current recommendation simply ranks the related tags based on the single modality of tag co-occurrence on the whole dataset, which ignores other modalities, such as visual correlation. This paper proposes a multi-modality recommendation based on both tag and visual correlation, and formulates the tag recommendation as a learning problem. Each modality is used to generate a ranking feature, and Rankboost algorithm is applied to learn an optimal combination of these ranking features from different modalities. Experiments on Flickr data demonstrate the effectiveness of this learning-based multi-modality recommendation strategy. CY - Madrid, Spain SN - 978-1-60558-487-4 AD - New York, NY, USA EP - 370 TI - Learning to tag PB - ACM T2 - WWW '09: Proceedings of the 18th international conference on World wide web SP - 361 KW - image KW - search KW - tagging KW - text N1 - [Talk] Std. image search: query -> annotations (text/content/tags) -> images -> results. Flickr tag distribution (more frequent to less frequent): uninformative, popular, specific, noise. Three types of features for tag similarity: tag coocurrence + tag content correlation (among the text of tags, e.g. synonyms -- handled from PicNet distance ACM MM'09) + tag visual correlation (among images) AU - Wu, Lei AU - Yang, Linjun AU - Yu, Nenghai AU - Hua, Xian PY - 2009/// UR - http://dx.doi.org/10.1145/1526709.1526758 ER -