QMUL @ MediaEval 2010 Tagging Task: Semantic query expansion for predicting user tags
|QMUL @ MediaEval 2010 Tagging Task: Semantic query expansion for predicting user tags|
|Author(s)||Chandramouli K., Kliegr T., Piatrik T., Izquierdo E.|
|Published in||MediaEval Benchmarking Initiative for Multimedia Evaluation - The "Multi" in Multimedia: Speech, Audio, Visual Content, Tags, Users, Context, MediaEval 2010 Working Notes Proceedings|
|Keyword(s)||Media annotation, Semantic similarity, Wikipedia, Wordnet (Extra: Media Annotations, Semantic expansion, Semantic query expansion, Semantic similarity, Social information, Visual content, Wikipedia, Wordnet, Benchmarking, Ontology, Semantics, Websites, User interfaces)|
|Article||BASE, CiteSeerX, Google Scholar|
|Web||Ask, Bing, Google (PDF), Yahoo!|
|Download and mirrors|
|Local copy||Not available|
|Remote mirror(s)||Not available|
|Export and share|
|BibTeX, CSV, RDF, JSON|
|Browse properties · List of publications|
QMUL @ MediaEval 2010 Tagging Task: Semantic query expansion for predicting user tags is a 2010publication by Chandramouli K., Kliegr T., Piatrik T., Izquierdo E. and published in MediaEval Benchmarking Initiative for Multimedia Evaluation - The "Multi" in Multimedia: Speech, Audio, Visual Content, Tags, Users, Context, MediaEval 2010 Working Notes Proceedings.
This paper describes our participation in "The Wild Wild Web Tagging Task @ MediaEval 2010", which aims to predict user tags based on features derived from video such as speech, audio, visual content or associated textual or social information. Two tasks were pursued: (i) closed-set annotations and (ii) open-set annotations. We have attempted to evaluate whether using only a limited number of features (video title, filename and description) can be compensated by semantic expansion with NLP tools and Wikipedia and Wordnet. This technique proved successful on the open-set task with approximately 20% generated tags being considered relevant by all manual annotators. On the closed-set task, the best result (MAP 0.3) was achieved on tokenized filenames combined with video descriptions, indicating that filenames are a valuable tag predictor.
- This section requires expansion. Please, help!
Probably, this publication is cited by others, but there are no articles available for them in WikiPapers.