Ying Zhou
School of Information Technologies, The University of Sydney
Mining socially tagged images
A Picture is worth a thousand words! What would be the worth of billions of pictures?
Pictures and what not
This is a photo taken in forbidden city, Beijing, China in October 2009. It is taken by a Canon EOS 450d camera with a Canon EF 70-200mm f/4.0 L IS USM lenses. It has been viewed 2000 times on flickr and is featured in 50+ galleries; 600+ flickr users called it a favourite; It is also included in several photo groups related with travel, camera, city life and so on; Its owner has 200+ contacts and is in the contact list of more than 300+ users, some of those commented, noted, favourited or galleried this photo. Its owner has uploaded 2000+ photos taken in various places around the world at various period of time....
Projects
A lot more can be said about a single picture without even look at the picture itself. The rich amount of contextual information, all captured by photo sharing sites such as flickr or Google Picasa enables all sorts of exciting researches in the social and technological area to understand the interweaving web of photos and users.
My research falls into two big categories: improve end user experience and trend discovery
Improve user experience in social sharing sites
The richness of contextual information is both a blessing and a curse to the users of and the visitors to a social sharing sites. Apart from some simple, fact-based information such as date, location, equipment, the owner of a photo has to choose wisely for tags to describe it, collections and sets to display it, groups to join and so on. The ultimate goal is to maximize a photo's visibility among billions of visitors out of billions of similar photos. Fears of missing out may drive many users to add indiscriminately along list of tags and to join in all sorts of group. This makes it hard for visitors, who are interested in finding high quality photos, to choose a proper set of query terms to filter out noisy results.
We are trying to apply clustering algorithms with heuristics from a photo's contextual information such as its groupship, its owner's contact to filter out query result to improve query user's experience by providing filtered results and to improve photo owner's experience by suggesting tags, sets and groups.
Trend Discovery
The temporal and spatial information makes it easy to track an individual user's moveabout. Simply aggregating many users' moveabout may reveal popular sites or hotspots on earch. A lot more can be discovered beyond simple aggregation. By investigating a group of similar users moveabouts on a time window can reveal interesting itinerary preference.
Members
- Xiaochen Huang (Mphil student)
- Mandy Cao (Honours student)
- Hao Zhong (Mphil student)
- Shirley Priyanka Lee (Master coursework student)
Publications
- JooHee Song, Ying Zhou, Hyiungsoo Jung and Joseph Davis,Adding context to social tagging systems, 21stAustralian Conference on Information Systems, Brisbane, QLD, Australia,December, 2010
- Ying Zhou, Xiaochen Huang and Shirley Priyanka Lee, Category Recommendation in User Specified Structure, EC-Web 2010, Bilbao, Spain, September 1-3, 2010
- Xiaochen Huang and Ying Zhou, An Asymmetric Similarity Measure for Tag Clustering on Flickr, accepted by APWeb2010, 6-8 April, 2010, Busan, Korea.
- H. Lin, J. Davis, and Y. Zhou, Integration of Computational and Crowd-Sourcing Methods for Ontology Extraction, Proceedings of the 5th International Conference on Semantics, Knowledge and Grid(SKG2009) Oct. 12-14, 2009, Zhuhai, China, pp 306-310
- Y. Zhou, Searching and Clustering on Social Tagging Site, Proceedings of the 5th International Conference on Semantics, Knowledge and Grid(SKG2009) Oct. 12-14, 2009, Zhuhai, China, pp 99-105
- H Lin, J. Davis and Ying Zhou. An Integrated Approach to Extracting Ontological Structures from Folksonomies, Proceedings of 6th Annual European Semantic Web Conference, May 31- June 4 2009, Heraklion, Greece