Qasim, Syed Sumair
Tag disambiguation based on social network information.
University of Southampton, School of Electronics and Computer Science,
PDF (Syed Sumair Qasim, thesis on Tag Disambiguation Based on Social Network Information)
Within 20 years the Web has grown from a tool for scientists at CERN into a global information space. While returning to its roots as a read/write tool, its entering a more social and participatory phase. Hence a new, improved version called the Social Web where users are responsible for generating and sharing content on the global information space, they are also accountable for replicating the information. This collaborative activity can be observed in two of the most widely practised Social Web services such as social network sites and social tagging systems. Users annotate their interests and inclinations with free form keywords while they share them with their social connections. Although these keywords (tag) assist information organization and retrieval, they
suffer from polysemy.
In this study we employ the effectiveness of social network sites to address the issue of ambiguity in social tagging. Moreover, we also propose that homophily in social network sites can be a useful aspect is disambiguating tags. We have extracted the ‘Likes’ of 20 Facebook users and employ them in disambiguation tags on Flickr. Classifiers are generated on the retrieved clusters from Flickr using K-Nearest-Neighbour algorithm and then their degree of similarity is calculated with user keywords. As tag disambiguation techniques lack gold standards for evaluation, we asked the users to indicate the contexts and used them as ground truth while examining the results. We analyse the performance of our approach by quantitative methods and report successful results. Our proposed method is able classify images with an accuracy of 6 out of 10 (on average). Qualitative analysis reveal some factors that affect the findings, and if addressed can produce more precise results.
Actions (login required)