The Faculty of Informatics is pleased to announce a seminar given by Anon Plangprasopchok
TITLE: Statistical Approaches for Inferring Category Knowledge from Social Annotation
SPEAKER: Anon Plangprasopchok, University of Southern California
DATE: Wednesday, April 29th, 2009
PLACE: USI Università della Svizzera italiana, room SI-006, Informatics building (Via G. Buffi 13)
Social annotation systems allow users to share metadata to describe and organize content of interests. This metadata, aggregated from those of many thousands of users, can potentially be used to infer category knowledge, for, for instances, classifying documents and recommending new relevant information. Since such metadata are voluntarily generated by different individuals, in order to exploit it, there are several issues needed to be considered -- sparsity, ambiguity and inconsistency.
In this talk, I will present two frameworks for inferring category knowledge from the user-generated content, while being aware of the previously mentioned challenges. One framework is for inferring flat concepts, and the other inferring concept hierarchies. By utilizing statistics of occurrences of all entities -- metadata, users and content items --, these frameworks combine these pieces of information; meanwhile, discriminate them according to their different usages by different users. Hence, the frameworks can extract high quality category knowledge. The other framework is for inferring relations between concepts. It aggregates user-specified shallow hierarchical relations and links them into deep taxonomies.
The frameworks were evaluated on the quality of the extracted knowledge. Such quality was measured according to the performance of the learned knowledge in the real-word tasks. For the first framework, I evaluated its performance on resource discovery task, where I used the learned concepts as resource descriptions. For the latter, the learned concept hierarchies were compared to hand-crafted web directories.
HOST: Prof. Fabio Crestani
URL 1: http://www.inf.unisi.ch