Discriminative graphical models for faculty homepage discovery |
| |
| Authors: | Yi Fang Luo Si Aditya P Mathur |
| |
| Institution: | (1) Department of Computer Science, Purdue University, West Lafayette, IN 47907, USA |
| |
| Abstract: | Faculty homepage discovery is an important step toward building an academic portal. Although the general homepage finding
tasks have been well studied (e.g., TREC-2001 Web Track), faculty homepage discovery has its own special characteristics and
not much focused research has been conducted for this task. In this paper, we view faculty homepage discovery as text categorization
problems by utilizing Yahoo BOSS API to generate a small list of high-quality candidate homepages. Because the labels of these
pages are not independent, standard text categorization methods such as logistic regression, which classify each page separately,
are not well suited for this task. By defining homepage dependence graph, we propose a conditional undirected graphical model
to make joint predictions by capturing the dependence of the decisions on all the candidate pages. Three cases of dependencies
among faculty candidate homepages are considered for constructing the graphical model. Our model utilizes a discriminative
approach so that any informative features can be used conveniently. Learning and inference can be done relatively efficiently
for the joint prediction model because the homepage dependence graphs resulting from the three cases of dependencies are not
densely connected. An extensive set of experiments have been conducted on two testbeds to show the effectiveness of the proposed
discriminative graphical model. |
| |
| Keywords: | |
| 本文献已被 SpringerLink 等数据库收录！ |
|
