Improving automatic edge selection for relational classification

Cristina Pérez-Solà, Jordi Herrera-Joancomartí

Research output: Contribution to journalArticleResearchpeer-review


In this paper, we address the problem of edge selection for networked data, that is, given a set of interlinked entities for which many different kinds of links can be defined, how do we select those links that lead to a better classification of the dataset. We evaluate the current approaches to the edge selection problem for relational classification. These approaches are based on defining a metric over the graph that quantifies the goodness of a specific link type. We propose a new metric to achieve this very same goal. Experimental results show that our proposed metric outperforms the existing ones.

Original languageAmerican English
Pages (from-to)284-295
Number of pages12
JournalLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Publication statusPublished - 2013


Dive into the research topics of 'Improving automatic edge selection for relational classification'. Together they form a unique fingerprint.

Cite this