System and method for using graph transduction techniques to make relational classifications on a single connected network

Abstract

A system and method for extending partially labeled data graphs to unlabeled nodes in a single network classification by weighting the data with a weight matrix that uses a modified graph Laplacian based regularization framework and applying graph transduction methods to the weighted data. The technique may be applied to data graphs that are directed or undirected, that may or may not have attributes and that may be homogeneous or heterogeneous.

Description

BACKGROUND OF THE INVENTION[0001]1. Field of the Invention[0002]The present invention generally relates to techniques for statistical relational learning, and more particularly to techniques for making relational classifications on a single connected network.[0003]2. Background Description[0004]Given the prevalence of large connected relational graphs across diverse domains, single or within network classification has been one of the popular endeavors in statistical relational learning (SRL) research. Ranging from social networking websites to movie databases to citation networks, large connected relational graphs are banal. In single network classification, we have a partially labeled data graph and the goal is to extend this labeling, as accurately as possible, to the unlabeled nodes. The nodes themselves may or may not have associated attributes. An example where within network classification could be useful is in forming common interest groups on social networking websites. For ...

AI Technical Summary

Benefits of technology

The present invention offers an efficient and accurate way to make within network relational classifications using graph transduction methods. It provides a procedure to learn a weight matrix for a directed or undirected graph that may show positive or negative auto-correlation and have edges between labeled and unlabeled nodes. This helps to effectively leverage the rich class of graph transduction methods, which are considered to be one of the most efficient and accurate in real applications.

Problems solved by technology

In many cases, the attributes may not accurately predict the labels, in which case, weighting the edges solely on them may not provide acceptable results.

Some of these intuitions are captured in the relational gaussian process model, but it is limited to undirected graphs and the suggested kernel function is not easy to adapt to relational settings where we may have heterogeneous data.

Images