Partial least squares regression for graph mining

Abstract

Attributed graphs are increasingly more common in many application domains such as chemistry, biology and text processing. A central issue in graph mining is how to collect informative subgraph patterns for a given learning task. We propose an iterative mining method based on partial least squares regression (PLS). To apply PLS to graph data, a sparse version of PLS is developed first and then it is combined with a weighted pattern mining algorithm. The mining algorithm is iteratively called with different weight vectors, creating one latent component per one mining call. Our method, graph PLS, is efficient and easy to implement, because the weight vector is updated with elementary matrix calculations. In experiments, our graph PLS algorithm showed competitive prediction accuracies in many chemical datasets and its efficiency was significantly superior to graph boosting (gBoost) and the naive method based on frequent graph mining.

On the Usefulness of Weight-Based Constraints in Frequent Subgraph Mining

“…On the other side, the selection process can be expensive as well. Recent studies to investigate scalable algorithms demonstrate this [7,10,17,18,21]. They deal with the direct mining of patterns satisfying an objective function, instead of following the two-step approach.…”

“…But they do not necessarily fall into any of the constraint classes mentioned before. Objective functions are either based on their ability to discriminate between classes or numerical values associated with the graphs [10,18] or on some topological similarity measures [7,17,21]. To sum up, various researchers have studied scalable mining of graph patterns, with much success.…”

Recent Advances and Trends in Large-Scale Kernel Methods

“…For example, an ε-insensitive loss function is adopted in the objective function [29], a smaller Gram matrix via a random selection of samples is used [30], and a fixed threshold on α is introduced to force many components to have the zero values in a graph regression task [31].…”

Fast Iterative Mining Using Sparsity-Inducing Loss Functions

“…More recently, substructure boosting approach has been successfully applied to different learning tasks on various kinds of data including RNA secondary structure clustering [9], video classification [10], and QSAR [11], [12]. These methods combine statistical learning algorithms with pattern mining algorithms to directly mine discriminative patterns which are optimal for the subsequent learning task in an iterative fashion [13].…”