Near-optimal supervised feature selection among frequent subgraphs

Abstract

Graph classification is an increasingly important step in numerous application domains, such as function prediction of molecules and proteins, computerised scene analysis, and anomaly detection in program flows.Among the various approaches proposed in the literature, graph classification based on frequent subgraphs is a popular branch: Graphs are represented as (usually binary) vectors, with components indicating whether a graph contains a particular subgraph that is frequent across the dataset.On large graphs, however, one faces the enormous problem that the number of these frequent subgraphs may grow exponentially with the size of the graphs, but only few of them possess enough discriminative power to make them useful for graph classification. Efficient and discriminative feature selection among frequent subgraphs is hence a key challenge for graph mining.In this article, we propose an approach to feature selection on frequent subgraphs, called CORK, that combines two central advantages. First, it optimizes a submodular quality criterion, which means that we can yield a near-optimal solution using greedy feature selection. Second, our submodular quality function criterion can be integrated into gSpan, the state-of-the-art tool for frequent subgraph mining, and help to prune the search space for discriminative frequent subgraphs even during frequent subgraph mining.

Output space sampling for graph patterns

“…We run the discriminatory pattern sampling algorithm with a minimum support value of 6 (20%). The leap search [34], and CORK [23] algorithms did not run on this dataset. Both failed with a segmentation fault.…”

“…Though majority of these algorithms consider the summarization of itemset patterns, [34,23] consider graph patterns. In another recent work in the graph domain, Hasan et.…”

“…[34] proposed an algorithm, which employs structural proximity and frequency descending mining to reduce the searchable portion of the candidate subgraph space to obtain a small set of subgraph patterns with higher discriminatory score. Very recently, a greedy subgraph feature selection algorithm, named CORK [23] was proposed. It is embedded in the gSpan [35] mining process and it can provide an approximation guaranty with respect to a sub-modular quality criteria.…”

k-Pattern Set Mining under Constraints

“…For instance, CBA [1] first computes all frequent itemsets (with their most frequent class label) and then induces an ordered rule-list classifier by removing redundant itemsets. Several alternative techniques (for instance, [28,30]) define measures of redundancy and ways to select only a limited number of patterns. Constructing a concise pattern set for use in classification can be seen as a form of feature selection.…”

Mining interesting subgraphs by output space sampling

“…However, their approach is prone to model overfitting as it does not provide any regularization capability. Very recently, a greedy subgraph feature selection algorithm, named CORK [64] was proposed. It is embedded in the gSpan [96] mining process and it can provide an approximation guaranty with respect to a sub-modular quality criteria.…”