Graph classification based on pattern co-occurrence

Abstract

Subgraph patterns are widely used in graph classification, but their effectiveness is often hampered by large number of patterns or lack of discrimination power among individual patterns. We introduce a novel classification method based on pattern cooccurrence to derive graph classification rules. Our method employs a pattern exploration order such that the complementary discriminative patterns are examined first. Patterns are grouped into co-occurrence rules during the pattern exploration, leading to an integrated process of pattern mining and classifier learning. By taking advantage of co-occurrence information, our method can generate strong features by assembling weak features. Unlike previous methods that invoke the pattern mining process repeatedly, our method only performs pattern mining once. In addition, our method produces a more interpretable classifier and shows better or competitive classification effectiveness in terms of accuracy and execution time.

LTS: Discriminative subgraph mining by learning from search history

“…Then, graphSig uses relatively high frequency thresholds to mine discriminative subgraph patterns from each group. COM [11] uses a heuristic subgraph exploration order to find discriminative patterns faster. It also takes into account cooccurrences of subgraph patterns to boost the discrimination power of features.…”

“…COM [11] and GAIA [12] compute a very loose estimated upper-bound by assuming a constant positive frequency and zero negative frequency of "descendant" patterns in the subgraph enumeration tree. LEAP [17] proposes the prune-by-structural-proximity strategy, which is based on an observation that subgraph patterns with similar structures tend to have similar scores.…”

“…Unless otherwise specified, we randomly select 256 outsider proteins (i.e., not members of the 16 families) as a common negative set used by all 16 protein datasets. The same datasets were used in [11] and [12]. To generate a protein graph, each graph node denotes an amino acid, whose location is represented by the location of its alpha carbon.…”

“…Each compound can be either active or inactive in a bioassay. The same datasets are used in [11]- [13], [17]. For each bioassay, we randomly select 400 active compounds as the positive set and 400 inactive : positive graph set : negative graph set F: a vector maintaining the optimal pattern for each positive graph Candidate_list: the set of subgraph patterns to be extended H: search history of the patterns visited by fast-probe 1.…”

“…We also compare fast-probe with three state-of-the-art discriminative subgraph mining algorithms, GAIA [12], COM [11] and graphSig [13], using the chemical compound datasets. The parameter settings (listed in Table III) for GAIA, COM and graphSig are the same as those used in [11], [12] and [13] for chemical datasets since we are using the same datasets.…”