Mining Attribute Evolution Rules in Dynamic Attributed Graphs

Abstract

A dynamic attributed graph is a graph that changes over time and where each vertex is described using multiple continuous attributes. Such graphs are found in numerous domains, e.g., social network analysis. Several studies have been done on discovering patterns in dynamic attributed graphs to reveal how attribute(s) change over time. However, many algorithms restrict all attribute values in a pattern to follow the same trend (e.g. increase) and the set of vertices in a pattern to be fixed, while others consider that a single vertex may influence its neighbors. As a result, these algorithms are unable to find complex patterns that show the influence of multiple vertices on many other vertices in terms of several attributes and different trends. This paper addresses this issue by proposing to discover a novel type of patterns called attribute evolution rules (AER). These rules indicate how changes of attribute values of multiple vertices may influence those of others with a high confidence. An efficient algorithm named AER-Miner is proposed to find these rules. Experiments on real data show AER-Miner is efficient and that AERs can provide interesting insights about dynamic attributed graphs.

Discovering Representative Attribute-stars via Minimum Description Length

“…Pasquier et al proposed to mine frequent trees in a forest of attributed trees [14], while Atzmueller et al designed the MinerLSD algorithm to find core subgraphs in an attributed graph [16]. Besides, algorithms were designed to find temporal patterns in dynamic attributed graphs [1], [19]. However, most graph pattern mining approaches have many parameters that users generally set by trial and error, which is time-consuming and prone to errors.…”

“…After calculating the gains, pairs with positive gains are added to candidates and the related information is recorded in rdict (line [10][11][12]. Finally, the pairs whose gains are influenced by the merge operation are updated in candidates (line [17][18][19][20][21]. It is a fact that frequencies of a-stars having partly merged leafsets are always reduced by the merge operation.…”

Towards an Efficient Approach to Manage Graph Data Evolution: Conceptual Modelling and Experimental Assessments

“…the addition and deletion of edges only [27,2] or both nodes and edges [24] over time, is the most studied evolution type. To meet more complex needs, some models include attributes of nodes [6,7] or both edges and nodes' attributes [11] to capture the temporal evolution in their values. In the previous cited works, they generally consider the attribute set of nodes or edges as fixed while it can evolve over time in real-world applications.…”

“…There is only one case where it has not yet got a pre-defined ending time: if some instances under the current state s ri u are current in the application and both connected entities' instances i h and i l do not have yet a pre-defined ending time. 7 where T i h is the valid time of an instance i h of e i and T i l is the valid time of an instance i l of e j . (T Example 1.…”

Mining Frequent Sequential Subgraph Evolutions in Dynamic Attributed Graphs

“…Existing pattern mining approaches in dynamic attributed graphs allow following sequential evolutions within an individual vertex [8] or a set of vertices [1,5,7] that occur frequently over time. None of these approaches allows finding frequent sequential evolutions for general sets of connected vertices (i.e., frequent subgraphs).…”