Frequent Subgraph Mining Algorithms in Static and Temporal Graph-Transaction Settings: A Survey

Subgraph mining in a large graph: A review

“…After determining that a candidate subgraph is a frequent subgraph, in order to further determine that a frequent subgraph is a closed subgraph (Takigawa & Mamitsuka, 2011; Yan & Han, 2003) the mining process needs to compare it with mined frequent subgraphs to check for the existence of a proper supergraph whose support is equal to its support. Some approaches for mining closed frequent subgraphs (Bifet et al, 2011; Jazayeri & Yang, 2021; Nguyen, Nguyen, et al, 2021; Takigawa & Mamitsuka, 2011) reduce the amount of computation time, and this effect can be more pronounced in some practical applications.…”

“…Almost all the popular approaches for FSM are computationally expensive and require two phases (Elseidy et al, 2014; Yan & Han, 2002): Generating phase : The mining process generates all available candidate subgraphs, in that frequent subgraphs with size k will generate candidate subgraphs with size ( k + 1); Testing phase : Checking and counting the number of isomorphisms of each candidate subgraph to determine whether that candidate subgraph is frequent or not. However, subgraph isomorphism processing (Ansari & Abulaish, 2021) is a well‐known NP‐complete problem (Elseidy et al, 2014; Nguyen et al, 2020; Yan & Han, 2002), and therefore the computational cost of this phase is significantly high. There are some new survey articles on the problem of FSM (Fournier‐Viger et al, 2020; Goyal & Ferrara, 2018; Jazayeri & Yang, 2021; Jiang et al, 2013; Ramraj & Prabhakar, 2015; Zeng et al, 2020), and these papers often focus on analysis and comparing the approaches and performance of the current algorithms. Thus, we do not try to find out or compare the differences between mining algorithms, but focus instead on the current problems of FSM as well as advanced algorithms in solving these problems.…”

Frequent Pattern Mining in Continuous-time Temporal Networks

“…In the literature of frequent subgraph mining, on the other hand, the common approach toward temporal network representation and analysis is converting the temporal dimension to a sequence of intervals and representing the continuous network as a sequence of aggregated static networks [44]. In this representation, for the range of temporal network, W, and aggregation window, agg w , the number of aggregation windows or static networks would be |seq| = W/agg w .…”

“…Frequent subgraph mining problem has attracted substantial attention in domains where the data can be represented as networks, such as in chemo-informatics [16], [17], [18], health informatics [19], [20], [21], [22], [23], public health [24], [25], [26], bioinformatics [27], [28], [29], social network analysis [30], [31], [32], computer vision [33], [34], [35], [36], [37], [38], and security [39], [40], [41], [42], [43]. The frequent subgraph mining in these discplines are either applied to a data set of small networks [44] or a data set of one large network [45]. These tasks are traditionally called networktransaction setting and motif discovery, respectively.…”

“…Besides, in some applications, the networks are added to the sequence in real-time, which creates a separate category of algorithms. For further details and a discussion of algorithms in each subcategory, refer to [44]. The algorithms can be classified based on the adopted approach for frequency computations in single static networks in the motif discovery problem.…”