Community evolution mode mining system of dynamic network and mining method thereof

Abstract

The invention provides a community evolution mode mining system of a dynamic network and a mining method thereof, and relates to the technical field of data mining. According to the method, a community evolution network is constructed based on a community division result and a community evolution matching method, a meta-community concept is proposed on this basis, and an evolution cluster recognition method with a strong evolution relationship is provided for representing a complex community evolution path. And finally, an evolution sequence frequent pattern mining module processes the evolution event sequence by adopting a frequent pattern mining method so as to identify the student social network community evolution pattern. Through the technical scheme provided by the invention, a community evolution mode recognition problem can be converted into a frequent mode mining problem of an event sequence, a corresponding evolution event sequence is extracted from a complex meta-community structure, and important evolution events and sequential relationships in an evolution relationship are reserved by reducing the complexity of an evolution structure; and an analyst is helped to identify the community evolution mode in the dynamic network.

Description

Technical Field

[0001] The present invention relates to the technical field of data mining, and in particular to a community evolution pattern mining system and a mining method of a dynamic network. Background Art

[0002] Dynamic networks are widely used to describe complex systems in the real world due to their high flexibility, covering many fields such as social networks, biological networks, and communication networks, and have become the research focus of complex network disciplines. In dynamic networks, the connection relationship and topological structure between nodes will continue to change over time. This complex dynamic characteristic poses a huge challenge to studying the structural evolution process of dynamic networks. Among them, the community is a close group relationship formed between network nodes, and its structure can remain relatively stable during the unstable dynamic network evolution process. Studying the evolutionary behavior of the community helps to better understand the structural characteristics of network composition, which is crucial to understanding the evolution process of dynamic networks.

[0003] The community evolution event structure of student social networks is complex and contains temporal characteristics, which increases the difficulty of tracking the community evolution path and makes it challenging to capture the complete community evolution path.

[0004] There are two main problems with existing methods: first, the community evolution relationship is time-varying and the paths overlap, which makes it extremely difficult to identify the evolution path; second, the complexity of the community evolution path makes it difficult to summarize the evolution pattern. These problems pose challenges to the identification of the evolution structure of dynamic network communities and the analysis of evolution patterns. Therefore, how to summarize and explore the community evolution path of student social networks and help researchers understand the structural characteristics and evolution patterns in the evolution of student social network communities has become an urgent problem to be solved.

[0005] Therefore, a community evolution pattern mining system and its mining method for dynamic networks are needed to solve the above technical problems. Summary of the invention

[0006] In order to solve the above technical problems, the present invention provides a community evolution pattern mining system for dynamic networks, comprising a data preprocessing unit, a community evolution network construction unit, an evolution structure perception unit and an evolution pattern mining unit;

[0007] The data preprocessing unit includes: a dynamic network modeling module, a time window division module and a community mining module; wherein the dynamic network modeling module models the data to be processed as dynamic network data; wherein the data to be processed is a student social network data set, and the student social network data set is a collection of face-to-face contact events between individuals based on wearable sensors;

[0008] The time window division module divides the dynamic network data into several static network snapshots on time slices according to the time slice processing strategy; wherein the time slice processing strategy is to divide the network into multiple static snapshots according to time, and then treat these snapshots as different networks for analysis;

[0009] The community mining module uses the community mining algorithm Louvain to discover and extract communities for the static network snapshot on each time slice, and obtains several communities in the static network snapshot;

[0010] The community evolution network construction unit includes: a community similarity calculation module, a community matching module and an evolution network construction module;

[0011] The community similarity calculation module is used to calculate the similarity between every two communities in adjacent time slices, and the community matching module identifies the evolution events between all communities in adjacent time slices based on the GED method;

[0012] The evolutionary network construction module regards each community as a node, constructs edges between communities with evolutionary events, and uses the evolutionary time process as the direction of the edge to generate a community evolutionary network;

[0013] The evolutionary structure perception unit is used to extract the meta-community structure;

[0014] The evolutionary pattern mining unit includes an evolutionary event sequence construction module and an evolutionary sequence frequent pattern mining module; wherein the evolutionary event sequence construction module extracts the evolutionary event sequence from the meta-community structure;

[0015] The evolution sequence frequent pattern mining module uses a frequent pattern mining method to process the evolution event sequence, thereby identifying the student social network community evolution pattern.

[0016] Preferably, the dynamic network modeling module models the data to be processed as dynamic network data, including:

[0017] The dynamic network modeling module converts a dynamic network G=(G0, G1..., G n ) is defined as a sequence of n graphs; the graph is constructed as a directed weighted graph:

[0018] G t =(V t , E t , t i );

[0019] Among them, the weighted graph G t A set of vertices V t and a set of edges Composition, t i represents the i-th time step; each edge ej ∈E t Represents a system consisting of two vertices A tuple consisting of each edge is assigned a weight Vertex number|V t | is the quantity that changes over time, and V represents the quantity contained in any V t The set of global vertices for all vertices v in .

[0020] Preferably, the community mining algorithm Louvain is used to perform community discovery and extraction to obtain several communities in a static network snapshot, including:

[0021] The community mining algorithm Louvain achieves the global optimal solution by optimizing the local modularity Q, where:

[0022]

[0023] In the formula, m represents the sum of the weights of all edges in the graph; A ij Represents the adjacency matrix value between nodes i and j. When two nodes are directly connected, A ij =1, otherwise A ij =0; k i represents the degree of node i, that is, the sum of the weight values ​​of all edges connected to the node within the community; c i With c j Respectively represent the communities where nodes i and j are located in the graph; δ(c i , c j ) means that if the community to which node i is assigned is the same as the community to which node j is assigned, its δ value is 1, otherwise δ value is set to 0.

[0024] Preferably, the community matching module identifies the evolution events between all communities in adjacent time slices based on the GED method, including:

[0025] The GED method of the community matching module determines community evolution events based on the inclusion degree I(C1, C2) to evaluate whether a community is contained in another community. Let C1 and C2 be subgraphs of the corresponding network snapshot G formed by the two community structures. The calculation formula for the inclusion degree I(C1, C2) of the community subgraph C1 is as follows:

[0026]

[0027] In the formula, the first term represents the quantitative part of the inclusion degree, and the second term is the qualitative part of the inclusion degree; is the social status of vertex x in the community subgraph C1, indicating the social status of vertex x in the community subgraph C1.

[0028] Preferably, the social status of the vertex x in the community subgraph C1 can be iteratively obtained by the following formula:

[0029]

[0030] Where, SP n and SP n+1 Respectively represent the social status of member x after the nth and n+1th iterations; ε represents a fixed coefficient ranging from 0 to 1; SP0(x)=1; C represents the mapping function of the relationship strength between point y and x, that is, the weight of the edge between node y and node x.

[0031] Preferably, the evolutionary structure perception unit is used to extract the meta-community structure, including:

[0032] The evolutionary structure perception unit uses the Infomap clustering algorithm to perform fine-grained division to form an evolutionary meta-community; wherein the specific steps of the Infomap clustering algorithm are as follows:

[0033] Random walk step: The Infomap clustering algorithm starts from any node in the network and selects the next node to jump to based on the probability model;

[0034] Steps for constructing Huffman codes: construct Huffman codes for each node based on the probability model of random walks; the length of the Huffman code of a node is related to the access frequency of the node during the random walk;

[0035] Hierarchical coding step: Huffman coding is performed on the nodes in a hierarchical manner to minimize the average coding length.

[0036] Select the optimal community division step: By combining the results of minimizing the average coding length, select the optimal community division as the final result of the Infomap clustering algorithm.

[0037] Preferably, the evolution sequence frequent pattern mining module uses a frequent pattern mining method to process the evolution event sequence, thereby identifying the student social network community evolution pattern, including:

[0038] The evolutionary sequence frequent pattern mining module uses the frequent pattern mining method based on the PreiixSpan algorithm to extract the evolutionary pattern of the student social network community.

[0039] A method for mining community evolution patterns of dynamic networks, using the above-mentioned system for mining community evolution patterns of dynamic networks, comprises:

[0040] Step 1: Use the dynamic network modeling module to model the student social network dataset into dynamic network data according to the dynamic network definition;

[0041] Step 2: The time window division module divides the dynamic network data into static network snapshots on several time slices according to the time slice processing strategy;

[0042] Step 3: The community mining module uses the community mining algorithm Louvain to discover and extract communities from the static network snapshot at each time slice, thereby obtaining several communities in the static network snapshot;

[0043] Step 4: Calculate the similarity between every two communities in adjacent time slices through the community similarity calculation module to provide a basis for the next step of determining evolution events;

[0044] Step 5: Identify the evolution events between all communities in adjacent time slices using the GED-based method through the community matching module;

[0045] Step 6: Through the evolutionary network construction module, each community is regarded as a node, and edges are constructed between communities with evolutionary events. The evolutionary time process is used as the direction of the edge, thereby generating a directed graph structure with community evolutionary relationships, that is, community evolution network construction;

[0046] Step 7: Use the Infomap clustering algorithm based on the meta-community to perform community discovery on the constructed community evolution network through the evolution structure perception unit to achieve fine-grained division, thereby obtaining the meta-community structure of community evolution;

[0047] Step 8: Simplify each evolution path in each meta-community through the evolution event sequence construction module, and extract the evolution event sequence from the meta-community structure;

[0048] Step 9: Use the PrefixSpan frequent pattern mining algorithm to extract the student social network community evolution pattern from the evolution event sequence through the evolution sequence frequent pattern mining module.

[0049] Compared with the related art, the community evolution pattern mining system and mining method of dynamic network provided by the present invention have the following beneficial effects:

[0050] This paper constructs a community evolution network based on the community division results of student social networks and the community evolution matching method, proposes the concept of meta-community on this basis, and provides an evolution cluster identification method with strong evolutionary relationships to characterize the complex community evolution path of student social networks. Finally, by modeling and revealing the community evolution characteristics of student social networks from the perspective of meta-community and event sequence, the community evolution pattern of student social networks is extracted. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A schematic diagram of a community evolution pattern mining system for a dynamic network provided by the present invention;

[0052] Figure 2 A schematic diagram of a complex network model construction framework provided by the present invention;

[0053] Figure 3 A schematic diagram of the execution flow of the Louvain algorithm provided by the present invention;

[0054] Figure 4 A schematic diagram of the evolutionary structure type provided by the present invention;

[0055] Figure 5 A schematic diagram of the community evolution pattern mining process framework provided by the present invention;

[0056] Figure 6 A schematic diagram of the metacommunity structure characterization process provided by the present invention;

[0057] Figure 7 Overview view of the student social network dataset;

[0058] Figure 8 Evolving schema views for student social network dataset;

[0059] Fig. 9 A timeline view of community evolution for the student social network dataset;

[0060] Fig.10 Community distribution view for student social network dataset;

[0061] Fig.11 Evolving views for meta-communities;

[0062] Fig.12 This is the structural view of the community evolution process of the 860th node;

[0063] Fig.13 Analyze views for similar meta-community patterns;

[0064] Fig.14 Projection view for the student social network dataset. DETAILED DESCRIPTION

[0065] The present invention will be further described below in conjunction with the accompanying drawings and implementation modes.

[0066] like Figure 1 As shown, a community evolution pattern mining system for dynamic networks includes a data preprocessing unit, a community evolution network construction unit, an evolution structure perception unit and an evolution pattern mining unit; wherein,

[0067] The data preprocessing unit includes: a dynamic network modeling module, a time window division module and a community mining module; wherein the dynamic network modeling module models the data to be processed as dynamic network data according to the dynamic network definition; the time window division module divides the dynamic network data into static network snapshots on a number of time slices according to the time slice processing strategy; the community mining module uses the community mining algorithm Louvain to perform community discovery and extraction on the static network snapshot on each time slice, thereby obtaining a number of communities in the static network snapshot;

[0068] The community evolution network construction unit includes: a community similarity calculation module, a community matching module and an evolution network construction module; wherein the community similarity calculation module is used to calculate the similarity between every two communities in adjacent time slices, and the community matching module identifies the evolution events between all communities in adjacent time slices based on the GED (i.e., Group Evolution Discovery) method; the evolution network construction module regards each community as a node, constructs edges between communities with evolution events, and uses the evolution time process as the direction of the edge, thereby generating a directed graph structure with community evolution relations, i.e., community evolution network construction;

[0069] The evolutionary structure perception unit is used to extract meta-communities; wherein, based on the definition concept of the meta-community, the evolutionary structure perception unit performs community discovery on the constructed community evolution network to achieve fine-grained division, thereby obtaining the meta-community structure of community evolution; wherein, the meta-community structure is a community evolution cluster with a close evolutionary relationship, which can characterize a complex evolutionary path;

[0070] The evolutionary pattern mining unit includes an evolutionary event sequence construction module and an evolutionary sequence frequent pattern mining module; wherein the evolutionary event sequence construction module simplifies each evolutionary path within each meta-community and extracts the evolutionary event sequence from the meta-community structure; the evolutionary sequence frequent pattern mining module uses a frequent pattern mining method to process the evolutionary event sequence, thereby identifying frequent community evolutionary patterns.

[0071] It should be noted that: this embodiment constructs a community evolution network based on the community division results and the community evolution matching method, and proposes the concept of meta-community on this basis, and provides an evolution cluster identification method with strong evolutionary relationships, which is used to characterize complex community evolution paths. Finally, by modeling and revealing the evolution characteristics of dynamic network communities from the perspective of meta-community and event sequence, the evolution pattern is extracted; among them, the community evolution pattern transforms the identification problem into a frequent pattern mining problem of event sequences, extracts the corresponding evolution event sequence from the complex meta-community structure, and retains the important evolutionary events and temporal relationships in the evolutionary relationship by reducing the complexity of the evolutionary structure. Based on this, a frequent sequence mining algorithm is introduced to summarize the frequent evolutionary patterns in the community evolution network, thereby helping analysts identify community evolution patterns in dynamic networks.

[0072] As a further solution, the dynamic network modeling module converts a dynamic network G=(G0, G1..., G n ) is defined as a sequence of n graphs; the graph is constructed as a directed weighted graph G t =(V t , E t , t i ), where the weighted graph G t A set of vertices V t and a set of edges Composition, t i represents the i-th time step; each edge e j ∈E t Represents a system consisting of two vertices A tuple consisting of each edge is assigned a weight Vertex number|V t | is the quantity that changes over time, and V represents the quantity contained in any V t The set of global vertices for all vertices v in .

[0073] It should be noted that time plays a very important role in the formation of network community structure. In order to analyze the dynamic network evolution behavior, it is necessary to introduce the time dimension in the network construction process. Among them, the most common method is to capture a single network snapshot during the dynamic network evolution process. These snapshots can show the dynamic phenomena of the network in a certain period of time. By splicing multiple network snapshots, a complete dynamic evolution process can be constructed. Dynamic networks are usually composed of activity event data containing timestamps. In order to analyze dynamic network data, this application constructs subsequent time points or event data within a period of time into a series of snapshots, each of which contains an instance of a dynamic network G. t .

[0074] When creating a static network snapshot, there are different choices for the time span of each time slice. In practice, it is best to choose based on domain knowledge or logical division of data time intervals, such as weekly, daily, or hourly. The quality of time slice division affects the quality of subsequent community division. We need to choose the appropriate time interval and time length in a targeted manner, and reasonably divide the time slices based on domain knowledge and analysis needs, so as to make full use of the data to extract meaningful community structures.

[0075] As a further solution, the community mining algorithm Louvain achieves a global optimal solution by optimizing the local modularity Q; wherein:

[0076]

[0077] In the formula, m represents the sum of the weights of all edges in the graph; A ij Represents the adjacency matrix value between nodes i and j. When two nodes are directly connected, A ij =1, otherwise A ij =0; k i represents the degree of node i, that is, the sum of the weight values ​​of all edges connected to the node within the community; c i With c j Respectively represent the communities where nodes i and j are located in the graph; δ(c i , c j ) means that if the community to which node i is assigned is the same as the community to which node j is assigned, its δ value is 1, otherwise δ value is set to 0.

[0078] It should be noted that the Louvain algorithm is a method for community detection in social networks and complex networks. The algorithm divides the graph structure into multiple communities or clusters to make the connections between nodes in the community closer and the connections between communities sparser. It can quickly divide communities for networks with millions of nodes. The detection results are relatively stable and the divided communities do not overlap. It is considered to be one of the fastest and best performing algorithms.

[0079] The Louvain algorithm is a multi-round iterative process. Its core idea is to continuously improve the division of the community by optimizing the objective function Q until an optimal division method is found that maximizes the modularity Q. The execution process of the Louvain algorithm is as follows: Figure 3 shown.

[0080] As an algorithm based on multi-level optimization of modularity, the Louvain algorithm achieves the global optimal solution by optimizing the local modularity Q. Modularity is a metric to measure the quality of a network community division. The modularity index is often expressed by the difference between the real connection relationship and the random connection relationship between nodes. The greater the edge weight in the community, the greater the modularity, the closer the connection between nodes in the community, and the better the quality of community division.

[0081] In community mining, local modularity can better reflect the connection and characteristics between nodes and communities, so it can better extract the effective community structure in the network. Its value range is [-1,1]. It is generally believed that modularity between 0.3 and 0.7 is a more appropriate community division result. This algorithm is considered to be a relatively accurate and efficient community mining algorithm, especially suitable for mining communities in large dynamic networks. It can be used as one of the basic algorithms for subsequent community evolution analysis. In addition, this algorithm does not require manual definition of the number of clusters, but achieves the global optimal solution by continuously iteratively updating the community to which the node belongs and the connection relationship between communities. It is considered to be one of the best performing community discovery algorithms. Therefore, we choose the Louvain algorithm to discover the basic community structure in network community evolution analysis.

[0082] This application divides the dynamic network into different time slices and uses traditional static community detection methods to extract communities. After the time slices are divided, the Louvain algorithm is used to mine communities for each snapshot of each time slice. After the community discovery algorithm is executed, a unique number is set for the community in each time slice, and the nodes in each time slice are assigned to different communities. However, over time, the structure of the community may change, and there may be a mismatch between the number of communities and community numbers in different time slices. Therefore, to analyze the dynamic evolution of communities over time, it is necessary to establish associations between adjacent time slice communities and track and identify the evolutionary behavior of these communities.

[0083] As a further solution, the GED method of the community matching module determines community evolution events based on the inclusion degree I(C1, C2) to evaluate whether one community is contained in another community; where C1 and C2 are subgraphs of the corresponding network snapshot G formed by two community structures, and the calculation formula of the inclusion degree I(C1, C2) of the community subgraph C1 is as follows:

[0084]

[0085] In the formula, the first term represents the quantitative part of the inclusion degree, and the second term is the qualitative part of the inclusion degree; is the social status of vertex x in the community subgraph C1, indicating the social status of vertex x in the community subgraph C1.

[0086] It should be noted that: as time goes by, the constant changes in nodes and edges in the dynamic network cause the members, structure and size of the community to change accordingly. Therefore, the community mining results on a certain time slice can only reflect the status of the community in a specific period of time, and cannot fully reflect the changes in the entire network evolution process. This pattern may change. Many real networks in life have a more obvious community structure, but such a community structure does not contain a time dimension, so we need to introduce a dynamic network with time information. This application applies the community evolution method to capture changes in community patterns.

[0087] In order to understand the evolution of the community, researchers describe the characteristics of its evolution by describing the life cycle of a specific community, that is, a series of events that the community experiences over time. This method determines the type of community event by finding the corresponding community in adjacent time slices and comparing it with the corresponding community. These events represent the evolution of the community over time in the entire network, which can help us better track and identify the structural changes and key evolution processes of the community. The community evolution discovery method (Group Evolution Discovery, GED) is currently the most commonly used strategy for tracking the evolution of network communities, which takes into account changes in the number and importance of community members. Therefore, this application chooses to use the GED model to calculate the similarity between communities in two adjacent time slices to determine community evolution events.

[0088] Specifically, the GED method proposes a metric method for matching community correlation, called inclusion degree I. This method judges community evolution events based on inclusion degree I (C1, C2), which evaluates whether one community is contained in another community.

[0089] As a further solution, the social status of the vertex x in the community subgraph C1 can be iterated by the following formula:

[0090]

[0091] Where, SP n and SP n+1 Respectively represent the social status of member x after the nth and n+1th iterations; ε represents a fixed coefficient ranging from 0 to 1; SP0(x)=1; C represents the mapping function of the relationship strength between point y and x, that is, the weight of the edge between node y and node x.

[0092] It should be noted that the GED method has a comprehensive classification of community evolution types, covering almost all the types of community evolution involved, including generation, persistence, growth, contraction, merger, splitting and disappearance. These types reflect the various states and changes that a community may experience during its evolution, providing researchers with a comprehensive perspective to understand the evolution of the community. The types of evolutionary structures it contains are as follows: Figure 4 shown.

[0093] exist Figure 4 In the data, the Forming state indicates a newly generated community, which means that the community appears for the first time, appears in the first time slice or there is no corresponding community in the previous time slice; Continuing indicates that the community is in a continuous state, and its structure remains relatively stable for a period of time with little change; Growing indicates that the community is expanding, which means that new nodes are joining the community and the community size is getting bigger; Shrinking indicates that the community is shrinking, which means that some nodes are leaving the community and the community size is getting smaller; Merging indicates that the community is in a merging state, which means that two or more independent communities are merging into a larger community; Splitting indicates that the community is splitting, which is manifested as a large community splitting into two or more smaller communities; Dissolving indicates that the community is in a disappearing state, that is, the community that exists in the current time slice does not exist in the next time slice. These different community evolution states reflect the dynamic changes of community structure and are of great significance for understanding the community evolution process and pattern.

[0094] GED defines seven types of evolution events. The specific evolution event determination methods are as follows:

[0095] (1) Forming (generating events): I(C1, C2) < 10% and I(C2, C1) < 10%, and C2 and all communities in the previous time slice meet this condition.

[0096] (2) Continuing (continuous event): I(C1, C2) ≥ α and I(C2, C1) ≥ β and |C1| = |C2|.

[0097] (3) Growing event: I(C1, C2) ≥ α and I(C2, C1) ≥ β and |C1| < |C2|, or I(C1, C2) ≥ α and I(C2, C1) < β and |C1| ≤ |C2|, and C2 can only find one community to match it in the previous time slice.

[0098] (4) Shrinking event: I(C1, C2) ≥ α and I(C2, C1) ≥ β and |C1| ≥ |C2|, or I(C1, C2) < α and I(C2, C1) ≥ β and |C1| ≥ |C2|, and C2 can only find one community to match it in the previous time slice.

[0099] (5) Merging event: I(C1, C2) ≥ α and I(C2, C1) < β and |C1| ≤ |C2|, and C2 can find multiple communities matching it in the previous time slice.

[0100] (6) Splitting: I(C1, C2) < α and I(C2, C1) ≥ β and |C1| ≥ |C2| are satisfied, and C1 can find multiple communities to match it in the next time slice.

[0101] (7) Dissolving (disappearance event): I(C1, C2) < 10% and I(C2, C1) < 10%, and C1 and all communities in the next time slice meet this condition.

[0102] In the above description, C1 and C2 represent two communities in different time slices, respectively, |C1| represents the number of nodes in community C1, and the value ranges of parameters α and β are not fixed. In a specific embodiment, it is recommended to set the value range within the range of [50%, 100%].

[0103] In the process of dynamic network evolution, the evolutionary path of the community has branching and merging structures. For example, in a dynamic network, a community can split into two or more communities, or multiple communities can merge into one community. The complete complex path of community evolution is composed of the evolutionary history paths of each community. Since the evolutionary classification model provides each community with the evolutionary relationship in adjacent time slices, we can trace the adjacent evolutionary relationship of each community and splice the communities with evolutionary relationships in each time slice in chronological order to obtain the complex evolutionary path of the community structure. However, due to the numerous associations between communities, if a single complex evolutionary path of a community is analyzed separately, the branches and the historical evolution of some communities in the branches will be ignored, resulting in the neglect of important evolutionary relationships.

[0104] To solve this problem, this application proposes a community evolution complex network model to describe the evolutionary relationship of all communities in a dynamic network, and gives the relevant definition of the network model, as described below.

[0105] Definition 1 Community evolution network model: A complex network model based on a directed weighted graph structure, used to describe the evolutionary relationship between all communities in a dynamic network. The construction method is shown in the following formula:

[0106] G N =(V N , E N , W)

[0107] Among them, V N represents the community in each time slice as a set of nodes; E N A set of directed edges representing the formation of association relationships (i.e., evolutionary events) between communities in two adjacent time slices, where the direction represents the time order; in addition, each edge is assigned a weight W, calculated as follows, which represents the number of common nodes in the community structure with association relationships between two adjacent time slices.

[0108] W=|C1∩C2|

[0109] Based on the above definition, we can construct the community evolution network required by the dynamic network dataset through the data preprocessing process. The complex network model construction framework is as follows: Figure 2 The specific framework process is described as follows.

[0110] Time slice division. Divide the dynamic network into continuous time slices according to specific time intervals. Figure 2 As shown, a long time period is divided into n+1 consecutive time slices t0, t1, ..., t n .

[0111] Community detection: In each time slice, the Louvain community mining algorithm is used to detect the community structure on each time slice and obtain the set of communities.

[0112] Identify community evolution events. Identify community evolution events between all communities in adjacent time slices, such as persistence, contraction, growth, and splitting.

[0113] Constructing a community evolution network. This application applies complex network analysis methods to community evolution networks, constructs community evolution networks based on community evolution events, and thus quantifies the dynamic community evolution process.

[0114] According to the above steps, we can intuitively observe the association relationship in the community evolution process by obtaining a community evolution network model. The nodes store the information of a single community, the edges describe the evolutionary relationship between communities, and each edge is assigned a weight, which represents the number of common nodes between two adjacent communities. In community evolution, two independent communities may merge into one community, or a community may split into several smaller communities. These interactions can be described as the in-degree or out-degree of a node in a directed graph. Using appropriate visual encoding (for example, using the width of the edge to represent the number of common nodes) helps improve our perception of the dynamic network evolution process.

[0115] Community evolution network: It is represented as a directed weighted graph of the constructed community evolution path, namely the community evolution network. In this network, each node represents a community, and the directed edge represents the evolutionary relationship from one community to another. The weight on the edge reflects the strength or relevance of this evolutionary relationship.

[0116] Meta-community: It is a community evolution unit with certain stability and internal connection obtained through community mining in the community evolution network. This special community structure is composed of a group of community structures with close evolutionary relationships. These related communities may experience different events such as branching and merging during the community evolution process and are distributed in several continuous time slices. The division of meta-communities can reveal the internal structure and change trend of the community and help depict the dynamic interaction relationship and evolution pattern between communities.

[0117] For example, in the air pollution transmission network dataset, the meta-community structure can be used to represent the evolution of community clusters with dense pollution transmission within a specific range. Through meta-community analysis, we can capture the cluster relationship of community evolution over a period of time and gain a deeper understanding of the dense transmission patterns and evolution trends of air pollution in a specific time and region.

[0118] Frequent evolution patterns: from multiple event fragment sequences {E0, E1, ..., E k} extracted from the frequent event fragments or event combinations P = [p0, p1, ..., p k ]. By analyzing and mining multiple event fragment sequences, we can determine which event fragments often appear at the same time, thereby finding these frequently occurring patterns.

[0119] In this application, the key to analyzing the evolution of dynamic networks is to explore the evolutionary laws of dynamic communities. As time goes by, the overall structure of the network continues to change. Some communities will merge into large communities, and the old communities will disappear or split. This dynamic change characteristic increases the complexity of pattern mining. Due to differences in community structures, node attributes, and behavioral characteristics, the evolutionary behaviors of different communities vary greatly. Therefore, an effective model is needed to extract frequent evolutionary patterns from complex and dynamic community evolution networks.

[0120] In order to help analysts better identify and analyze the community evolution process, this application adopts a community evolution network model to describe the evolutionary relationship of communities in dynamic networks. However, this evolutionary network contains a large number of nodes and directed edges, and it is difficult for analysts to quickly and effectively discover frequent and meaningful patterns in the graph. In addition, the dynamic community evolution path has a complex topological structure. How to obtain a simplified representation of the path in an orderly and scalable way so that analysts can easily identify potential patterns through this representation is still an important and challenging task.

[0121] This application proposes a new process framework for mining community evolution patterns, which aims to address the challenges of mining evolution patterns in dynamic communities. The process is designed to effectively capture the evolution patterns of dynamic communities. The entire process framework summarizes frequent patterns from the dynamic community evolution network. Its structure is as follows: Figure 5 As shown. First, we use the constructed community evolution network as input and apply the Infomap clustering algorithm for fine-grained division to form evolutionary meta-communities. Then, the evolutionary path within each meta-community is simplified and converted into a representation of an event sequence. Finally, the frequent pattern mining method PreiixSpan algorithm is used to extract frequent evolutionary patterns. The introduction of this application process framework will help deepen the understanding of dynamic community evolutionary pattern mining.

[0122] As a further solution, the evolutionary structure perception unit uses the Infomap clustering algorithm to perform fine-grained division to form an evolutionary meta-community; wherein the specific steps of the Infomap clustering algorithm are as follows:

[0123] Random walk step: The Infomap clustering algorithm starts from any node in the network and selects the next node to jump to based on the probability model;

[0124] Steps for constructing Huffman codes: construct Huffman codes for each node based on the probability model of random walks; the length of the Huffman code of a node is related to the access frequency of the node during the random walk;

[0125] Hierarchical coding step: Huffman coding is performed on the nodes in a hierarchical manner to minimize the average coding length.

[0126] Select the optimal community division step: By combining the results of minimizing the average coding length, select the optimal community division as the final result of the Infomap clustering algorithm.

[0127] It should be noted that: In order to solve the problem that the community evolution path is complex and contains temporal relationships, which makes it difficult to effectively identify, we conducted research based on the community evolution network and proposed the concept of meta-community. The community evolution network is composed of multiple independent sub-evolution networks and is not a connected graph. Community mining is performed on the dynamic community evolution network and divided into independent community structures. We define this structure as a meta-community. The internal connections of the meta-community are close, but the connections with the outside are relatively few. Therefore, the community evolution relationship in the same meta-community is more significant. Based on this feature, in-depth exploration of the community evolution behavior within a single meta-community and the community evolution pattern between multiple meta-communities will help to conduct a more detailed evolutionary sub-network analysis, thereby revealing more sophisticated evolutionary laws, which is of great significance for understanding the evolutionary characteristics of the community.

[0128] The community evolution network is a directed graph structure with temporal relationships, so this application applies the research method of directed graphs to analyze the community evolution network. Considering the powerful ability and high efficiency of Infomap in processing network structures with a large number of vertices and edges, we use the Infomap algorithm based on minimum information entropy to segment the entire evolution network into community structure units with strong spatial interactions. The community structure unit in the community evolution network is called a meta-community. Therefore, this application chooses the Infomap algorithm based on minimum information entropy to explore the community evolution network, and introduces the community evolution meta-community construction algorithm to mine the meta-community. The specific algorithm flow is given in Table 1.

[0129] Table 1 Community evolution meta-community construction algorithm

[0130]

[0131] The core idea of ​​the Infomap algorithm is to minimize the expected distance of information flow in the entire network. It regards nodes as random walkers and transfers them to the edges of the network in a probabilistic manner. The algorithm will iterate this process repeatedly until it finally converges. In each iteration, Infomap will evaluate the modularity of the current partition and try to optimize the partition result by redirecting the walk path. Suppose there is a community partitioning method that divides the network G into M communities. Based on this, Infomap proposes a metric formula L(M) to measure the average encoding length of random walks, which can relatively concisely describe the path of random walks in the network. The specific formula is shown below.

[0132]

[0133] The first term of the above formula represents the average byte length required for random walks to transfer between communities, and the second term represents the average byte length required for walks within a community. The symbol represents the proportion of the encoded community number; H(F) represents the average weighted byte length required to encode the community number; represents the proportion of codes of all nodes (including nodes that jump out) belonging to a specific community i; H(p i ) represents the average weighted byte length required to encode all nodes in community i, p i represents all nodes belonging to community i.

[0134] A meta-community represents a community cluster with a strong evolutionary relationship over a period of time. Based on this, this application proposes a new indicator to describe the duration of the evolution of a meta-community, namely, the meta-community life cycle l. The size of the meta-community life cycle l can be calculated using the following formula.

[0135] l=|t n -t m |·|s|,t n >t m

[0136] The size of the meta-community life cycle l represents the time span experienced by the current meta-community structure, which includes the various stages and changes that the community structure in the meta-community has experienced in its development process, including birth, growth, maturity and extinction. The larger the life cycle of a meta-community, the longer it has existed and the more development stages and changes it has experienced. n It is represented as the time slice of the last community in the meta-community, t m It is represented as the time slice where the first community in the metacommunity is located.

[0137] The life cycle of a metacommunity can measure the stability and persistence of the metacommunity structure. Metacommunities with longer life cycles may have stronger organizational structures. By analyzing the life cycle sizes of different metacommunities, we can compare the stability and persistence of different metacommunity organizations, thereby evaluating their importance and influence in dynamic networks. This also helps researchers gain a deeper understanding of the dynamic changes in social networks and organizations.

[0138] The types of metacommunity evolutionary events mark structural changes in the complex pathways of metacommunity evolution. Figure 6A typical metacommunity path is shown in , which includes branches formed by splitting and merging processes. In order to more accurately describe the complex structure of the community evolution path, this application uses event sequences to simplify the complex structure of the evolutionary metacommunity. By converting the event sequence, the structural characteristics of the evolutionary path are summarized into different event types, and each event reflects the structural evolution process of the evolutionary path at each stage. This transformation converts the community evolution network into an event sequence, greatly simplifies the complexity of the metacommunity structure, and retains the structural type information that is beneficial to feature recognition and pattern summarization, thus providing convenience for mining community evolution patterns.

[0139] Figure 6 The figure shows a meta-community structure characterization process. Figure 6 -a, Figure 6 -b and Figure 6 -c respectively represent the evolution paths between different communities in a meta-community. Since the structure of the community evolution path is complex and contains time series characteristics, we can use the event sequence represented by the symbol to describe the community evolution process. Therefore, we can transform a complex evolution path into a series of symbols (evolution event symbols), that is, an evolution event sequence. Each symbol represents a specific behavior or stage in its evolution process. For example, Figure 6 -a, Figure 6 -b and Figure 6 The three complex paths in -c can be represented by a series of evolutionary event symbols "5-5-5-2", "5-2-5-2" and "5-2" respectively. Therefore, we transform the extraction of community evolution patterns into a frequent pattern mining problem of event sequences, completing the transformation process from complex evolutionary structures to event sequences.

[0140] As a further solution, the evolution sequence frequent pattern mining module adopts a frequent pattern mining method based on the PreiixSpan algorithm to extract frequent evolution patterns.

[0141] It should be noted that sequence pattern mining, as an important research direction of data mining, aims to detect ordered patterns that frequently appear in sequence databases. This technology has practical application value in many fields, such as behavioral pattern analysis, disease progression pattern, and DNA sequence analysis. As a prefix-based sequence pattern mining method, the core concept of the PreiixSpan algorithm is to discover patterns based on prefixes. There are several key concepts involved in the mining process, which will be explained in detail in the following sections.

[0142] Prefix: For two given sequences Q = <a1,a2,a3,...,a n > and W = <b1,b2,b3,...,b n>, if there is a way to sort integers 1 <k1<k2<...<k n <n, so that Then Q is called the prefix of W.

[0143] Projection: For a given sequence Q, W and Y, and satisfying If sequence Y is a prefix of sequence W, and there is no subsequence larger than W that satisfies the above conditions, then sequence W is said to be a projection of prefix Y relative to sequence Q.

[0144] Frequent sequence: In the community evolution sequence data set, if a positive integer k is set as the minimum support threshold, if the frequency of occurrence when accessing sequence q is greater than or equal to the minimum support threshold k, then sequence q is called a frequent sequence.

[0145] Projection database: The set of sequences with prefix x is called the x-projection database. If sequence x is a frequent sequence pattern in sequence database D, then the set of all sequences with prefix x in sequence database D is called the projection database of x.

[0146] The PrefixSpan algorithm is an effective sequential pattern mining algorithm, and its core design concept is divide and conquer. The algorithm first divides the original database into multiple smaller projection databases, and then mines sequential patterns on each projection database. During the mining process, the algorithm first scans the original database to identify frequent prefixes of length 1, and then searches the corresponding projection database for each frequent prefix to obtain frequent sequences of length 1. Next, it mines the frequent sequences corresponding to prefixes of length 2 in a recursive manner, and then, by analogy, gradually mines the frequent sequences corresponding to longer prefixes until no longer prefixes can be mined.

[0147] Table 2 Sequence database D

[0148]

[0149] For example, assuming that the minimum support threshold is 2, the PrefixSpan algorithm is used to mine based on database D (shown in Table 2). The specific mining results are shown in Table 3, which shows the prefix, projection database and corresponding sequence pattern data.

[0150] Table 3 Projection database and sequence pattern

[0151]

[0152] The PrefixSpan algorithm can reveal local evolutionary patterns in the process of community evolution, regardless of whether they overlap with each other. It meets the needs of mining community evolution patterns and can mine frequent subsequences from community evolution event sequences. The PrefixSpan-based frequent pattern mining algorithm used in this application mainly includes the following steps.

[0153] (1) First, we traverse all community evolution event sequence data in the meta-community event sequence dataset and manually set the minimum support threshold. During the traversal process, we search for all frequent evolution pattern item sets with a length of 1.

[0154] (2) Secondly, based on the sequence patterns in the frequent evolving pattern itemsets as prefixes, we partition the original dataset and form the corresponding projected datasets.

[0155] (3) Finally, we adopt a recursive method to repeat step (1) and traverse each projection data set. This process continues until we have mined frequent community evolution subsequences of all community evolution projection sets.

[0156] According to the above algorithm steps, we obtained multiple frequently occurring event sequence fragments, which are represented as the community evolution patterns P mined in the dynamic community evolution event sequence set E. Each pattern contains a support value, which indicates the number of times the current pattern appears in different meta-communities. Among them, the pattern with higher support indicates that the evolution pattern appears more frequently in the event sequence, so we believe that it is a more frequent evolution pattern in the current dataset; on the contrary, the fragment with low support indicates that the evolution pattern appears less frequently. The evolution pattern obtained by the above method can well summarize the evolutionary structure in the dynamic community evolution and reflect the common evolutionary patterns in the dynamic community evolution event sequence.

[0157] Analysis of the evolution of student social network communities

[0158] (1) Overview of community evolution model

[0159] Analysts adopted a step-by-step in-depth visual analysis strategy. They first mined the communities in the dataset to get an overview of the distribution of the community structure, thereby gaining an overall perception of the entire student social network evolution data. The system then allowed analysts to track, identify and analyze the community structure from multiple angles, from overview to detail, to explore the evolutionary relationships and paths between communities. Finally, they gained an in-depth understanding of the node evolution behavior within the community evolution model from the node level.

[0160] In order to summarize the evolution pattern of student social networks, parameters are set in the control panel, the Louvain algorithm is used to extract the community structure, and the PCA method is used for dimensionality reduction. After selecting the algorithm, click the run button to execute the operation. After the calculation is completed, the dynamic network data analysis module presents the community mining results, and the system summarizes the frequent evolution patterns.

[0161] like Figure 7 Student social network dataset network data overview view,First, the data distribution of the dataset can be obtained in the student social network dataset network data overview view. From the figure, we can see that the current dataset contains 180 nodes and 34,258 edges, where nodes represent students and edges represent contact events between students. From the line graph, we can see that the number of edges in this dataset is distributed more regularly over time. From the bar graph representing the community mining results, we find that the small community structure accounts for the highest proportion of this dataset, among which most communities are mainly composed of 1-5 nodes, and other scale distributions account for a small proportion. At the same time, there are also some large community structures with more than 20 people.

[0162] like Figure 8 The evolution pattern view of the student social network dataset shown in the figure gives an overview of the dataset from the perspective of frequent patterns of community evolution, and the system summarizes a total of 17 frequent patterns of community evolution. By analyzing the frequent evolution pattern view of the student social network dataset, we found that most of the evolution sequences are composed of light gray "1" type merging structures and dark gray "2" type splitting structures. This shows that most of the student social groups in the dataset often experience merger and splitting events, and may first experience splitting to form new independent communities, and then these independent communities will experience merger events within a certain period of time, and then merge into a larger organizational structure.

[0163] like Fig. 9 The community evolution timeline view of the student social network dataset is shown in the figure. In order to further explore the evolution of the community structure in the dataset over time, we use the evolution timeline view to observe the distribution of community evolution events in the dataset. This figure intuitively presents the changing trend of the distribution of community evolution events over time. After observing the view, we use the gray rectangular box to filter the abnormal time period to be explored. For example, from the figure we find that in the evolution process from the 32nd to the 33rd time slice on the second day, the event distribution is single and all are merging evolution events.

[0164] like Fig.10Community distribution view of the student social network dataset. We further explored the community distribution view and found that the number of community structures contained in the 32nd time slice was large, and all communities flowed out to the next time slice. Through the floating prompt box, we can verify that the communities in the current window are all small communities with less than 10 nodes; while in the next time slice, the number of communities is small, there are a few large communities, and the number of communities flowing into the current window is large. This also proves that during this time period, all communities have merged events, and multiple small community structures have merged into several large communities.

[0165] The community distribution view only provides an overall overview of the community, but cannot explore changes in the internal structure of the community. We want to further understand how large communities are formed. Therefore, in the community distribution view, we click on a specific community, and its meta-community structure is highlighted in the view. We select the community numbered 261, which has 22 community nodes. From the capsule glyph, we can see that this community has only inflows but no outflows. Fig.11 The meta-community evolution view shows the meta-community evolution from 12:00 to 15:00. The meta-community started with two communities. As time went on, the community structure began to split, and the number of communities gradually increased, forming multiple small communities and merging into the currently selected large community in the later stage. From the node link diagram, we can see that the community nodes have the same color, and they are all students in the same class. Therefore, we speculate that the class has public courses or collective activities during the 15:00 period, which brings scattered students together.

[0166] In addition, our visual analysis system also provides further exploration of the detailed structure of node evolution within the community. Through the node link diagram, we can observe an important node in the current community structure. This node has edges with multiple other nodes, indicating that the student has had contact with these students. After selecting this node, the system will highlight the node in all community structures to better show the community situation in different time slices. According to the time sequence of community evolution, we can observe that in Fig.12 In the structural view of the community evolution process where the 860th node is located, as time goes by, the student always remains at the core of the community structure and has close connections with other classmates. Therefore, the student can be regarded as a core figure in the community evolution process.

[0167] Later, we became interested in another community structure in the 33rd time slice and clicked on it to explore further. Fig.13As shown in the figure, through visual analysis of the meta-community structure, we found that there are also a large number of merger and split events in the current meta-community structure. It is particularly noteworthy that many community merger events occurred between 2 pm and 3 pm. The currently selected community structure (enlarged circular node) is formed by the merger of multiple small communities, and is split into several smaller communities again in the following time slice. This finding prompted us to further explore the possible reasons for the gathering of students during this time period. From the perspective of social networks, 3 pm may be the time for students to participate in club activities, extracurricular activities or discussion groups, which may prompt students to gather on a large scale at a specific time to form a large community structure.

[0168] After conducting an in-depth study of the above data, we Fig.14 In the projection view, we found that the structures of the current two metacommunities are similar, and most of the evolutionary event sequences are composed of splitting and merging events, and the composition is relatively similar. Therefore, we are interested in more complex evolutionary patterns, especially those with similar evolutionary patterns to the current metacommunity. By observing the projection view, we found that the node distances of the current two communities in the metacommunity projection view are very close, which indicates that they have similar evolutionary structures. Therefore, it is proved that the two metacommunities have similar evolutionary structures, and this law is further verified by the projection view.

[0169] Comprehensive research results show that 3 pm is the time when community splits frequently occur in student social networks, which may be due to the concentrated time when students participate in campus club activities, extracurricular gatherings or discussion groups. In-depth research on the evolutionary pattern of community structure in student social networks helps to reveal the underlying laws and patterns in social networks. By analyzing the formation, development and disappearance of different communities, we can gain insight into the evolutionary trend of students' social behavior and better understand their behavior patterns and interaction methods in social networks. By integrating the application of visual analysis systems, we can intuitively explore the evolutionary paths of various communities in social networks and clarify their changing laws and trends. This not only provides analysts with an effective means of mining social network data, but also provides decision makers with a more scientific and effective basis for decision-making.

Claims

1. A community evolution pattern mining system for dynamic networks, characterized by: It includes data preprocessing unit, community evolution network construction unit, evolution structure perception unit and evolution pattern mining unit; The data preprocessing unit includes: a dynamic network modeling module, a time window division module and a community mining module; wherein the dynamic network modeling module models the data to be processed as dynamic network data; wherein the data to be processed is a student social network data set, and the student social network data set is a collection of face-to-face contact events between individuals based on wearable sensors; The time window division module divides the dynamic network data into several static network snapshots on time slices according to the time slice processing strategy; wherein the time slice processing strategy is to divide the network into multiple static snapshots according to time, and then treat these snapshots as different networks for analysis; The community mining module uses the community mining algorithm Louvain to discover and extract communities for the static network snapshot on each time slice, and obtains several communities in the static network snapshot; The community evolution network construction unit includes: a community similarity calculation module, a community matching module and an evolution network construction module; The community similarity calculation module is used to calculate the similarity between every two communities in adjacent time slices, and the community matching module identifies the evolution events between all communities in adjacent time slices based on the GED method; The evolutionary network construction module regards each community as a node, constructs edges between communities with evolutionary events, and uses the evolutionary time process as the direction of the edge to generate a community evolutionary network; The evolutionary structure perception unit is used to extract the meta-community structure; The evolutionary pattern mining unit includes an evolutionary event sequence construction module and an evolutionary sequence frequent pattern mining module; wherein the evolutionary event sequence construction module extracts the evolutionary event sequence from the meta-community structure; The evolution sequence frequent pattern mining module uses a frequent pattern mining method to process the evolution event sequence, thereby identifying the student social network community evolution pattern.

2. According to claim 1, a community evolution pattern mining system for dynamic networks is characterized in that: The dynamic network modeling module models the data to be processed as dynamic network data, including: The dynamic network modeling module converts a dynamic network G=(G0, G1..., G n ) is defined as a sequence of n graphs; the graph is constructed as a directed weighted graph: G t =(V t ,E t ,t i ); Among them, the weighted graph G t A set of vertices V t and a set of edges Composition, t i represents the i-th time step; each edge e j ∈E t Represents a system consisting of two vertices A tuple consisting of each edge is assigned a weight Vertex number|V t | is the quantity that changes over time, and V represents the quantity contained in any V t The set of global vertices for all vertices v in .

3. The community evolution pattern mining system of a dynamic network according to claim 1, characterized in that: The community mining algorithm Louvain is used to discover and extract communities, and several communities in the static network snapshot are obtained, including: The community mining algorithm Louvain achieves the global optimal solution by optimizing the local modularity Q, where: In the formula, m represents the sum of the weights of all edges in the graph; A ij Represents the adjacency matrix value between nodes i and j. When two nodes are directly connected, A ij =1, otherwise A ij =0; k i represents the degree of node i, that is, the sum of the weight values ​​of all edges connected to the node within the community; c i With c j Respectively represent the communities where nodes i and j are located in the graph; δ(c i ,c j ) means that if the community to which node i is assigned is the same as the community to which node j is assigned, its δ value is 1, otherwise δ value is set to 0.

4. The community evolution pattern mining system for dynamic networks according to claim 1, characterized in that: The community matching module identifies the evolution events between all communities in adjacent time slices based on the GED method, including: The GED method of the community matching module determines community evolution events based on the inclusion degree I(C1, C2) to evaluate whether a community is contained in another community. Let C1 and C2 be subgraphs of the corresponding network snapshot G formed by the two community structures. The calculation formula for the inclusion degree I(C1, C2) of the community subgraph C1 is as follows: In the formula, the first term represents the quantitative part of the inclusion degree, and the second term is the qualitative part of the inclusion degree; is the social status of vertex x in the community subgraph C1, indicating the social status of vertex x in the community subgraph C1.

5. The community evolution pattern mining system of a dynamic network according to claim 4, characterized in that: The social status of the vertex x in the community subgraph C1 can be iterated by the following formula: Where, SP n and SP n+1 Respectively represent the social status of member x after the nth and n+1th iterations; ε represents a fixed coefficient ranging from 0 to 1; SP0(x)=1; C represents the mapping function of the relationship strength between point y and x, that is, the weight of the edge between node y and node x.

6. The community evolution pattern mining system for dynamic networks according to claim 1, characterized in that: The evolutionary structure-aware unit is used to extract meta-community structures, including: The evolutionary structure perception unit uses the Infomap clustering algorithm to perform fine-grained division to form an evolutionary meta-community; wherein the specific steps of the Infomap clustering algorithm are as follows: Random walk step: The Infomap clustering algorithm starts from any node in the network and selects the next node to jump to based on the probability model; Steps for constructing Huffman codes: construct Huffman codes for each node based on the probability model of random walks; the length of the Huffman code of a node is related to the access frequency of the node during the random walk; Hierarchical coding step: Huffman coding is performed on the nodes in a hierarchical manner to minimize the average coding length. Select the optimal community division step: By combining the results of minimizing the average coding length, select the optimal community division as the final result of the Infomap clustering algorithm.

7. The community evolution pattern mining system for dynamic networks according to claim 1, characterized in that: The evolution sequence frequent pattern mining module uses a frequent pattern mining method to process the evolution event sequence, thereby identifying the student social network community evolution pattern, including: The evolutionary sequence frequent pattern mining module uses a frequent pattern mining method based on the PrefixSpan algorithm to extract the evolutionary pattern of the student social network community.

8. A method for mining community evolution patterns in dynamic networks, applied to a system for mining community evolution patterns in dynamic networks as claimed in any one of claims 1 to 7, characterized in that: include: Step 1: Use the dynamic network modeling module to model the student social network dataset into dynamic network data according to the dynamic network definition; Step 2: The time window division module divides the dynamic network data into static network snapshots on several time slices according to the time slice processing strategy; Step 3: The community mining module uses the community mining algorithm Louvain to discover and extract communities from the static network snapshot at each time slice, thereby obtaining several communities in the static network snapshot; Step 4: Calculate the similarity between every two communities in adjacent time slices through the community similarity calculation module to provide a basis for the next step of determining evolution events; Step 5: Identify the evolution events between all communities in adjacent time slices using the GED-based method through the community matching module; Step 6: Through the evolutionary network construction module, each community is regarded as a node, and edges are constructed between communities with evolutionary events. The evolutionary time process is used as the direction of the edge, thereby generating a directed graph structure with community evolutionary relationships, that is, community evolution network construction; Step 7: Use the Infomap clustering algorithm based on the meta-community to perform community discovery on the constructed community evolution network through the evolution structure perception unit to achieve fine-grained division, thereby obtaining the meta-community structure of community evolution; Step 8: Simplify each evolution path in each meta-community through the evolution event sequence construction module, and extract the evolution event sequence from the meta-community structure; Step 9: Use the PrefixSpan frequent pattern mining algorithm to extract the student social network community evolution pattern from the evolution event sequence through the evolution sequence frequent pattern mining module.

Images