A dynamic community detection method and system considering the daily rhythm of crowd movement

Abstract

The present invention discloses a dynamic community detection method and system that takes into account the daily rhythm of crowd movement. First, the original mobile big data is used to generate a time series set of start-destination network snapshots according to a preset time window as a dynamic crowd movement network; then, static community detection is performed at each time interval to obtain a time series set of community structures for the time interval; then, the information of the time series set of community structures is integrated to obtain a consensus network, recording the connection strength attributes of the community member relationships; finally, the consensus community structure is detected and the time dimension is added to generate the final dynamic community structure. The present invention creatively proposes a method for integrating all crowd movement data within a day using consensus clustering technology to obtain a robust network community structure. The present invention can provide quantitative data support for the field of urban planning. The consistency of the results of the present invention is significantly better than the existing "two-stage" dynamic community detection method, and has a promising future for promotion and application.

Description

Technical Field

[0001] The present invention belongs to the technical field of urban geography and relates to a dynamic community detection method and system, and in particular to a dynamic community detection method and system based on an origin-destination (OD) network that takes into account the daily rhythm of crowd movement. Background Art

[0002] Social network analysis and mining have become a hot research area in academia and industry in recent years, with community structure research being a crucial topic. Current research on community structure often focuses on the static level: finding hidden community structures within a given network. However, due to the constant flow of people, social networks are constantly changing, and community structures evolve with them. This has led to numerous shortcomings in existing static community detection methods.

[0003] Static community detection methods can be divided into three categories based on the type of network: methods based on undirected and unweighted networks, methods based on undirected and weighted networks, and methods based on directed and weighted networks. However, these static community detection methods ignore the dynamic nature of human mobility. The daily rhythm of human activity is the basis for organizing human activities, as the human circadian clock oscillates with an endogenous 24-hour cycle. Obviously, using static methods for a specific time period (also called a snapshot) cannot fully capture the characteristics of human mobility throughout the day.

[0004] Many studies have developed dynamic community detection methods by integrating the temporal dimension of human mobility. A classic approach, the Independent Community Detection and Matching (ICDM) method, applies a classic static community detection method to detect communities at snapshot t and then matches them with communities detected in previous snapshots using a specific similarity metric. If the similarity between two subsequent communities exceeds a given threshold, they are identified as members of the same dynamic community; otherwise, they are identified as members of two independent communities. While existing methods are effective in capturing continuously evolving community structures with high similarity between two subsequent time periods, they are not well suited for revealing the community structure of human mobility networks with daily rhythms, as mobility patterns can change significantly during certain snapshots (such as morning and evening rush hours and midnight). In this case, existing dynamic community detection methods often overidentify communities due to the low similarity between the community structures detected before and after certain snapshots. Consequently, these existing methods fail to effectively capture the spatial interaction patterns between different locations throughout the day, resulting in significant bias in the detected dynamic community structure. Therefore, it is necessary to develop effective methods for detecting dynamic community structure by considering the daily rhythms of human mobility. Summary of the Invention

[0005] In order to solve the problem of low consistency of community structure and failure to consider the daily rhythm of crowd movement, the present invention proposes a dynamic community detection method and system that considers the daily rhythm of crowd movement.

[0006] The technical solution adopted by the method of the present invention is: a dynamic community detection method considering the daily rhythm of crowd movement, comprising the following steps:

[0007] Step 1: Generate a time series set of network snapshots of the origin and destination according to the preset time window using the original mobile big data G = {G 1 ,…,G t ,…,G τ}, as a dynamic crowd mobile network;

[0008] Given a time interval Δ, a day is divided into discrete time intervals {Δ,…,tΔ,…τΔ}, τ=24 / Δ; each graph G at the tth time interval t is a directed weighted graph G t =(V t ,E t ), where V t is a set of nodes, E t is the set of edges on the tth time interval; each node Represents a geographical location, each edge Indicates that from the original node To the target node Directed movement with weight attributes as the number of trips counted in the t-th time interval; where the time interval is called a snapshot;

[0009] Step 2: Perform static community detection at each time interval to obtain a time series set of community structures at each time interval;

[0010] Step 3: Based on the consensus clustering method, the information of the time series set of community structure is integrated to obtain the consensus network and record the connection strength attributes of the community member relationships;

[0011] Step 4: Apply the static community detection method on the consensus network to obtain the consensus community structure, and add the time dimension to generate the final dynamic community structure;

[0012] Step 5: Extract evolution patterns based on the dynamic community structure and apply clustering algorithms to obtain pattern classification.

[0013] The technical solution adopted by the system of the present invention is: a dynamic community detection system that considers the daily rhythm of population movement, including the following modules:

[0014] Module 1 is used to generate a time series set G of the origin-destination network snapshot according to a preset time window using the original mobile big data. 1 ,…,G t ,…,G τ}, as a dynamic crowd mobile network;

[0015] Given a time interval Δ, a day is divided into discrete time intervals {Δ,…,tΔ,…τΔ}, τ=24 / Δ; each graph G at the tth time interval t is a directed weighted graph G t =(V t ,E t ), where V t is a set of nodes, E t is the set of edges on the tth time interval; each node Represents a geographical location, each edge Indicates that from the original node To the target node Directed movement with weight attributes as the number of trips counted in the t-th time interval; where the time interval is called a snapshot;

[0016] Module 2 is used to perform static community detection at each time interval to obtain a time series set of community structures at each time interval;

[0017] Module 3 is used to integrate the information of the time series set of community structure based on the consensus clustering method to obtain the consensus network and record the connection strength attributes of the community member relationships;

[0018] Module 4 is used to apply the static community detection method on the consensus network to obtain the consensus community structure and add the time dimension to generate the final dynamic community structure.

[0019] Module 5 is used to extract evolutionary patterns on dynamic community structures and apply clustering algorithms to obtain pattern classification.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] This paper innovatively proposes a method that uses consensus clustering to integrate all human mobility data within a day to derive robust network community structures. This method can provide quantitative data support for urban planning. The consistency of the results obtained by this method is significantly superior to existing "two-stage" dynamic community detection methods, demonstrating promising applications. Compared to previous "two-stage" dynamic community methods, the proposed method possesses two unique features. First, it is robust to dynamic communities in human mobility networks that exhibit significant pattern changes at specific snapshots. Unlike "two-stage" methods, which detect dynamic communities by matching similar communities across two subsequent snapshots, the proposed method uses consensus clustering to detect dynamic communities based on the duration of strong interactions between nodes throughout the day. Thus, the proposed method extends previous methods by explicitly accounting for the daily rhythms of human activity. Second, the proposed method is able to classify the underlying evolutionary patterns of the detected dynamic communities. In "two-stage" methods, evolutionary events between static communities at different snapshots are identified to reveal dynamic urban structure. However, in the proposed method, evolutionary events are identified for subcommunities within detected dynamic communities. The number of identified evolutionary events and interaction strengths are used to classify the potential evolutionary patterns of the detected dynamic communities. Therefore, the proposed method extends previous methods and provides a new method for classifying community evolution patterns to understand dynamic urban structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0023] Figure 2 This is a schematic diagram of the second step of the embodiment of the present invention;

[0024] Figure 3 This is a schematic diagram of the third and fourth steps of the embodiment of the present invention;

[0025] Figure 4 This is a test data area diagram for an example of the present invention;

[0026] Figure 5 This is the second step of the present invention example - the static community result graph on each snapshot;

[0027] Figure 6 This is the third step of the present invention's example - consensus network result diagram;

[0028] Figure 7 This is the fourth step of the present invention - consensus community result diagram;

[0029] Figure 8 This is the fourth step of the present invention - dynamic community result graph;

[0030] Figure 9This is a hierarchical clustering example of the evolutionary characteristics of the dynamic community structure example of the present invention. DETAILED DESCRIPTION

[0031] In order to facilitate ordinary technicians in this field to understand and implement the present invention, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the implementation examples described herein are only used to illustrate and explain the present invention and are not used to limit the present invention.

[0032] Based on the consideration of data availability, this example selects N city public data sets to carry out experimental demonstration. The example data space range is as follows: Figure 4 The study area was artificially divided into 263 geographical areas.

[0033] Please see Figure 1 The present invention provides a dynamic community detection method that considers the daily rhythm of crowd movement, comprising the following steps:

[0034] Step 1: Dynamic origin-destination network (OD network for short) construction.

[0035] Given a time interval Δ, a day can be divided into discrete time intervals τ = 24 / Δ, which are {Δ,…,tΔ,…τΔ}. Therefore, the OD network can be expressed as G = {G 1 ,…,G t ,…,G τ}Time series graph. Each graph G at the tth time interval (also called snapshot) t It can be represented as a directed weighted graph G t =(V t ,E t ), where V t is a set of nodes, E t is the set of edges on the tth time interval. Each node Represents a geographical location, such as an artificially divided traffic area or grid. Each edge Indicates that from the original node To the target node Directed movement with weight attributes As the statistical number of trips in the tth time interval.

[0036] This embodiment first considers each area within the research scope as a node of the OD network. The original mobile big data provides the geographical coordinates of the user's starting point, the geographical coordinates of the end point, the departure time, and the arrival time. The corresponding OD network nodes of the starting point and the end point are determined according to the geographical coordinates. Given a time interval Δ = 0.5 hours, a day is divided into 48 time periods. In each time period, the nodes are connected with edges, and the number of trips between node i and node j is counted. Each edge Indicates that from the original node To the target node Directed movement with weight attributes As the number of trips in the tth time interval, the OD network snapshot G in the corresponding time period can be obtained t Finally, the time series of 48 OD network snapshots constitutes the dynamic OD network data G required for the experiment. 1 ,…,G t ,…,G τ}.

[0037] Step 2: Static community detection on each network snapshot.

[0038] At each time interval t, the corresponding snapshot G t Apply any static community detection method (such as InfoMap algorithm) to detect G t Community structure in It's G t The set of k communities detected in Contains a group of nodes. The spatial interaction strength between nodes within a community is greater than the interaction between nodes in different communities.

[0039] Please see Figure 2 , shows a schematic diagram of the static community detection process. At each time interval t, the snapshot G t InfoMap algorithm is applied to detect G t The community structure in It's G t The set of k communities detected in . Each community Contains a group of nodes with strong spatial interactions, while the interactions between nodes in different communities are weak. Figure 5 , showing an example result graph of static community detection.

[0040] Step 3: Consensus network construction.

[0041] For each detected community A community membership graph is constructed by connecting an undirected edge between any two nodes in the same community. After constructing all community membership graphs of all snapshots, they are projected onto geographic space to form a consensus network, denoted as G c =(V C ,E C ), G c is an undirected weighted graph, where is the union of all nodes in all communities E Cis the union of all edges in all snapshots of all community member graphs; edges The weight of It is defined as:

[0042]

[0043] in is a binary indicator, Representation node and All in the community Otherwise Representation node and All in the community The number of snapshots in , and normalized by the total number of snapshots τ.

[0044] Please see Figure 3 , the left side shows how to get the community membership graph through the static community set. For each detected community A community membership graph is constructed by connecting an undirected edge between any two nodes in the same community. After constructing all community membership graphs of all snapshots, they are projected onto geographic space to form an undirected weighted graph consensus network G c .G c The edge connecting node i and node j Indicates that two nodes belong to the same community. Weight Represents the probability that two nodes belong to the same community, calculated by dividing the number of snapshots of the two nodes belonging to the same community by the total number of snapshots. Figure 6 , showing the visualization of consensus network instances on a map.

[0045] Step 4: Consensus community structure detection.

[0046] Consensus Network G c The static community detection method is applied to detect the consensus community structure, which is recorded as For each snapshot G t , construct the same community set in and Have the same set of nodes; for the consensus community All its promoted consensus communities Building the ultimate dynamic community

[0047] Please see Figure 3 The right side shows a schematic diagram of consensus community detection. The consensus network G obtained in step 3 cThe InfoMap algorithm is applied to detect the consensus community structure, which is denoted as Please see Figure 7 , showing an example of consensus community structure.

[0048] Then, the time dimension is added to generate the final dynamic community structure, for each snapshot G t , construct the same community set, in and Have the same set of nodes. For the consensus community All its promoted consensus communities Forming the ultimate dynamic community Please see Figure 8 , showing the resulting example of dynamic community structure.

[0049] Step 5: Extract evolution patterns based on the dynamic community structure and apply clustering algorithms to obtain pattern classification.

[0050] The specific implementation includes three sub-steps:

[0051] Step 5.1: Quantify the time series of interaction strength for each dynamic community It refers to the sum of the edge weights of the community at each time interval;

[0052] Step 5.2: Detect subcommunities within each dynamic community and utilize the total number of evolution events Quantify the time series of subcommunity evolution events;

[0053] Step 5.3: Perform hierarchical clustering classification on the evolution patterns based on the time series metrics obtained in the previous two steps.

[0054] Please see Figure 9 , a hierarchical clustering example of the evolutionary characteristics of a dynamic community structure instance is presented, demonstrating the ability of the proposed method to detect dynamic communities and analyze their evolutionary patterns. Figure 9 As shown, four different evolutionary patterns are obtained, and the vertical axis represents the difference measure between the patterns:

[0055] 1) Pattern 1: The travel intensity of the dynamic communities {C2, C3, C4, and C9} located in the suburbs of City N reaches its peak during the morning rush hour, likely driven by commuting. Their subcommunity evolution patterns are also characterized by the morning and evening rush hours, with more evolution events occurring during these times than at other times.

[0056] 2) Model 2: The travel intensity of the dynamic community {C5, C6, C7} located in the E administrative district of City N reaches its peak intensity during the evening rush hour, and the number of community evolution events is less than that of Model 1.

[0057] 3) Pattern 3: The vibrant community {C1} is located in District D, the downtown area of City N. Daytime travel intensity remains high. Except for the morning and evening rush hours, where evolutionary events are rarely detected, no evolutionary events are detected at other times.

[0058] 4) Pattern 4: The dynamic community {C8} located in Administrative District B of City N reaches its peak intensity during the evening rush hour. The most evolutionary events occur from 5:30 to 6:00 in the morning.

[0059] This paper innovatively proposes a method that uses consensus clustering technology to integrate all daily population movement data to derive robust network community structures. This method can provide quantitative data support for urban planning. The consistency of the results obtained by this method significantly outperforms existing "two-stage" dynamic community detection methods, demonstrating promising prospects for widespread application.

[0060] It should be understood that the above description of the preferred embodiment is relatively detailed and cannot be regarded as limiting the scope of protection of the patent of the present invention. Under the guidance of the present invention, ordinary technicians in this field can also make substitutions or modifications without departing from the scope of protection of the claims of the present invention, which all fall within the scope of protection of the present invention. The scope of protection requested by the present invention shall be based on the attached claims.

Claims

1. A dynamic community detection method considering the daily rhythm of crowd movement, characterized by: The following steps are involved: Step 1: Generate a time series set of network snapshots of the origin and destination according to the preset time window using the original mobile big data G = {G 1 ,…,G t ,…,G τ }, as a dynamic crowd mobile network; Given a time interval Δ, a day is divided into discrete time intervals {Δ,…,tΔ,…τΔ}, τ=24 / Δ; each graph G at the tth time interval t is a directed weighted graph G t =(V t ,E t ), where V t is a set of nodes, E t is the set of edges on the tth time interval; each node Represents a geographical location, each edge Indicates that from the original node To the target node Directed movement with weight attributes as the number of trips counted in the t-th time interval; where the time interval is called a snapshot; Step 2: Perform static community detection at each time interval to obtain a time series set of community structures at each time interval; Step 3: Based on the consensus clustering method, the information of the time series set of community structure is integrated to obtain the consensus network and record the connection strength attributes of the community member relationships; For each detected community A community membership graph is constructed by connecting an undirected edge between any two nodes in the same community. After constructing all community membership graphs at all time intervals, they are projected onto geographic space to form a consensus network, denoted as G c =(V C ,E C ), G c is an undirected weighted graph, where is the union of all nodes in all communities E C is the union of all edges in all time intervals in all community member graphs; edges The weight of It is defined as: in is a binary indicator, Representation node and All in the community Otherwise Representation node and All in the community The number of time intervals in , and normalized by the total number of time intervals τ; Step 4: Apply the static community detection method on the consensus network to obtain the consensus community structure, and add the time dimension to generate the final dynamic community structure; Step 5: Extract evolution patterns based on the dynamic community structure and apply clustering algorithms to obtain pattern classification.

2. The dynamic community detection method considering the daily rhythm of human mobility according to claim 1 is characterized by: In step 2, at each time interval t, the G t Apply static community detection method to detect G t Community structure in It's G t The set of k communities detected in Contains a group of nodes; the strength of spatial interactions between nodes within a community is greater than the interactions between nodes in different communities.

3. The method for dynamic community detection considering the daily rhythm of human mobility according to any one of claims 1 to 2, characterized in that: In step 4, the consensus network G c The static community detection method is applied to detect the consensus community structure, which is recorded as For each time interval G t , construct the same community set in and Have the same set of nodes; for the consensus community All its promoted consensus communities Building the ultimate dynamic community 4. The method for dynamic community detection considering the daily rhythm of human mobility according to any one of claims 1-2, characterized in that: The specific implementation of step 5 includes three sub-steps: Step 5.1: Quantify the time series of interaction strength for each dynamic community It refers to the sum of the edge weights of the community at each time interval; Step 5.2: Detect subcommunities within each dynamic community and utilize the total number of evolution events Quantify the time series of subcommunity evolution events; Step 5.3: Perform hierarchical clustering classification on the evolution patterns based on the time series metrics obtained in the previous two steps.

5. A dynamic community detection system that considers the daily rhythm of crowd movement, characterized by: Includes the following modules: Module 1 is used to generate a time series set G of the origin-destination network snapshot according to a preset time window using the original mobile big data. 1 ,…,G t ,…,G τ }, as a dynamic crowd mobile network; Given a time interval Δ, a day is divided into discrete time intervals {Δ,…,tΔ,…τΔ}, τ=24 / Δ; each graph G at the tth time interval t is a directed weighted graph G t =(V t ,E t ), where V t is a set of nodes, E t is the set of edges on the tth time interval; each node Represents a geographical location, each edge Indicates that from the original node To the target node Directed movement with weight attributes as the number of trips counted in the t-th time interval; where the time interval is called a snapshot; Module 2 is used to perform static community detection at each time interval to obtain a time series set of community structures at each time interval; Module 3 is used to integrate the information of the time series set of community structure based on the consensus clustering method to obtain the consensus network and record the connection strength attributes of the community member relationships; For each detected community A community membership graph is constructed by connecting an undirected edge between any two nodes in the same community. After constructing all community membership graphs at all time intervals, they are projected onto geographic space to form a consensus network, denoted as G c =(V C ,E c ), G c is an undirected weighted graph, where is the union of all nodes in all communities E C is the union of all edges in all time intervals in all community member graphs; edges The weight of It is defined as: in is a binary indicator, Representation node and All in the community Otherwise Representation node and All in the community The number of time intervals in , and normalized by the total number of time intervals τ; Module 4 is used to apply the static community detection method on the consensus network to obtain the consensus community structure and add the time dimension to generate the final dynamic community structure; Module 5: Used to extract evolutionary patterns in dynamic community structures and apply clustering algorithms to obtain pattern classification.

6. The dynamic community detection system considering the daily rhythm of human mobility according to claim 5 is characterized by: In module 2, at each time interval t, the G t Apply static community detection method to detect G t Community structure in It's G t The set of k communities detected in Contains a group of nodes; the strength of spatial interactions between nodes within a community is greater than the interactions between nodes in different communities.

7. The dynamic community detection system considering the daily rhythm of human mobility according to any one of claims 5-6, characterized in that: In Module 4, the consensus network G c The static community detection method is applied to detect the consensus community structure, which is recorded as For each time interval G t , construct the same community set in and Have the same set of nodes; for the consensus community All its promoted consensus communities Building the ultimate dynamic community 8. The dynamic community detection system considering the daily rhythm of human movement according to any one of claims 5-6, characterized in that: Module 5 consists of three submodules: Module 5.1, time series for quantifying the interaction strength of each dynamic community It refers to the sum of the edge weights of the community at each time interval; Module 5.2 is used to detect subcommunities within each dynamic community and utilize the total number of evolution events Quantify the time series of subcommunity evolution events; Module 5.3 is used to perform hierarchical clustering classification on the evolution patterns based on the time series metrics obtained in the first two steps.

Images