Trajectory flow node community partitioning and role recognition method and device based on dependency hypergraph

Abstract

The application relates to a trajectory flow node community division and role recognition method and device based on a dependency hypergraph. The method comprises the following steps: acquiring original trajectory flow data containing a plurality of sequence samples composed of physical nodes, extracting high-order dependency rules with statistical support from the trajectory flow data, taking each rule as a dependency motif composed of a historical context node and a target node; constructing corresponding hyperedges according to all the nodes involved in each dependency motif, forming a dependency hypergraph and assigning weights, performing community detection on the dependency hypergraph to obtain community division results of each physical node, calculating a community internal core degree score and a cross-community connection dispersion degree score of each node based on the community division results, and determining the community role of each physical node. The method can retain the structural integrity of a multi-node joint participation high-order transfer event, and improve the accuracy of community division and node role recognition in trajectory flow data.

Description

Technical Field

[0001] This application relates to the field of trajectory data / stream data mining technology, and in particular to a method and apparatus for trajectory stream node community partitioning and role recognition based on dependency hypergraph. Background Technology

[0002] With the increasing informatization of global shipping, intelligent transportation, and logistics supply chains, massive amounts of trajectory flow data have been generated. Trajectory flow data consists of continuous position sequences of numerous moving objects, such as ship port berthing sequences and vehicle station passage sequences. By modeling and analyzing trajectory flow data, we can uncover transfer dependencies between nodes and identify community structures and node roles in transportation or logistics networks, thus providing technical support for applications such as route planning, traffic scheduling, and hub node identification.

[0003] Currently, mainstream trajectory flow network modeling methods are typically based on the first-order Markov assumption, which assumes that the transition of the next node depends only on the current node, ignoring longer historical path information. However, real-world trajectory flow data often exhibits significant higher-order dependencies. For example, in shipping networks, the sequence of previous port calls of a vessel has a significant impact on the prediction of the next destination port, and relying solely on first-order transitions would result in a severe loss of path information.

[0004] To address these issues, existing research has proposed methods for network modeling based on higher-order dependency rules. However, most current methods use virtual nodes to represent higher-order dependency rules, resulting in a mixture of virtual and physical nodes in the network model and reduced physical interpretability. A few methods that only use physical nodes to construct the network decompose dependency rules into paired edges, losing structural information about the joint participation of multiple nodes. Secondly, in the projection and community partitioning process, dependency rules of different orders are treated equally without distinction, and the strong contextual constraint information carried by higher-order sparse rules is easily diluted by lower-order dense rules. Thirdly, there is a lack of quality assessment indicators for community partitioning oriented towards the characteristics of flowing data, making it difficult to measure the degree of autonomy of communities in the context of the flowing process and the clarity of higher-order boundaries. Summary of the Invention

[0005] Based on this, it is necessary to provide a method and apparatus for trajectory flow node community partitioning and role recognition based on dependency hypergraph, which can preserve the multi-node joint participation relationship of higher-order dependency rules, distinguish the structural contributions of different-order dependency rules, and evaluate the quality of community partitioning from both the flow process and higher-order structure dimensions, in order to address the above-mentioned technical problems.

[0006] A method for trajectory flow node community partitioning and role recognition based on dependency hypergraph, the method comprising: Obtain raw trajectory stream data, which contains multiple sequence samples composed of physical nodes; Extract statistically supported higher-order dependency rules from the trajectory flow data, and treat each higher-order dependency rule as a dependency module, which consists of a historical context node and a target node; A corresponding hyperedge is constructed based on all context nodes and target nodes involved in each dependent modifier. A dependency hypergraph is constructed based on the physical nodes and hyperedges, and each hyperedge is assigned a weight representing the dependency strength. Community detection is performed on the dependency hypergraph to obtain the community division results of each physical node; Based on the community division results, the coreness score within the community and the cross-community connectivity dispersion score of each physical node are calculated respectively. Based on the coreness score within the community and the cross-community connectivity dispersion score, the community role of each physical node is determined, thereby realizing the community division and role recognition of trajectory flow nodes.

[0007] In one embodiment, extracting statistically supported higher-order dependency rules from the trajectory stream data includes: Extract historical sequence segments composed of continuous physical nodes from the trajectory stream data; A high-order dependency test is performed on the trajectory flow data to determine whether the transition probability between each physical node depends on historical path information outside the current physical node; When a higher-order dependency is determined, the degree of difference in the conditional transition distribution of each historical sequence segment at adjacent orders is statistically determined, and the maximum effective order at which each sequence segment is allowed to be expanded is adaptively determined. Extract the conditional transition rules that are statistically determined and retained at each effective order, and use them as the higher-order dependency rules.

[0008] In one embodiment, when assigning weights to each hyperedge to characterize the dependence strength, the weights of each hyperedge are determined according to the following formula: ; In the above formula, For the weight of the superedge, The reliability weight of the historical context in the higher-order dependency rule. This is used to estimate the conditional transition probability of the target node under the given historical context.

[0009] In one embodiment, community detection is performed on the dependency hypergraph to obtain the community partitioning results of each physical node, including: Perform a degree-preserving projection on the dependent hypergraph to obtain a weighted projection graph; Community division is performed on the weighted projection map to obtain the first community division result of each physical node, and the first community division result is used as the final community division result.

[0010] In one embodiment, community detection is performed on the dependency hypergraph to obtain the community partitioning results of each physical node, and the method further includes: Based on the first community partitioning result, the weights of each hyperedge in the dependency hypergraph are iteratively reweighted. After each iteration, the node degree-preserving projection and community partitioning are performed again based on the updated hyperedge weights until the preset convergence condition is met, and the second community partitioning result to which each physical node belongs is obtained. The second community partitioning result is taken as the final community partitioning result.

[0011] In one embodiment, the weights of each hyperedge in the dependency hypergraph are iteratively reweighted, including: In each iteration, the community label of each node within a hyperedge is determined based on the current community division result, and the homogeneity of each hyperedge is calculated. The homogeneity is the maximum percentage of nodes within a hyperedge belonging to the same community. Calculate the mean homogeneity of all superedges in the current round, weight superedges with homogeneity higher than the mean, subtract superedges with homogeneity lower than the mean, and perform global normalization on the weights of all superedges after reweighting.

[0012] In one embodiment, based on the community division results, calculating the intra-community coreness score and cross-community connectivity dispersion score for each physical node includes: Project the dependency hypergraph into a weighted graph. Based on the community partitioning results, calculate the intra-community core score and cross-community connectivity dispersion score of each physical node on the weighted graph. Alternatively, traverse each hyperedge in the dependency hypergraph, calculate the hyperedge-level core score and hyperedge-level dispersion score of each hyperedge based on the community partitioning results, and then for each physical node, weight and aggregate the hyperedge-level scores of all hyperedges in which the physical node participates, using the hyperedge weight as the weight, to obtain the intra-community core score and cross-community connectivity dispersion score of the corresponding physical node.

[0013] In one embodiment, the community role of each physical node is determined based on the coreness score within the community and the cross-community connectivity dispersion score, including: For each community, physical nodes whose coreness score exceeds the first preset threshold are identified as community kernel group nodes; For each community, physical nodes whose cross-community connectivity dispersion score exceeds the second preset threshold are identified as cross-community connectivity nodes; The fusion role score of each physical node is calculated by multiplying the core score within the community and the cross-community connection dispersion score. Physical nodes whose fusion role scores exceed a third preset threshold are identified as connection-type hub nodes that have both community core group and cross-community connection functions.

[0014] In one embodiment, the method further includes: Based on the community division results, the flow retention degree of each community is calculated, which is the conditional probability that a person remains in the community after a random walk from the community; and / or, based on the community division results, the higher-order boundary sharpness of each community is calculated, which is determined according to the ratio of the amount of dependency motif cut across the community boundary to the volume of dependency motif within the community. The quality of the community division results is evaluated based on the flow retention and / or the higher-order boundary clarity.

[0015] This application also provides a device for trajectory flow node community partitioning and role recognition based on dependency hypergraph, the device comprising: The data acquisition module is used to acquire raw trajectory flow data, which includes multiple sequence samples composed of physical nodes; The dependency module is used to extract statistically supported higher-order dependency rules from the trajectory flow data, and to treat each higher-order dependency rule as a dependency module, which consists of a historical context node and a target node. The dependency hypergraph construction module is used to construct a corresponding hyperedge based on all context nodes and target nodes involved in each dependency module, construct a dependency hypergraph based on the physical nodes and hyperedges, and assign a weight representing the dependency strength to each hyperedge. The community partitioning module is used to perform community detection on the dependency hypergraph and obtain the community partitioning results of each physical node. The node role score calculation module is used to calculate the community coreness score and cross-community connection dispersion score of each physical node based on the community division results. The community role determination module is used to determine the community role of each physical node based on the coreness score within the community and the cross-community connection dispersion score, thereby realizing the community division and role recognition of trajectory flow nodes.

[0016] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the specific steps in the above-described method for trajectory flow node community partitioning and role recognition based on dependency hypergraph.

[0017] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the specific steps in the above-described method for trajectory flow node community partitioning and role recognition based on dependency hypergraph.

[0018] The aforementioned method and apparatus for trajectory flow node community partitioning and role recognition based on dependency hypergraphs extract statistically supported high-order dependency rules from trajectory flow data containing multiple sequence samples composed of physical nodes. Each high-order dependency rule is treated as a dependency motif, which consists of historical context nodes and target nodes. A corresponding hyperedge is constructed based on all context nodes and target nodes involved in each dependency motif. A dependency hypergraph is constructed based on the physical nodes and hyperedges, and each hyperedge is assigned a weight representing the dependency strength. Community detection is performed on the dependency hypergraph to obtain the community partitioning results of each physical node. Based on the community partitioning results, the intra-community core score and cross-community connectivity dispersion score of each physical node are calculated. Based on the intra-community core score and the cross-community connectivity dispersion score, the community role of each physical node is determined, thereby realizing trajectory flow node community partitioning and role recognition.

[0019] Beneficial effects: This method uses high-order dependency rules extracted from trajectory flow data as dependency motifs and constructs a hyperedge by integrating all context nodes and target nodes involved in each dependency motif. This preserves the structural integrity of multi-node joint participation in high-order transition events and overcomes the interpretability deficiencies of existing methods that use a mix of virtual and physical nodes to represent high-order dependencies, or the structural information loss caused by decomposing relationships into paired edges in physical node modeling. The dependency hypergraph constructed on this basis uses physical nodes as the node set and dependency motifs as the hyperedge set, enabling community detection to be performed directly on multi-node collaborative relationships. The community partitioning results more realistically reflect the high-order organizational patterns in trajectory flow data. By calculating the intra-community coreness score and cross-community connectivity dispersion score of each physical node, the node role can be quantitatively characterized from two dimensions: intra-community hub and cross-community connectivity. This identifies community core group nodes, cross-community connecting nodes, and connecting hub nodes that possess both functions, providing a more refined basis for node functional classification for applications such as route planning and hub port identification. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a trajectory flow node community partitioning and role recognition method based on a dependency hypergraph in one embodiment. Figure 2 This is a schematic diagram illustrating the construction process of a hypergraph in one embodiment, wherein, Figure 2 (a) shows a schematic diagram of the sample stream sequence and D-Motif extraction. Figure 2(b) shows a schematic diagram of a dependency hypergraph composed of D-Motifs; Figure 3 This is a schematic diagram of the DMEdge2Node path calculation process in one embodiment; Figure 4 This is a schematic diagram comparing the network collapse curves of high-order features and first-order features of nodes in an experiment. Figure 4 (a) is a schematic diagram showing the number of sequences that can still be completely traversed in the original trajectory flow. Figure 4 (b) A schematic diagram showing the number of connected components in the physical reference network; Figure 5 This is a structural block diagram of a trajectory flow node community partitioning and role recognition device based on a dependency hypergraph in one embodiment; Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0022] To address the problems in existing technologies, such as the loss of multi-node joint participation due to decomposing high-order dependency rules into paired edges, the dilution of high-order sparse information due to treating different order dependency rules equally in community partitioning, and the lack of community partitioning quality evaluation indicators oriented towards the characteristics of streaming data, this application, as follows... Figure 1 As shown, a method for trajectory flow node community partitioning and role recognition based on dependency hypergraph is provided, which specifically includes the following steps: Step S100: Obtain the original trajectory stream data, which contains multiple sequence samples composed of physical nodes.

[0023] Step S110: Extract statistically supported high-order dependency rules from the trajectory flow data, and treat each high-order dependency rule as a dependency module. The dependency module consists of historical context nodes and target nodes.

[0024] Step S120: Construct a corresponding hyperedge based on all context nodes and target nodes involved in each dependent modifier, construct a dependency hypergraph based on physical nodes and hyperedges, and assign a weight representing the dependency strength to each hyperedge.

[0025] Step S130: Perform community detection on the dependent hypergraph to obtain the community division results of each physical node.

[0026] Step S140: Based on the community division results, calculate the intra-community coreness score and cross-community connectivity dispersion score for each physical node.

[0027] Step S150: Based on the coreness score within the community and the cross-community connectivity dispersion score, determine the community role of each physical node to realize the community division and role identification of trajectory flow nodes.

[0028] In step S100, container shipping network analysis is used as an application scenario for explanation. The raw trajectory flow data comes from historical ship navigation data recorded by the Automatic Identification System (AIS) or from ship voyage registration data provided by shipping companies. Each sequence sample corresponds to a complete ship voyage, consisting of the port identifiers that the ship calls at in sequence during that voyage, arranged chronologically. For example, a typical sequence sample can be represented as "Port A → Port B → Port C → Port D", where each element is a physical node, corresponding to a real geographically existing port.

[0029] Specifically, the set of physical nodes is denoted as Each node Each port has a unique identifier and corresponding geographic coordinates (longitude and latitude). The set of all sequence samples in the trajectory stream data is denoted as […]. , ,in, The total number of sequence samples. For the first sequence.

[0030] It should be noted that the above description using a container shipping network as an example is merely illustrative and is not intended to limit the scope of protection of this application. In other embodiments, physical nodes can also be any type of trajectory flow network node, such as bus stops, subway stations, logistics transfer nodes, or airports, as long as they are sequence samples composed of continuous geographical location entity nodes, or user information in a data stream.

[0031] In step S110, extracting statistically supported higher-order dependency rules from trajectory flow data includes: firstly, extracting historical sequence segments composed of continuous physical nodes from the trajectory flow data; performing higher-order dependency tests on the trajectory flow data to determine whether the transition probability between each physical node depends on historical path information outside the current physical node; when higher-order dependency is determined, statistically determining the degree of difference in conditional transition distribution of each historical sequence segment under adjacent orders; adaptively determining the maximum effective order that each sequence segment is allowed to expand; and extracting the conditional transition rules retained by statistical determination under each effective order as higher-order dependency rules.

[0032] In this embodiment, before performing high-order modeling on the trajectory flow data, it is first necessary to determine whether the data truly exhibits high-order dependencies. The high-order dependency test employs three complementary statistical testing methods, providing criteria from the perspectives of consistency of first-order closure constraints, nested hypothesis testing, and the information criterion. If any method supports rejecting the first-order Markov hypothesis, it is considered that the data possesses a structure requiring high-order dependency modeling.

[0033] Specifically, before conducting the verification, basic quantities are statistically analyzed from the trajectory flow data. Let the set of physical nodes be... For order , No. The sequence at time , length is The historical context is defined as: (1) In formula (1), All belong to the physical node set , express of The second Cartesian product.

[0034] To facilitate the subsequent description and quantification of higher-order dependencies, we will first explain the meaning of order. Used to characterize the length of historical nodes upon which the transition behavior of the current node depends in a trajectory sequence. When When , it is of the first order, meaning that the transition of the current node depends only on the previous historical node, corresponding to the traditional first-order Markov assumption. When When the order is higher, it indicates that the transition dependencies of the current node are consecutive. The path information formed by the collective historical nodes is called higher-order dependency.

[0035] To quantitatively test the first-order Markov hypothesis and provide a unified statistical basis for subsequent higher-order dependency analysis, a definition is defined for trajectory flow data. Rank fragment counting and The basic statistics of the conditional transition distribution. For the historical context shown in formula (1), the historical context that satisfies the following... And the current node is transferred to The total number of occurrences of the trajectory segment is defined as The order segment count is represented as: (2) In formula (2), This represents the number of indexes that meet the conditions. Based on the above segment count, maximum likelihood estimation is used to obtain... Order-conditional transition distribution, used to describe the conditional transition distribution in a given historical context. Under the given conditions, the current node will transfer to the target node. The probability is expressed as: (3) In formula (3), at the order The set of observable contexts on is denoted as .

[0036] The above Rank fragment counting and The conditional transition distribution constitutes the Chapman distribution in this embodiment. The Kolmogorov consistency test, likelihood ratio test, and BIC model selection method form a unified statistical basis for three higher-order dependency testing methods. These three complementary methods provide criteria from the perspectives of consistency of first-order closure constraints, nested hypothesis testing, and information criteria, respectively. When any method supports the rejection of the first-order hypothesis, it is considered that the data has a structure that needs to be modeled with higher-order dependencies.

[0037] Once a higher-order dependency is determined, the degree of difference in conditional transition distribution among historical sequence segments at adjacent orders is statistically assessed, and the maximum effective order for each sequence segment to be extended is adaptively determined. Specifically, the context is extended upwards order by order, and the current historical sequence segment is calculated... Rank and The KL divergence between conditional transition distributions of different orders is calculated; simultaneously, a dynamic threshold is obtained that increases with increasing order and decreases with increasing sample support. When the KL divergence exceeds this threshold, the historical sequence segment is considered expandable to... Rank, and with If the order is used as the new current order, the expansion continues upwards; otherwise, the expansion stops, and the maximum effective order is [order number missing]. In this process, regardless of whether it is expanded, the lower-order rules are retained at their respective orders to fully depict the start-end relationships under different historical lengths.

[0038] Finally, the conditional transition rules retained by the above statistical determination at each effective order are extracted as the higher-order dependency rules output in this step. Each rule is represented as follows: (4) In formula (4), For length Historical context, This refers to the corresponding target node. In this method, each retained rule is called a dependency motif, denoted as D-motif. A D-motif is a basic structural unit obtained through statistical determination, characterizing the higher-order dependency relationship where the target node is accessed under the condition that context nodes co-occur. Among these, historical context nodes are the consecutive nodes in the higher-order dependency rules that precede the appearance of the target node. A sequence segment composed of physical nodes For the order of the modifier, and The target node is the next physical node that appears after the historical context node in the higher-order dependency rule.

[0039] In step S120, each higher-order dependency rule output in step S110 is treated as a dependency motif. A corresponding hyperedge is constructed based on all context nodes and target nodes involved in each dependency motif. A dependency hypergraph is constructed based on physical nodes and hyperedges, and each hyperedge is assigned a weight representing the dependency strength.

[0040] In this embodiment, for each dependent motif All context nodes involved With the target node They are all included in the same superedge, that is: (5) For ease of description, the multi-node set constructed from the dependent motifs will be referred to as a "hyperedge" below. This hyperedge is constructed based on all context nodes and target nodes involved in each dependent motif.

[0041] Specifically, the hyperedge shown in formula (5) maps a complete higher-order transition event to a set of multiple nodes, directly preserving the joint participation relationship between the context node and the target node at the structural level. It should be noted that the higher-order transition rules... While nodes inherently possess directionality from context to target node, in the structural representation of a dependency hypergraph, the relationships between context nodes and between context nodes and the target node are considered as collaborative relationships involving participation in the same higher-order transition event, without explicitly distinguishing directions. This convention emphasizes the collaborative participation relationships between nodes in higher-order transitions, aligning with the community structure that subsequent community detection and node role analysis focus on.

[0042] Furthermore, by traversing all dependent modalities output in step S110 and constructing their corresponding hyperedges one by one, a dependent hypergraph is obtained, formally defined as: (6) In formula (6), The set of nodes is all port nodes (or other geographically located entity nodes) in the trajectory flow data. For D-motif hyperedge set, The hyperedge weighting function is used to quantitatively characterize the statistical strength of each dependent motif.

[0043] Furthermore, to quantitatively characterize the statistical strength of each dependent modality, the hyperedge weights are defined as follows: (7) In formula (7), For the weight of the superedge, For the reliability weight of historical context in higher-order dependency rules, For estimating the conditional transition probability of a target node under historical context, i.e., based on length The conditional transition probability estimate is obtained from the fragment count.

[0044] like Figure 2 As shown, a schematic diagram of the construction of a dependency hypergraph is presented. Figure 2 (a) Starting from the sample stream sequence, the higher-order rules to be retained are determined by order determination, and then the D-motif hyperedge is generated by formula (5). Figure 2 (b) The membership relationships between each D-motif and physical node are represented using a hypergraph association representation. Compared to the method of decomposing rules into cross-order connections between node pairs in multi-order networks, this construction directly preserves the set of multiple nodes involved in the rule, providing a foundation for subsequent community structure analysis.

[0045] In step S130, community detection is performed on the dependency hypergraph constructed in step S120. This method provides two community detection methods: one is a one-time community partitioning based on improved node degree-preserving projection, and the other is an iterative reweighted community partitioning based on the one-time partitioning. The two methods can be selected according to the application scenario.

[0046] In this embodiment, community detection is performed on the dependency hypergraph to obtain the community partitioning results of each physical node. This includes: performing node degree-preserving projection on the dependency hypergraph to obtain a weighted projection graph; performing community partitioning on the weighted projection graph to obtain the first community partitioning results of each physical node; and using the first community partitioning results as the final community partitioning results.

[0047] Specifically, dependency hypergraphs are a type of weighted undirected hypergraph. To input them into existing graph community detection algorithms while preserving the higher-order structural information of hyperedges, the hypergraph needs to be projected as a weighted graph first. Traditional node-degree-preserving projection methods distribute the contribution of each hyperedge to the connection between node pairs according to the number of nodes it contains, but do not distinguish the dependency order corresponding to the hyperedge. For dependency modal hyperedges obtained from higher-order dependency rules, the historical context length reflected by hyperedges of different orders is different, and the stability of statistical estimation also varies with order.

[0048] To preserve order information during the projection phase, this embodiment proposes an improved nodal degree-preserving projection method, which introduces additional cross-order weights for each dependent phantom hyperedge. ,in Let the order of the rule corresponding to this hyperedge be: (8) In formula (8), In single-parameter form, it is represented as: (9) In formula (9), This is the cross-order weight amplification parameter. Time weights are concentrated in lower orders. When all orders are weighted equally, the improved NDP degenerates into the original NDP. Time weights are concentrated towards higher orders. This is achieved by adjusting... The value of can be used to control the contribution of higher-order dependency rules in the projection graph according to specific task requirements.

[0049] Furthermore, the weighted graph obtained by maintaining the projection of the improved node degree described above... The community detection algorithm is then executed to perform partitioning. This embodiment uses the Leiden algorithm for one-time community partitioning, which iteratively optimizes the modularity function through three stages: local movement, merging, and refinement. The resulting community partitioning is denoted as... The community tag for each physical node is .

[0050] In this embodiment, based on the above partitioning, an improved iterative reweighted community partitioning method is further provided, that is, an optimized IRMM is obtained by improving the original IRMM. Traditional iterative reweighting methods update hyperedge weights using a global moving average, and the update magnitude does not depend on the average performance of the hyperedge group under the current partitioning. For dependency hypergraphs, the retained dependency motif hyperedges are determined statistically, with their size distribution concentrated in low-order regions, while high-order hyperedges are sparse but carry strong contextual constraints. If hyperedges of different orders are treated equally during reweighting, high-order signals are easily diluted by low-order signals. Therefore, this embodiment proposes the following three improvements: Specifically, the projection used in each iteration is first replaced by the improved node-degree-preserving projection described above, so that the iteration process inherits the cross-order weights. This ensures the transmission of order information. Among these, the initial projection diagram... The weights calculated according to formula (8) are still in the form of formula (8) after each round of updating the superedge weights (with the updated weights). replace ) to obtain the next round of weighted graph .

[0051] Furthermore, given the first Round division With node community tags For each superedge Define homogeneity: (10) In formula (10) For super-edge The community tag with the largest internal share, This characterizes the degree to which nodes within a hyperedge belong to the same community under the current partitioning. This definition is computationally concise and applicable to the size of hyperedges in a hypergraph (…). ,and The percentage of the largest community is consistent with other homogeneity measures in the case of uniform division of multiple communities in terms of ranking.

[0052] Furthermore, based on the relative reweighting of the population mean and the conservation of total weight, using the homogeneity mean of all hyperedges in the current round, the calculation process is expressed as follows: (11) Based on formula (11), the reweighting rule is rewritten as follows: (12) In formula (12), The step size coefficient is updated for reweighting. Formula (12) only applies corrections to superedges that deviate from the group mean: superedges with homogeneity higher than the mean are weighted, superedges with homogeneity lower than the mean are weighted, and superedges near the mean remain approximately unchanged. The update magnitude is determined by... Control. Compared with the original IRMM's approach of assigning absolute values ​​to all hyperedges and then smoothing them historically, formula (12) uses relative deviation as the driving force. In a single iteration, the weights of most hyperedges close to the mean are not modified. Only the hyperedges at the two ends are adjusted, so that the effective effect of reweighting is concentrated on the hyperedges that are sensitive to partitioning.

[0053] Furthermore, to avoid the accumulation of relative reweighting over multiple iterations leading to a drift in the total weight of the superedge group, the updated weights are further globally normalized, the process of which is expressed as follows: (13) In formula (13), such that The modulus is conserved in each iteration. This operation ensures that reweighting only redistributes weights among hyperedges without changing their total amount, and the absolute scale of modularity remains stable during iteration, making it convenient to use the difference in modularity between two adjacent iterations as a convergence criterion.

[0054] Based on the three improvements mentioned above, the complete process of the improved iterative reweighted community partitioning is as follows: Based on the first community partitioning result, the weights of each hyperedge in the dependent hypergraph are iteratively reweighted. After each iteration, the node degree-preserving projection and community partitioning are re-performed based on the updated hyperedge weights until the preset convergence condition is met, resulting in the second community partitioning result for each physical node. This second community partitioning result is then used as the final community partitioning result. Specifically, the improved IRMM framework, through relative reweighting iteration, gradually strengthens the tendency for hyperedges to fall into the same community, making the community structure closer to the semantics of strict hypergraph modularity. Unlike the OS framework, which favors dense organization of the projected graph, the improved IRMM places greater emphasis on the overall community affiliation of hyperedges.

[0055] In this embodiment, the second community partitioning result obtained after partitioning based on the improved IRMM is represented as follows: Community tags for physical nodes are .

[0056] In this embodiment, the hypergraph is relied upon. The community partitioning process is organized in the order of improved NDP projection, Leiden single-shot partitioning OS framework, and improved IRMM framework as the community detection process. The partitioning obtained from the first Leiden solution is the output of the OS (One-Shot) framework, and it also serves as the initial partitioning for the improved IRMM iteration. The improved IRMM iterates through formulas (10), (12), and (13) on this initial partitioning until the difference in modularity between two adjacent rounds is less than the tolerance. Or reach the maximum number of iterations This also leads to the community division within the OS framework. With node labels And improve community partitioning under the IRMM framework. With node labels .

[0057] In this paper, pseudocode for implementing step S130 above is also presented, as shown in Table 1:

[0058] The two community detection methods described above each output the first community division results. Second Community Division Results In practical applications, one of the partitioning results can be selected as the final output of step S130 based on task requirements, and used for the node role score calculation in subsequent step S140. The two partitioning methods each have their own emphasis and can be used in combination in different analysis scenarios.

[0059] In step S140, based on the community partitioning results obtained in step S130, the intra-community coreness score and cross-community connectivity dispersion score of each physical node are calculated. This step provides two types of mapping paths from the dependency hypergraph to node scores, denoted as the NDP2Node path and the DMEdge2Node path, which can be used individually or in combination. The intra-community coreness score and cross-community connectivity dispersion score are calculated based on these two paths.

[0060] In this embodiment, the NDP2Node path first projects the dependent hypergraph into a weighted graph according to the improved node degree preservation projection rule described in step S130, and then directly calculates the score at the node level of the projected graph.

[0061] Specifically, based on the NDP2Node path calculation, ,in, , These represent the node's coreness score within its community and its cross-community connectivity score, respectively.

[0062] In this embodiment, in the DMEdge2Node path, each hyperedge in the dependency hypergraph is traversed, and the hyperedge-level core score and hyperedge-level dispersion score of each hyperedge are calculated according to the community partitioning results. Then, for each physical node, the hyperedge-level scores of all hyperedges in which the physical node participates are weighted and aggregated with the hyperedge weight as the weight to obtain the community core score and cross-community connection dispersion score of the corresponding physical node.

[0063] Specifically, the computation of DMEdge2Node consists of two steps: hyperedge-level scoring and node-level aggregation. Let the node community tag be denoted as... It is acceptable or First, for each superedge... Record the nodes that fall into the community. The quantity is: (14) And denote the size of the superedge as Super-edge The main community is defined as the community with the largest proportion within the superedge, denoted as: (15) The corresponding number of nodes within the main community is denoted as The hyperedge-level cross-community connectivity dispersion score is defined as follows: (16) The cross-community connectivity dispersion score calculated by formula (16) takes a larger value when the nodes inside the hyperedge belong to different communities, and takes zero when the entire hyperedge falls into the same community, thus characterizing the degree of cross-community dispersion of the hyperedge under the current partition.

[0064] Furthermore, the coreness score within the super-border level community is standardized for super-border groups within the same main community, using the main community as a reference: (17) In formula (17), , respectively The entire super-border of the main community The mean and standard deviation. The super edge fusion index is: (18) Subsequently, for each physical node The hyperedge-level scores of all hyperedges in which it participates are weighted and aggregated using the hyperedge weights: (19) In formula (19), The superedge weights are defined in step S120.

[0065] In this embodiment, the core difference between the NDP2Node path and the DMEdge2Node path lies in the selection of the contribution unit: the NDP2Node path uses node pairs as the basic contribution unit, first projects the dependency hypergraph into a weighted graph, and then calculates the intra-community coreness score and cross-community connectivity dispersion score of each node on the projected graph. The result is biased towards diffusion and bridging semantics. The DMEdge2Node path uses dependency motif hyperedges as the basic contribution unit, first calculates the hyperedge-level score of each hyperedge at the hyperedge level, and then aggregates the hyperedge-level scores to each node it contains using the hyperedge weight. The node score reflects the cumulative contribution of all dependency motifs in which the node participates, and is biased towards higher-order collaborative constraint semantics.

[0066] The complete calculation process of the DMEdge2Node path is as follows: Figure 3 As shown, Figure 3 China and Israel's three dependent modal hyperedges , , For example, the left side shows the hyperedge-level score of each hyperedge calculated using the hyperedge-level score calculation method. The middle section performs a weighted summation based on the hyperedge weights, aggregating the hyperedge-level scores to the nodes contained in each hyperedge; the right side section gives the final score for each node. Nodes are distinguished by coloring based on their current community tags. It should be noted that the hyperedge-level fusion score... Node fusion score All according to The definition and role determination are detailed in subsequent steps S150.

[0067] In this implementation, the NDP2Node path and the DMEdge2Node path correspond to two different ways of using higher-order information in the dependency hypergraph: the former simplifies the dependency hypergraph to a binary projection, using the projection intensity of node pairs as the basis for calculation; the latter preserves the integrity of hyperedges, using dependency motifs as the basic contributing units. The node scores output by the two paths have different structural semantics, which are retained as complementary features in this paper. In practical applications, one can be selected or used in combination depending on the specific task requirements.

[0068] In step S150, the community role of each physical node is determined based on the core score within the community and the cross-community connection dispersion score, including: for each community, identifying physical nodes with a core score within the community exceeding a first preset threshold as community kernel group nodes; for each community, identifying physical nodes with a cross-community connection dispersion score exceeding a second preset threshold as cross-community connection nodes; calculating the fusion role score of each physical node based on the product of the core score within the community and the cross-community connection dispersion score; and identifying physical nodes with a fusion role score exceeding a third preset threshold as connection-type hub nodes that have both community kernel group and cross-community connection functions.

[0069] In this embodiment, the community core score of each physical node output in step S140 is used as the basis. And cross-community connectivity dispersion score The community role of each physical node is determined. The zP two-dimensional role analysis framework is adopted, and a fusion role score is introduced on this basis to identify community kernel group nodes, cross-community connection nodes, and connection hub nodes that have both functions.

[0070] Specifically, the community coreness score It characterizes the deviation of the connectivity of physical nodes within their respective communities from the average level within the community. A higher value indicates a more tightly connected node within its community, placing it at the core of the community and assuming a role within the community's core group. (Cross-community connectivity dispersion score) It depicts the evenness of the distribution of physical node connections across communities. The higher the value, the more dispersed the node's connections are across multiple communities, and the more it has the attribute of connecting nodes across communities.

[0071] Furthermore, to quantitatively identify nodes that possess both of the above characteristics, this embodiment defines the fusion role score as: (20) The above-mentioned fusion character scores are only in and At the same time, when the value is higher, a larger value is taken, and the degree to which a node plays both the role of community kernel group and cross-community connection is characterized in the form of a continuous quantity, thus avoiding the sensitivity of empirical parameters caused by setting independent thresholds and then taking the intersection.

[0072] Based on the above scores, this embodiment determines roles according to the following rules: Nodes exceeding the first preset threshold are identified as community kernel group nodes; Nodes exceeding the second preset threshold are identified as cross-community connection nodes; Nodes exceeding a third preset threshold are identified as connecting hub nodes that function as both community kernel groups and cross-community connections. These three criteria are not mutually exclusive; a single node can satisfy multiple conditions simultaneously. In shipping network scenarios, the identified connecting hub nodes correspond to core ports undertaking cross-regional transshipment functions, and have a crucial impact on the overall network connectivity.

[0073] To verify the effectiveness of the node high-order feature analysis method based on dependency hypergraph proposed in this application in the evaluation of node importance in trajectory flow, experiments were conducted on simulated flow data.

[0074] The experiment uses network disintegration as the evaluation framework and employs a static ranking strategy: each method calculates node scores on the initial data, generates a fixed removal order, and then gradually removes the top-ranked nodes. Network performance changes are evaluated on the same physical baseline network (a first-order undirected weighted graph composed of the cumulative adjacent hop counts of the original trajectories). Evaluation metrics include: the number of still feasible sequences (RS(k)): measures the network's functional robustness; a faster decrease in RS(k) indicates a higher efficiency in disrupting functional paths. The number of connected components (Comp(k)): measures the network's structural connectivity; a faster increase in Comp(k) indicates a stronger ability to disintegrate the network structure.

[0075] This paper proposes a node high-order feature method based on dependency hypergraphs, which includes two types of community partitioning methods (OS and improved IRMM), two types of mapping paths (NDP2Node and DMEdge2Node), and three types of node high-order feature scores (z, P, zP), totaling 12 node rankings. In the dependency hypergraph construction, the maximum rule order is set to K=4, the cross-order weight amplification parameter of the improved NDP projection is set to α=1 / 2, and the projection denominator adopts a combined mode. The Leiden algorithm resolution parameter ρ=1.0, the improved IRMM reweighting step size η=0.8, and the modularity convergence tolerance ε_Q=10. -5 The maximum number of iterations, T_max, is 8.

[0076] Figure 4The collapse curves of RS(k) and Comp(k) as the number of node removal steps are shown. To highlight the early differences, magnified boxes are set in the x∈[3,20], y∈[0,30000] region of RS(k) and the x∈[65,100], y∈[70,100] region of Comp(k). The results show that several ranking strategies proposed in this paper exhibit significant curve separation from the first-order baseline method in the magnified regions, indicating that the higher-order features of nodes based on community structure have a stronger destructive efficiency on network function and structure.

[0077] Table 2 summarizes the AUC results (raw AUC and normalized AUC) of each method on the RS(k) metric. As can be seen from the table, DMEdge2Node-zP(OS) achieved the lowest Norm_AUC (0.0159), significantly outperforming the best-performing first-order method, Betweenness (0.0268). NDP2Node-P(OS) and NDP2Node-P(IRMM) achieved Norm_AUCs of 0.0179 and 0.0163, respectively, also outperforming all first-order metrics. Regarding functional robustness, all community-structure-based methods in this paper outperformed first-order network centrality, with DMEdge2Node-zP(OS) showing the best performance.

[0078] Table 3 summarizes the AUC results of each method on the Comp(k) metric. The Norm_AUC of our method ranges from 52.5 to 53.6, all higher than that of first-order methods (e.g., Weighted Degree is 51.14, Betweenness is 51.80). Among them, NDP2Node-zP(IRMM) achieves the highest Norm_AUC of 53.64, indicating that our method has a significant advantage in accelerating network splitting.

[0079] Table 2. AUC results of each method on the RS(k) index

[0080] Table 3. AUC results of each method on the Comp(k) index (structure as above)

[0081] Further analysis is conducted on the three design dimensions within the method of this application. In terms of mapping path comparison: NDP2Node and DMEdge2Node alternately lead in different metrics, with DMEdge2Node performing better in functional disruption tasks but incurring higher computational costs; NDP2Node exhibits better stability in structural disintegration tasks. These two paths correspond to different structural signals, which are both retained in this application. In terms of community partitioning method comparison: OS and improved IRMM each have advantages under different metrics and mapping paths, indicating that they correspond to community semantics at the projection level and the hyperedge global level, respectively, and can be used complementaryly. In terms of node feature comparison: P (cross-community connectivity dispersion) is most sensitive to functional disruption, zP (fusion metric) shows stability in structural disintegration, and z (intra-community hubliness) also performs well in some combinations. In practical applications, appropriate features should be selected based on the task objectives.

[0082] The results show that the proposed high-order node feature method based on dependency hypergraph can effectively identify nodes that play a key role in disrupting flow path function and splitting network structure, and is significantly better than traditional first-order network methods in the task of evaluating the importance of trajectory flow nodes.

[0083] The aforementioned method for trajectory flow node community partitioning and role recognition based on dependency hypergraphs uses high-order dependency rules extracted from trajectory flow data as dependency motifs. It constructs a hyperedge by integrating all context nodes and target nodes involved in each dependency motif. This preserves the structural integrity of multi-node joint participation in high-order transition events, overcoming the structural information loss caused by existing methods that decompose high-order dependencies into paired edges. During community detection, cross-order weights are introduced to distinguish the contributions of dependency rules of different orders, and a relative reweighting mechanism based on population mean is used to strengthen the overall community affiliation tendency of the hyperedge, effectively preserving the strong path constraint information carried by high-order sparse rules. Furthermore, by calculating the coreity score within a community and the cross-community connectivity dispersion score, the method quantitatively identifies community core group nodes, cross-community connecting nodes, and connecting hub nodes with both functions from two dimensions: internal community hub and cross-community connectivity. This provides a more refined node functional classification basis for port community partitioning and hub identification in shipping networks.

[0084] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0085] In one embodiment, such as Figure 5 As shown, a trajectory flow node community partitioning and role recognition device based on dependency hypergraph is provided, including: a data acquisition module 200, a dependency phantom construction module 210, a dependency hypergraph construction module 220, a community partitioning module 230, a node role score calculation module 240, and a community role determination module 250, wherein: The data acquisition module 200 is used to acquire raw trajectory flow data, which includes multiple sequence samples composed of physical nodes; The dependency module 210 is used to extract statistically supported higher-order dependency rules from the trajectory flow data, and to treat each higher-order dependency rule as a dependency module, wherein the dependency module consists of a historical context node and a target node. The dependency hypergraph construction module 220 is used to construct a corresponding hyperedge based on all context nodes and target nodes involved in each dependency module, construct a dependency hypergraph based on the physical nodes and hyperedges, and assign a weight representing the dependency strength to each hyperedge. The community partitioning module 230 is used to perform community detection on the dependency hypergraph and obtain the community partitioning results of each physical node. The node role score calculation module 240 is used to calculate the community coreness score and cross-community connection dispersion score of each physical node based on the community division results. The community role determination module 250 is used to determine the community role of each physical node based on the coreness score within the community and the cross-community connection dispersion score, thereby realizing the community division and role recognition of trajectory flow nodes.

[0086] Specific limitations regarding the trajectory flow node community partitioning and role recognition device based on dependency hypergraphs can be found in the limitations of the trajectory flow node community partitioning and role recognition method based on dependency hypergraphs mentioned above, and will not be repeated here. Each module in the aforementioned trajectory flow node community partitioning and role recognition device based on dependency hypergraphs can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0087] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When executed by the processor, the computer program implements a method for trajectory flow node community partitioning and role recognition based on a dependency hypergraph. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0088] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0089] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps: Obtain raw trajectory stream data, which contains multiple sequence samples composed of physical nodes; Extract statistically supported higher-order dependency rules from the trajectory flow data, and treat each higher-order dependency rule as a dependency module, which consists of a historical context node and a target node; A corresponding hyperedge is constructed based on all context nodes and target nodes involved in each dependent modifier. A dependency hypergraph is constructed based on the physical nodes and hyperedges, and each hyperedge is assigned a weight representing the dependency strength. Community detection is performed on the dependency hypergraph to obtain the community division results of each physical node; Based on the community division results, the coreness score within the community and the cross-community connectivity dispersion score of each physical node are calculated respectively. Based on the coreness score within the community and the cross-community connectivity dispersion score, the community role of each physical node is determined, thereby realizing the community division and role recognition of trajectory flow nodes.

[0090] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Obtain raw trajectory stream data, which contains multiple sequence samples composed of physical nodes; Extract statistically supported higher-order dependency rules from the trajectory flow data, and treat each higher-order dependency rule as a dependency module, which consists of a historical context node and a target node; A corresponding hyperedge is constructed based on all context nodes and target nodes involved in each dependent modifier. A dependency hypergraph is constructed based on the physical nodes and hyperedges, and each hyperedge is assigned a weight representing the dependency strength. Community detection is performed on the dependency hypergraph to obtain the community division results of each physical node; Based on the community division results, the coreness score within the community and the cross-community connectivity dispersion score of each physical node are calculated respectively. Based on the coreness score within the community and the cross-community connectivity dispersion score, the community role of each physical node is determined, thereby realizing the community division and role recognition of trajectory flow nodes.

[0091] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0092] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0093] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for trajectory flow node community partitioning and role recognition based on dependency hypergraph, characterized in that, The method includes: Obtain raw trajectory stream data, which contains multiple sequence samples composed of physical nodes; Extract statistically supported higher-order dependency rules from the trajectory flow data, and treat each higher-order dependency rule as a dependency module, which consists of a historical context node and a target node; A corresponding hyperedge is constructed based on all context nodes and target nodes involved in each dependent modifier. A dependency hypergraph is constructed based on the physical nodes and hyperedges, and each hyperedge is assigned a weight representing the dependency strength. Community detection is performed on the dependency hypergraph to obtain the community division results of each physical node; Based on the community division results, the coreness score within the community and the cross-community connectivity dispersion score of each physical node are calculated respectively. Based on the coreness score within the community and the cross-community connectivity dispersion score, the community role of each physical node is determined, thereby realizing the community division and role recognition of trajectory flow nodes.

2. The method for trajectory flow node community partitioning and role recognition based on dependency hypergraph as described in claim 1, characterized in that, Extracting statistically supported higher-order dependency rules from the trajectory flow data includes: Extract historical sequence segments composed of continuous physical nodes from the trajectory stream data; A high-order dependency test is performed on the trajectory flow data to determine whether the transition probability between each physical node depends on historical path information outside the current physical node; When a higher-order dependency is determined, the degree of difference in the conditional transition distribution of each historical sequence segment at adjacent orders is statistically determined, and the maximum effective order at which each sequence segment is allowed to be expanded is adaptively determined. Extract the conditional transition rules that are statistically determined and retained at each effective order, and use them as the higher-order dependency rules.

3. The method for trajectory flow node community partitioning and role recognition based on dependency hypergraph as described in claim 1, characterized in that, When assigning weights to each hyperedge to represent the strength of dependency, the weights of each hyperedge are determined according to the following formula: ； In the above formula, For the weight of the superedge, The reliability weight of the historical context in the higher-order dependency rule. This is used to estimate the conditional transition probability of the target node under the given historical context.

4. The method for trajectory flow node community partitioning and role recognition based on dependency hypergraph as described in claim 1, characterized in that, Community detection is performed on the dependency hypergraph to obtain the community partitioning results of each physical node, including: Perform a degree-preserving projection on the dependent hypergraph to obtain a weighted projection graph; Community division is performed on the weighted projection map to obtain the first community division result of each physical node, and the first community division result is used as the final community division result.

5. The method for trajectory flow node community partitioning and role recognition based on dependency hypergraph as described in claim 4, characterized in that, The community detection process for the dependency hypergraph, which yields the community partitioning results for each physical node, also includes: Based on the first community partitioning result, the weights of each hyperedge in the dependency hypergraph are iteratively reweighted. After each iteration, the node degree-preserving projection and community partitioning are performed again based on the updated hyperedge weights until the preset convergence condition is met, and the second community partitioning result to which each physical node belongs is obtained. The second community partitioning result is taken as the final community partitioning result.

6. The method for trajectory flow node community partitioning and role recognition based on dependency hypergraph as described in claim 5, characterized in that, Iterative reweighting of the weights of each hyperedge in the dependency hypergraph includes: In each iteration, the community label of each node within a hyperedge is determined based on the current community division result, and the homogeneity of each hyperedge is calculated. The homogeneity is the maximum percentage of nodes within a hyperedge belonging to the same community. Calculate the mean homogeneity of all superedges in the current round, weight superedges with homogeneity higher than the mean, subtract superedges with homogeneity lower than the mean, and perform global normalization on the weights of all superedges after reweighting.

7. The method for trajectory flow node community partitioning and role recognition based on dependency hypergraph according to claim 1, characterized in that, Based on the community division results, the coreity score within the community and the cross-community connectivity dispersion score of each physical node are calculated, including: Project the dependency hypergraph into a weighted graph. Based on the community partitioning results, calculate the intra-community core score and cross-community connectivity dispersion score of each physical node on the weighted graph. Alternatively, traverse each hyperedge in the dependency hypergraph, calculate the hyperedge-level core score and hyperedge-level dispersion score of each hyperedge based on the community partitioning results, and then for each physical node, weight and aggregate the hyperedge-level scores of all hyperedges in which the physical node participates, using the hyperedge weight as the weight, to obtain the intra-community core score and cross-community connectivity dispersion score of the corresponding physical node.

8. The method for trajectory flow node community partitioning and role recognition based on dependency hypergraph according to claim 1, characterized in that, Based on the coreness score within the community and the cross-community connectivity dispersion score, the community role of each physical node is determined, including: For each community, physical nodes whose coreness score exceeds the first preset threshold are identified as community kernel group nodes; For each community, physical nodes whose cross-community connectivity dispersion score exceeds the second preset threshold are identified as cross-community connectivity nodes; The fusion role score of each physical node is calculated by multiplying the core score within the community and the cross-community connection dispersion score. Physical nodes whose fusion role scores exceed a third preset threshold are identified as connection-type hub nodes that have both community core group and cross-community connection functions.

9. The method for trajectory flow node community partitioning and role recognition based on dependency hypergraph according to any one of claims 1-8, characterized in that, The method further includes: Based on the community division results, the flow retention degree of each community is calculated, which is the conditional probability that a person remains in the community after a random walk from the community; and / or, based on the community division results, the higher-order boundary sharpness of each community is calculated, which is determined according to the ratio of the amount of dependency motif cut across the community boundary to the volume of dependency motif within the community. The quality of the community division results is evaluated based on the flow retention and / or the higher-order boundary clarity.

10. A device for trajectory flow node community partitioning and role recognition based on dependency hypergraph, characterized in that, The device includes: The data acquisition module is used to acquire raw trajectory flow data, which includes multiple sequence samples composed of physical nodes; The dependency module is used to extract statistically supported higher-order dependency rules from the trajectory flow data, and to treat each higher-order dependency rule as a dependency module, which consists of a historical context node and a target node. The dependency hypergraph construction module is used to construct a corresponding hyperedge based on all context nodes and target nodes involved in each dependency module, construct a dependency hypergraph based on the physical nodes and hyperedges, and assign a weight representing the dependency strength to each hyperedge. The community partitioning module is used to perform community detection on the dependency hypergraph and obtain the community partitioning results of each physical node. The node role score calculation module is used to calculate the community coreness score and cross-community connection dispersion score of each physical node based on the community division results. The community role determination module is used to determine the community role of each physical node based on the coreness score within the community and the cross-community connection dispersion score, thereby realizing the community division and role recognition of trajectory flow nodes.

Images