Community analysis method, apparatus, and electronic device

Abstract

The application discloses a community analysis method and device and electronic equipment. The method comprises the following steps: receiving time series data containing entity relationships, and dividing the time series data into multiple time slice time series data according to a preset step length; determining multiple knowledge graphs corresponding to multiple time slices respectively according to the multiple time slice time series data, and storing the knowledge graphs as graph snapshots to obtain a graph sequence; determining a community evolution event log and an entity migration path of each entity according to at least the graph sequence, wherein the community evolution event log is used to indicate the life cycle of all communities involved in the graph sequence; determining graph features according to at least the community evolution event log and the entity migration path of each entity; and generating an analysis report according to the graph features. The application solves the technical problems that the community analysis method in the related art cannot capture the dynamic changes of community structures over time and cannot effectively quantify the connection value of entities among different communities.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, in particular, to a community analysis method and device and electronic equipment. BACKGROUND

[0002] The community analysis system based on graph retrieval-augmented generation (GraphRAG) has become a core tool in financial risk control, technology intelligence and other scenarios. However, the community analysis method of such systems generally adopts a static analysis paradigm, that is, only analyzing a graph snapshot at a fixed time point, which leads to the following key defects: unable to capture the time sequence evolution of community structure, that is, the dynamic change of community structure over time, limited to analysis within the community, and lacking effective means to detect relationships such as cooperation, competition or integration between communities, and unable to effectively quantify the connection value and influence of entities between different communities.

[0003] At present, there is no effective solution to the above problems. SUMMARY

[0004] The embodiments of the present application provide a community analysis method, device and electronic equipment to at least solve the technical problems that the community analysis method in the related art cannot capture the dynamic change of community structure over time and cannot effectively quantify the connection value of entities between different communities.

[0005] According to an aspect of an embodiment of the present application, a community analysis method is provided, comprising: receiving time series data containing entity relationships, and dividing the time series data into multiple time slice time series data according to a preset step size; determining multiple knowledge graphs corresponding to multiple time slices respectively according to the multiple time slice time series data, and storing the knowledge graphs as graph snapshots to obtain a graph sequence, wherein one community in each knowledge graph corresponds to a group of associated entities and relationships between the entities, the graph sequence contains multiple graph snapshots, and each graph snapshot corresponds to a time slice; determining a community evolution event log and an entity migration path of each entity at least according to the graph sequence, wherein the community evolution event log is used to indicate the life cycle of all communities involved in the graph sequence; determining graph features at least according to the community evolution event log and the entity migration path of each entity, wherein the graph features are used to reflect the entity migration rule and the community evolution rule; and generating an analysis report according to the graph features, wherein the analysis report is used to at least indicate the reason for the evolution of each community.

[0006] According to some embodiments of the present application, the knowledge graph corresponding to each time slice is determined according to the time sequence data of the plurality of time slices, comprising: for each time slice, mapping the entities and the relationships between the entities in the time sequence data containing entity relationships into a knowledge graph to obtain the knowledge graph corresponding to each time slice, wherein the knowledge graph at least contains one community; storing the knowledge graph as a graph snapshot to obtain a graph sequence, comprising: storing the knowledge graph corresponding to each time slice as a graph snapshot, wherein the snapshot metadata of the graph snapshot at least contains the time stamp of the snapshot establishment, the number of entities, the relationship density, and the community base; sorting all graph snapshots in the order of time slice from later to earlier to obtain the graph sequence.

[0007] According to some embodiments of the present application, the community evolution event log is determined at least according to the graph sequence, comprising: identifying the evolution event of each community in each graph snapshot in the graph sequence, wherein the evolution event at least includes the birth, merger, split, and extinction of the community; determining the life cycle of each community according to the community identification and the evolution event in the order of time slice from earlier to later, and determining the community evolution event log according to the evolution event information in the life cycle of all communities.

[0008] According to some embodiments of the present application, the entity migration path of each entity is determined by: identifying the community attribution information of each entity in each graph snapshot in the graph sequence, wherein the same entity has only one unique entity identification in different graph snapshots; determining the entity migration path of each entity according to the community attribution information of each entity in the order of time slice from earlier to later, wherein the entity migration path is used to indicate the community migration of the entity from the first time slice to the last time slice.

[0009] According to some embodiments of the present application, the graph features are determined at least according to the community evolution event log and the entity migration path of each entity, comprising: determining the centrality distribution of each community in the graph snapshot of each time slice, wherein the centrality distribution of each community is determined by the centrality of each node in the community; determining the relationship entropy value of each community in the graph snapshot of each time slice, wherein the relationship entropy value is the entropy value of the relationship type between the entities in the community; determining the community centrality distribution and the relationship entropy value distribution at least according to the centrality distribution of each community in the graph snapshot of each time slice and the relationship entropy value of each community in the graph snapshot of each time slice, wherein the community centrality distribution at least reflects the trend of the centrality distribution of each community over time, and the relationship entropy value distribution reflects the trend of the relationship entropy value of each community over time; determining the structured summary, the community centrality distribution and the relationship entropy value distribution, the community evolution event log, and the entity migration path of each entity as the graph features, wherein the structured summary is at least used to summarize and present the community evolution process.

[0010] According to some embodiments of the present application, the structured summary is determined by: determining an entity cross-community connection value score of each entity in each time slice in the graph sequence, determining an average value of the entity cross-community connection value scores of each entity in all time slices as a value score of the entity, wherein the entity cross-community connection value score is determined at least according to the number of edges within the community of the entity, the size of the community associated with the entity, and the number of different communities associated with the entity; determining all entities with a value score greater than a preset threshold as super hub entities, grouping all super hub entities into a super hub entity list, and determining the communities associated with all super hub entities in the super hub entity list as target communities; and determining the entities and relationships between entities in all target communities as associated community features; and determining the structured summary at least according to the super hub entity list and the associated community features.

[0011] According to some embodiments of the present application, the analysis report is generated according to the graph features, including: performing deterministic analysis on the graph features by a rule engine to obtain a decision label; and performing analysis on the graph features and the decision label by a semantic engine to obtain an analysis report, wherein the semantic engine is configured to perform semantic interpretation on the graph features and the decision label and convert the graph features and the decision label into the analysis report.

[0012] According to another aspect of the embodiments of the present application, a community analysis apparatus is also provided, including: a division module configured to receive time series data containing entity relationships, and divide the time series data into time series data of multiple time slices according to a preset step size; a first determination module configured to determine multiple knowledge graphs corresponding to the multiple time slices respectively according to the time series data of the multiple time slices, and store the knowledge graphs as graph snapshots to obtain a graph sequence, wherein one community in each knowledge graph corresponds to a group of associated entities and relationships between the entities, and the graph sequence contains multiple graph snapshots, each graph snapshot corresponding to one time slice; a second determination module configured to determine a community evolution event log and an entity migration path of each entity at least according to the graph sequence, wherein the community evolution event log is configured to indicate the life cycle of all communities involved in the graph sequence; a third determination module configured to determine graph features at least according to the community evolution event log and the entity migration path of each entity, wherein the graph features are configured to reflect the entity migration law and the community evolution law; and a generation module configured to generate an analysis report according to the graph features, wherein the analysis report is configured to indicate at least the reason for the evolution of each community.

[0013] According to another aspect of the embodiments of the present application, a non-volatile storage medium is also provided, which stores a program, wherein when the program is running, the device in which the non-volatile storage medium is located is controlled to perform the above community analysis method.

[0014] According to a further aspect of the embodiments of the present application, an electronic device is also provided, comprising a memory and a processor, the processor being configured to execute a program stored in the memory, wherein the program performs the above community analysis method when executed.

[0015] According to a further aspect of the embodiments of the present application, a computer program product is also provided, comprising computer instructions, which, when executed by a processor, implement the above community analysis method.

[0016] In the embodiments of the present application, time series data containing entity relationships is received, and the time series data is divided into time series data of multiple time slices according to a preset time length; knowledge graphs corresponding to the multiple time slices are determined according to the time series data of the multiple time slices, and the knowledge graphs are stored as graph snapshots to obtain a graph sequence, wherein one community in each knowledge graph corresponds to a group of entities having correlations and relationships between the entities, the graph sequence contains multiple graph snapshots, and each graph snapshot corresponds to a time slice; a community evolution event log and an entity migration path of each entity are determined at least according to the graph sequence, the community evolution event log is used to indicate the life cycle of all communities involved in the graph sequence; graph features are determined at least according to the community evolution event log and the entity migration path of each entity, wherein the graph features are used to reflect entity migration rules and community evolution rules; and an analysis report is generated according to the graph features, wherein the analysis report is used to at least indicate the reasons for the evolution of each community. Through the determination of the community evolution event log and the entity migration path of each entity from the graph sequence, the dynamic changes of community structures over time are captured, the graph features are determined based on the community evolution event log and the entity migration path of each entity, the graph features effectively quantify the connection values of entities between different communities, and finally the analysis report is generated, thereby solving the technical problems in the related art that the community analysis method cannot capture the dynamic changes of community structures over time and cannot effectively quantify the connection values of entities between different communities. BRIEF DESCRIPTION OF DRAWINGS

[0017] The accompanying drawings, which are included to provide a further understanding of the present application and constitute a part of this application, illustrate certain illustrative embodiments of the present application and together with the description serve to explain the present application. In the drawings:

[0018] Figure 1 FIG. 1 is a hardware structure block diagram of a computer terminal for implementing a community analysis method according to an embodiment of the present application;

[0019] Figure 2 FIG. 2 is a flowchart of a community analysis method according to an embodiment of the present application;

[0020] Figure 3is a system architecture diagram provided by an embodiment of the present application;

[0021] Figure 4 is a dynamic community evolution flowchart provided by an embodiment of the present application;

[0022] Figure 5 is a cross-community relationship mining schematic diagram provided by an embodiment of the present application;

[0023] Figure 6 is a decision engine interaction logic diagram provided by an embodiment of the present application;

[0024] Figure 7 is a structural schematic diagram of a community analysis device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0025] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0026] The information collected by the embodiments of the present application is information and data authorized by the user or authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of related data comply with relevant laws, regulations and standards in the relevant region, necessary security measures are taken, public order is not violated, and appropriate operation entrances are provided for the user to choose authorization or refuse automatic decision results; if the user chooses to refuse, the expert decision process is entered.

[0027] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily have to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device that includes a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0028] In order to better understand the embodiments of the present application, the technical terms involved in the embodiments of the present application are explained as follows:

[0029] Graph Retrieval-Augmented Generation (GraphRAG): A technique that combines knowledge graph retrieval with generative AI to enhance the semantic understanding and data correlation capabilities of community analysis. It retrieves relevant information from the knowledge graph to provide a richer context for the generative model, enabling community analysis to more accurately capture the semantic relationships between data.

[0030] Community mining algorithm: An algorithm used to discover community structure in complex networks, which can divide the nodes in the network into different communities, making the connections within the community more dense and the connections between communities more sparse.

[0031] Large model technology: Large model technology is an important breakthrough in the field of artificial intelligence in recent years, mainly relying on deep learning and large-scale data training to build models with strong language understanding and generation capabilities. The current mainstream large models are usually based on the Transformer architecture, trained through two stages of pre-training and fine-tuning. The pre-training stage uses massive text data to learn language patterns and semantic relationships, and the fine-tuning stage optimizes for specific tasks to improve the practicality of the model. The large model's generation summary ability lies in its ability to extract key information from a large amount of text and summarize it in a concise and accurate manner. This capability is widely used in text summarization, information retrieval, intelligent customer service, and other fields.

[0032] In related technologies, community analysis methods generally use a static analysis paradigm, i.e., only analyzing a snapshot of the graph at a fixed time point. Therefore, there is a technical problem that the community analysis method in related technologies cannot capture the dynamic changes of community structure over time and cannot effectively quantify the connection value of entities between different communities. To solve this problem, the related solutions provided in the embodiments of the present application are described in detail below.

[0033] According to the embodiments of the present application, an embodiment of a community analysis method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0034] The method embodiments provided by the embodiments of the present application can be executed in a computer terminal or similar computing device. Figure 1 A hardware structure block diagram of a computer terminal for implementing the community analysis method is shown. As Figure 1As shown, the computer terminal 10 may include one or more processors 102 (shown as 102a, 102b, ..., 102n in the figure) 102 (processor 102 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0035] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10. As involved in the embodiments of this application, the data processing circuits serve as processor control (e.g., selection of a variable resistor termination path connected to an interface).

[0036] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the community analysis method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned community analysis method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0037] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0038] The display can be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10.

[0039] In the above operating environment, this application provides an embodiment of a community analysis method. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Also, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0040] like Figure 2 The diagram shown is a flowchart of a community analysis method provided according to an embodiment of this application, including:

[0041] Step S202: Receive time series data containing entity relationships, and divide the time series data into multiple time slices according to a preset step size.

[0042] In the technical solution provided in step S202, the time-series data containing entity relationships records a dataset of entities, relationships between entities, and information about their changes over time within a preset time interval. An entity can be any identifiable object, such as a person, organization, product, location, etc., while entity relationships are the connections or interactions between these objects. The time-series data containing entity relationships can originate from various scenarios across different industries, such as communication records, scientific collaborative publications, and supply chain transaction records. In financial transaction scenarios, entities can be different company or individual accounts, while entity relationships are reflected in transaction activities between accounts. The time-series data records the time, parties involved, and transaction amount of each transaction. In social media analysis, entities can be users, and entity relationships are the interactions between users, such as comments, likes, and shares. The time-series data records entities, relationships between entities, and the time, type, and user IDs of the interacting parties. For technology intelligence monitoring, entities can be research institutions, researchers, or patents, while entity relationships are research collaborations, patent citations, or technological associations.

[0043] In some embodiments of the present application, after receiving the time series data containing entity relationships, the time series data is divided into multiple time slice time series data according to a preset step length using a time slicing mechanism (i.e., a method for capturing the phased change characteristics of network structure by dividing the continuous time axis into equal length intervals (such as by month / week) and constructing an independent knowledge graph snapshot at each interval endpoint). For example, for the time series data containing entity relationships in the preset time interval [T_start, T_end] (T_start and T_end represent the start time point and the end time point respectively), the time interval (i.e., time slice) is divided according to the preset time slicing step length Δt (i.e., the above-mentioned preset step length), and multiple time intervals are obtained. For each time interval (i represents the serial number of any one time interval), for example, if the time series data lasts from January to December, and Δt is set to one month, 12 time slices will be created, each representing the time series data of one month.

[0044] Step S204, determining the knowledge graph corresponding to each of the multiple time slices according to the time series data of the multiple time slices, and storing the knowledge graph as a graph snapshot to obtain a graph sequence.

[0045] In the technical solution provided in step S204, a community in each knowledge graph corresponds to a group of associated entities and the relationships between the entities, and the graph sequence contains multiple graph snapshots, each of which corresponds to a time slice.

[0046] There are multiple implementation methods for determining the knowledge graph corresponding to each of the multiple time slices according to the time series data of the multiple time slices, for example: for each time slice, mapping the entities and the relationships between the entities in the time series data containing entity relationships to a knowledge graph to obtain the knowledge graph corresponding to each time slice, wherein the knowledge graph contains at least one community; there are multiple implementation methods for storing the knowledge graph as a graph snapshot to obtain a graph sequence, for example: storing the knowledge graph corresponding to each time slice as a graph snapshot, wherein the snapshot metadata of the graph snapshot contains at least the time stamp of the snapshot establishment, the number of entities, the relationship density, and the community base; sorting all graph snapshots in the order of time slice from late to early to obtain a graph sequence.

[0047] For each time slice, when performing entity recognition and relation extraction on the corresponding time series data, entities (such as individuals, companies, etc.) are represented as nodes in the knowledge graph, and the relationships between entities (such as transactions, references, cooperation, etc.) are represented as edges in the knowledge graph (nodes (also known as entities) store detailed information such as entity identification, type, attributes, etc.; edges need to store the type of relationship, strength (such as transaction amount, reference times, or cooperation degree), etc., and after mapping, an initial knowledge graph describing the entity relationship network in the time slice is formed.

[0048] After obtaining the initial knowledge graph by mapping, community detection algorithms (such as modularity maximization (Louvain method, Louvain optimizes the modularity function to identify communities, and modularity is an index to measure the quality of community structure, which reflects the difference between the dense connections within the community and the sparse connections between communities), label propagation (Label Propagation, Label Propagation method determines community attribution through the propagation of labels between entities, each entity has a unique label at the beginning, through iteration and update, entities tend to adopt the label of the most closely connected entity, and finally reach a stable state to form a community) can be used to analyze the community structure of the initial knowledge graph of each time slice, and a group of entities with close connections can be divided, and a group of entities belongs to a community. A community is a relatively independent subnetwork in a knowledge graph composed of closely connected entities (connected by direct edges or through multi-hop paths) and their relationships. Each community collects a group of entities with mutual connectivity and their relationships. These entities may belong to different types, but they form a relatively independent subnetwork. Entities within a community have the same attributes or characteristics, and the relationships between entities reflect collaborative behavior, such as jointly completing a project. For example, a school community contains entities (nodes in the knowledge graph) related to school staff and students.

[0049] Then, using graph database technology (for example, NebulaGraph, a high-performance, distributed, and easy-to-scale graph database), the knowledge graph of each time slice is stored as a graph snapshot, including the knowledge graph and its community division results. The metadata of each graph snapshot includes at least the timestamp of the snapshot establishment, the number of entities, the relationship density (the richness of the relationship between entities, calculated as the ratio of the number of edges in the graph to the number of nodes (entities)), and the community base (the total number of communities in the knowledge graph). All graph snapshots are sorted to ensure that they are arranged in chronological order, forming a graph sequence that can track the changes in communities over time.

[0050] Step S206, at least according to the graph sequence, determine the community evolution event log and the entity migration path of each entity.

[0051] In the technical solution provided in step S206, the community evolution event log is used to indicate the life cycle of all communities involved in the graph sequence. There are various implementation manners for determining the community evolution event log according to at least the graph sequence, for example: identifying the evolution event of each community in each graph snapshot in the graph sequence, wherein the evolution event at least includes the birth, merger, split, and death of the community; determining the life cycle of each community according to at least the community identity and the evolution event in chronological order according to time slices, and determining the community evolution event log according to the evolution event information in the life cycle of all communities.

[0052] In the above steps, the entity migration path of each entity is determined by: identifying the community affiliation information of each entity in each graph snapshot in the graph sequence, wherein the same entity has only one unique entity identifier in different graph snapshots; and determining the entity migration path of each entity according to at least the community affiliation information of each entity in chronological order from the first to the last time slice, wherein the entity migration path is used to indicate the community migration of the entity from the first time slice to the last time slice.

[0053] In some embodiments of the present application, when identifying the evolution event of each community in each graph snapshot in the graph sequence, the graph snapshots of adjacent time slices are compared to identify the evolution event of each community (the evolution of the community refers to the change of the community structure). Specifically, it includes: birth: a community that appears for the first time in the current time slice. Merger: two or more communities merge into one community in the subsequent time slice. Split: a community splits into two or more communities in the subsequent time slice. Death: a community that exists in the current time slice no longer appears in the subsequent time slice, indicating that the community has been dissolved or completely absorbed by other communities. In chronological order from the first to the last time slice, the community identifiers are aligned, which means that in the graph snapshots of different time slices, the community identifiers of those that are actually the same community are matched and standardized. Thus, all evolution events are associated to obtain the life cycle of each community, and the occurrence time stamp, the affected community identifier (for example, community ID), and the change type of all community evolution events (birth, merger, split, and death) in the entire life cycle of each community are recorded as the evolution event information of the community life cycle. The evolution event information of all communities is determined as the community evolution event log.

[0054] For each time-slice snapshot of the graph, the entity identifier (e.g., entity ID) and its associated community identifier are recorded as the entity's community affiliation information. By comparing the community affiliation information of entities in adjacent time slices, the time point and path of an entity's migration from one community to another are determined. If the entity's associated community identifier changes compared to the previous time slice, the entity is considered to have migrated. After obtaining the community affiliation information of each entity in each graph snapshot, entity identifiers are aligned according to the order of time slices from first to last, using the entity identifier as the association. From the community affiliation information of the same entity, the changes in its community affiliation from the first time slice to the last time slice in the graph sequence are recorded as the entity migration path. The entity migration path includes the migration record (entity identifier, community identifier, and migration time, etc.) for each community migration of the entity. For example, an improved FacetNet algorithm can be used to process the graph sequence, perform dynamic community detection, dynamically track the community structure evolution path, output a community evolution event log and the migration path of each entity, and the improved FacetNet algorithm is a time-smooth community detection method based on a probabilistic model, with the objective function being:

[0055] ,

[0056] Where T corresponds to the last time slice. for The adjacency matrix at time 10:00. Community affiliation matrix For time series smoothing coefficients, To reconstruct the loss term, , Represents nodes (i.e., entities). and The actual connection strength, where n is the total number of entities. For time smoothing term, ,in, Represents a node In time slice Belongs to the community The probability, Represents a node In time slice t-1, the community is involved. The probability of belonging to a community is given by the algorithm, where K is the total number of communities. This algorithm unifies network connection strength and community membership probability into the same metric space for analysis, thereby obtaining community evolution event logs and migration paths for each entity. The community affiliation matrix contains the probability of each entity (i.e., node) belonging to a certain community, which can be determined by the following formula:

[0057] ,

[0058] in, Indicates time slice middle node belongs to community probability, time slice any one node (also referred to as entity), denotes a set of nodes directly connected to node u (node u is a node different from node denotes a set of nodes directly connected to node u (node u is a node different from node denotes a community identifier of any one community in a time slice denotes a community identifier of any one community in a time slice denotes the connection strength of node , denotes the connection strength of node denotes all communities traversed when summing.

[0059] In order to intuitively show the process of community dynamic evolution and entity migration, heat maps and Sankey diagrams are used for visualization. For example, a heat map is used to demonstrate the community size at different times, and a Sankey diagram is used to demonstrate the entity migration flow from a source community to a target community (the target community is the community after dynamic change is completed every other time slice). The X-axis of the heat map represents different time points (time slices) in the time sequence, and the Y-axis represents the unique community identifier (community ID) of each community. The color depth of the heat map is directly related to the community size (i.e., the number of nodes in the community). The deeper the color, the larger the community size, and vice versa. The left and right columns of the Sankey diagram represent the community state under different time slices. Each flow line represents the migration of entities from one community to another, and the width of the flow line maps the number of entity migrations, i.e., the amount of entity migration. The wider the width, the more entities migrate. The Sankey diagram can intuitively show how entities transfer between different communities over time and how these migration activities affect the structure and relevance of the communities. Both the heat map and the Sankey diagram contain a time axis dragging function. Users can quickly view the graph snapshot details of the knowledge graph at any time point by simply dragging the time axis slider.

[0060] Step S208, determining the graph features according to at least the community evolution event log and the entity migration path of each entity.

[0061] In the technical solution provided in step S208, the graph features are used to reflect the entity migration law and the community evolution law. There are various implementation manners of determining the graph features according to at least the community evolution event log and the entity migration path of each entity, for example: determining the centrality distribution of each community in the graph snapshot of each time slice, wherein the centrality distribution of each community is determined by the centrality of each node in the community; determining the relationship entropy value of each community in the graph snapshot of each time slice, wherein the relationship entropy value is the entropy value of the relationship type between entities in the community; determining the community centrality distribution and the relationship entropy value distribution according to at least the centrality distribution of each community in the graph snapshot of each time slice, the relationship entropy value of each community in the graph snapshot of each time slice, wherein the community centrality distribution is at least used to reflect the change trend of the centrality distribution of each community over time, and the relationship entropy value distribution is used to reflect the change trend of the relationship entropy value of each community over time; determining the structured summary, the community centrality distribution and the relationship entropy value distribution, the community evolution event log, and the entity migration path of each entity as the graph features, wherein the structured summary is at least used to summarize and present the community evolution process.

[0062] The above step calculates the centrality distribution of each community c in the graph snapshot KG_t of each time slice t. The centrality distribution reflects the importance of each node (node is the entity mentioned above) in the community, for example, the centrality of each node can be determined by calculating the degree centrality. By calculating the centrality of the node, it can be determined which nodes are in the core position in the community and have an important influence on information dissemination and community structure stability. For each node u in each community c, after calculating the centrality of each node, the centrality distribution of each community is calculated, that is, the distribution of the value of the centrality of each node (also called entity) in each community, forming the centrality distribution table of the community at time slice (t), recording the number of entities in each centrality value interval. The above operation is performed on all time slices in the graph sequence, and the centrality distribution data of each community at different time points is collected, and a time series graph of community centrality distribution containing all communities is drawn, such as a histogram or a density curve, to obtain the community centrality distribution, and to visually show the change trend of the centrality distribution of each community over time.

[0063] For each graph snapshot KG_t in each time slice t, the relationship entropy value of each community c (c refers to any one community) is calculated. The relationship entropy value reflects the diversity or uncertainty of the relationship types within the community. In the community network, the higher the relationship entropy value, the more complex and diverse the relationship types within the community, and vice versa. The calculation of the relationship entropy value is based on the frequency of each relationship type within the community. In the graph snapshot of each time slice, first, all relationship types within the community are identified, then the frequency of each relationship type within the community is counted, and the total number of edges in the community is counted. The number of edges of each relationship type is divided by the total number of edges in the community to obtain the probability of each relationship type within the community. Finally, the relationship entropy value of each community is determined according to the probability of each relationship type within the community. The relationship entropy value of each time slice (t) community (c) is recorded to form a complete entropy value time sequence. The entropy value time sequence of all communities is plotted against time, and key time points such as the time when the entropy value fluctuates significantly are identified to obtain the relationship entropy value distribution.

[0064] Finally, the structured summary, the community centrality distribution, the relationship entropy distribution, the community evolution event log, and the entity migration path of each entity are determined as the graph features, which include the above-mentioned multi-dimensional data.

[0065] In the above steps, the structured summary can be realized by the following method: determining the entity cross-community connection value score of each entity in each time slice in the graph sequence, determining the average value of the entity cross-community connection value score of each entity in all time slices as the value score of the entity, wherein the entity cross-community connection value score is determined at least according to the number of edges of the entity within the community, the size of the community associated with the entity, and the number of different communities associated with the entity; determining all entities with a value score greater than a preset threshold as super hub entities, grouping all super hub entities into a super hub entity list, and determining the communities associated with all super hub entities in the super hub entity list as target communities; determining all entities and relationships between entities in all target communities as associated community features; and determining the structured summary based on at least the super hub entity list and the associated community features.

[0066] In some embodiments of the present application, for each entity, the entity cross-community connection value score (used to quantify the entity cross-community connection value) is calculated once in each time slice in the graph sequence, and then the average value of the entity cross-community connection value score of each entity in all time slices is determined as the value score of the entity. For example, the entity cross-community connection value score of any entity in any time slice can be determined by the following formula of the Bridge_Score model:

[0067] ​

[0068] wherein, represents the entity cross-community connection value score of any one entity e in any one time slice, represents the number of edges of entity e connecting different communities (i.e. the number of different communities associated with the entity), represents the number of edges of entity e connecting with other entities within the community to which it belongs (i.e. the number of edges of the entity within the community), , represents the size of the community (the number of entities in the community), i.e. the size of the community associated with the entity, represents any two communities connected to entity e.

[0069] After calculating the value score of each entity, all entities with a value score greater than a preset threshold are determined as super hub entities, all super hub entities are grouped into a super hub entity list, and all communities associated with the super hub entities in the super hub entity list are determined as target communities (the target communities include the communities directly or indirectly associated with the super hub entities in all time slices). The entities and relationships between the entities in all target communities (i.e. the entities and the edges connecting the entities in the knowledge graph indicated by the graph snapshot) are determined as community features. At least according to the super hub entity list and the associated community features, a structured summary is determined, for example, the associated community features are input into a trained large language model (Large Language Model, LLM for short), the associated community features are analyzed by the LLM, and a structured summary is automatically generated by combining a preset summary generation template. The summary generation template includes but is not limited to the following contents: the name, type and attribute of the super hub entity; the super hub entity plays a connecting role between which communities, and its importance in cross-community connection, exploration of inter-community collaboration opportunities: such as technology fusion, resource complementation; prediction of potential risks (prediction of cross-community events in future multiple periods): such as risk transmission path, conflict probability; based on community dynamics and super hub entity role, provide action suggestions. The structured summary is at least used to summarize and present the community evolution process description, and the structured summary includes the community evolution process description, and can further include trend prediction, decision prediction, risk warning and cooperation suggestion, and other analysis suggestions and summarized collaboration opportunities, such as cross-community technology fusion, resource complementation scheme, risk transmission, and inter-community crisis transmission path warning. The structured summary can be a cross-community relationship summary, and the intelligence of the cross-community relationship summary generated by the LLM driven strategy is improved, for example, an executable cooperation scheme can also be automatically output, which greatly reduces the cost of manual analysis.

[0070] Step S210, generating an analysis report according to the graph features.

[0071] In the technical solution provided in step S210, the analysis report is used to indicate at least the reason for the evolution of each community. There are various implementation manners for generating the analysis report based on the graph features, for example: deterministic analysis of the graph features by a rule engine to obtain a decision label; analysis of the graph features and the decision label by a semantic engine to obtain an analysis report, wherein the semantic engine is used to semantically interpret the graph features and the decision label and convert them into the analysis report.

[0072] The rule engine is used to perform rule-based logical reasoning and decision-making. The rule engine serves as a bridge between the graph features and the decision label, and performs deterministic analysis (deterministic analysis refers to evaluating the structural changes of the community and the behavior patterns of the super hub entity by parsing and applying pre-defined business rules, to ensure that the analysis result conforms to the business logic). For example, the rule engine is used to perform deterministic analysis on the graph features to identify the centrality distribution of each community in different time slices, identify entities with declining or rising centrality, and judge the stability of the community structure. A community size fluctuation threshold is set, for example, if the community size changes by more than 25% in two consecutive time slices, it is determined to be a "size mutation". The rule engine is used to detect whether there is a significant change in the community connected to the target entity, such as a significant decrease or increase in the number of connected communities, and finally obtain a decision label. The decision label is a standardized decision information generated after the rule engine analyzes the graph features. It summarizes the key findings and decision points in community analysis in the form of a label, such as "high-risk migration", "loss of key nodes", "technology fusion opportunity", "centrality decline", "size mutation", etc. The rule engine can perform a matching process between the graph features and pre-defined business rules, and once the data in the graph features meet the rule conditions, the corresponding decision label is generated. The LLM+rule engine dual-drive decision-making enhances the explainability of the decision-making, generates an analysis report by combining deterministic rules and semantic reasoning, and supports interactive verification of the graph-chart bidirectionally. Through the method of the embodiment of the present application, a three-level intelligent support closed loop of "describing phenomena → attribution analysis → predicting decisions (also known as describing-attribution-predicting)" is finally achieved. The community structured summary generated by the LLM in the related art lacks rule engine verification and has the risk of factual errors (such as misjudging competitive relationships as cooperative); it cannot support the three-level decision chain of "describing-attribution-predicting", especially lacking the early warning ability based on community evolution (such as a prediction accuracy of less than 30% six months before the risk cluster spreads).

[0073] The semantic engine (e.g., an LLM semantic engine, which can be a component built on LLM for processing and parsing the deep semantic meaning of text or data) receives graph features and decision labels. It then uses deep semantic understanding techniques (e.g., natural language processing and knowledge graph parsing methods) to analyze the graph features behind each decision label, understanding their specific meaning and potential impact in community evolution. This results in attribution analysis, which explains the root causes of community changes (i.e., the reasons for the evolution of each community), such as "a sudden change in community size" due to "a weakening of the bridging effect of the super-hub entity." The semantic engine can also combine historical evolution paths and the current community state to predict future trends and risks, as well as recommended decisions. For example, "It is expected that the activity of community X will further decline within the next three months, with potential risks including..., and it is recommended to take... measures." The semantic engine fills the attribution analysis, predictions of future trends and risks, and recommended decisions into the corresponding positions according to a template structure to obtain an analysis report, which is a natural language report. The template can include multiple parts such as a title, summary, attribution analysis, trend prediction, and action recommendations. To enhance the usability and interactivity of the analysis report, the semantic engine also embeds relevant visualization elements such as graph snapshot links, heatmaps, and Sankey diagrams. Users can click on these charts to jump to specific time points and views, observe the details of community changes and entity migrations indicated by decision tags, and further verify and understand the report's conclusions. Heatmaps and Sankey diagrams can be linked with the knowledge graph. For example, when a user clicks on community C in the Sankey diagram, all entities and relationship paths of community C in the knowledge graph will be automatically highlighted. When a user selects a super-hub entity E in the knowledge graph, the heatmap will highlight the curve showing the change in the size of the community where E is located. Real-time generation of related evidence chains is also achieved. Finally, the data from steps S202-S210 involving the entire process can be exported as a multimodal structured report (e.g., PDF / HTML format), with the report automatically embedding graph query links for result reproduction and verification.

[0074] Figure 3is a system architecture diagram provided by an embodiment of the present application. The intelligent visual community analysis system using the method of the embodiment of the present application is an integrated system with functions of dynamic graph construction, community evolution tracking, cross-domain relationship mining and visual decision support. Starting from the reception of time series data, a series of graph snapshots KGt (i.e. the above-mentioned graph snapshots) are generated through time series slicing processing, and then the dynamic community evolution analysis module performs dynamic community detection to identify community splitting / merging tracks (i.e. the process of determining the community evolution event log and the entity migration path of each entity), and then visualizes through the heat map and the Sankey diagram. In the cross-community relationship mining + interpretable decision engine module, the Bridge_Score (Bridge_Score refers to the value score) is calculated for the current graph KGt (which can refer to any graph snapshot), and then the super hub entity is screened, the LLM generates a relationship summary (i.e. the above-mentioned structured summary), and then performs community analysis (the centrality is calculated through the rule engine (i.e. the process of analyzing the centrality of the entities in the graph snapshot), and the risk rule refers to a series of risk assessment criteria preset in the rule engine, which is used to automatically detect and mark the risk situation occurring in the community analysis process), and then the LLM performs semantic interpretation, and finally generates decision insights (i.e. the above-mentioned analysis report), and finally supports multi-modal report generation (i.e. the above-mentioned multi-modal structured report).

[0075] The system dynamically tracks the community structure evolution path, quantifies the cross-community hub entity and predicts the cooperation opportunity, generates an interpretable decision suggestion and supports interactive report generation. It can be applied in various industries, for example, for the financial risk control scene, the time series data containing entity relationship is the transaction flow (timestamp / transaction party / amount) of a bank's enterprise customers in a year, the node = enterprise / person, and the edge = transaction frequency + fund flow direction. After dividing the data by time slicing, it is divided into 12 graph snapshots by month. It is found that in March, a large commodity transaction community (with a size of 120 nodes) is detected, and the value score of the super hub entity (enterprise X) is 8.7, connecting the mineral / logistics / financial community. The generated structured summary is "Enterprise X plays the role of fund transfer in the mineral-logistics chain, and it is recommended to monitor its cross-border transactions...". Through the rule engine, the monthly decrease of the logistics community centrality is 42%, and the decision label is marked as "funding chain risk". When the user clicks on the logistics community marked with "funding chain risk" on the heat map, all entities and their guarantee relationship network data related to the community are automatically displayed. The user can place the guarantee network graph in the risk report generation template through the drag-and-drop operation on the report template editing interface. The analysis report outputs all entities and their guarantee relationship network data related to the community X with the decision label "funding chain risk", displays the mutual guarantee situation of key enterprises in the logistics community, and the risk analysis text, including the risk cause, potential impact and recommended coping strategy.

[0076] For the technology intelligence monitoring scenario (e.g., the semiconductor field), the time series data containing entity relationships is collected from global patent databases and academic paper libraries for multiple years, identifying research institutions, enterprises, and technology keywords, a total of 120,000 nodes, obtaining cooperative applications, paper citations, and technology associations, a total of 450,000 edges. Six graph snapshots are generated according to the year. Key findings are obtained through community evolution event logs and entity migration paths of each entity: a single "lithography technology community" (with a size of 8200 nodes) appears in a certain year, which splits into an Extreme Ultraviolet Lithography (EUV) community and another lithography community (the sunburst chart shows that 32% of the entities migrate); a new quantum chip manufacturing community is added in a certain year (formed by 15% of the entities of the original community). The value of a certain entity (a certain school) is divided into 9.2, connecting the above three communities. The LLM generates collaborative suggestions: "The school plays a technical bridge role in the lithography-quantum field, and it is recommended to track the migration trend of its new mask technology patents to quantum computing." The rule engine identifies that when the technical similarity between communities > 0.7 and the patent growth of the hub node > 50 / year, the decision label is marked as a technology fusion opportunity. The semantic engine generates an analysis report covering future technology development directions, potential cooperation opportunities, and possible challenges. Users can click on the quantum chip manufacturing community in the heat map, and the heat map automatically highlights the school and its directly associated entities and relationships. Users can drag and drop the technology path diagram directly into the designated area of the analysis report generation template, and the LLM module receives the technology path diagram and related data, combined with community evolution analysis and hub entity role, the analysis report will include a natural language description of the technology roadmap prediction.

[0077] For supply chain risk early warning (e.g., the automotive manufacturing industry), data such as vehicle manufacturers' capacity utilization, module supplier-geographical distribution, and raw material supplier-alternative supplier quantity are obtained from global supplier databases and logistics information flows, and graph snapshots are generated according to the quarter. Key findings include the formation of a stable "lithium battery supply chain community" (including 120 suppliers) in a certain quarter of a certain year, a sudden split event (the sunburst chart shows that 35% of the nodes flow to a new community) in a certain quarter of a certain year, and finally the community reorganization caused by a lithium mine sanction. The value score of an identified entity (a logistics enterprise) is 7.8, connecting four regional sub-communities, with a central degree drop of 62%, triggering a risk scan. The LLM generates a report summary: "Enterprise Y bears the key transportation, and its route disruption will lead to: <15 days of inventory at certain electric chip factories; alternative path cost increase of 40%." The rule engine identifies that when the hub node degree drops by > 60% and the community size fluctuates by > 25%, the analysis report will provide an emergency plan. At this time, the user clicks on the supply chain sunburst chart broken link, which automatically highlights the affected vehicle manufacturers and alternative paths.

[0078] Figure 4 This application provides a dynamic community evolution flowchart, illustrating a dynamic community evolution process, including time-series slice generation, community identifier alignment algorithm, Sankey diagram to demonstrate community merging and entity migration, and heatmap to demonstrate community identifiers, time slices, and community size. First, time-series slice generation is performed. For time-series data containing entity relationships within the interval [T1, Tn] (T1 and Tn represent the start and end times respectively), time slices are obtained by dividing the data into time slices according to a preset step size Δt. Then, the data of each time slice is mapped to a knowledge graph (…). The knowledge graph is segmented according to time slices (where i ranges from 1 to n, and n is the total number of time slices). Finally, a snapshot is stored (the knowledge graph is stored as a snapshot). When determining the community evolution event log, community identifier alignment is required, as shown in the figure. The community ID for a given moment (belonging to any time slice) is: The presence of three communities indicates the existence of three distinct communities within this knowledge graph. The next moment of time ( The community ID for the time slice corresponding to the moment is: , This represents the communities that evolved from the previous time slice, C1 and C... The similarity between them was 0.8, C2, C3 and C4. The similarity between the two is 0.7 and 0.6. Community identifier alignment matches communities with similar structures and member compositions at different points in time, ensuring they are linked even if their community identifiers change in the snapshot sequence due to community evolution. With community identifier alignment complete, the system can record historical events such as community generation, development, splitting, merging, or disappearance, forming a community evolution event log and entity migration paths. Finally, these are displayed using Sankey diagrams and heatmaps. For example, the Sankey diagram displays community merger information and entity migration information, while the heatmap displays community ID, time slices, and community size.

[0079] Figure 5 This is a schematic diagram of cross-community relationship mining provided in an embodiment of this application, illustrating an example of community relationship mining. It calculates the Bridge_Score (value score) of the hub entity E of communities A and B, and combines the information of E with the attributes of communities A and B to generate an LLM summary (i.e., generate a structured summary). For example, it generates a summary suggesting that communities A and B... field( The structured summary is “Cooperation and Association (Confidence Level W%) in a certain field”, where W refers to a specific data point.

[0080] Figure 6A decision engine interaction logic diagram is provided according to an embodiment of the present application, which shows an interaction process of a rule engine and a semantic engine. After a community summary (i.e., a structured summary) is generated, for community information (i.e., the above graph features), a deterministic risk value evaluation (i.e., the above deterministic analysis of the graph features by the rule engine) is determined according to the rule engine, a semantic interpretation is performed according to an LLM (for example, an LLM semantic engine), and finally a decision insight (i.e., the above analysis report) is obtained by integration. Finally, the integration is a multi-modal analysis report, i.e., the above multi-modal structured report.

[0081] An embodiment of the present application also provides a structural diagram of a community analysis device, as shown in Figure 7 The community analysis device comprises:

[0082] The division module 702 is configured to receive time series data containing entity relationships, and divide the time series data into time series data of multiple time slices according to a preset step size.

[0083] The first determination module 704 is configured to determine a knowledge graph corresponding to each time slice according to the time series data of the multiple time slices, and store the knowledge graph as a graph snapshot to obtain a graph sequence, wherein one community in each knowledge graph corresponds to a group of entities and relationships between the entities, and the graph sequence contains multiple graph snapshots, and each graph snapshot corresponds to a time slice.

[0084] The second determination module 706 is configured to determine a community evolution event log and an entity migration path of each entity according to at least the graph sequence, wherein the community evolution event log is used to indicate the life cycle of all communities involved in the graph sequence.

[0085] The third determination module 708 is configured to determine graph features according to at least the community evolution event log and the entity migration path of each entity, wherein the graph features are used to reflect the entity migration rule and the community evolution rule.

[0086] The generation module 710 is configured to generate an analysis report according to the graph features, wherein the analysis report is used to indicate at least the reason for the evolution of each community.

[0087] It should be noted that Figure 7 The community analysis device shown in Figure 2 is configured to execute the community analysis method shown in Figure 2 Therefore the related explanations and descriptions in the community analysis method in

[0088] are also applicable to the community analysis device, which will not be described here again.It should be noted that each module in the community analysis apparatus described above can be a program module (for example, a program instruction set for implementing a certain specific function) or a hardware module. For the latter, it can be in the form of, but not limited to, a processor or a processor that implements the functions of the modules.

[0089] The embodiments of the present application further provide a non-volatile storage medium, which comprises a stored program, wherein the program, when executed, controls a device in which the non-volatile storage medium is located to perform the community analysis method in any one of the above embodiments.

[0090] The embodiments of the present application further provide an electronic device, which comprises a processor configured to execute a program, wherein the program, when executed, performs the community analysis method in any one of the above embodiments.

[0091] According to another aspect of the embodiments of the present application, a computer program product is further provided, which comprises a computer program configured to implement the community analysis method in any one of the above embodiments when executed by a processor.

[0092] In the above embodiments of the present application, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0093] In the several embodiments provided by the present application, it should be understood that the disclosed technology can be implemented in other ways. The device embodiments described above are only illustrative, for example, the division of the units can be a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, unit or module, and can be electrical or other forms.

[0094] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on multiple units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiments of the present application.

[0095] In addition, each function unit in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software function unit.

[0096] When the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application, essentially or in part, or all or part of the technical solutions, can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a mobile hard disk, a magnetic disk or an optical disk, and various media that can store program codes.

[0097] The above is only the preferred embodiment of the present application. It should be pointed out that, for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which should be considered as the protection scope of the present application.

Claims

1. A community analysis method, characterized in that, include: Receive time-series data containing entity relationships, and divide the time-series data into multiple time slices according to a preset step size; Based on the time series data of the multiple time slices, the knowledge graphs corresponding to the multiple time slices are determined, and the knowledge graphs are stored as graph snapshots to obtain a graph sequence. In each knowledge graph, a community corresponds to a group of related entities and the relationships between the entities. The graph sequence contains multiple graph snapshots, and each graph snapshot corresponds to a time slice. At least based on the graph sequence, the community evolution event log and the entity migration path of each entity are determined, wherein the community evolution event log is used to indicate the life cycle of all communities involved in the graph sequence; The graph features are determined at least based on the community evolution event log and the entity migration path of each entity. The graph features are used to reflect the entity migration pattern and the community evolution pattern. The graph features include a structured summary, community centrality distribution and relation entropy distribution, community evolution event log, and the entity migration path of each entity. The structured summary is used at least to summarize and present the community evolution process. The community centrality distribution is used at least to reflect the changing trend of the centrality distribution of each community over time. The relation entropy distribution is used to reflect the changing trend of the relation entropy of each community over time. An analysis report is generated based on the map features, wherein the analysis report is used to indicate at least the reasons for the evolution of each community.

2. The method according to claim 1, characterized in that, The step of determining the knowledge graph corresponding to each of the multiple time slices based on the time series data of the multiple time slices includes: for each time slice, mapping the entities and the relationships between entities in the time series data containing entity relationships to a knowledge graph to obtain the knowledge graph corresponding to each time slice, wherein the knowledge graph contains at least one community; The knowledge graph is stored as a graph snapshot to obtain a graph sequence, including: storing the knowledge graph corresponding to each time slice as a graph snapshot, wherein the snapshot metadata of the graph snapshot includes at least the timestamp of the snapshot creation, the number of entities, the relation density, and the community cardinality; sorting all graph snapshots in the order of time slices from beginning to end to obtain the graph sequence.

3. The method according to claim 1, characterized in that, At least based on the aforementioned map sequence, a community evolution event log should be determined, including: Identify the evolutionary events of each community in each snapshot of the map sequence, wherein the evolutionary events include at least the birth, merger, split, and extinction of the community; The lifecycle of each community is determined in chronological order of time slices, based at least on the community identifier and the evolutionary events, and the community evolutionary event log is determined based on the evolutionary event information in the lifecycles of all communities.

4. The method according to claim 1, characterized in that, The migration path for each entity is determined in the following way: Identify the community affiliation information of each entity in each map snapshot in the map sequence, wherein the same entity has only one unique entity identifier in different map snapshots; The entity migration path of each entity is determined based on the community affiliation information of each entity, in accordance with the time slices from first to last. The entity migration path is used to indicate the migration status of the entity from the first time slice to the last time slice.

5. The method according to claim 1, characterized in that, The determination of graph features based at least on community evolution event logs and the entity migration paths of each entity includes: Determine the centrality distribution of each community in the map snapshot of each time slice, wherein the centrality distribution of each community is determined by the centrality of each entity in the community; Determine the relational entropy value of each community in the map snapshot of each time slice, wherein the relational entropy value is the entropy value of the relation type between entities within the community; The distribution of community centrality and the distribution of relational entropy are determined at least based on the centrality distribution of each community in the map snapshot of each time slice and the relational entropy value of each community in the map snapshot of each time slice. The structured summary, the distribution of community centrality, the distribution of relation entropy, the community evolution event log, and the entity migration path of each entity are determined as the graph features.

6. The method according to claim 5, characterized in that, The structured summary is determined in the following way: Determine the entity cross-community connectivity value score for each entity in each time slice of the graph sequence, and determine the average of the entity cross-community connectivity value scores for each entity across all time slices as the entity's value score, wherein the entity cross-community connectivity value score is determined at least based on the number of edges of the entity within the community, the size of the community associated with the entity, and the number of different communities associated with the entity; All entities whose value scores are greater than a preset threshold are identified as super hub entities. All super hub entities are combined into a super hub entity list. The communities associated with all super hub entities in the super hub entity list are identified as target communities. The entities in all target communities and the relationships between entities are identified as associated community features. The structured summary is determined based at least on the list of super hub entities and the characteristics of the associated communities.

7. The method according to claim 1, characterized in that, An analysis report is generated based on the aforementioned spectral features, including: The decision labels are obtained by performing deterministic analysis on the graph features using a rule engine. The semantic engine analyzes the graph features and the decision labels to obtain the analysis report, wherein the semantic engine is used to semantically interpret the graph features and the decision labels and transform them into the analysis report.

8. A community analysis device, characterized in that, include: The partitioning module is used to receive time-series data containing entity relationships and divide the time-series data into multiple time slices according to a preset step size. The first determining module is used to determine the knowledge graph corresponding to each of the multiple time slices based on the time series data of the multiple time slices, and store the knowledge graph as a graph snapshot to obtain a graph sequence. In each knowledge graph, a community corresponds to a group of related entities and the relationship between the entities. The graph sequence contains multiple graph snapshots, and each graph snapshot corresponds to a time slice. The second determining module is used to determine, at least based on the graph sequence, the community evolution event log and the entity migration path of each entity, wherein the community evolution event log is used to indicate the life cycle of all communities involved in the graph sequence; The third determining module is used to determine graph features based at least on the community evolution event log and the entity migration path of each entity. The graph features are used to reflect the entity migration pattern and the community evolution pattern. The graph features include a structured summary, community centrality distribution and relation entropy distribution, community evolution event log, and the entity migration path of each entity. The structured summary is used at least to summarize and present the community evolution process. The community centrality distribution is used at least to reflect the trend of centrality distribution of each community over time. The relation entropy distribution is used to reflect the trend of relation entropy of each community over time. A generation module is used to generate an analysis report based on the map features, wherein the analysis report is used to indicate at least the reasons for the evolution of each community.

9. A non-volatile storage medium, characterized in that, The non-volatile storage medium stores a program, wherein when the program is executed, it controls the device where the non-volatile storage medium is located to perform the community analysis method according to any one of claims 1 to 7.

10. An electronic device, characterized in that, include: A memory and a processor, the processor being configured to run a program stored in the memory, wherein the program, when running, executes the community analysis method according to any one of claims 1 to 7.

Images