Knowledge graph-based influence prediction method and system

Abstract

The invention relates to the technical field of organization transformation management and data analysis, and discloses a knowledge graph-based influence prediction method and system, and the method comprises the steps: constructing a double-layer knowledge graph which integrates an organization formal hierarchy and an informal social network; processing the double-layer atlas structure by applying a hierarchical graph convolutional network, and calculating multi-dimensional influence features; analyzing based on the multi-dimensional influence features, capturing an organization structure evolution rule by applying a time-varying graph structure learning algorithm, and identifying key opinion leaders; the key opinion leader information is utilized, the organization influence propagation process is simulated based on the Agent technology, and different intervention strategy effects are evaluated; generating a multi-dimensional influence prediction result; compared with a traditional method with a single network structure, the method has the advantages that the group influence prediction accuracy is improved, the key opinion leader recognition accuracy is improved, and the time sequence prediction error is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of organizational change management and data analysis, and more specifically, to a knowledge graph-based influence prediction method and system. Background Art

[0002] During organizational change, accurately predicting the spread of influence is crucial for the successful implementation of change. Currently, influence prediction in organizational change is mainly based on static network analysis methods, which identify key nodes by calculating centrality indicators or simulate influence spread using traditional information diffusion models.

[0003] These methods face multiple challenges in practical application: First, the internal flow of knowledge and information changes dramatically during organizational change, and traditional static network analysis methods cannot timely reflect the dynamic evolution of organizational structure; second, the formal hierarchical structure and informal social network in the organization interact with each other, and existing methods usually only focus on a single network structure, resulting in inaccurate predictions of cross-departmental knowledge influence; third, there is a lack of accurate identification algorithms for key opinion leaders in the organization, making it impossible to quantitatively assess the extent to which different people influence or hinder change; in addition, multimodal information within the organization is not effectively integrated, resulting in one-sided results of group-level influence analysis; finally, the dynamic evolution of interactions between organizational members has not been effectively modeled, affecting the timeliness and accuracy of the prediction results. Summary of the Invention

[0004] The present invention provides a knowledge graph-based influence prediction method and system to solve the technical problems in related technologies such as inaccurate influence prediction during organizational change and difficulty in capturing the dynamic evolution of network structure.

[0005] The present invention provides an influence prediction method based on knowledge graph, comprising: Construct a two-layer knowledge graph that integrates the formal organizational hierarchy and informal social networks; Apply hierarchical graph convolutional networks to process double-layer graph structures and calculate multi-dimensional influence features; Based on the analysis of multi-dimensional influence characteristics, a time-varying graph structure learning algorithm is applied to capture the evolution of organizational structure and identify key opinion leaders; Using key opinion leader information and agent-based technology to simulate the organizational influence dissemination process and evaluate the effectiveness of different intervention strategies; Integrate the two-layer knowledge graph structure, multi-dimensional influence characteristics, key opinion leaders and influence propagation simulation results to generate multi-dimensional influence prediction results.

[0006] Furthermore, the construction of the two-layer knowledge graph includes: Collect organizational structure data, employee interaction data, and multimodal content data; Construct a formal organizational hierarchy map; Constructing informal social network graphs; Integrate multimodal node features.

[0007] Furthermore, the step of calculating the multi-dimensional influence characteristics includes: Construct a hierarchical graph convolutional network HGCN; Implement cross-network attention mechanism and fuse representations of different network nodes; Construct an influence propagation function based on organizational roles.

[0008] Furthermore, the hierarchical graph convolutional network HGCN is constructed, in which the node representation is updated as follows: ; in and Representation Spectrum Middle Layer and The employee nodes of a layer represent the matrix, Indicates the network layer index, Represents the index of the next layer, represents the adjacency matrix after adding self-connection, Representation Spectrum middle The corresponding degree matrix, Representation Spectrum Middle Trainable weight matrices for layers, used in graph convolutional networks; Represents the activation function.

[0009] Furthermore, the application of the time-varying graph structure includes: Construct a time-varying graph structure learning algorithm to capture the evolution of graph structure over time; Calculate multidimensional centrality and bridging indices and ; Comprehensive scoring to identify key opinion leaders; Calculate tissue inertia coefficient , assessing organizational resistance to change.

[0010] Furthermore, the comprehensive score identifies key opinion leaders and calculates the comprehensive influence score of the employee node: ; in Represents employee nodes Comprehensive score of key opinion leaders, represents the scoring calculation function, Represents a normalization operation, which converts a vector into a standardized scalar value. 、 represents the weight coefficient of the centrality index and the bridging index, and represents the centrality index and the bridging index, and Represents employee nodes respectively The bridging index and centrality index vector of .

[0011] Furthermore, the simulation of the organizational influence propagation process includes: Build a behavioral agent for each organization member, including state vector, behavior space and transfer function; Simulate the spread of influence under different intervention strategies.

[0012] Furthermore, the simulation of influence propagation under different intervention strategies includes: Define the intervention strategy space: ; in represents the intervention strategy space, 、 、 Respectively represent , , intervention strategies, represents the total number of intervention strategies; For each strategy , intervene by modifying the state or behavior rules of a specific agent.

[0013] Furthermore, generating the influence prediction result includes: Generate individual influence distribution predictions; Generate group influence distribution predictions; Output key opinion leaders ranking and potential role analysis; Generate visualization results of influence propagation paths; Output evaluation of the effectiveness of intervention strategies.

[0014] The present invention provides an influence prediction system based on a knowledge graph, which is used to execute the above-mentioned influence prediction method based on a knowledge graph, including: A two-layer knowledge graph construction module is used to build a two-layer knowledge graph that integrates the formal organizational hierarchy and informal social networks; Multi-dimensional influence feature calculation module, used to apply hierarchical graph convolutional networks to process double-layer graph structures and calculate multi-dimensional influence features; The time-varying structure analysis module is used to apply time-varying graph structure learning algorithms to capture organizational structure evolution patterns and identify key opinion leaders; The influence dissemination simulation module is used to simulate the organizational influence dissemination process based on agent technology and evaluate the effects of different intervention strategies; The prediction result generation module is used to integrate the two-layer knowledge graph structure, multi-dimensional influence characteristics, key opinion leaders and influence propagation simulation results to generate multi-dimensional influence prediction results.

[0015] The beneficial effects of the present invention are: improved prediction accuracy, compared with the traditional single network structure method, improved group influence prediction accuracy, improved key opinion leader identification accuracy, and reduced time series prediction error; Enhanced change management effectiveness, enabling accurate assessment of the impact of different intervention strategies during organizational change. By identifying key opinion leaders and optimizing intervention strategies, it reduces resistance to change and increases the success rate of change. It achieves multi-information fusion, effectively integrating multiple information sources such as organizational structure data, employee interaction behavior, and multimodal content. Compared with the analysis method of a single data source, it improves information integrity and enhances noise resistance. It provides dynamic adaptability. Through the time-varying graph structure learning algorithm, it can adapt to the dynamic changes of organizational structure, maintain high prediction accuracy during organizational changes, and improve response speed. It supports explainable analysis, which not only provides prediction results, but also provides clear explanations for the prediction results through visualization of influence propagation paths and analysis of key nodes, thereby enhancing decision makers' understanding and trust in the prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flow chart of an influence prediction method based on knowledge graph of the present invention; Figure 2 It is a flowchart of the steps of constructing a double-layer knowledge graph of the present invention; Figure 3 is a flow chart of the steps of calculating multi-dimensional influence features of the present invention; Figure 4 is a flow chart of the steps of performing time-varying graph structure analysis of the present invention; Figure 5 is a flow chart of the steps of the influence propagation simulation process of the present invention; Figure 6 It is a flow chart of the steps of generating influence prediction results of the present invention. DETAILED DESCRIPTION

[0017] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. Furthermore, features described in some examples may be combined in other examples.

[0018] At least one embodiment of the present invention discloses a method for predicting influence based on knowledge graph, such as Figures 1 to 6 As shown, the following steps are included: Step 1: Construct a two-layer knowledge graph that integrates the formal organizational hierarchy and informal social networks; This step achieves a complete organizational knowledge network representation: Step 1.1, collect multi-source data; Acquire organizational structure data (including department divisions, reporting relationships, position information, etc.), employee interaction data (including communication records, collaboration history, meeting participation, etc.), and multimodal content data (including text documents, images, employee behavior logs, etc.).

[0019] Step 1.2: Build a formal organizational hierarchy map; Converting organizational structure data into a formal hierarchy map: ; in Represents the formal organizational hierarchy map, Represents the set of employee nodes in the formal organizational hierarchy graph, Represents the set of edges based on formal reporting relationships in the formal organizational hierarchy graph.

[0020] Step 1.3, construct an informal social network graph; Build an informal social network graph based on employee interaction data: ; in represents the informal social network graph, represents the set of employee nodes in the informal social network graph, Represents the edge set based on actual interaction relationships in the informal social network graph; Each edge Indicates employees With employees The interaction relationship between them is calculated by the function: ; in Represents an edge The weight of represents the weight calculation function, Indicates employees With employees The frequency of interaction between Indicates employees With employees The duration of the interaction between Indicates employees With employees The type of interaction between them.

[0021] Step 1.4, integrating multimodal node features; Process multimodal content data into node feature vectors , for each employee node , the feature vector consists of text features, behavior features and image features, namely: ; in Represents employee nodes The eigenvector of 、 and Represents employee nodes respectively Text features, behavioral features and image features, Represents a vector concatenation operation.

[0022] The outputs of this step include: Formal organizational hierarchy map , representing the organization's formal reporting relationships and hierarchical structure; informal social network map , representing informal interactions and social relationships among organizational members; the multimodal feature vector of each node , which contains employee characteristic information extracted from various data sources.

[0023] These outputs together constitute the organization's two-layer knowledge graph, providing a complete data foundation for subsequent analysis.

[0024] Step 2: Apply a hierarchical graph convolutional network to process the two-layer graph structure and calculate multi-dimensional influence features; This step applies a hierarchical graph convolutional network to the two-layer graph structure constructed in step 1, capturing the interaction patterns between networks and calculating multi-dimensional influence features. The two-layer knowledge graph, as input, provides complete topological information about the organization's formal hierarchical structure and informal social network. Combined with multimodal employee node features, it can comprehensively analyze the influence propagation channels within the organization: Step 2.1, construct a hierarchical graph convolutional network HGCN; The network contains two parallel graph convolution branches, which process the formal hierarchical graph in step 1.2 respectively. and the informal social network graph in step 1.3 , and then fuse the outputs of the two branches through the attention mechanism.

[0025] For the atlas (in Indicates the type of graph, which can be formal or informal). The employee node representation of the layer is updated as: ; in and Representation Spectrum Middle Layer and The employee nodes of a layer represent the matrix, Indicates the network layer index, Represents the index of the next layer, represents the adjacency matrix after adding self-connection, Representation Spectrum middle The corresponding degree matrix, Representation Spectrum Middle Trainable weight matrices for layers, used in graph convolutional networks; represents the activation function; The calculation formula is: ; in Representation Spectrum The original adjacency matrix of Representation Spectrum The identity matrix of (for self-connection).

[0026] Step 2.2, implement the cross-network attention mechanism; For each employee node , calculate the formal atlas representation after step 2.1 and informal graph representation Fusion representation of , using the attention mechanism to adaptively determine the importance weights of different graphs: ; ; in Represents employee nodes In the atlas The attention weights in represents the learnable parameter vector of the attention mechanism, represents the learnable parameter matrix of the attention mechanism, Represents employee nodes In the atlas In the expression, Represents employee nodes The query vector, Represents vector concatenation operation, represents the exponential function, Indicates the sum of the two graph types, formal and informal. Indicates the graph type index, Represents employee nodes The fusion representation of Represents employee nodes In the atlas In the representation; Indicates the sum of the two graph types, formal and informal. Indicates the graph type index, Represents employee nodes The fusion representation of Represents employee nodes In the atlas In the expression, Represents employee nodes In the atlas The attention weights in .

[0027] Step 2.3: Construct an influence propagation function based on organizational roles; Considering the influence transfer characteristics between different roles in the organization, we use the employee node features integrated in step 1.4 and the fusion representation generated in step 2.2 to define a role-based influence propagation function: ; ; in Represents employee nodes For employee nodes The probability of influence spreading, represents the influence propagation function, Represents employee nodes role, Represents employee nodes role, Represents employee nodes The role embedding vector, Represents employee nodes The role embedding vector, Represents employee nodes and The relationship between represents the role-related learnable parameter matrix (used for influence propagation function), A learnable parameter matrix representing the relationship representation (used for the influence propagation function), Represents the activation function.

[0028] The output of this step includes: the fusion representation of each employee node , which integrates the employee node characteristics of formal and informal networks; the probability of influence transmission between employee nodes describes the possibility of influence transmission between organizational members.

[0029] These outputs reflect the influence characteristics after considering organizational role factors in the two-layer network structure, providing basic data for subsequent time-varying analysis and opinion leader identification.

[0030] Step 3: Analyze the multi-dimensional influence characteristics and apply the time-varying graph structure learning algorithm to capture the evolution of organizational structure and identify key opinion leaders; This step applies a time-varying graph structure learning algorithm to capture organizational structure evolution patterns, identify key opinion leaders, and assess resistance to organizational change. This step uses the two-layer knowledge graph constructed in step 1 as the initial state and, combined with the multi-dimensional influence features calculated in step 2, expands the static network structure into a dynamic evolution process, enabling modeling and prediction of structural changes during organizational transformation. Step 3.1, construct a time-varying graph structure learning algorithm; The algorithm captures the evolution of the graph structure over time in a recursive way. Graph structure , based on the graph structure at the previous moment and current external features Update, where Initially comes from the two-layer knowledge graph constructed in step 1, Contains the multimodal node features integrated in step 1.4: ; in Indicates a time point The graph structure of Indicates a time point The graph structure of (the graph structure at the previous moment), represents the graph structure update function, Indicates a time point external features, represents the learnable weight matrix in graph structure update, represents the learnable weight matrix in feature update, represents the bias term, Represents the activation function.

[0031] The algorithm can capture both the continuity and mutation of organizational structure changes.

[0032] Step 3.2, calculate the multidimensional centrality and bridging index; Based on the graph representation fused in step 2.2, calculate the multidimensional centrality and bridging indicators of employee nodes to identify potential opinion leaders: ; ; in Represents employee nodes The centrality index vector of Represents a vector construction operation, Represents employee nodes The degree centrality of Represents employee nodes The betweenness centrality of Represents employee nodes The closeness centrality, Represents employee nodes The eigenvector centrality of Represents employee nodes The bridging index, Represents employee nodes The neighbor set of Represents employee nodes and The similarity of Represents employee nodes All neighbor pairs of Perform the summation, Represents employee nodes The number of possible pairings of neighbors is calculated as follows: ; in Represents employee nodes The number of neighbors.

[0033] Step 3.3: Identify key opinion leaders through comprehensive scoring; Combine the centrality index and bridging index calculated in step 3.2 to calculate the comprehensive influence score of the employee node: ; in Represents employee nodes Comprehensive score of key opinion leaders, represents the scoring calculation function, Represents a normalization operation, which converts a vector into a standardized scalar value. 、 represents the weight coefficient of the centrality index and the bridging index, and represents the centrality index and the bridging index.

[0034] Step 3.4, calculate the tissue inertia coefficient; Based on the differences in the structure of the before-after moment graphs obtained in step 3.1, assess the resistance to organizational change: ; in represents the tissue inertia coefficient, represents the inertia coefficient calculation function, Represents the operation of converting a graph structure into a vector, represents the vector norm (a function used to measure the size of a vector), Indicates the time interval, The drag coefficient of the graph structure at the previous moment is calculated based on the graph characteristics; 、 Indicates a time point and Graph structure.

[0035] The output of this step includes: time-varying graph structure sequence , describing the evolution of organizational structure over time; key opinion leader ratings , identifying key influencers in the organization; organizational inertia coefficient , quantifying the degree of organizational resistance to change. These outputs together characterize the structural dynamics and key impact points of the organization during the transformation period, providing the necessary parameters and initial states for subsequent influence propagation simulations.

[0036] Step 4: Using key opinion leader information, simulate the organizational influence diffusion process based on agent technology and evaluate the effectiveness of different intervention strategies; This step uses agent technology to simulate the organizational influence propagation process, achieving collaborative optimization of individual and group influence predictions. This step leverages the results of the previous steps: using the two-layer knowledge graph constructed in step 1 as the network topology foundation, using the multidimensional influence features calculated in step 2 as the initial node attributes, and combining the key opinion leaders identified in step 3 with the time-varying graph structure to construct a dynamic influence propagation model.

[0037] Step 4.1, construct an agent-based organizational member behavior model; Build a behavioral agent for each organization member, including a state vector, a behavior space, and a transfer function. The initial representation of the state vector directly uses the node fusion representation of step 2.2 and the node features of step 1.4: ; ; ; in Indicates the identifier of the Agent, used to distinguish different organization member Agents. Indicates organization members The state vector of Represents a vector construction operation, Represents employee nodes The fusion representation of Indicates organization members Character information, Indicates organization members The current influence value of Indicates organization members acceptance; Indicates organization members behavior, Represents a sampling operation, Indicates organization members The policy function, Indicates that the status Possible behavior , Indicates organization members The next state, Indicates organization members The state transition function, Represents employee nodes All neighbors The set of behaviors; The specific implementation of the organizational member behavior model includes: State representation, the state vector of each agent contains 16-dimensional node representation, 4-dimensional role encoding, 1-dimensional influence value and 1-dimensional acceptance value; Behavioral space, including information dissemination behaviors (active sharing, passive reception, transmission modification, rejection) and influence behaviors (support, opposition, neutrality); The policy network uses a two-layer fully connected network, inputs a state vector, and outputs the probability distribution of each behavior. The structure is the state dimension 32 The number of behaviors, including Represents connections between network layers; The state transition function updates the agent's influence value and acceptance based on the current state, the actions taken, and the neighbor actions, using an attention-weighted neighbor influence aggregation mechanism.

[0038] In specific application scenarios, the model can simulate the following situations: During the implementation of corporate transformation, identify the influence transmission paths of key opinion leaders on their department members and predict possible information distortion or blockage points in the process of transmitting transformation information from top to bottom. In organizational culture integration, the dynamic evolution of employees from different backgrounds’ acceptance of new policies is simulated, the potential formation mechanism of resistance groups is predicted, and targeted intervention plans are provided.

[0039] Step 4.2: simulate the spread of influence under different intervention strategies; Define the intervention strategy space: ; in represents the intervention strategy space, 、 、 Respectively represent , , intervention strategies, represents the total number of intervention strategies; For each strategy , intervene by modifying the state or behavior rules of a specific agent: ; ; in Indicates the intervention strategies, Represents a conversion operation, Indicates organization members The initial state and strategy, Indicates that the organization members after the intervention status and strategy, Indicates that in the strategy Next, through Network after round propagation The influence distribution of each node in represents the influence propagation simulation function, Represents the network structure, Represents the state set of all Agents, Represents the set of policy functions of all agents, Indicates the number of propagation rounds or iterations; The specific implementation steps of the influence propagation simulation algorithm include: Initialization: Initialize the state vector of each agent based on the node features and relationships in the knowledge graph in step 1 ; Intervention Apply: Apply intervention strategies to selected nodes , modify its status or rules of behavior; Iterative update: In each round of iteration, each agent updates its strategy function Select a behavior and update the state after interacting with the neighbors. The formula is: ; ; ; in Indicates a time point Members of the organization behavior, Indicates a time point Members of the organization status, Indicates a time point Members of the organization The influence value, Indicates a time point Members of the organization The influence value of Represents employee nodes All neighbors Perform the summation, Indicates that from the employee node To employee node The influence weight of represents the behavior interaction function, a function for evaluating behavior compatibility, represents the addition operation, Indicates a time point Members of the organization status, Indicates a time point Time employee node The expression, Indicates a time point Members of the organization acceptance; Termination condition: When the preset number of iterations is reached Or stop when the network status changes below the threshold; Result output: Returns the influence distribution and change trajectory of each node in the final network.

[0040] Taking the reorganization of corporate departments as an example, we can simulate the spread of influence: Initial data: A company has 100 employees distributed across 5 departments. Build an organizational knowledge graph with 100 nodes. Initial influence distribution: Set the initial influence value based on employee rank and centrality; Intervention strategy design: Strategy 1 (Information Transparency): Increase the frequency of information sharing among management nodes and modify their policy functions to favor proactive sharing. Strategy 2 (Opinion Leader Activation): Identify and activate five key opinion leaders to increase their initial influence and acceptance. Strategy 3 (Grassroots Participation): Increase the acceptance and feedback weight of grassroots employee nodes; Simulation results: 30 rounds of iterations simulate the influence spread under each strategy: Strategy 1 improved overall acceptance, but there was an imbalance in information reception among two departments; Strategy 2 increased acceptance, with less variation across departments; Strategy 3 improves the efficiency of uploading grassroots feedback, but slows down the overall advancement speed.

[0041] The outputs of this step include: the influence propagation trajectory under different intervention strategies ;Influence distribution results of each strategy and changes in organizational members’ acceptance. These outputs comprehensively describe the dynamic process of influence flowing through the organization and the effects of interventions, providing direct input for generating the final prediction results.

[0042] Step 5: Integrate the two-layer knowledge graph structure, multi-dimensional influence characteristics, key opinion leaders, and influence propagation simulation results to generate multi-dimensional influence prediction results; This step integrates the above analysis results to generate multi-dimensional prediction results, providing data support for organizational change decisions. This step comprehensively utilizes the outputs of all previous steps: based on the two-layer knowledge graph structure of step 1, combined with the multi-dimensional influence characteristics of step 2, referring to the key opinion leaders identified in step 3, and based on the influence propagation simulation results of step 4, a comprehensive prediction report is finally generated: Step 5.1, generate individual influence distribution prediction; Based on the simulation results of step 4.2, calculate the influence change curve of each organization member over time , and predict future time points Influence value ,in Indicates the time span of the forecast.

[0043] Step 5.2, generate group influence distribution prediction; Aggregate the individual influences from step 5.1, combine them with the organizational structure information constructed in steps 1.2 and 1.3, and calculate the overall influence index of different departments, teams, or functional groups. and predicted values .

[0044] Step 5.3: Output the ranking of key opinion leaders; Based on the comprehensive impact score calculated in step 3.3 , combined with the actual influence demonstrated in the simulation in step 4, generate a ranked list of key opinion leaders in the organization and provide an analysis of their potential role in the change process.

[0045] Step 5.4: Generate visualization of influence propagation paths; Based on the influence propagation simulation results from step 4.2 and the network topology constructed in step 1, visualize the propagation path of key information in the organization and identify bottlenecks and acceleration points of information flow; Step 5.5: Output the effectiveness evaluation of the intervention strategy; Based on the simulation results of different intervention strategies in step 4.2, calculate the intervention strategy effectiveness index , rank different intervention strategies, and provide specific and executable intervention suggestions, including key personnel management, information dissemination path optimization, organizational structure adjustment, etc.

[0046] The final output of this step is a complete set of influence prediction results, including influence prediction data at the individual and group levels; identification and role analysis of key opinion leaders; visualization of influence propagation paths; and evaluation and recommendations for intervention strategies.

[0047] These outputs provide comprehensive, accurate and actionable data support for organizational change decisions, and can guide organizations to optimize change implementation plans and improve the success rate of change.

[0048] A knowledge graph-based influence prediction system, used to implement the above-mentioned knowledge graph-based influence prediction method, includes: A two-layer knowledge graph construction module is used to build a two-layer knowledge graph that integrates the formal organizational hierarchy and informal social networks; Multi-dimensional influence feature calculation module, used to apply hierarchical graph convolutional networks to process double-layer graph structures and calculate multi-dimensional influence features; The time-varying structure analysis module is used to apply time-varying graph structure learning algorithms to capture organizational structure evolution patterns and identify key opinion leaders; The influence dissemination simulation module is used to simulate the organizational influence dissemination process based on agent technology and evaluate the effects of different intervention strategies; The prediction result generation module is used to integrate the two-layer knowledge graph structure, multi-dimensional influence characteristics, key opinion leaders and influence propagation simulation results to generate multi-dimensional influence prediction results.

[0049] Here, the present invention provides an implementation example: reorganization of a large bank's business departments; A major bank decided to restructure its business units, merging six previously independent departments (Personal Finance, Corporate Finance, Financial Markets, FinTech, Risk Management, and Back-office Operations) into four departments (Retail Integrated Finance, Corporate and Institutional Finance, Financial Markets and Technology, and Risk Control and Operations Support). This transformation affected 1,200 employees, and management sought to predict the impact of the restructuring, identify potential resistance points, and develop effective intervention strategies to ensure a smooth transition.

[0050] First, the following multi-source data were collected: Organizational structure data: including department information, reporting relationships, job levels, etc., as shown in Table 1: Table 1: Sample organizational structure data

[0051] Employee interaction data: This includes email frequency, meeting participation, project collaboration, etc. over the past six months, as shown in Table 2: Table 2: Sample employee interaction data

[0052] Multimodal content data: including email text, internal forum speeches, performance evaluations, etc.

[0053] Based on the collected data, a two-layer knowledge graph was constructed. The formal hierarchical graph contains 1,200 nodes and 1,427 edges, and the informal social network graph contains 1,200 nodes and 5,813 edges. The characteristics of the fused graph are shown in Table 3: Table 3: Statistics of characteristics of two-layer knowledge graph

[0054] The HGCN model was applied to calculate multi-dimensional influence characteristics and identify key opinion leaders. The characteristics of the top five key opinion leaders and their influence types are shown in Table 4: Table 4: Key opinion leader identification results

[0055] Using an agent-based organizational member behavior model, we simulated the influence diffusion effects of three intervention strategies. We conducted 30 independent simulations for each strategy, each with 100 diffusion iterations. Table 5 compares the effects of different intervention strategies. Table 5: Comparison of intervention strategies

[0056] This implementation method, applied to the banking sector restructuring scenario, achieved significant technical results, mainly in two aspects: Prediction accuracy verification: The prediction accuracy was verified by comparing the system prediction results with the influence propagation data during the actual transformation process. The prediction accuracy comparison of different methods is shown in Table 6: Table 6: Comparison of prediction accuracy

[0057] Verification of change management effectiveness: The intervention strategy guided by the prediction results of this system was compared with the originally planned standard change strategy, which improved the effectiveness of change management. The comparison of change management effectiveness indicators is shown in Table 7: Table 7: Comparison of change management effectiveness indicators

[0058] Through verification in the above two aspects, this implementation method has demonstrated obvious technical advantages in real application scenarios, especially in accurately predicting the propagation path of organizational influence and improving the effectiveness of change management, providing strong data support for organizational change decision-making.

[0059] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make more forms of equivalent embodiments based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. A knowledge graph-based influence prediction method, characterized in that: include: Construct a two-layer knowledge graph that integrates the formal organizational hierarchy and informal social networks; Apply hierarchical graph convolutional networks to process double-layer graph structures and calculate multi-dimensional influence features; Based on the analysis of multi-dimensional influence characteristics, a time-varying graph structure learning algorithm is applied to capture the evolution of organizational structure and identify key opinion leaders; Using key opinion leader information and agent-based technology to simulate the organizational influence dissemination process and evaluate the effectiveness of different intervention strategies; Integrate the two-layer knowledge graph structure, multi-dimensional influence characteristics, key opinion leaders and influence propagation simulation results to generate multi-dimensional influence prediction results.

2. The influence prediction method based on knowledge graph according to claim 1 is characterized in that: The construction of the two-layer knowledge graph includes: Collect organizational structure data, employee interaction data, and multimodal content data; Construct a formal organizational hierarchy map; Constructing informal social network graphs; Integrate multimodal node features.

3. The influence prediction method based on knowledge graph according to claim 1 is characterized in that: The step of calculating the multi-dimensional influence characteristics includes: Construct a hierarchical graph convolutional network HGCN; Implement cross-network attention mechanism and fuse representations of different network nodes; Construct an influence propagation function based on organizational roles.

4. The influence prediction method based on knowledge graph according to claim 3 is characterized in that: The hierarchical graph convolutional network HGCN is constructed, where the node representation is updated as follows: ； in and Representation Graph Middle Layer and The employee nodes of a layer represent the matrix, Indicates the network layer index, Represents the index of the next layer, represents the adjacency matrix after adding self-connection, Representation Graph middle The corresponding degree matrix, Representation Graph Middle Trainable weight matrices for layers, used in graph convolutional networks; Represents the activation function.

5. The influence prediction method based on knowledge graph according to claim 1 is characterized in that: The application of the time-varying graph structure includes: Construct a time-varying graph structure learning algorithm to capture the evolution of graph structure over time; Calculate multidimensional centrality and bridging indices and ; Comprehensive scoring to identify key opinion leaders; Calculate tissue inertia coefficient , assessing organizational resistance to change.

6. The influence prediction method based on knowledge graph according to claim 5 is characterized in that: The comprehensive score identifies key opinion leaders and calculates the comprehensive influence score of employee nodes: ； in Represents employee nodes Comprehensive score of key opinion leaders, represents the scoring calculation function, Represents a normalization operation, which converts a vector into a standardized scalar value. 、 represents the weight coefficient of the centrality index and the bridging index, and represents the centrality index and the bridging index, and Represents employee nodes respectively The bridging index and centrality index vector of .

7. The influence prediction method based on knowledge graph according to claim 1 is characterized in that: The simulated organizational influence propagation process includes: Build a behavioral agent for each organization member, including state vector, behavior space and transfer function; Simulate the spread of influence under different intervention strategies.

8. The influence prediction method based on knowledge graph according to claim 1 is characterized in that: The simulation of influence spread under different intervention strategies includes: Define the intervention strategy space: ； in represents the intervention strategy space, 、 、 Respectively represent , , intervention strategies, represents the total number of intervention strategies; For each strategy , intervene by modifying the state or behavior rules of a specific agent.

9. The influence prediction method based on knowledge graph according to claim 1 is characterized in that: Generating the influence prediction result includes: Generate individual influence distribution predictions; Generate group influence distribution predictions; Output key opinion leaders ranking and potential role analysis; Generate visualization results of influence propagation paths; Output evaluation of the effectiveness of intervention strategies.

10. An influence prediction system based on knowledge graph, characterized in that: A method for predicting influence based on a knowledge graph, used to execute any one of claims 1 to 9, comprising: A two-layer knowledge graph construction module is used to build a two-layer knowledge graph that integrates the formal organizational hierarchy and informal social networks; Multi-dimensional influence feature calculation module, used to apply hierarchical graph convolutional networks to process double-layer graph structures and calculate multi-dimensional influence features; The time-varying structure analysis module is used to apply time-varying graph structure learning algorithms to capture organizational structure evolution patterns and identify key opinion leaders; The influence dissemination simulation module is used to simulate the organizational influence dissemination process based on agent technology and evaluate the effects of different intervention strategies; The prediction result generation module is used to integrate the two-layer knowledge graph structure, multi-dimensional influence characteristics, key opinion leaders and influence propagation simulation results to generate multi-dimensional influence prediction results.

Images