A BI decision management system and method based on big data

Abstract

The present invention discloses a BI decision management system and method based on big data, which relates to the technical field of BI decision management. The Louvain algorithm divides a graph structure into several communities, optimizes decision strategies based on community feature information using the Q-learning algorithm, updates Q values ​​and selects optimal actions, performs iterative optimization, and outputs an optimized Q-learning model after reaching the convergence condition. After the Q-learning model is evaluated based on the performance value, the reward function value of the Q-learning model is adjusted according to the performance evaluation result. By constructing a graph structure and using the Louvain algorithm for community discovery, the management system can integrate multi-source data, identify the natural group structure in the data, provide a more comprehensive overall analysis perspective, perform adaptive learning, better capture the complex relationships and dynamic changes in the data, and improve the flexibility and accuracy of decision-making.

Description

Technical Field

[0001] The present invention relates to the technical field of BI decision management, and in particular to a BI decision management system and method based on big data. Background Art

[0002] The concept of BI can be traced back to the 1950s. At that time, BI systems mainly relied on statistical analysis and report generation. With the continuous development of computer technology and database technology, BI systems have gone through multiple stages of development from data warehouses, online analytical processing (OLAP) to big data and cloud computing. BI (Business Intelligence) decision management system is a comprehensive information system that integrates data collection, data processing, data analysis and data presentation. Its main goal is to help enterprises and organizations make effective decision management through data-driven methods.

[0003] The prior art has the following deficiencies:

[0004] Traditional BI systems often rely on static reports and simple OLAP operations, which make it difficult to integrate and analyze large-scale, multi-source data, resulting in data silos and insufficient analysis depth. In addition, when dealing with complex relationships and dynamically changing data, they are often unable to effectively capture and reflect hidden patterns and dynamic changes in the data.

[0005] Based on this, the present invention proposes a BI decision management system and method based on big data. By constructing a graph structure and using the Louvain algorithm for community discovery, it can integrate multi-source data, identify the natural group structure in the data, and provide a more comprehensive overall analysis perspective. It also combines the Louvain algorithm and the Q-learning algorithm for adaptive learning to better capture the complex relationships and dynamic changes in the data, thereby improving the flexibility and accuracy of decision-making. Summary of the invention

[0006] The purpose of the present invention is to provide a BI decision management system and method based on big data to address the deficiencies in the background technology.

[0007] In order to achieve the above object, the present invention provides the following technical solution: a BI decision management method based on big data, the management method comprising the following steps:

[0008] The management system collects relevant business data through the API interface of the enterprise big data database, extracts feature data after preprocessing the relevant business data, takes the nodes in the feature data as nodes, and constructs a graph structure with the relationships between nodes as edges;

[0009] The graph structure is divided into several communities through the Louvain algorithm, and community feature information is extracted from the community information generated by the Louvain algorithm. The decision strategy is optimized based on the Q-learning algorithm using the community feature information, the Q value is updated and the optimal action is selected. The Q-learning model is iteratively optimized so that it can select the optimal decision action under different states.

[0010] When the convergence conditions are reached, the optimized Q-learning model is output, and the performance value of the Q-learning model in different communities is calculated. After the Q-learning model is evaluated based on the performance value, the reward function value of the Q-learning model is adjusted according to the performance evaluation results. The Q-learning model with adjusted reward function value is applied to the subsequent BI decision management of the enterprise.

[0011] In a preferred embodiment, the graph structure is divided into several communities by using the Louvain algorithm, and community feature information is extracted from the community information generated by the Louvain algorithm, including the following steps:

[0012] The constructed graph structure is used as the input of the Louvain algorithm, including nodes, edges and edge weights. The Louvain algorithm is run to discover communities on the graph structure and obtain the community label of each node.

[0013] Generate a feature vector for each community, and calculate the scale coefficient of the community based on the feature vector of the community. The feature vector includes the total edge weight, community density, and cohesion index. The total edge weight, community density, and cohesion index are comprehensively calculated to obtain the scale coefficient. The expression is: Where GM is the scale factor, is the total edge weight, n represents the number of edges in the community, QZ i represents the weight of the ith edge, SM is the community density, NJ is the cohesion index, α, β, and γ are the proportional coefficients of the total edge weight, community density, and cohesion index, respectively, and α, β, and γ are all greater than 0;

[0014] After obtaining the scale coefficient, compare the scale coefficient with the preset coefficient threshold. The coefficient threshold is used to distinguish the size of the community. If the scale coefficient is greater than the coefficient threshold, the community is defined as a large community. If the scale coefficient is less than or equal to the coefficient threshold, the community is defined as a small community. The community's feature vector and scale coefficient are used to establish community feature information, and the community feature information is displayed using a graph visualization tool.

[0015] In a preferred embodiment, the decision strategy is optimized based on the Q-learning algorithm using community feature information, the Q value is updated and the optimal action is selected, and the Q-learning model is iteratively optimized, including the following steps:

[0016] Extract the community feature information of each community, use the extracted community feature information to construct the state representation, and select the initial state s from the current state space c , based on the greedy strategy ∈-select action a, randomly select an action with probability ∈, select the action with the maximum current Q value with probability 1-∈, in the current state s d Execute action a and get the new state s new And reward r, then update the Q value, the expression is: a z is all selected actions, τ is the learning rate, is the discount factor, and the new state s new As the current state d , according to the updated Q value table, select the optimal action in each state, and optimize the current decision-making strategy based on the selected optimal action.

[0017] In a preferred embodiment, calculating the performance values ​​of the Q-learning model in different communities and evaluating the Q-learning model based on the performance values ​​comprises the following steps:

[0018] When the number of iterations is greater than or equal to the number threshold, it is judged that the convergence condition is met, and the Q value table updated during the training process is saved as a file or variable as the optimized Q-learning model. A group of communities are selected from the data set as test samples, and the task completion rate and resource utilization efficiency of the Q-learning model are obtained. The task completion rate and resource utilization efficiency are normalized to obtain the normalized value of the task completion rate and the normalized value of the resource utilization efficiency. The normalized value of the task completion rate and the normalized value of the resource utilization efficiency are summed to obtain the performance value. If the performance value is greater than or equal to the performance threshold, the performance of the Q-learning model is evaluated to be up to standard. If the performance value is less than the performance threshold, the performance of the Q-learning model is evaluated to be not up to standard.

[0019] In a preferred embodiment, adjusting the reward function value of the Q-learning model according to the performance evaluation result includes the following steps:

[0020] When the performance of the Q-learning model is evaluated to be up to standard, the reward function value of the Q-learning model is increased. When the performance of the Q-learning model is evaluated to be not up to standard, the reward function value of the Q-learning model is reduced. The adjustment algorithm is: In the formula, R currentis the current reward function value, bx s is the performance value, bx y is the performance threshold.

[0021] In a preferred embodiment, the nodes in the feature data are used as nodes, and the relationships between the nodes are used as edges to construct a graph structure, including the following steps:

[0022] Determine the nodes in the graph and the relationship between the nodes based on the feature data, extract the attributes of each node from the feature data, the node attributes include user ID, age, gender, purchase history, and assign a unique node ID to each node;

[0023] Define the relationship between nodes and use the relationship between nodes as the edge in the graph. Calculate the weight of the edge according to business needs.

[0024] Create a node list containing the IDs and attributes of all nodes, create an edge list containing the start points, end points, and weights of all edges, and import the node list and edge list into the graph database to build the graph structure.

[0025] In a preferred embodiment, calculating the weight of an edge according to business requirements includes the following steps:

[0026] Get the interaction frequency and interaction duration of the edge, normalize the interaction frequency and interaction duration, map the value range of the interaction frequency and interaction duration to [0, 1], get the normalized value of the interaction frequency and the normalized value of the interaction duration, and sum the normalized value of the interaction frequency and the normalized value of the interaction duration to get the weight of the edge.

[0027] A BI decision management system based on big data, including a graph structure building module, a community generation module, and a model optimization module:

[0028] Graph structure construction module: collect relevant business data through the API interface of the enterprise big data database, extract feature data after preprocessing the relevant business data, use the nodes in the feature data as nodes, and use the relationship between nodes as edges to build a graph structure, and send the graph structure to the community generation module;

[0029] Community generation module: The Louvain algorithm is used to divide the graph structure into several communities, which represent the natural group structure in the data. Community feature information is extracted from the community information generated by the Louvain algorithm and sent to the model optimization module.

[0030] Model optimization module: Based on the Q-learning algorithm, the community feature information is used to optimize the decision-making strategy, update the Q value and select the optimal action, and iteratively optimize the Q-learning model so that it can select the optimal decision-making action under different conditions. When the convergence conditions are reached, the optimized Q-learning model is output and the performance value of the Q-learning model in different communities is calculated. After evaluating the Q-learning model based on the performance value, the reward function value of the Q-learning model is adjusted according to the performance evaluation results to improve the decision-making effect of the model. The Q-learning model with adjusted reward function value is applied to the subsequent BI decision management of the enterprise.

[0031] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0032] The present invention constructs a graph structure by taking nodes in feature data as nodes and taking the relationship between nodes as edges, divides the graph structure into several communities by Louvain algorithm, and these communities represent the natural group structure in the data, extracts community feature information from the community information generated by the Louvain algorithm, optimizes the decision strategy based on the Q-learning algorithm, updates the Q value and selects the optimal action, iteratively optimizes the Q-learning model so that it can select the optimal decision action under different states, outputs the optimized Q-learning model after reaching the convergence condition, calculates the performance value of the Q-learning model in different communities, evaluates the Q-learning model according to the performance value, and adjusts the reward function value of the Q-learning model according to the performance evaluation result. The management system can integrate multi-source data, identify the natural group structure in the data, provide a more comprehensive overall analysis perspective, and combine the Louvain algorithm and the Q-learning algorithm for adaptive learning, better capture the complex relationship and dynamic changes in the data, and improve the flexibility and accuracy of decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0034] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0036] Example 1: Please refer to Figure 1 As shown, the BI decision management method based on big data described in this embodiment includes the following steps:

[0037] The management system collects relevant business data through the API interface of the enterprise big data database, extracts feature data after preprocessing the relevant business data, takes the nodes in the feature data as nodes, and constructs a graph structure with the relationship between nodes as edges. The graph structure is divided into several communities through the Louvain algorithm. These communities represent the natural group structure in the data. Community feature information is extracted from the community information generated by the Louvain algorithm. Based on the Q-learning algorithm, the decision strategy is optimized using community feature information, the Q value is updated and the optimal action is selected. The Q-learning model is iteratively optimized so that it can select the optimal decision action under different states. When the convergence condition is reached, the optimized Q-learning model is output, and the performance value of the Q-learning model in different communities is calculated. After evaluating the Q-learning model based on the performance value, the reward function value of the Q-learning model is adjusted according to the performance evaluation results to improve the decision-making effect of the model. The Q-learning model with adjusted reward function value is applied to the subsequent BI decision management of the enterprise.

[0038] This application constructs a graph structure by taking nodes in feature data as nodes and the relationships between nodes as edges, and divides the graph structure into several communities through the Louvain algorithm. These communities represent the natural group structure in the data, extract community feature information from the community information generated by the Louvain algorithm, and optimize the decision strategy based on the Q-learning algorithm using community feature information, update the Q value and select the optimal action, iteratively optimize the Q-learning model so that it can select the optimal decision action under different states, and when the convergence condition is reached, output the optimized Q-learning model, calculate the performance value of the Q-learning model in different communities, evaluate the Q-learning model based on the performance value, and adjust the reward function value of the Q-learning model based on the performance evaluation result. By constructing a graph structure and using the Louvain algorithm for community discovery, the management system can integrate multi-source data, identify the natural group structure in the data, provide a more comprehensive overall analysis perspective, and combine the Louvain algorithm and the Q-learning algorithm for adaptive learning, so as to better capture the complex relationships and dynamic changes in the data and improve the flexibility and accuracy of decision-making.

[0039] Embodiment 2: The management system collects relevant business data through the API interface of the enterprise big data database, pre-processes the relevant business data and extracts feature data, including the following steps:

[0040] Use the API interface provided by the enterprise big data database to collect relevant business data from various business systems (such as ERP, CRM, inventory management system, etc.), and obtain original data through the API, including transaction data, user behavior data, inventory data, market data, etc.

[0041] Check and process missing data, use methods such as filling, deletion or interpolation, identify and process outliers in the data, use statistical methods or machine learning methods to delete duplicate records, ensure data uniqueness and accuracy, convert data to a unified format, ensure that data from different data sources are compatible, convert data types to appropriate formats (such as integers, floating point numbers, dates, etc.), scale data to a unified scale range, such as standardizing numerical data to the 0-1 range, and encode categorical variables, such as using One-Hot Encoding or Label Encoding.

[0042] Based on business needs and data analysis goals, relevant features are selected from the original data, and new features are generated based on the existing data, such as time features (week, month, season, etc.) and interaction features (product and sum of features). Aggregate calculations are performed on the data, such as average, maximum, minimum, standard deviation, etc. Stop words and punctuation marks are removed, stemming and word form restoration are performed, and text data is converted into numerical vectors. For example, methods such as TF-IDF and Word2Vec are used, and the extracted feature data is stored in the feature library for subsequent analysis and modeling.

[0043] The nodes in the feature data are taken as nodes, and the relationships between the nodes are taken as edges to construct a graph structure, including the following steps:

[0044] Determine the nodes in the graph based on the feature data. For example, in a user behavior analysis system, the nodes can be users, products, transactions, etc. Determine the relationship between the nodes, such as the social relationship between users, the purchase relationship between users and products, etc. Extract the attributes of each node from the feature data. For example, the attributes of a user node can include user ID, age, gender, purchase history, etc. Assign a unique node ID to each node for identification in the graph structure.

[0045] Define the relationship between nodes and use it as the edge in the graph. For example, the social relationship between user A and user B can be used as an edge. Calculate the weight of the edge according to business needs, create a node list containing the IDs and attributes of all nodes, and create an edge list containing the start point, end point and weight of all edges.

[0046] Graph databases (such as Neo4j and ArangoDB) are used to store and manage graph structure data. Graph databases can efficiently store and query node and edge information, import node lists and edge lists into graph databases, and build a complete graph structure.

[0047] Calculate the edge weight according to business requirements, including the following steps:

[0048] Get the interaction frequency and interaction duration of the edge, normalize the interaction frequency and interaction duration, map the value range of the interaction frequency and interaction duration to [0, 1], get the normalized value of the interaction frequency and the normalized value of the interaction duration, sum the normalized value of the interaction frequency and the normalized value of the interaction duration to get the weight of the edge. The larger the weight, the more important the edge is.

[0049] The Louvain algorithm is used to divide the graph structure into several communities, which represent the natural group structure in the data. The community feature information is extracted from the community information generated by the Louvain algorithm, including the following steps:

[0050] The constructed graph structure is used as the input of the Louvain algorithm, including nodes, edges and edge weights. The Louvain algorithm is run to discover communities on the graph structure and obtain the community label of each node.

[0051] Count the number of edges and total weight within the community, analyze the internal connection strength of the community, count the number of edges and total weight between communities, analyze the connection strength between communities, generate a feature vector for each community, and calculate the scale coefficient of the community based on the feature vector of the community. The feature vector includes the total edge weight, community density and cohesion index. The total edge weight, community density and cohesion index are comprehensively calculated to obtain the scale coefficient. The expression is: Where GM is the scale factor, is the total edge weight, n represents the number of edges in the community, QZ i represents the weight of the ith edge, SM is the community density, NJ is the cohesion index, α, β, and γ are the proportional coefficients of the total edge weight, community density, and cohesion index, respectively, and α, β, and γ are all greater than 0;

[0052] After obtaining the scale coefficient, compare the scale coefficient with the preset coefficient threshold. The coefficient threshold is used to distinguish the size of the community. If the scale coefficient is greater than the coefficient threshold, the community is defined as a large community. If the scale coefficient is less than or equal to the coefficient threshold, the community is defined as a small community. The community's feature vector and scale coefficient are used to establish community feature information. Use graph visualization tools to display the community feature information to help understand the community division results.

[0053] The calculation logic of the cohesion index is: for a node in a community, the clustering coefficient calculation formula is as follows: Where C(i) is the clustering coefficient of the i-th node, E i is the actual number of edges between neighbors of the i-th node, k i is the number of neighbors of the i-th node, and the cohesion index of all nodes in the community is calculated, expressed as: Where |C| represents the total number of nodes in the community.

[0054] The calculation expression of community density is: In the formula, E C is the actual number of edges in the community, and |C| represents the total number of nodes in the community.

[0055] Larger cohesion indices and community densities are usually associated with smaller communities because nodes are more likely to form dense connections in small communities.

[0056] Based on the Q-learning algorithm, the community feature information is used to optimize the decision-making strategy, update the Q value and select the optimal action. The Q-learning model is iteratively optimized so that it can select the optimal decision-making action under different states, including the following steps:

[0057] Extract the community feature information of each community, use the extracted community feature information to construct the state representation, and select the initial state s from the current state space c , based on the greedy strategy ∈-select action a, randomly select an action with probability ∈, select the action with the maximum current Q value with probability 1-∈, in the current state s d Execute action a and get the new state s new And reward r, then update the Q value, the expression is: a z is all selected actions, τ is the learning rate, is the discount factor, and the new state s new As the current state d .

[0058] According to the updated Q-value table, the optimal action in each state is selected, and based on the selected optimal action, the current decision-making strategy is optimized so that it can select the optimal action in the future.

[0059] In order to better illustrate the Q value table update solution, this application example is as follows:

[0060] Define the state space: Each state can represent a combination of features of a community, such as the initial state s c Including the number of nodes, total edge weight, and scale factor of the community;

[0061] Define the action space: This can include different decision options, such as choosing actions including increasing resources, decreasing resources, or keeping them unchanged;

[0062] Initialize the Q table: initialize to zero;

[0063] Setting parameters: for example, τ = 0.1, ∈=0.1.

[0064] Extract features such as number of nodes, total edge weight, and scale coefficient from the community structure;

[0065] Select the initial state: for example, the feature combination s of the current community d =(50,1000,1.3);

[0066] Select action: Based on the greedy strategy ∈-select action a to increase resources;

[0067] Execute action: get the new state s after execution new=(52,1100,1.5) and reward r=10.

[0068] After updating the Q value according to the Q value update formula, the action with the largest Q value is selected in each state, and the optimized strategy is applied in actual business according to the community feature information.

[0069] When the convergence condition is reached, the optimized Q-learning model is output, and the performance value of the Q-learning model in different communities is calculated. The Q-learning model is evaluated based on the performance value, including the following steps:

[0070] When the number of iterations is greater than or equal to the number threshold, it is judged that the convergence condition is met, and the Q value table updated during the training process is saved as a file or variable as the optimized Q-learning model, and a group of communities are selected from the data set as test samples. These communities should cover communities with different characteristics and sizes, obtain the task completion rate and resource utilization efficiency of the Q-learning model, normalize the task completion rate and resource utilization efficiency, obtain the normalized value of the task completion rate and the normalized value of the resource utilization efficiency, sum the normalized value of the task completion rate and the normalized value of the resource utilization efficiency to obtain the performance value, if the performance value is greater than or equal to the performance threshold, the Q-learning model performance is evaluated to meet the standard, if the performance value is less than the performance threshold, the Q-learning model performance is evaluated to be unsatisfactory.

[0071] Adjust the reward function value of the Q-learning model according to the performance evaluation results to improve the decision-making effect of the model, and apply the Q-learning model with the adjusted reward function value to the subsequent BI decision management of the enterprise, including the following steps:

[0072] When the performance of the Q-learning model is evaluated to be up to standard, the reward function value of the Q-learning model is increased. When the performance of the Q-learning model is evaluated to be not up to standard, the reward function value of the Q-learning model is reduced. The adjustment algorithm is: In the formula, R current is the current reward function value, bx s is the performance value, bx y is the performance threshold.

[0073] Embodiment 3: The BI decision management system based on big data described in this embodiment includes a graph structure construction module, a community generation module, and a model optimization module:

[0074] Graph structure construction module: collect relevant business data through the API interface of the enterprise big data database, extract feature data after preprocessing the relevant business data, use the nodes in the feature data as nodes, and use the relationship between nodes as edges to build a graph structure, and send the graph structure to the community generation module;

[0075] Community generation module: The Louvain algorithm is used to divide the graph structure into several communities, which represent the natural group structure in the data. Community feature information is extracted from the community information generated by the Louvain algorithm and sent to the model optimization module.

[0076] Model optimization module: Based on the Q-learning algorithm, the community feature information is used to optimize the decision-making strategy, update the Q value and select the optimal action, and iteratively optimize the Q-learning model so that it can select the optimal decision-making action under different conditions. When the convergence conditions are reached, the optimized Q-learning model is output and the performance value of the Q-learning model in different communities is calculated. After evaluating the Q-learning model based on the performance value, the reward function value of the Q-learning model is adjusted according to the performance evaluation results to improve the decision-making effect of the model. The Q-learning model with adjusted reward function value is applied to the subsequent BI decision management of the enterprise.

[0077] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0078] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0079] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0080] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0081] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A BI decision management method based on big data, characterized by: The management method comprises the following steps: The management system collects relevant business data through the API interface of the enterprise big data database, extracts feature data after preprocessing the relevant business data, takes the nodes in the feature data as nodes, and constructs a graph structure with the relationships between nodes as edges; The graph structure is divided into several communities through the Louvain algorithm, and community feature information is extracted from the community information generated by the Louvain algorithm. The decision strategy is optimized based on the Q-learning algorithm using the community feature information, the Q value is updated and the optimal action is selected. The Q-learning model is iteratively optimized so that it can select the optimal decision action under different states. When the convergence condition is reached, the optimized Q-learning model is output, and the performance value of the Q-learning model in different communities is calculated. After evaluating the Q-learning model based on the performance value, the reward function value of the Q-learning model is adjusted according to the performance evaluation result, and the Q-learning model with the adjusted reward function value is applied to the subsequent BI decision-making management of the enterprise; The graph structure is divided into several communities by the Louvain algorithm, and community feature information is extracted from the community information generated by the Louvain algorithm, including the following steps: The constructed graph structure is used as the input of the Louvain algorithm, including nodes, edges and edge weights. The Louvain algorithm is run to discover communities on the graph structure and obtain the community label of each node. Generate a feature vector for each community, and calculate the scale coefficient of the community based on the feature vector of the community. The feature vector includes the total edge weight, community density, and cohesion index. The total edge weight, community density, and cohesion index are comprehensively calculated to obtain the scale coefficient. The expression is: , where is the scale factor, is the total edge weight, represents the number of edges within the community, Indicates The weight of the edge, is the community density, is the cohesion index, , , are the proportional coefficients of the total edge weight, community density, and cohesion index, respectively, and , , All are greater than 0; After obtaining the scale coefficient, compare the scale coefficient with the preset coefficient threshold. The coefficient threshold is used to distinguish the size of the community. If the scale coefficient is greater than the coefficient threshold, the community is defined as a large community. If the scale coefficient is less than or equal to the coefficient threshold, the community is defined as a small community. The community feature vector and the scale coefficient are used to establish community feature information, and the community feature information is displayed using a graph visualization tool. The calculation logic of the cohesion index is: for a node in a community, the clustering coefficient calculation formula is as follows: , where For the The clustering coefficient of nodes, For the The actual number of edges between neighbors of a node, For the The number of neighbors of a node is used to calculate the cohesion index of all nodes in the community. The expression is: , where Represents the total number of nodes in the community; The calculation expression of community density is: , where is the actual number of edges in the community, Represents the total number of nodes in the community.

2. The BI decision management method based on big data according to claim 1, characterized in that: Based on the Q-learning algorithm, the community feature information is used to optimize the decision-making strategy, update the Q value and select the optimal action, and iteratively optimize the Q-learning model, including the following steps: Extract community feature information of each community, construct state representation using the extracted community feature information, and select the initial state from the current state space , based on the greedy strategy Select Action , with probability Randomly choose an action with probability Select the action with the maximum current Q value, in the current state Execute an action , get the new state and rewards , and then update the Q value, the expression is: , For all selection actions, is the learning rate, is the discount factor, and the new state As current status , according to the updated Q value table, select the optimal action in each state, and optimize the current decision-making strategy based on the selected optimal action.

3. The BI decision management method based on big data according to claim 2 is characterized by: Calculate the performance value of the Q-learning model in different communities and evaluate the Q-learning model based on the performance value, including the following steps: When the number of iterations is greater than or equal to the number threshold, it is judged that the convergence condition is met, and the Q value table updated during the training process is saved as a file or variable as the optimized Q-learning model. A group of communities are selected from the data set as test samples, and the task completion rate and resource utilization efficiency of the Q-learning model are obtained. The task completion rate and resource utilization efficiency are normalized to obtain the normalized value of the task completion rate and the normalized value of the resource utilization efficiency. The normalized value of the task completion rate and the normalized value of the resource utilization efficiency are summed to obtain the performance value. If the performance value is greater than or equal to the performance threshold, the performance of the Q-learning model is evaluated to be up to standard. If the performance value is less than the performance threshold, the performance of the Q-learning model is evaluated to be not up to standard.

4. The BI decision management method based on big data according to claim 3 is characterized by: Adjust the reward function value of the Q-learning model according to the performance evaluation results, including the following steps: When the performance of the Q-learning model is evaluated to be up to standard, the reward function value of the Q-learning model is increased. When the performance of the Q-learning model is evaluated to be not up to standard, the reward function value of the Q-learning model is reduced. The adjustment algorithm is: , where is the current reward function value, is the performance value, is the performance threshold.

5. The BI decision management method based on big data according to claim 4 is characterized by: The nodes in the feature data are taken as nodes, and the relationships between the nodes are taken as edges to construct a graph structure, including the following steps: Determine the nodes in the graph and the relationship between the nodes based on the feature data, extract the attributes of each node from the feature data, the node attributes include user ID, age, gender, purchase history, and assign a unique node ID to each node; Define the relationship between nodes and use the relationship between nodes as the edge in the graph. Calculate the weight of the edge according to business needs. Create a node list containing the IDs and attributes of all nodes, create an edge list containing the start points, end points, and weights of all edges, and import the node list and edge list into the graph database to build the graph structure.

6. The BI decision management method based on big data according to claim 5 is characterized by: Calculate the edge weight according to business requirements, including the following steps: Get the interaction frequency and interaction duration of the edge, normalize the interaction frequency and interaction duration, map the value range of the interaction frequency and interaction duration to [0, 1], get the normalized value of the interaction frequency and the normalized value of the interaction duration, and sum the normalized value of the interaction frequency and the normalized value of the interaction duration to get the weight of the edge.

7. A BI decision management system based on big data, used to implement the management method according to any one of claims 1 to 6, characterized in that: Including graph structure building module, community generation module, and model optimization module: Graph structure construction module: collect relevant business data through the API interface of the enterprise big data database, extract feature data after preprocessing the relevant business data, use the nodes in the feature data as nodes, and use the relationship between nodes as edges to build a graph structure, and send the graph structure to the community generation module; Community generation module: The Louvain algorithm is used to divide the graph structure into several communities, which represent the natural group structure in the data. Community feature information is extracted from the community information generated by the Louvain algorithm and sent to the model optimization module. Model optimization module: Based on the Q-learning algorithm, the community feature information is used to optimize the decision-making strategy, update the Q value and select the optimal action, and iteratively optimize the Q-learning model so that it can select the optimal decision-making action under different conditions. When the convergence conditions are reached, the optimized Q-learning model is output and the performance value of the Q-learning model in different communities is calculated. After evaluating the Q-learning model based on the performance value, the reward function value of the Q-learning model is adjusted according to the performance evaluation results to improve the decision-making effect of the model. The Q-learning model with adjusted reward function value is applied to the subsequent BI decision management of the enterprise.

Images