A method and apparatus for constructing a decision knowledge graph based on event flow

Abstract

This invention discloses a method and apparatus for constructing a decision knowledge graph based on event streams. The method includes: collecting an event stream data set; the event stream data set includes several event stream data sets; the event stream data includes event logs, event-related sensor data, event subject information, event action information, event time information, time context information, and event object information; preprocessing the event stream data set to obtain a preprocessed event stream data set; and performing decision knowledge graph construction processing on the preprocessed event stream data set to obtain a decision knowledge graph; the decision knowledge graph includes decision entities and the relationship information between each decision entity.

Description

Technical Field

[0001] This invention relates to the fields of knowledge graphs, industrial data processing, and intelligent decision-making technologies, specifically to a method and apparatus for constructing a decision knowledge graph based on event flow. Background Technology

[0002] In engineering decision-making, such as industrial production scheduling, intelligent traffic management, and emergency response, the scientific nature and timeliness of decisions directly determine the effectiveness and safety level of project implementation. With the development of industrial internet and IoT technologies, event data generated during the decision-making process exhibits characteristics of multi-source, real-time, and high-dimensionality. This data encompasses various types, including event logs, sensor-collected data, subject action information, and temporal context, providing a rich foundation for the mining and reuse of decision-making knowledge.

[0003] In current engineering practice, the accumulation and application of decision knowledge mainly rely on traditional methods such as manually compiling decision cases and building rule bases, which have many limitations. On the one hand, manual processing of multi-source event stream data is inefficient, unable to meet the real-time processing needs of massive amounts of data, and prone to errors due to subjective judgment. On the other hand, existing methods for constructing decision knowledge graphs generally lack targeted processing mechanisms for multi-source event stream data, and are insufficient in handling issues such as noise and missing values ​​in sensor data, resulting in poor quality of the constructed knowledge graph data. Furthermore, in the entity relationship identification stage, traditional methods struggle to effectively utilize the temporal and attribute association features of event data, often leading to misjudgments and omissions of relationships. This makes it difficult for the constructed decision knowledge graph to accurately reflect the actual decision-making logic and provide reliable support for subsequent decisions.

[0004] Furthermore, existing technologies for constructing decision knowledge graphs often rely on a single data source, lacking a comprehensive acquisition mechanism for open-source decision process databases, existing decision texts, and real-time sensor data, resulting in incomplete knowledge coverage. In the relationship identification process, the absence of a scientific mechanism for fusion distance calculation and multi-model fusion identification makes it difficult to accurately distinguish different types of entity relationships, and the accuracy of relationship existence judgments is insufficient, further impacting the practicality of decision knowledge graphs. These problems make existing decision knowledge graph construction methods difficult to implement in engineering practice, failing to meet the demands for precision and intelligence in engineering decision-making. Therefore, there is an urgent need for a method that can efficiently process multi-source event stream data and accurately construct decision knowledge graphs. Summary of the Invention

[0005] This invention primarily addresses the problem of how to accurately construct a decision knowledge graph based on multi-source event stream data. This invention discloses a method and apparatus for constructing a decision knowledge graph based on event streams.

[0006] In a first aspect, this invention discloses a method for constructing a decision knowledge graph based on event flow, comprising: S1, Collect an event stream data set; the event stream data set includes several event stream data; the event stream data includes event logs, event-related sensor data, event subject information, event action information, event time information, time context information, and event object information; S2, preprocess the event stream data set to obtain a preprocessed event stream data set; S3, perform decision knowledge graph construction processing on the preprocessed event stream data set to obtain a decision knowledge graph; the decision knowledge graph includes decision entities and the relationship information between each decision entity.

[0007] The preprocessing of the event stream data set to obtain a preprocessed event stream data set includes: S21, perform data cleaning processing on the event stream data set to obtain a first data set; S22, perform data category discrimination processing on the first data set to obtain a preprocessed event stream data set.

[0008] The step of constructing a decision knowledge graph from the preprocessed event stream data set includes: S31, perform entity recognition processing on the preprocessed event stream data set to obtain a decision entity set; the decision entity set includes each decision entity and its corresponding attribute information set; the attribute information set includes a data sequence for each type of attribute; the data sequence is a sequence of data for each type of attribute collected at a preset sampling time interval; S32, perform relationship identification processing on the set of decision entities to obtain a set of relationship information; the set of relationship information includes the relationship information between each decision entity; S33, using the set of decision entities and the set of relational information, a decision knowledge graph is constructed.

[0009] The process of performing relation identification processing on the set of decision entities to obtain a set of relation information includes: S321, For every two decision entities in the decision entity set, perform relationship category discrimination processing to obtain the corresponding relationship category information; S322, For every two decision entities in the decision entity set, perform relationship existence discrimination processing to obtain the corresponding relationship existence information; S323. Using the relationship category information and relationship existence information of all two decision entities, a set of relationship information is constructed.

[0010] The process of performing relationship category discrimination on every two decision entities in the decision entity set to obtain corresponding relationship category information includes: S3211, Subtract the data sequences of the same type of attribute in the attribute information sets of every two decision entities in the decision entity set to obtain the difference data sequence of each type of attribute; S3212, Perform statistical processing on the differential data sequences of each attribute to obtain the corresponding mean and variance; S3213, using the mean and variance of all class attributes, construct a set of statistical distribution values; S3214, Using the set of statistical distribution values, a spatial distribution vector is constructed; the elements of the spatial distribution vector are the mean and variance of all class attributes; S3215, calculate the fusion distance between the spatial distribution vector and the preset spatial standard vector to obtain the fusion distance value; S3216, Determine the value range to which the fusion distance value belongs based on a preset set of value ranges; S3217, Based on the value range to which it belongs, determine the relationship type corresponding to the value range to which it belongs, which is the determined relationship category information.

[0011] The expression for calculating the fusion distance is: , in, Let be the norm value of the i-th vector of the difference vector between the spatially distributed vector and the preset spatial standard vector. The first to third vector norms are respectively... Norms, Euclidean norms, Frobenius norms Let i be the Legendre polynomial of order i. This represents the sum of the norm values ​​of all vectors. This represents the fusion distance value.

[0012] The step of performing relationship existence determination processing on every two decision entities in the decision entity set to obtain corresponding relationship existence information includes: S3221, Subtract the data sequences of the same type of attribute in the attribute information sets of every two decision entities in the decision entity set to obtain the difference data sequence of each type of attribute; S3222, perform statistical processing on the differential data sequences of each attribute to obtain the corresponding mean and variance; S3223, using the mean and variance of all class attributes, construct a set of statistical distribution values; S3224, Using the set of statistical distribution values, a spatial distribution vector is constructed; the elements of the spatial distribution vector are the mean and variance of all class attributes; S3225, using a pre-set, trained first network model, the spatial distribution vector is processed to obtain a first output value and a corresponding first confidence value; S3226, Using a pre-set, trained second network model, the spatial distribution vector is processed to obtain a second output value and a corresponding second confidence value; S3227, Perform a weighted fusion process on the first output value and the second output value to obtain the relationship fusion value; S3228, Determine whether the relationship fusion value is greater than a preset first discrimination threshold, and obtain a first discrimination result; S3229, if the first discrimination result is greater than, the relationship information between the two decision entities is confirmed to exist; if the first discrimination result is not greater than, the relationship information between the two decision entities is confirmed to not exist.

[0013] A second aspect of this invention discloses an apparatus for constructing a decision knowledge graph based on event flow, the apparatus comprising: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the event-flow-based decision knowledge graph construction method.

[0014] In a third aspect, the present invention discloses a computer-readable storage medium storing computer instructions, which, when invoked by a computer, are used to execute the event-flow-based decision knowledge graph construction method.

[0015] A fourth aspect of this invention discloses an information data processing terminal, which is used to implement the event-stream-based decision knowledge graph construction method.

[0016] The beneficial effects of this invention are as follows: This invention constructs a data foundation by collecting multi-source event stream data, covering various types such as event logs, sensor-collected data, and information on event subjects and objects. It also clarifies multiple channels for data collection, including open-source decision process databases, existing decision text recognition, and sensor data collection. This comprehensively covers various key data in the decision-making process, solving the problems of single data sources and incomplete knowledge coverage in the construction of decision knowledge graphs in existing technologies. It provides rich and comprehensive data support for the subsequent construction of accurate decision knowledge graphs, and improves the versatility and coverage of decision knowledge graphs.

[0017] This invention employs a combined data cleaning and data category discrimination approach in the data preprocessing stage. Data cleaning, tailored to the characteristics of sensor data, accurately identifies and smooths noisy data while filling in missing values, effectively improving the quality of the original data. Data category discrimination compares the data with preset categories, deleting inconsistent data to ensure the validity and standardization of the preprocessed data. This targeted preprocessing mechanism solves the problem of inconsistent data quality in multi-source event streams in engineering practice, avoids interference from poor-quality data in subsequent knowledge graph construction, and ensures the accuracy of subsequent entity and relationship recognition.

[0018] This invention, in the entity recognition process, combines a pre-defined decision entity name database with a named entity recognition algorithm to accurately extract decision entities and their attribute information. Simultaneously, it presents the attribute information as a data sequence formed at pre-defined sampling time intervals, fully preserving the temporal characteristics of the attribute data and providing a crucial basis for the accurate determination of subsequent entity relationships. Compared to existing technologies that only extract entity names while ignoring the temporal characteristics of attributes, this invention can more comprehensively mine the feature information of entities, improving the completeness and accuracy of entity recognition.

[0019] This invention refines relationship information into relationship category information and relationship existence information during the relationship identification process. It first constructs a spatial distribution vector by calculating the statistical characteristics of attribute difference data sequences, then uses fusion distance calculation to determine the relationship category. Simultaneously, it employs a dual-network model to fuse and determine relationship existence, forming a scientifically rigorous relationship identification mechanism. The fusion distance calculation comprehensively considers the influence of multiple vector norms and Legendre polynomials, accurately distinguishing different types of entity relationships. The weighted fusion processing of the dual-network model combines the outputs and confidence levels of both models, improving the accuracy of relationship existence determination. It also explicitly excludes relationship category information where relationship existence information is nonexistent, avoiding interference from invalid information in knowledge graph construction, and significantly improving the accuracy and reliability of relationship identification. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the implementation of the method of the present invention. Detailed Implementation

[0021] To better understand the content of this invention, an embodiment is provided here.

[0022] Figure 1 This is a flowchart illustrating the implementation of the method of the present invention.

[0023] In a first aspect, this invention discloses a method for constructing a decision knowledge graph based on event flow, comprising: S1, Collect an event stream data set; the event stream data set includes several event stream data; the event stream data includes event logs, event-related sensor data, event subject information, event action information, event time information, time context information, and event object information; S2, preprocess the event stream data set to obtain a preprocessed event stream data set; S3, perform decision knowledge graph construction processing on the preprocessed event stream data set to obtain a decision knowledge graph; the decision knowledge graph includes decision entities and the relationship information between each decision entity; The definitions of the event subject information, event action information, event time information, time context information, and event object information are all based on the relevant definitions in text data processing.

[0024] The collected event stream data set can be obtained from an open-source decision process database, or by text recognition of existing decision texts, or by collecting sensor data.

[0025] The preprocessing of the event stream data set to obtain a preprocessed event stream data set includes: The event stream data set is cleaned to obtain a first data set; The first data set is subjected to data category discrimination processing to obtain a preprocessed event stream data set; The data cleaning process includes filling in missing values ​​and smoothing noisy data. The smoothing of noisy data involves first identifying the noisy data, and then smoothing the noisy data based on the data before and after the noisy data. The noisy data is a value that is less than the sensor's detection sensitivity for the observed data, or greater than the sensor's measurement limit for the observed data.

[0026] The data category discrimination process involves determining whether the category of each type of data in the first data set is consistent with a preset data category, deleting inconsistent data from the first data set, and obtaining a preprocessed event stream data set.

[0027] The step of constructing a decision knowledge graph from the preprocessed event stream data set includes: S31, perform entity recognition processing on the preprocessed event stream data set to obtain a decision entity set; the decision entity set includes each decision entity and its corresponding attribute information set; the attribute information set includes a data sequence for each type of attribute; the data sequence is a sequence of data for each type of attribute collected at a preset sampling time interval; S32, perform relationship identification processing on the set of decision entities to obtain a set of relationship information; the set of relationship information includes the relationship information between each decision entity; S33, using the set of decision entities and the set of relational information, a decision knowledge graph is constructed.

[0028] The relationship information includes relationship category information and relationship existence information; in particular, if the relationship existence information is zero for some relationships, then the relationship category information has no reference value and is not used in the knowledge graph construction process.

[0029] The entity recognition process is achieved by using a pre-defined decision entity name database and a named entity recognition algorithm to process the text data. The named entity recognition algorithm can be implemented using a CRF model.

[0030] The process of performing relation identification processing on the set of decision entities to obtain a set of relation information includes: S321, For every two decision entities in the decision entity set, perform relationship category discrimination processing to obtain the corresponding relationship category information; S322, For every two decision entities in the decision entity set, perform relationship existence discrimination processing to obtain the corresponding relationship existence information; S323. Using the relationship category information and relationship existence information of all two decision entities, a set of relationship information is constructed.

[0031] The process of performing relationship category discrimination on every two decision entities in the decision entity set to obtain corresponding relationship category information includes: S3211, Subtract the data sequences of each type of attribute in the attribute information set of every two decision entities in the decision entity set to obtain the difference data sequence of each type of attribute; S3212, Perform statistical processing on the differential data sequences of each attribute to obtain the corresponding mean and variance; S3213, using the mean and variance of all class attributes, constructs a set of statistical distribution values.

[0032] S3214, Using the set of statistical distribution values, a spatial distribution vector is constructed; the elements of the spatial distribution vector are the mean and variance of all class attributes; S3215, calculate the fusion distance between the spatial distribution vector and the preset spatial standard vector to obtain the fusion distance value; S3216, Determine the value range to which the fusion distance value belongs based on a preset set of value ranges; S3217, Based on the value range to which it belongs, determine the relationship type corresponding to the value range to which it belongs, which is the determined relationship category information.

[0033] The preset set of value ranges includes the value ranges corresponding to relation types ranging from 1 to 6, with each value range corresponding to a relation type.

[0034] The expression for calculating the fusion distance is: , in, Let be the norm value of the i-th vector of the difference vector between the spatially distributed vector and the preset spatial standard vector. The first to third vector norms are respectively... Norms, Euclidean norms, Frobenius norms Let i be the Legendre polynomial of order i. This represents the sum of the norm values ​​of all vectors. This represents the fusion distance value.

[0035] This invention introduces a fusion distance calculation formula that comprehensively considers multiple methods of measuring the difference between spatially distributed vectors and preset standard vectors when determining relation categories. This formula achieves nonlinear fusion of different geometric features in the vector space by weighting vector norms of different orders and introducing Legendre polynomials as weighting coefficients. This multi-dimensional distance calculation method overcomes the limitations of single distance metrics under specific data distributions, more accurately capturing the distributional differences of different relation categories in the feature space, thereby significantly improving the accuracy and robustness of relation category determination and providing a solid foundation for the subsequent construction of high-quality decision knowledge graphs.

[0036] The step of performing relationship existence determination processing on every two decision entities in the decision entity set to obtain corresponding relationship existence information includes: S3221, Subtract the data sequences of each type of attribute in the attribute information set of every two decision entities in the decision entity set to obtain the difference data sequence of each type of attribute; S3222, perform statistical processing on the differential data sequences of each attribute to obtain the corresponding mean and variance; S3223, using the mean and variance of all class attributes, constructs a set of statistical distribution values.

[0037] S3224, Using the set of statistical distribution values, a spatial distribution vector is constructed; the elements of the spatial distribution vector are the mean and variance of all class attributes; S3225, using a pre-set, trained first network model, the spatial distribution vector is processed to obtain a first output value and a corresponding first confidence value; S3226, Using a pre-set, trained second network model, the spatial distribution vector is processed to obtain a second output value and a corresponding second confidence value; S3227, Perform a weighted fusion process on the first output value and the second output value to obtain the relationship fusion value; S3228, Determine whether the relationship fusion value is greater than a preset first discrimination threshold, and obtain a first discrimination result; S3229, if the first discrimination result is greater than, the relationship information between the two decision entities is confirmed to exist; if the first discrimination result is not greater than, the relationship information between the two decision entities is confirmed to not exist. The expression for the fusion weighting process is: in, and These are the first output value and the second output value, respectively. and These are the first confidence level value and the second confidence level value, respectively. This is the value for relational fusion.

[0038] This invention employs an exponential fusion weighted calculation formula to dynamically adjust the weight ratio of the outputs of the first and second network models in the final decision when determining the existence of relationships. This formula utilizes the characteristics of the exponential function to automatically focus on and enhance the output of the high-confidence model while effectively suppressing interference from the low-confidence model. This confidence-based adaptive fusion mechanism fully combines the complementary advantages of the two network models in feature extraction and pattern recognition, effectively reducing the misjudgment rate of a single model in complex decision-making scenarios and significantly improving the reliability and stability of relationship existence information determination.

[0039] The first network model is a multi-layer feedforward neural network structure; the hidden layer nodes of the neural network adopt piecewise linear activation functions and introduce an adaptive learning rate adjustment mechanism based on gradient magnitude to achieve sensitive response to different relation types; during network training, a backpropagation algorithm with momentum term is adopted, combined with L2 regularization term to suppress overfitting; the output layer of the neural network generates a first output value and a corresponding confidence value. The first network model employs an adaptive decision-making framework that integrates multi-scale feature extraction and dynamic weight adjustment mechanisms, including: Wavelet packet decomposition is performed on the input vector to obtain its energy distribution characteristics in different frequency bands; Calculate the standard deviation of energy in each frequency band and Shannon entropy to form a frequency domain sensitivity index; Based on the frequency domain sensitivity index, the weighting coefficients of each feature channel in the neural network input layer are dynamically adjusted so that highly sensitive features have a higher weight in the decision-making process. Meanwhile, a gated attention module is introduced in the intermediate layer of the network. This module generates an attention weight map by calculating the cosine similarity matrix between feature vectors and performs weighted aggregation on the hidden states, thereby enhancing the ability to identify key relational components. The output of the gated attention module is used as the input of the next layer to achieve adaptive focusing and enhancement of features.

[0040] The second network model includes a first convolutional module, a second convolutional module, a third convolutional module, a fourth convolutional module, a fifth convolutional module, a sixth convolutional module, a seventh convolutional module, an eighth convolutional module, a thirteenth convolutional module, a first normalization module, a second normalization module, a third normalization module, a fourth normalization module, a first upsampling module, a second upsampling module, a first fusion module, a first pooling module, a first multidimensional feature extraction module, a second multidimensional feature extraction module, a third multidimensional feature extraction module, a fourth multidimensional feature extraction module, a first deep feature extraction module, a second deep feature extraction module, and a third deep feature extraction module; The input of the first convolutional module is configured as the input of the second network model. The output of the first convolutional module is connected to the input of the first normalization module. The output of the first normalization module is connected to the input of the second convolutional module. The output of the second convolutional module is connected to the input of the second normalization module. The output of the second normalization module is connected to the input of the first multidimensional feature extraction module. The output of the first multidimensional feature extraction module is connected to the input of the first deep feature extraction module and the input of the seventh convolutional module. The output of the first deep feature extraction module is connected to the input of the second multidimensional feature extraction module. The output of the second multidimensional feature extraction module is connected to the input of the second deep feature extraction module. The output of the second deep feature extraction module is connected to the input of the third multidimensional feature extraction module. The output of the third multidimensional feature extraction module is connected to the input of the third deep feature extraction module. The output of the third deep feature extraction module is connected to the input of the fourth multidimensional feature extraction module. The output of the fourth multidimensional feature extraction module is connected to the input of the thirteenth convolutional module. The output of the third convolutional module is connected to the input of the third convolutional module; the output of the third convolutional module is connected to the input of the fourth convolutional module; the output of the fourth convolutional module is connected to the input of the fifth convolutional module; the output of the fifth convolutional module is connected to the input of the first pooling module; the output of the first pooling module is connected to the input of the sixth convolutional module; the output of the sixth convolutional module is connected to the input of the first upsampling module; the output of the first upsampling module is connected to the input of the first fusion module; the output of the seventh convolutional module is connected to the input of the first activation module; the output of the first activation module is connected to the input of the first fusion module; the output of the first fusion module is connected to the input of the fourth normalization module; the output of the fourth normalization module is connected to the input of the second activation module; the output of the second activation module is connected to the input of the eighth convolutional module; the output of the eighth convolutional module is connected to the input of the second upsampling module; the output of the second upsampling module is configured as the model output of the second network model, used to output a second output value and a corresponding second confidence value.

[0041] The multidimensional feature extraction module can be composed of multiple convolutional modules connected in series; The deep feature extraction module can be implemented through self-attention feature extraction.

[0042] The training of the first and second network models can be achieved using the steepest descent method based on the training dataset.

[0043] The first network model in this invention employs an adaptive decision-making framework that integrates multi-scale feature extraction and dynamic weight adjustment. It obtains frequency domain features through wavelet packet decomposition and enhances key feature recognition capabilities by combining a gated attention module, enabling adaptive responses to different types of relationship features. The second network model, through the synergistic effect of multi-layer convolution, normalization, feature extraction, and upsampling modules, fully mines the deep features of spatially distributed vectors. This targeted design of the dual network models improves relationship discrimination capabilities from two perspectives: adaptive feature response and deep feature mining, ensuring the stability and accuracy of relationship recognition in various engineering decision-making scenarios.

[0044] This invention establishes a complete decision-making knowledge graph construction process by completing event stream data acquisition, preprocessing, entity recognition, relationship recognition, and knowledge graph construction in stages. Each stage is closely linked and progressively advanced, enabling efficient processing of massive amounts of multi-source event stream data and automated construction of the decision-making knowledge graph. This method requires minimal manual intervention, effectively improving the efficiency of decision-making knowledge graph construction, reducing labor costs, and ensuring that the constructed knowledge graph accurately reflects the relationships and decision-making logic between decision entities. This provides reliable knowledge support for engineering decisions, enhancing the scientific rigor and timeliness of engineering decisions, and demonstrating significant engineering application value.

[0045] In all embodiments of the present invention, the variables involved in all computational expressions or mathematical functions have been dimensionlessized before computation.

[0046] In all embodiments of the present invention, the values ​​of the independent variables in the input of all computational expressions or mathematical functions meet the reasonable requirements of the input range of the computational expressions or mathematical functions, and can ensure that the computational expressions or mathematical functions can be calculated smoothly without violating physical laws or mathematical rules.

[0047] A second aspect of this invention discloses an apparatus for constructing a decision knowledge graph based on event flow, the apparatus comprising: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the event-flow-based decision knowledge graph construction method.

[0048] In a third aspect, the present invention discloses a computer-readable storage medium storing computer instructions, which, when invoked by a computer, are used to execute the event-flow-based decision knowledge graph construction method.

[0049] A fourth aspect of this invention discloses an information data processing terminal, which is used to implement the event-stream-based decision knowledge graph construction method.

[0050] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A method for constructing a decision knowledge graph based on event flow, characterized in that, include: S1, Collect an event stream data set; the event stream data set includes several event stream data; the event stream data includes event logs, event-related sensor data, event subject information, event action information, event time information, time context information, and event object information; S2, preprocess the event stream data set to obtain a preprocessed event stream data set; S3, perform decision knowledge graph construction processing on the preprocessed event stream data set to obtain a decision knowledge graph; The decision knowledge graph includes decision entities and information about the relationships between them.

2. The method for constructing a decision knowledge graph based on event flow as described in claim 1, characterized in that, The preprocessing of the event stream data set to obtain a preprocessed event stream data set includes: S21, perform data cleaning processing on the event stream data set to obtain a first data set; S22, perform data category discrimination processing on the first data set to obtain a preprocessed event stream data set.

3. The method for constructing a decision knowledge graph based on event flow as described in claim 1, characterized in that, The step of constructing a decision knowledge graph from the preprocessed event stream data set includes: S31, perform entity recognition processing on the preprocessed event stream data set to obtain a decision entity set; the decision entity set includes each decision entity and its corresponding attribute information set; the attribute information set includes a data sequence for each type of attribute; the data sequence is a sequence of data for each type of attribute collected at a preset sampling time interval; S32, perform relationship identification processing on the set of decision entities to obtain a set of relationship information; the set of relationship information includes the relationship information between each decision entity; S33, using the set of decision entities and the set of relational information, a decision knowledge graph is constructed.

4. The method for constructing a decision knowledge graph based on event flow as described in claim 3, characterized in that, The process of performing relation identification processing on the set of decision entities to obtain a set of relation information includes: S321, For every two decision entities in the decision entity set, perform relationship category discrimination processing to obtain the corresponding relationship category information; S322, For every two decision entities in the decision entity set, perform relationship existence discrimination processing to obtain the corresponding relationship existence information; S323. Using the relationship category information and relationship existence information of all two decision entities, a set of relationship information is constructed.

5. The method for constructing a decision knowledge graph based on event flow as described in claim 4, characterized in that, The process of performing relationship category discrimination on every two decision entities in the decision entity set to obtain corresponding relationship category information includes: S3211, Subtract the data sequences of the same type of attribute in the attribute information sets of every two decision entities in the decision entity set to obtain the difference data sequence of each type of attribute; S3212, Perform statistical processing on the differential data sequences of each attribute to obtain the corresponding mean and variance; S3213, using the mean and variance of all class attributes, construct a set of statistical distribution values; S3214, Using the set of statistical distribution values, a spatial distribution vector is constructed; the elements of the spatial distribution vector are the mean and variance of all class attributes; S3215, calculate the fusion distance between the spatial distribution vector and the preset spatial standard vector to obtain the fusion distance value; S3216, Determine the value range to which the fusion distance value belongs based on a preset set of value ranges; S3217, Based on the value range to which it belongs, determine the relationship type corresponding to the value range to which it belongs, which is the determined relationship category information.

6. The method for constructing a decision knowledge graph based on event flow as described in claim 5, characterized in that, The expression for calculating the fusion distance is: ， in, Let $\frac{i}{i}$ be the norm value of the difference vector between the spatially distributed vector and the preset spatial standard vector. The first to third vector norms are respectively: Norms, Euclidean norms, Frobenius norms Let i be the Legendre polynomial of order i. This represents the sum of the norm values ​​of all vectors. This represents the fusion distance value.

7. The method for constructing a decision knowledge graph based on event flow as described in claim 4, characterized in that, The step of performing relationship existence determination processing on every two decision entities in the decision entity set to obtain corresponding relationship existence information includes: S3221, Subtract the data sequences of the same type of attribute in the attribute information sets of every two decision entities in the decision entity set to obtain the difference data sequence of each type of attribute; S3222, perform statistical processing on the differential data sequences of each attribute to obtain the corresponding mean and variance; S3223, using the mean and variance of all class attributes, construct a set of statistical distribution values; S3224, Using the set of statistical distribution values, a spatial distribution vector is constructed; the elements of the spatial distribution vector are the mean and variance of all class attributes; S3225, using a pre-set, trained first network model, the spatial distribution vector is processed to obtain a first output value and a corresponding first confidence value; S3226, Using a pre-set, trained second network model, the spatial distribution vector is processed to obtain a second output value and a corresponding second confidence value; S3227, Perform a weighted fusion process on the first output value and the second output value to obtain the relationship fusion value; S3228, Determine whether the relationship fusion value is greater than a preset first discrimination threshold, and obtain a first discrimination result; S3229, if the first discrimination result is greater than, the relationship information between the two decision entities is confirmed to exist; if the first discrimination result is not greater than, the relationship information between the two decision entities is confirmed to not exist.

8. A decision knowledge graph construction device based on event flow, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the event-flow-based decision knowledge graph construction method as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, which, when invoked by a computer, are used to execute the event-flow-based decision knowledge graph construction method as described in any one of claims 1 to 7.

10. An information data processing terminal, characterized in that, The information data processing terminal is used to implement the event-flow-based decision knowledge graph construction method as described in any one of claims 1 to 7.

Images