Intelligent decision graph construction method based on dynamic time sequence event data

Abstract

The invention is suitable for the technical field of data analysis, and provides an intelligent decision graph construction method based on dynamic time sequence event data, which comprises the following steps: reading and cleaning original event data, processing fields, expanding nested information through a structured analysis and entity alignment technology, and generating cleaned structured data; the frequency of different field combinations is calculated, a frequency feature column is generated in combination with a timestamp and a weight factor, and a multi-dimensional relation is combined; screening event pairs based on a dynamic frequency threshold, constructing a directed graph data structure, taking unique identifiers of the events as nodes, establishing edges of a graph, setting edge weights, and optimizing a time sequence relationship between the events through time sequence matching and a dynamic weighting mechanism; and outputting a graph containing the node number, the edge number and the graph edge data screened based on the frequency. According to the method, the accurate, efficient and intelligent graph construction method is realized, the technical progress in the field of dynamic event data analysis is promoted, and the method can be applied to an actual scene in which dynamic event data needs to be processed and an associated graph needs to be constructed.

Description

Technical Field

[0001] The present invention belongs to the technical field of data analysis, and particularly relates to a method for constructing an intelligent decision-making graph based on dynamic time-series event data. Background Art

[0002] With the advent of the big data era, the amount of data in various fields of society has grown exponentially. Especially in the fields of event analysis, public opinion monitoring, and social network analysis, the processing and mining of massive data have become important research directions. Traditional data analysis methods focus on structured data and usually analyze a single data source or static data. However, in practical applications, data is often highly complex, not only containing rich association relationships and spatio-temporal characteristics, but also often spanning multiple sources and dimensions, which undoubtedly poses a huge challenge to the existing technologies.

[0003] Currently, network analysis, text mining, and graph data processing technologies have been widely applied in the fields of public opinion analysis, social network analysis, and event prediction. Especially the graph data processing technology can effectively reveal the potential connections and patterns between data items by constructing a relationship graph between data. In the graph, nodes represent different data objects, and edges represent the relationships between them. This graph-based construction and analysis method provides a new perspective for data mining.

[0004] However, there are still certain limitations in the existing technologies. In terms of the graph construction process, existing research generally adopts manual rules or named entity recognition (NER) and relation extraction technologies based on natural language processing (NLP) in the data processing stage to identify event subjects, event contents, and their associations. However, such methods have obvious limitations: on the one hand, manual rules rely on the experience of domain experts and are difficult to adapt to different scenarios and complex data; on the other hand, NLP-based methods usually rely on large-scale labeled datasets, are limited by the quality and coverage of training data, and are prone to introducing noise and errors. In addition, due to the diversity of natural language expressions, the same event may be described in different ways in different texts, resulting in difficult information matching and affecting the accuracy of time series analysis. On this basis, there are often many problems in the constructed graphs. Due to over-reliance on the above defective methods, the timeliness of event occurrence and multi-level relationships are often ignored in the process of graph construction. In the graph data analysis process, existing graph data analysis methods usually take static graphs as the analysis basis and lack consideration of dynamic event streams, resulting in analysis results that may be lagging or lack timeliness in a rapidly changing social environment. Even though some relatively advanced technologies currently attempt to combine time series analysis with network graph construction to improve this dilemma, there are still intractable problems such as extremely high computational complexity and large resource consumption when dealing with large-scale heterogeneous data sources and multi-dimensional relationships. Therefore, how to improve the automation and intelligence level of graph construction while ensuring timeliness has become one of the difficulties in current technologies. In view of this, the present invention proposes an intelligent decision-making graph construction method based on dynamic time series event data. Summary of the Invention

[0005] The purpose of the present invention is to provide an intelligent decision-making graph construction method based on dynamic time series event data, aiming to solve the problems proposed in the above background technology.

[0006] The purpose of the present invention is achieved through the following technical solutions: An intelligent decision-making graph construction method based on dynamic time series event data, comprising the following steps: Step S1: Data preprocessing; Read and clean the original event data, perform formatting conversion and normalization processing on the time fields, fill and unify the formats of the text fields, and expand the nested information through structured parsing and entity alignment technologies to generate the cleaned structured data; Step S2: Frequency calculation and multi-dimensional relationship merging; Calculate the frequencies of different field combinations, generate frequency feature columns in combination with timestamps and weight factors, and merge multi-dimensional relationships to reflect the association strengths between different field combinations; Step S3: Graph construction; Screen event pairs based on a dynamic frequency threshold, construct a directed graph data structure, use the unique event identifier as a node, establish the edges of the graph and set edge weights, and optimize the temporal relationship between events through temporal matching and dynamic weighting mechanisms; Step S4: Output and application; Output a graph spectrum containing the number of nodes, the number of edges, and graph edge data filtered by frequency, supporting visualization display and analysis applications.

[0007] Furthermore, the data preprocessing step specifically includes: Reading and cleaning: Read the original event data, clean null values, remove invalid data, and perform outlier detection operations; Time field processing: Use natural language processing technology to format and convert the time fields in the data, and filter out eligible event data according to the set time range; Text field processing: Fill null values in the text fields with "unknown", uniformly convert them to string format, use a deep learning-based text parsing method to process the text fields and event descriptions, and use the safe_literal_eval method to parse dictionary or list type data in the fields; Data expansion: Perform an "expansion" operation on each column of data, split fields containing multiple values, and use a structured parsing strategy to decompose nested information into independent data units. Normalize the same entities in different data sources through entity alignment technology to finally obtain the cleaned structured data.

[0008] Furthermore, the frequency calculation and multi-dimensional relationship merging step specifically includes: Frequency calculation: Calculate the frequencies of different field combinations in the data. Introduce a weight factor in the calculation process by combining event timestamp information, calculate the associations at different levels, and generate a new frequency feature column for each combination; Multi-dimensional relationship merging: Merge the generated frequency feature columns to obtain data with enhanced frequencies to reflect the association strength between different field combinations.

[0009] Furthermore, the graph construction step specifically includes: Screen event pairs: Based on the data with enhanced frequencies, set a dynamic frequency threshold to screen event pairs that meet the preset frequency threshold; Establish a graph structure: In the directed graph data structure, use the unique identifier of the event data as a node, and establish the edges of the graph according to the frequencies between the screened event pairs. The prerequisite for establishing an associated edge in the graph is that the frequency value of the event pair is greater than the preset threshold; Temporal Matching and Relationship Optimization: Perform temporal matching on the event relationships in the graph, construct graph edges that conform to the temporal order based on the event timestamps, and adopt a dynamic weighting mechanism to dynamically assign weights to the event relationships; Edge Weight Setting: The weight value of the edge is determined by the frequency information and represents the strength of the relationship between events.

[0010] Furthermore, the graph learns the association patterns between events through a graph neural network and supports real-time updating of causal relationships.

[0011] A computer device for implementing the above-mentioned intelligent decision-making graph construction method based on dynamic temporal event data, including: A central processing unit for performing data preprocessing and event analysis tasks; A graphics processing unit for accelerating the operations of the graph neural network during the construction of the event graph; Memory, serving as a temporary data storage space; A storage unit for storing the original data, calculation intermediate results, and the finally constructed event graph; A network communication module for realizing the communication between the computer device and a remote server or cloud computing resources, achieving data sharing and distributed computing; An input device for providing an operation interface; A display device for visualizing the event graph and presenting the data analysis results; A data interface for connecting an external storage device or sensor; An operating system and software environment for providing a running environment.

[0012] Furthermore, the central processing unit reads the Excel data file in the storage unit, executes the data preprocessing steps, and temporarily stores the cleaned data in the memory; the central processing unit executes the frequency calculation and multi-dimensional relationship merging steps, stores the calculation results in the storage unit, and displays the preliminary analysis results through the display device; the graphics processing unit processes large-scale data, performs graph neural network calculations, and executes the graph construction steps, during which the network communication module supports the interaction between the computer device and the remote server; finally, the constructed event graph is stored in the storage unit, and the display device visualizes the graph results through Matplotlib and NetworkX.

[0013] Compared with the prior art, the beneficial effects of the present invention are: Construction of a temporally enhanced graph: Different from traditional methods for constructing static data graphs, the present invention is based on temporal event data, capable of effectively capturing the temporal dependence and dynamic changes between events, ensuring that the nodes and edges in the graph conform to the propagation order of actual events, thereby improving the accuracy and practicality of the graph, effectively overcoming the static nature and limitations of traditional methods, and providing a more accurate and intelligent data analysis framework.

[0014] Frequency analysis of multi-dimensional relationships: The present invention performs frequency calculation and dynamic merging through multi-dimensional data relationships (such as the main character and event type, the main character and economic impact, etc.), enhancing the expressive power of the graph and the depth of data mining.

[0015] Automated and intelligent processing: This method automatically filters event pairs that meet the frequency threshold, reducing manual intervention, enhancing the efficiency and intelligence level of data processing, and meeting the requirements of large-scale data processing.

[0016] In summary, through the combination of temporality, multi-dimensional analysis, and automated processing, the present invention has achieved a more accurate, efficient, and intelligent graph construction method, promoting technological progress in the field of dynamic event data analysis. It can be widely applied to various practical scenarios that require processing dynamic event data and constructing association graphs, such as social network analysis, crisis prediction, public opinion monitoring, etc. Brief Description of the Drawings

[0017] Figure 1 It is a schematic diagram of the computer device structure relied on by the method of the present invention.

[0018] Figure 2 It is a flowchart of the method of the present invention.

[0019] Figure 3 It is a schematic diagram of the data preprocessing step.

[0020] Figure 4 It is a schematic diagram of the frequency calculation and multi-dimensional relationship merging step.

[0021] Figure 5 It is a schematic diagram of the graph construction step.

[0022] Figure 6 It is a schematic diagram of the finally constructed event graph.

[0023] Figure 7 It is a schematic diagram of the event chain mined by the method of the present invention.

[0024] In the figure: Central Processing Unit 1001, Graphics Processing Unit 1002, Memory 1003, Storage Unit 1004, Network Communication Module 1005, Input Device 1006, Display Device 1007, Data Interface 1008, Operating System and Software Environment 1009. Detailed Implementation Modes

[0025] For a clearer understanding of the technical features, objectives, and beneficial effects of the present invention, the technical solution of the present invention will be described in detail below, but it should not be construed as a limitation on the scope of implementation of the present invention.

[0026] To solve the problems existing in the prior art in the background art, the present invention has carried out innovative designs in many aspects. First, automated data cleaning and formatting methods are adopted to reduce human intervention, and combined with domain-adapted parsing technologies to improve the availability and consistency of data. In terms of data cleaning, unstructured data is standardized to ensure data integrity and consistency. In terms of data formatting, by standardizing event-related fields, the operability of data is enhanced, making it suitable for subsequent event frequency calculation and time series analysis. In addition, compared with traditional NLP-based methods, the present invention realizes the efficient processing of event data through a multi-level data parsing strategy, combining technologies such as keyword matching, pattern recognition, and statistical analysis. For example, during the event parsing process, the system will automatically identify entities in the text and match the standard event library to ensure the uniqueness and traceability of events. At the same time, combined with a dynamic data screening mechanism based on frequency analysis, low-frequency events and noise data are filtered to improve data reliability. Through these optimization measures, the present invention has significantly improved in terms of efficiency and accuracy compared with traditional methods, laying a solid foundation for subsequent time series analysis and providing data support for more complex event reasoning and prediction.

[0027] The following describes the specific implementation of the present invention in detail with reference to specific embodiments.

[0028] An embodiment of the present invention provides a method for constructing an intelligent decision-making graph based on dynamic time series event data, and its flowchart is as Figure 2 shown, and the method includes the following steps: Step S1: Data preprocessing; the specific steps are as follows (see Figure 3 ): Reading and cleaning: Read the original event data, clean null values, remove invalid data, and perform outlier detection operations to screen out data that may be incorrect.

[0029] Time field processing: Use natural language processing (NLP) technology to format and convert the time fields in the data, accurately parse and normalize time information. Such as the conversion between relative time (such as "yesterday", "last month") and absolute time (such as "2022-06-15") to ensure that the time information in the data is in a unified format. Then, according to the set time range, filter out the event data within the time range.

[0030] Text field processing: Due to the diversity of data sources, there may be cases of missing fields or format mismatches. Therefore, first, fill the empty values in the text fields in the data (such as "main character", "event type", etc.) with "unknown", and then uniformly convert them to string format. During this process, use methods such as synonym replacement to process text fields and event descriptions to reduce information loss caused by different expression methods (such as synonyms, abbreviations). At the same time, use the safe_literal_eval method to parse dictionary or list type data in the fields.

[0031] Data expansion: Perform an "expansion" operation on each column in the data to split fields containing multiple values. For the complex nested information that may be included in the event description, such as an event associated with multiple relevant persons or multiple influencing factors, adopt a structured parsing strategy to disassemble the nested information into independent data units, and through entity alignment technology, normalize the same entities in different data sources, and finally obtain the cleaned structured data.

[0032] The data preprocessing steps of the present invention integrate various advanced technologies such as automatic parsing, format conversion, and data cleaning, overcoming many drawbacks of traditional rule-based methods (such as regular expression matching, stop word removal, keyword extraction, etc.). Traditional methods can play a certain role when dealing with structured data, but when faced with unstructured text data, they are easily affected by factors such as grammar variations, spelling mistakes, and ambiguities. Moreover, when dealing with multilingual data or professional field terms, they rely too much on manual annotation and dictionary matching, resulting in poor scalability. In contrast, this method has a high degree of automation, can significantly reduce the dependence on manual annotation, and improve data processing efficiency. It is not only applicable to unstructured text such as social media data and news reports, but also can be applied to the analysis scenarios of structured data such as government reports and research papers. Due to its strong generalization ability, it can play an important role in time series event analysis tasks in different fields such as economic dynamic analysis, international political event tracking, and enterprise crisis management.

[0033] Step S2: Frequency calculation and multi-dimensional relationship merging; the specific steps are as follows (see Figure 4 ): Frequency calculation: Based on the cleaned structured data, calculate the frequencies of different field combinations in the data (such as the combination of "main character" and "event type", the combination of "main character" and "economic impact", the combination of "event type" and "economic impact"). During the calculation process, combine the event timestamp information and introduce a weight factor to calculate the associations at different levels such as "main character - event type" and "event type - economic impact". Generate a new frequency feature column for each combination to measure the influence of events in different dimensions. For example, calculate the frequency of events involving a specific main character to measure the activity level of this main character and the degree of influence on a certain type of event; calculate the co-occurrence frequency of an event with different economic impact factors to detect the direct or indirect effects that the event may have on the economic system.

[0034] Multi-dimensional relationship merging: Merge the generated frequency feature columns to obtain data with enhanced frequencies to reflect the association strength between different field combinations. For example, if A is the main character and B is the event type, the merged frequency is 1; if C is the main character and D is the economic impact, the merged frequency is 2; frequencies 1 and 2 together form the frequency.

[0035] Compared with the traditional frequency analysis methods that rely on keyword statistics and TF-IDF calculation, the multi-dimensional frequency calculation method proposed in the present invention has obvious advantages. Traditional methods often have great limitations in dealing with unstructured data. Especially when the data volume is large and contains a lot of noise, the statistical results are easily interfered by factors such as extreme values and data imbalance, resulting in large deviations in the analysis results. In addition, the TF-IDF calculation method is mainly based on the word occurrence frequency and lacks the ability to model the temporal relationship between events, making its effect limited in time-sensitive event analysis tasks. While this method can more comprehensively reveal the complex associations in the event system, accurately capture key event relationships, adapt to different data scales and application scenarios, and has high generality. At the same time, the frequency analysis combined with time information can further reveal the evolution trend of events and provide support for temporal relationship modeling. For example, in policy analysis and public event tracing, it can help identify high-impact events and provide data basis for decision-making, and has great application potential in fields such as social science research, financial analysis, and public opinion monitoring.

[0036] Step S3: Graph construction; the specific steps are as follows (see Figure 5 ) Screen event pairs: Based on the data with enhanced frequencies, set a dynamic frequency threshold, automatically determine the data screening criteria based on the overall distribution of events, automatically filter out low-frequency and invalid associations, remove outliers in the statistical sense, and improve the accuracy and credibility of the data. On this basis, screen event pairs that meet the preset frequency threshold.

[0037] Building a Graph Structure: In a directed graph data structure, use the unique identifier of event data as nodes, and establish the edges of the graph based on the frequencies between the filtered event pairs to represent the relationships between events. Only event pairs with frequency values greater than a preset threshold will have associated edges established in the graph.

[0038] Temporal Matching: Perform temporal matching on the event relationships in the graph. Construct graph edges that conform to the temporal order based on the event timestamps to ensure that the relationships between events are in line with the time sequence and avoid time confusion. At the same time, adopt a dynamic weighting mechanism to dynamically assign weights to event relationships based on multi-dimensional factors such as time interval, event influence, and historical co-occurrence patterns to more accurately measure the association strength between events. Specifically, first, calculate the time interval between any two events. The closer the time, the higher the weight, showing a decaying trend as the time difference increases. Second, calculate the average influence of the event pair according to the importance indicators of the events themselves (such as attention received, level score, etc.). The greater the influence, the higher the weight. Then, count the frequency of the co-occurrence or adjacent occurrence of this event pair in the historical data. The higher the frequency, the stronger the association trend between the two, and a higher weight is given. In the comprehensive stage, standardize these three types of indicators and dynamically adjust the weight coefficients of the three according to the current data distribution characteristics to ensure that the weight assignment is sensitive and adaptable to the data characteristics.

[0039] Edge Weight Setting: The weight value of the edge is determined by the frequency information and represents the strength of the relationship between events.

[0040] In past research, the construction of event graphs mostly relied on text-based event relationship extraction, such as dependency syntactic analysis, knowledge triple construction, etc. These methods usually extract entities and their relationships from text data based on pre-trained models or rule templates and construct knowledge graphs through relationship linking. However, there are significant limitations in dealing with temporal information. On the one hand, traditional event relationship extraction methods lack accurate modeling of time factors. The event time information in natural language text has various formats, such as absolute time ("March 15, 2022"), relative time ("last month"), time span ("the past two years"), etc., which makes it difficult for the knowledge graph directly extracted from text to form accurate temporal relationships. Moreover, existing methods often focus on static relationships and ignore the time constraints in the event development process, unable to correctly depict causal relationships, which limits the practical application value of the graph. On the other hand, traditional methods for constructing event graphs rely on preset relationship templates or manually annotated datasets. It may be effective in specific domains, but when extended to cross-domain analysis scenarios, the predefined event relationship types are difficult to cover all possible associations, resulting in poor performance of the knowledge graph in complex event analysis. And the cost of manual annotation is high, and the scale of the training dataset is limited. It is difficult for machine learning-based event relationship extraction methods to ensure high accuracy when dealing with massive unstructured text data.

[0041] In view of the many drawbacks of traditional methods, this method proposes a brand-new construction strategy. First, core events and their association relationships are identified through event frequency calculation, and a temporal relationship network is constructed based on time information. The statistical laws of historical data are used to ensure the reliability of the relationships between events. By introducing time attributes, a causal sequence is established in the event graph, enabling the connections between events to not only reflect "correlation", but also express "sequence" and potential causal relationships. This temporal processing can effectively enhance the application value of the knowledge graph in predictive analysis, such as predicting the possible chain reactions or long-term impacts of certain types of events. During the graph construction process, the data screening mechanism is crucial. Actual event data contains a large amount of noise, such as abnormal data caused by input errors, information redundancy, duplicate records, etc. If low-frequency events are not screened, they will interfere with the analysis results and obscure the relationships of key events. By setting a dynamic frequency threshold to automatically screen data, high-quality data input can be provided for graph construction. In addition, the edge weights are determined by frequency information, intuitively reflecting the strength of the relationships between events and providing a quantitative basis for analyzing event associations. At the same time, the dynamic weighting mechanism further optimizes the modeling of the temporal relationships between events. In traditional event graphs, the weights of edges are often the same or based only on simple metrics, while this method dynamically assigns weights to event relationships from multiple-dimensional factors. If event A often triggers event B in a specific environment and has been observed multiple times, the association weight will increase, helping the model to more accurately identify high-confidence causal relationships during prediction.

[0042] Step S4: Output and application; Output the finally constructed event graph (see Figure 6 ), the graph contains the number of nodes, the number of edges, and graph edge data filtered based on frequency, and these data can support subsequent graph analysis, event prediction, or network analysis applications. Using network structure analysis and visualization techniques, the connection relationships between event nodes are intuitively displayed based on the networkx library. By highlighting key events by combining different attributes such as colors and node sizes, the constructed temporal event graph becomes more interpretable. By intuitively displaying the node connection relationships and highlighting key events, it can assist analysts in quickly grasping the complex associations between events and help decision-makers formulate targeted strategies. Compared with traditional event extraction methods, the temporal event graph constructed in this study can more accurately reflect the development context of events and has broad application prospects in fields such as policy research, social sentiment analysis, and economic prediction.

[0043] Another embodiment of the present invention provides a computer device (see Figure 1 ), for implementing the intelligent decision-making graph construction method based on dynamic temporal event data described above, including: Central Processing Unit (CPU) 1001: It mainly performs data preprocessing and event analysis tasks, including data cleaning, text parsing, outlier detection, and time format conversion.

[0044] Graphics Processing Unit (GPU) 1002: It is used to accelerate large-scale data processing, especially to accelerate network operations during the construction of the event graph, such as graph neural network (GNN) calculations.

[0045] Memory (RAM) 1003: It serves as a temporary data storage space, supporting batch data loading and calculations to improve processing efficiency.

[0046] Storage unit (SSD / HDD) 1004: It is used to store raw data (such as Excel files), intermediate calculation results, and the finally constructed event graph.

[0047] Network communication module 1005: It realizes the communication between the computer device and remote servers or cloud computing resources to achieve data sharing and distributed computing.

[0048] Input device (keyboard / mouse) 1006: It provides an operation interface for users. Users can set parameters, query analysis results, or adjust calculation tasks through the input device.

[0049] Display device (monitor) 1007: It visualizes the event graph and presents the data analysis results.

[0050] Data interface (USB / PCIe) 1008: It is used to connect external storage devices or sensors to obtain more data sources.

[0051] Operating system and software environment 1009: It provides a running environment to support the operation of software libraries such as Python, Pandas, and NetworkX to achieve data processing, frequency calculation, and graph construction.

[0052] The central processing unit 1001 reads the Excel data file in the storage unit 1004, executes the data preprocessing step, and temporarily stores the cleaned data in the memory 1003 to avoid repeatedly reading the hard disk and improve the processing efficiency. The central processing unit 1001 executes the frequency calculation and multi-dimensional relationship merging step, stores the calculation result in the storage unit 1004, and displays the preliminary analysis result through the display device 1007. The graphics processing unit 1002 processes large-scale data, performs graph neural network (GNN) calculations, and executes an efficient graph construction step. During this process, the network communication module 1005 plays a role in supporting the interaction between the computer device and the remote server, obtaining a larger-scale data set or performing distributed calculations as needed. Finally, the constructed event graph is stored in the storage unit 1004 to support subsequent query analysis. The display device 1007 visualizes the graph result through Matplotlib and NetworkX to help users understand the data relationship.

[0053] As Figure 7 shown, the method of the present invention has successfully mined hidden high-frequency long event chains. In the figure, the label of each node represents a specific event ID. By combining the statistical and graph network analysis perspectives, not only can frequent event sequences be identified, but also the robustness of the connection relationships between these events is enhanced.

[0054] In summary, the temporal event graph constructed by the present invention has significant effects in many aspects and can effectively solve the limitations of traditional static knowledge graphs: Through the multi-dimensional frequency calculation in step S2, combined with the timestamp and weight factor, and the directed graph structure constructed in step S3 (where the edge weights are determined by frequencies and optimized through temporal matching), the graph can not only clearly present the causal paths of events but also perform trend prediction based on historical data, providing support for practical applications such as policy decision-making and economic analysis. The processing of the time field in step S1 lays the foundation for time feature modeling. The temporal matching operation in step S3 can automatically identify the temporal dependence relationships between events, and then construct an event chain based on the passage of time, revealing the logical evolution trajectory of events. Compared with traditional static knowledge graphs that only focus on entity relationships and ignore the time dimension, this temporal event graph incorporates time factors. Through operations such as time feature modeling and temporal matching, it can more accurately convey event information, especially in terms of time sequence and causal relationships. At the same time, it combines historical data for trend prediction, significantly enhancing the trend analysis ability and providing a new perspective for the study of temporal events. The dynamic weighting mechanism adopted in step S3 dynamically assigns weights based on multi-dimensional factors such as time interval, event influence, and historical co-occurrence patterns, covering event occurrence frequency and correlation strength, etc., and can more accurately reflect the true connections between events. Since the construction process takes into account time factors and event associations, when a new event occurs, based on the timestamp and the existing event relationship network, combined with the dynamic weighting mechanism, the system can automatically adjust the event chain and update the causal relationships, enabling the graph to maintain real-time performance and effectiveness. Due to its dynamic adjustment characteristics, the graph can promptly reflect changes in the social environment, adapt to rapidly changing situations, and greatly enhance its value in practical applications. In addition, combined with advanced machine learning methods such as graph neural networks (GNNs), the graph model proposed in this invention can also automatically learn the complex association patterns between events, mine potential hidden relationships from the data, and improve the accuracy of prediction.

[0055] In summary, the temporal event graph provides an efficient, accurate, and interpretable solution for fields such as social event analysis, policy evaluation, and risk prediction. By integrating time factors, this method not only improves the applicability of the knowledge graph in a dynamic environment but also enhances the logical reasoning ability between events, providing strong support for research and applications in multiple fields such as social science, economic research, and artificial intelligence. In the future, with the broadening of data acquisition channels and the continuous advancement of algorithm optimization, the application scenarios of the temporal event graph will be more extensive and will play a greater role in intelligent decision-making, risk warning, public governance, etc. Considering the above advantages, the temporal event graph can display causal paths, predict trends, combine external data sources, dynamically adjust, etc., providing efficient and accurate analysis for fields such as social event analysis, policy evaluation, and risk prediction. Moreover, the construction process and relationship establishment are based on clear steps and are interpretable. The multi-step construction of the graph takes into account time factors, enabling it to better adapt to dynamic environmental changes. By establishing temporal dependencies and causal relationships, it enhances the logical reasoning ability and provides strong support for research and applications in fields such as social science, economic research, and artificial intelligence. In the future, with the broadening of data acquisition channels and the continuous advancement of algorithm optimization, the application scenarios of the temporal event graph will be more extensive and will play a greater role in intelligent decision-making, risk warning, public governance, etc.

[0056] The above is only the preferred embodiment of the present invention. It should be noted that for those skilled in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, which should also be regarded as the protection scope of the present invention, and these will not affect the implementation effect of the present invention and the practicality of the patent.

Claims

1. An intelligent decision-making graph construction method based on dynamic time-series event data, characterized in that, It includes the following steps: Step S1: Data preprocessing; Read and clean the original event data, perform formatting conversion and normalization processing on the time field, fill and unify the format of the text field, expand the nested information through structured parsing and entity alignment technologies, and generate the cleaned structured data; Step S2: Frequency calculation and multi-dimensional relationship merging; Calculate the frequencies of different field combinations, generate frequency feature columns by combining the timestamp and weight factor, and merge the multi-dimensional relationships to reflect the association strength between different field combinations; Step S3: Graph construction; Based on the dynamic frequency threshold, screen event pairs, construct a directed graph data structure, use the unique identifier of the event as the node, establish the edges of the graph and set the edge weights, and optimize the temporal relationship between events through temporal matching and dynamic weighting mechanisms; Step S4: Output and application; Output a graph spectrum containing the number of nodes, the number of edges, and the graph edge data filtered based on frequency, which supports visual display and analysis applications.

2. The method for constructing an intelligent decision-making graph based on dynamic time-series event data according to claim 1, wherein The data preprocessing step specifically includes: Reading and cleaning: Read the original event data, clean the null values, remove the invalid data, and perform outlier detection operations; Time field processing: Use natural language processing technology to perform formatting conversion on the time field in the data, and screen out the eligible event data according to the set time range; Text field processing: Fill the null values in the text field with "unknown", uniformly convert them to string format, use the text parsing method based on deep learning to process the text field and event description, and use the safe_literal_eval method to parse the dictionary or list type data in the field; Data expansion: Perform an "expansion" operation on each column of data, split the fields containing multiple values, and use the structured parsing strategy to disassemble the nested information into independent data units, and normalize the same entities in different data sources through entity alignment technology to finally obtain the cleaned structured data.

3. The method for constructing an intelligent decision-making graph based on dynamic time-series event data according to claim 2, wherein The frequency calculation and multi-dimensional relationship merging step specifically includes: Frequency calculation: Calculate the frequencies of different field combinations in the data, introduce the weight factor by combining the event timestamp information during the calculation process, calculate the associations at different levels, and generate a new frequency feature column for each combination; Multi-dimensional relationship merging: Merge the generated frequency feature columns to obtain the data with enhanced frequency to reflect the association strength between different field combinations.

4. The method for constructing an intelligent decision-making graph based on dynamic time-series event data according to claim 3, wherein The graph construction step specifically includes: Screening event pairs: Based on the data with enhanced frequency, set a dynamic frequency threshold to screen the event pairs that meet the preset frequency threshold; Establishing the graph structure: In the directed graph data structure, use the unique identifier of the event data as the node, and establish the edges of the graph according to the frequency between the screened event pairs; the prerequisite for establishing the associated edges in the graph is that the frequency value of the event pair is greater than the preset threshold; Temporal matching and relationship optimization: Perform temporal matching on the event relationships in the graph spectrum, construct the graph edges that conform to the temporal order based on the event timestamps, and use the dynamic weighting mechanism to dynamically assign weights to the event relationships; Edge weight setting: The weight value of the edge is determined by the frequency information and represents the strength of the relationship between events.

5. The method for constructing an intelligent decision-making graph based on dynamic time-series event data according to claim 1, characterized in that The said event graph learns the association patterns between events through a graph neural network and supports real-time update of causal relationships.

6. A computer device for implementing the intelligent decision-making graph construction method based on dynamic time-series event data according to any one of claims 1-5, characterized in that, It includes: A central processing unit for performing data preprocessing and event analysis tasks; A graphics processing unit for accelerating the operations of the graph neural network during the construction of the event graph; Memory, serving as a temporary data storage space; A storage unit for storing raw data, calculation intermediate results, and the finally constructed event graph; A network communication module for realizing the communication between the computer device and a remote server or cloud computing resources to achieve data sharing and distributed computing; An input device for providing an operation interface; A display device for visualizing the event graph and presenting the data analysis results; A data interface for connecting to an external storage device or sensor; An operating system and software environment for providing a running environment.

7. The computer device according to claim 6, characterized in that The central processing unit reads the Excel data file in the storage unit, performs the data preprocessing steps, and temporarily stores the cleaned data in the memory; The central processing unit performs the frequency calculation and multi-dimensional relationship merging steps, stores the calculation results in the storage unit, and displays the preliminary analysis results through the display device; the graphics processing unit processes large-scale data, performs graph neural network calculations, and executes the graph construction steps. During this period, the network communication module supports the interaction between the computer device and the remote server; finally, the constructed event graph is stored in the storage unit, and the display device visualizes the graph results through Matplotlib and NetworkX.

Images