Method for constructing event process model based on enhanced Alpha algorithm

By introducing a graph attention mechanism into the Alpha algorithm to identify and process dependencies in complex event logs, the problem of insufficient recognition accuracy and adaptability of traditional Alpha algorithms when processing complex and noisy data is solved, and a higher process is achieved Model accuracy and robustness, suitable for a variety of application scenarios.

CN119046339APending Publication Date: 2024-11-29QUFU NORMAL UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411050436.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-01
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

When processing complex and noise-containing event log data, the traditional Alpha algorithm and its extension face the problems of low recognition accuracy and poor adaptability.

Method used

Using an enhanced Alpha algorithm method, by introducing a graph attention mechanism, analyzing and processing event log data, identifying dependencies between activities, and building an event process model. The specific steps include data acquisition, graph structure training, dependency recognition and Process model construction, using GATA algorithm to optimize and verify the model.

Benefits of technology

Improve the discovery accuracy and robustness of the process model, enable more accurate identification of dependencies in complex event logs, better tolerance and adaptability, and is suitable for event log data of various types and scales.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119046339A_ABST
    Figure CN119046339A_ABST
Patent Text Reader

Abstract

The invention discloses a method for constructing an event process model based on an enhanced Alpha algorithm, and is applied to the technical field of data mining and process mining. The method comprises the following steps: S1, acquiring an original event log, converting the original event log into a graph structure, and defining attributes of activities and edges; s2, introducing an attention mechanism into the graph structure, training the graph structure through the attention mechanism to obtain a graph attention network, and outputting the graph attention network; s3, based on the output of the graph attention network, identifying a dependency relationship between activities; and S4, based on the identified dependency relationship, constructing an event process model by adopting a rule of an enhanced Alpha algorithm. According to the method, event log data are analyzed and processed by introducing a graph attention mechanism, so that the discovery precision and robustness of the process model are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of data mining and process mining, and more specifically to a method for constructing an event process model based on an enhanced Alpha algorithm. Background Art

[0002] Process mining technology is a branch of data mining that focuses on extracting process knowledge from event logs. It is a technology that mines knowledge from existing event logs to discover, monitor and improve actual processes. It is widely used in finance, medical care, logistics, ERP, CRM, retail, industrial quality inspection, public services and other industries. As a classic algorithm in process mining, the Alpha algorithm constructs a corresponding process model by analyzing the activity sequence relationship in the event log. However, the traditional Alpha algorithm and its extensions face the challenges of low recognition accuracy and poor adaptability when processing complex and noisy log data.

[0003] Therefore, an enhanced Alpha algorithm is proposed, which may involve the application of reinforcement learning technology. Reinforcement learning is a machine learning method that learns the most likely goal through the interaction between the agent and the environment. In the process of building a process model, reinforcement learning can help optimize the search strategy of the Alpha algorithm to improve the accuracy and efficiency of the model.

[0004] Therefore, it is an urgent problem for those skilled in the art to propose a method for constructing an event process model based on an enhanced Alpha algorithm to solve the difficulties existing in the prior art. Summary of the invention

[0005] In view of this, the present invention provides a method for constructing an event process model based on the enhanced Alpha algorithm, which improves the discovery accuracy and robustness of the process model by introducing a graph attention mechanism to analyze and process event log data.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] A method for constructing an event process model based on an enhanced Alpha algorithm comprises the following steps:

[0008] S1, data acquisition step: obtain the original event log and convert it into a graph structure to define the attributes of activities and edges;

[0009] S2, graph structure training step: introduce the attention mechanism into the graph structure, train the graph structure through the attention mechanism to obtain the graph attention network and output it;

[0010] S3, dependency identification step: Based on the output of the graph attention network, identify the dependency between activities;

[0011] S4, process model construction steps: Based on the identified dependencies, the event process model is constructed using the rules of the enhanced Alpha algorithm.

[0012] In the above method, optionally, S2 includes:

[0013] For each pair of connected activities v and u, calculate the attention coefficient e vu The formula is as follows:

[0014] e vu = leakyRelU(a T [Wh v ||Wh u ])

[0015] Among them, [·||·] represents the concatenation operation of the vector, LeakyReLU is an activation function, W is the weight matrix, and a T is the weight vector used to calculate the attention coefficient, h v is the feature vector of activity v, h u is the feature vector of activity u.

[0016] In the above method, optionally, S2 includes:

[0017] The attention coefficients of each activity's neighboring activities are normalized using the softmax function:

[0018]

[0019] Among them, N v is the set of adjacent activities of activity v, e vk is the unnormalized attention coefficient between activity v and its neighbor activity k, where k is N v A neighbor activity in .

[0020] The above method, optionally, S2 also includes: using a graph attention network to learn dependencies between activities, specifically learning and strengthening key dependencies through an attention mechanism.

[0021] In the above method, optionally, the dependency relationships between the activities identified in S3 include direct successor relationships, causal relationships, and parallel relationships.

[0022] In the above method, optionally, the enhanced Alpha algorithm used in S4 is specifically a GATA algorithm.

[0023] The above method, optionally, also includes optimization and verification of the process model in S4:

[0024] Model optimization: merge redundant places or transfers to simplify the event process model without changing its behavior;

[0025] Model verification: Verify whether the constructed event process model accurately reflects the process recorded in the event log data, which is done by simulating the Petri net and comparing it with the original log.

[0026] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method for constructing an event process model based on an enhanced Alpha algorithm, which has the following beneficial effects:

[0027] (1) The GATA algorithm used in the present invention has the following advantages: through the graph attention mechanism, the algorithm can more accurately identify the dependencies in complex event logs and improve the accuracy of the process model; the algorithm has better tolerance to noise and abnormal data and can process complex log data without reducing the recognition accuracy; the GATA algorithm is applicable to event logs of various types and sizes and has good versatility and adaptability;

[0028] (2) The enhanced Alpha algorithm based on graph attention network provides a new technical solution for the field of process mining, which is particularly suitable for processing complex and noisy event log data. By introducing the graph attention mechanism, the algorithm not only improves the accuracy and efficiency of process model discovery, but also enhances the processing capability and robustness of complex data. This has important practical significance for various application scenarios that need to extract process knowledge from complex system logs, such as financial fraud detection, production process optimization, medical process analysis, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0030] Figure 1 A flow chart of a method for constructing an event process model based on an enhanced Alpha algorithm provided by the present invention;

[0031] Figure 2 It is the technical roadmap of the present invention. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0033] The core of this invention is to represent the event log as a graph structure, where nodes represent events or activities, and edges represent the relationship between events. The graph structure is learned using a graph attention network to extract the dependencies between events, including parallel, selective, and cyclic dependencies. Figure 2 shown.

[0034] Reference Figure 1 As shown, the present invention discloses a method for constructing an event process model based on an enhanced Alpha algorithm, comprising the following steps:

[0035] S1, data acquisition step: obtain the original event log and convert it into a graph structure to define the attributes of activities and edges;

[0036] S2, graph structure training step: introduce the attention mechanism into the graph structure, train the graph structure through the attention mechanism to obtain the graph attention network and output it;

[0037] S3, dependency identification step: Based on the output of the graph attention network, identify the dependency between activities;

[0038] S4, process model construction steps: Based on the identified dependencies, the event process model is constructed using the rules of the enhanced Alpha algorithm.

[0039] Specifically, the identified dependencies are used to construct a process model using the rules of the enhanced Alpha algorithm. This usually involves converting causal relationships into structural elements in the event process model, such as places and transitions in Petri nets.

[0040] Among them, using the rules of the enhanced Alpha algorithm to build an event process model mainly involves identifying and applying the activity relationships in the log to build an event process model, such as a Petri net. The enhanced Alpha algorithm provides better ability to handle parallel behavior, loops, and other complex patterns by improving the limitations of the original Alpha algorithm. The following are the general steps and rules used when building an event process model:

[0041] 1. The basic elements for constructing a Petri net include:

[0042] (1) Place: Indicates the state or condition of a process.

[0043] (2) Transition: refers to an activity or event in a process.

[0044] (3) Arc: connects Place and Transition, indicating the relationship between state and activity.

[0045] 2. Apply the Enhanced Alpha algorithm rules including:

[0046] Initial and End Activities: Identify the initial and end activities in the log, which will serve as the starting and ending points of the Petri net, respectively.

[0047] Creation of Transitions and Places: For each identified causal relationship A->B, create a Transition (representing activity B) and two Places, one representing the state in which A is completed and the other representing the condition under which B can start. In this way, the model maps the logical flow between activities.

[0048] Handling parallel relationships: Parallel activities are handled in a special way, and you need to ensure that the model can represent situations where activities can start at the same time. This usually involves introducing parallel structures, such as using split and merge nodes in Petri nets.

[0049] Loops and Complex Structures: The enhanced Alpha algorithm handles loops and other complex structures through specific rules and techniques, such as by identifying loop patterns and adjusting the Petri net structure accordingly.

[0050] Through the above steps and rules, the enhanced Alpha algorithm can effectively build a model that reflects the real business process from complex and noisy log data. This approach improves the accuracy and applicability of the model, especially when dealing with complex structures such as parallel behaviors and loops.

[0051] Furthermore, S2 includes:

[0052] (1) For each pair of connected activities v and u, calculate the attention coefficient e vu , indicating the importance of the relationship between nodes. The attention mechanism is implemented through a feedforward neural network to calculate e vu The formula is as follows:

[0053] e vu = leakyRelU(a T [Wh v ||Wh u ])

[0054] Among them, [·||·] represents the concatenation operation of the vector, LeakyReLU is an activation function, W is the weight matrix, and a T is the weight vector used to calculate the attention coefficient, h v is the feature vector of activity v, h u is the feature vector of activity u.

[0055] (2) Attention coefficient normalization: Normalize the attention coefficients of the adjacent activities of each activity using the softmax function:

[0056]

[0057] Among them, N v is the set of adjacent activities of activity v, e vk is the unnormalized attention coefficient between activity v and its neighbor activity k, where k is N v A neighbor activity in .

[0058] (3) Feature aggregation: The features of neighboring nodes are weighted and aggregated according to the attention coefficient, and the feature representation of each node is updated. The formula is as follows:

[0059]

[0060] Here, σ is a non-linear activation function (such as ReLU).

[0061] (4) In order to improve the feature representation capability and model stability, a multi-head attention mechanism is applied. Each head independently calculates the attention coefficient and feature aggregation, and finally the outputs of all heads are concatenated or averaged. The formula is as follows:

[0062]

[0063] Here, K is the number of attention heads.

[0064] Furthermore, S2 also includes: using the graph attention network to learn the dependencies between activities, specifically learning and strengthening key dependencies through the attention mechanism.

[0065] Furthermore, the dependencies between activities identified in S3 include direct successor relationships, causal relationships, and parallel relationships.

[0066] 1) Direct successor relationship: If activity A occurs immediately after activity B in the log, we say that A directly causes B. This relationship is used to identify the basic sequence flow between activities.

[0067] 2) Causal relationship: By analyzing the direct successor relationship, we can determine the causal relationship between activities. If activity A often occurs before activity B, and there is no counterexample (that is, B does not occur before A), then it can be considered that A causes B.

[0068] 3) Parallel relationship: If activities A and B are direct successors of each other, then they may be executed in parallel.

[0069] By using the output of the graph attention network, the enhanced Alpha algorithm can more accurately identify these relationships, especially when the data is complex or noisy.

[0070] Furthermore, the enhanced Alpha algorithm used in S4 is specifically the GATA algorithm.

[0071] Furthermore, S4 also includes the optimization and verification of the event process model:

[0072] Model optimization: merge redundant places or transfers to simplify the event process model without changing its behavior;

[0073] Model verification: Verify whether the constructed event process model accurately reflects the process recorded in the event log data, which is done by simulating the Petri net and comparing it with the original log.

[0074] In a specific embodiment, the GATA algorithm is applied to the fraud detection process in the financial industry. By analyzing transaction logs, the event process model of fraudulent behavior is effectively identified, thereby improving detection efficiency and accuracy. The details are as follows:

[0075] First, transaction events and their attributes are extracted from a large number of financial transaction logs. Then, the α algorithm is applied to identify the sequential, causal, parallel and selection relationships between transaction events, and a preliminary transaction process model is constructed. Next, a causal graph is constructed and enhanced with a graph attention network. Through node feature integration, attention coefficient calculation and normalization, and multi-head attention mechanism, transaction events and relationships are modeled and represented more finely. Finally, the trained graph attention network model can efficiently process new transaction log data and automatically identify potential fraud.

[0076] GATA can not only quickly discover complex fraud patterns, but also reduce false positives and false negatives, and improve the accuracy of detection. In addition, the model can dynamically adapt to changes in transaction behavior, capture new fraud patterns, and provide financial institutions with real-time, accurate fraud detection and early warning, which can effectively identify potential fraudulent behavior, thereby enhancing the security and reliability of the financial system.

[0077] In the manufacturing industry, the GATA algorithm is used to optimize production processes. By analyzing production logs, the GATA algorithm can identify bottlenecks and inefficient links in the production process, provide a scientific basis for process optimization, and ultimately improve production efficiency and product quality. The details are as follows:

[0078] In manufacturing enterprises, there are multiple processes in the production process, including raw material processing, parts assembly, finished product inspection and packaging, etc. Enterprises accumulate a large amount of production data by recording production logs.

[0079] By applying the GATA algorithm, the following goals can be achieved:

[0080] 1. Analyze the relationship between each process and identify which processes are production bottlenecks (for example, certain equipment frequently breaks down, causing the entire production line to stagnate).

[0081] 2. Discover inefficient links in parallel processes (for example, the production speed of some processes is significantly lower than that of other processes, resulting in reduced overall efficiency).

[0082] 3. Propose optimization plans, such as adding or replacing equipment in bottleneck processes, redesigning the process sequence, and making resource allocation more reasonable.

[0083] The technical solution of the present invention overcomes the limitations of the traditional Alpha algorithm in processing complex structures and noisy data by deeply analyzing the subtle relationships in the event log. The introduction of the GATA algorithm not only contributes a new perspective and method to the development of process mining technology, but also provides a powerful tool for research and application in related fields.

[0084] In future research, we can further explore the potential of graph attention networks in extracting deeper process models, such as by introducing time series analysis, multi-dimensional analysis and other technologies to further improve the accuracy and applicability of the model. In addition, considering the complexity and computational cost of the algorithm, studying how to optimize the algorithm's operating efficiency to adapt to the processing of larger-scale data is also an important direction for future research.

[0085] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0086] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for constructing an event process model based on the enhanced Alpha algorithm, It is characterized in that The following steps are involved: S1, data acquisition step: obtain the original event log and convert it into a graph structure to define the attributes of activities and edges; S2, graph structure training step: introduce the attention mechanism into the graph structure, train the graph structure through the attention mechanism to obtain the graph attention network and output it; S3, dependency identification step: Based on the output of the graph attention network, identify the dependency between activities; S4, process model construction steps: Based on the identified dependencies, the event process model is constructed using the rules of the enhanced Alpha algorithm.

2. A method for constructing an event process model based on an enhanced Alpha algorithm according to claim 1, It is characterized in that S2 includes: For each pair of connected activities v and u, calculate the attention coefficient e vu The formula is as follows: e vu =LeakyRelU(a T [Wh v ||Wh u ]) Among them, [·||·] represents the concatenation operation of the vector, LeakyReLU is an activation function, W is the weight matrix, and a T is the weight vector used to calculate the attention coefficient, h v is the feature vector of activity v, h u is the feature vector of activity u.

3. A method for constructing an event process model based on an enhanced Alpha algorithm according to claim 2, It is characterized in that S2 includes: The attention coefficients of each activity's neighboring activities are normalized using the softmax function: Among them, N v is the set of adjacent activities of activity v, e vk is the unnormalized attention coefficient between activity v and its neighbor activity k, where k is N v A neighbor activity in .

4. A method for constructing an event process model based on an enhanced Alpha algorithm according to claim 3, It is characterized in that S2 also includes: using graph attention networks to learn dependencies between activities, specifically learning and strengthening key dependencies through attention mechanisms.

5. The method for constructing an event process model based on the enhanced Alpha algorithm according to claim 1, It is characterized in that The dependencies between activities identified in S3 include direct successor relationships, causal relationships, and parallel relationships.

6. A method for constructing an event process model based on an enhanced Alpha algorithm according to claim 1, It is characterized in that The enhanced Alpha algorithm used in S4 is specifically the GATA algorithm.

7. A method for constructing an event process model based on an enhanced Alpha algorithm according to claim 6, It is characterized in that S4 also includes optimization and validation of process models: Model optimization: merge redundant places or transfers to simplify the event process model without changing its behavior; Model verification: Verify whether the constructed event process model accurately reflects the process recorded in the event log data, which is done by simulating the Petri net and comparing it with the original log.

Citation Information

Patent Citations

  • Incremental event log-oriented process model mining method and system

    CN115525693A

  • RPA process mining method and device based on multi-modal event extraction

    CN117236676A