Business process compliance inspection method, computer device, and readable storage medium

Abstract

The present invention discloses a business process compliance checking method, computer equipment and readable storage medium. Among them, the method of the present invention first extracts a fully fitted event log and activities from the actual business process recorded in the information system, takes the activity sequence as an attribute, records the activities that can take values ​​for each attribute according to the activity sequence, and constructs a decision tree with the activity sequence as an attribute; then, the trajectory to be judged is judged according to the constructed decision tree, and a method for judging whether the trajectory is compliant based on the decision tree is used and pruned, so that it can more efficiently implement trajectory compliance checking. The present invention also conducts simulation experiments using earthquake emergency plan processes and real data as examples. Through simulation experiments, it is proved that the method of the present invention significantly improves the efficiency of business process compliance checking compared with traditional methods while ensuring a high accuracy rate.

Description

Technical Field

[0001] The present invention relates to a business process compliance checking method, computer equipment and a readable storage medium. Background Art

[0002] With the continuous development of computer hardware and software, as well as computing technology and the Internet of Things, the data generated by software is growing rapidly, and big data technology is developing rapidly. The advent of the big data era has provided a foundation for the development of business process management. Furthermore, due to the increasing competition among enterprises, each strives to accurately execute every process. More and more enterprises are establishing their own log management systems to record the execution of each process in real time. The development of information systems provides important guarantees and functions for enterprises to better manage their businesses. This not only improves management efficiency but, more importantly, ensures the correctness of business management. Process mining is of great significance to the research of business process management, which has led to its development. Process mining plays a significant role in business process management. Process mining mainly analyzes the large amount of event logs generated by business process management, using mining algorithms to identify and detect erroneous behaviors and improve them, thereby continuously improving business processes. In information management systems, the integrity of event logs plays a crucial role in business process management. Especially in the management of natural disasters, which pose a significant threat to the safety of people's lives and property, compliance checks on business processes are even more important. Because there are deviations between event logs and process models, some businesses and event-related businesses are unique in that they cannot tolerate process errors during the execution of business activities, such as emergency response to natural disasters such as earthquakes and tsunamis. Therefore, detecting deviations is extremely important.

[0003] In recent years, numerous scholars have conducted extensive research in the field of seismology, applying artificial intelligence (AI) and big data analysis techniques to earthquake prediction and seismic data analysis. Since the 1970s, scholars worldwide have been identifying and analyzing earthquake precursor data. Earthquake precursors are effective analytical data for earthquake prediction and form the foundation of this analysis. After years of research, my country has achieved intelligent and digitized analysis of earthquake precursor data, effectively predicting the onset of earthquakes. However, due to the complexity of earthquake information, earthquake prediction still has shortcomings. While earthquake predictions can be delayed and inaccurate, earthquake disasters can cause significant damage to people's lives, making post-earthquake preparedness equally important.

[0004] Under the current circumstances where earthquake prediction technology is still insufficient and immature, earthquake emergency rescue work is an effective way to reduce the safety of people's lives and property. Earthquake emergency rescue work is an important means to deal with earthquake disasters. Over the years, in major earthquake disaster relief work, the earthquake emergency plan business process has played a significant guiding role in earthquake emergency rescue work. Our country puts the safety of people's lives and property first. After the earthquake occurred, the Earthquake Bureau responded immediately. Carrying out rescue work in an orderly manner according to the emergency plan has played a significant role in saving people's lives and property safety. Therefore, consistency testing of earthquake emergency plan logs is imminent, and process mining methods are used to detect abnormal behaviors, so as to improve the deficiencies in the emergency plan business process. Nowadays, there are many different research methods for consistency testing, and its measurement standards have multiple aspects. The most important measurement standard is fit, such as:

[0005] Token-based replay is an effective method for consistency detection, and provides a method for incrementally checking the consistency of process models and event logs. Petkovic et al. proposed a framework that can detect whether the traces in the event log conform to the BPMN process model. In addition, alignment is one of the more advanced methods, and a lot of research has been done on this technology. Cook et al. provided a comparison method that obtains alignment results by quantifying the similarity between process execution and process models. Song et al. proposed heuristic rules and trace replay techniques to optimize the search space.

[0006] However, the current method of checking compliance by sequentially judging trajectories from front to back has the problem of low checking efficiency. Summary of the Invention

[0007] The purpose of the present invention is to propose a business process compliance checking method based on an activity sequence decision tree. The method generates a decision tree according to the fully fitted event log extracted from the actual business process, and performs compliance checking based on the generated decision tree, thereby realizing compliance checking of the trajectory starting from the event log and improving the inspection efficiency of the business process.

[0008] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solutions:

[0009] The business process compliance checking method based on the activity sequence decision tree includes the following steps:

[0010] Step 1. Preprocess the fully fitted event logs extracted from the actual business process;

[0011] Extract fully fitting event logs and activities from the actual business processes recorded in the information system, concatenate the extracted activities into a track according to the event ID, create an activity set according to the order of the activities in the track, store the activities in the track in the order in the activity set, and then create an attribute set;

[0012] Define the sequence as: elements σ1, σ2, σ3, ..., σ n An ordered and finite list consisting of σ=<σ1,σ2,σ3,…,σ n > indicates that A is an activity set, A * represents the set of finite sequences on A;

[0013] If there exists a process instance σ, that is, an activity sequence σ∈A * σ is called a trajectory if there is a finite non-empty multiset L∈β(A * ), then L is called an event log;

[0014] Step 2. Perform attribute partitioning based on the number of elements in each activity set. After the attribute set is constructed, define the unique set of activities in the activity set as the key activity set. This activity is defined as a key activity, and the priority of the key activities is defined based on the activity position. The later the activity position, the higher the priority.

[0015] Step 3. Pre-prune the tree by removing redundant judgments according to Occam's razor principle, and then build the decision tree;

[0016] This step specifically involves pruning the attributes from all attributes to only the attributes corresponding to key activities, and selecting the optimal classification attribute based on the priority of the key activities. Iteratively select the optimal attribute to generate a tree node, and remove the selected optimal classification attribute from the attribute set. This iteration continues until the attribute set is empty, meaning all key activities have been traversed, resulting in a classification decision tree. The decision tree construction process is as follows:

[0017] Initialize the decision tree to generate the initial root node, and execute the decision tree node insertion operation. The decision tree node insertion function is a recursive function, which executes the decision tree node insertion recursively until the attribute set is empty and the decision tree is built.

[0018] A decision tree is defined as a four-tuple DT = (N, P, C, FA), where N is the name of the tree node, P is the parent node of the node, C is the node's child list, and FA is the attribute feature of the parent node to which the node belongs;

[0019] Step 4. Replay the event logs with abnormal trajectories in the fitted event logs extracted from the actual business process on the decision tree, or replay the new event logs generated in the actual business process. For each trajectory in the event log, the decision tree internal decision nodes are used to continuously determine whether the activities at each position in the trajectory are compliant, and finally output the predicted label of the leaf node, that is, the business process compliance prediction result.

[0020] In addition, based on the business process compliance checking method based on the activity sequence decision tree, the present invention also proposes a computer device for implementing the above-mentioned business process compliance checking method based on the activity sequence decision tree.

[0021] The computer device includes a memory and a processor. The memory stores executable code. When the processor executes the executable code, it is used to implement the business process compliance checking method based on the activity sequence decision tree mentioned above.

[0022] In addition, based on the business process compliance checking method based on the activity sequence decision tree, the present invention also proposes a computer-readable storage medium for implementing the above-mentioned business process compliance checking method based on the activity sequence decision tree.

[0023] The computer-readable storage medium stores a program thereon, and when the program is executed by a processor, is used to implement the above-mentioned business process compliance checking method based on the activity sequence decision tree.

[0024] The present invention has the following advantages:

[0025] As described above, the present invention describes a business process compliance checking method based on an activity sequence decision tree. This method breaks the traditional idea of ​​conducting business process compliance checking in sequence, and innovatively proposes a method for implementing business process compliance checking using a decision tree. This method will fully traverse the log once, record the attributes and attribute characteristics of the desirable activities in each sequence position, and finally construct a decision tree. According to the characteristics of the abnormal activities in the abnormal trajectory, the decision tree is pruned according to the defined key activities, which greatly reduces the scale of the decision tree and improves the detection efficiency. Finally, experimental analysis is performed based on the pruned decision tree, and experiments are conducted on earthquake emergency response process simulation data sets and real event logs respectively. The final experimental results show that the business process compliance detection method based on the activity sequence decision tree proposed in the present invention can more efficiently implement compliance checking. This method greatly improves the inspection efficiency within an acceptable accuracy range. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Flowchart of a business process compliance checking method based on an activity sequence decision tree in an embodiment of the present invention.

[0027] Figure 2 Generate a schematic diagram for the earthquake emergency response event decision tree.

[0028] Figure 3 This is a curve chart of the average number of comparisons of simulated data.

[0029] Figure 4 The accuracy comparison curve of simulated data is shown in Figure 2.

[0030] Figure 5 This is a curve chart of the average number of comparisons for the real dataset Document_Processing_Event_Logs.

[0031] Figure 6 This is a curve chart of the average number of comparisons for the real dataset Electronic_Invoicing_Event_Logs.

[0032] Figure 7 This is the accuracy curve of the real dataset Document_Processing_Event_Logs.

[0033] Figure 8 This is the accuracy curve of the real dataset Electronic_Invoicing_Event_Logs. DETAILED DESCRIPTION

[0034] The present invention discloses a business process compliance checking method based on an activity sequence decision tree. The method uses fully fitted event logs extracted from actual business processes recorded in an information system as a training set, constructs a decision tree in the offline stage, and adds abnormal trajectories to the fully fitted event logs as a test set in the online stage to perform compliance prediction.

[0035] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0036] like Figure 1 As shown in the figure, the business process compliance checking method based on the activity sequence decision tree has the following specific steps:

[0037] Step 1. Preprocess the fully fitted event logs extracted from the actual business process.

[0038] Extract fully fitting event logs and activities from the actual business processes recorded in the information system, concatenate the extracted activities into tracks according to the event IDs, create an activity set according to the order of activities in the track, store the activities in the track in the activity set in order, and then create an attribute set.

[0039] To better understand the definition of trajectory, the definition of sequence is given before the formal definition of trajectory.

[0040] Define a sequence: elements σ1, σ2, σ3, ..., σ n An ordered and finite list consisting of σ=<σ1,σ2,σ3,…,σ n > indicates that A is a set, A * represents the set of finite sequences on A.

[0041] Defining event logs and trajectories: With the development of big data, a large number of event descriptions are recorded in the information systems of various enterprises and stored in logs. Each recorded event describes the trace it left in the system. That is, the sequence of events recorded in the log is defined as a trajectory. The formal definition is as follows:

[0042] Let A be an activity set. If there exists a σ, that is, an activity sequence σ∈A * σ is called a trajectory. If there is a finite non-empty multiset L∈β(A * ), then L is called a log.

[0043] Definition of decision tree: A decision tree is defined as a four-tuple DT = (N, P, C, FA), where N is the name of the tree node, P is the parent node of the node, C is the node child list, and FA is the attribute feature of the parent node to which the node belongs.

[0044] Trees have many applications. A decision tree, also known as a judgment tree, is a predictive model. One important application is describing classification problems, which aim to divide data according to different criteria. Each non-leaf node in the tree represents a judgment condition, indicating a test on an attribute. Leaf nodes represent the classification results, indicating the output of the test.

[0045] It plays an important role in decision analysis, providing a strategy that is most likely to achieve the goal.

[0046] The basic principle of a decision tree is that the decision problem is represented by a decision point, the possible options are represented by option branches, the possible results are represented by probability branches, and finally the entire tree is used as the basis for the decision maker's decision.

[0047] Decision trees are mainly used for classification problems. They can be applied to trajectory compliance checks and can classify trajectories efficiently and accurately.

[0048] Below is a simple event log. Table 1 describes the records of emergency plans after multiple earthquake disasters using a more compact log format. Table 2 describes the meaning of each activity in the log.

[0049] Table 1 Activities and their meanings

[0050] id Trajectory id Trajectory <![CDATA[σ1]]> <a,b,d,e,f,g,h> <![CDATA[σ5]]> <a,b,d,e,g,f,h> <![CDATA[σ2]]> <a,d,b,e,f,g,h> <![CDATA[σ6]]> <a,d,b,e,g,f,h> <![CDATA[σ3]]> <a,c,d,e,f,g,h> <![CDATA[σ7]]> <a,c,d,e,g,f,h> <![CDATA[σ4]]> <a,d,c,e,f,g,h> <![CDATA[σ8]]> <a,d,c,e,g,f,h>

[0051] Table 2 Activities and their meanings

[0052]

[0053] The above is simulated data from earthquake emergency response procedures following multiple earthquakes. This event log analysis illustrates the process of constructing a decision tree. A test set is constructed by adding non-fitting trajectories containing unusual activity to the fitted data trajectories to test the efficiency and accuracy of the decision tree. Figure 4 is the event log after adding abnormal trajectories, where trajectories σ9, σ 10 , σ 11 , σ 12 Missing, extra, misplaced, and duplicate activities were added.

[0054] Taking this dataset as an example, the process of event log compliance checking using a decision tree is explained.

[0055] Table 3 Simulation test set log records

[0056] id Trajectory id Trajectory <![CDATA[σ1]]> <a,b,d,e,f,g,h> <![CDATA[σ7]]> <a,c,d,e,g,f,h> <![CDATA[σ2]]> <a,d,b,e,f,g,h> <![CDATA[σ8]]> <a,d,c,e,g,f,h> <![CDATA[σ3]]> <a,c,d,e,f,g,h> <![CDATA[σ9]]> <a,d,b,e,g,h> <![CDATA[σ4]]> <a,d,c,e,f,g,h> <![CDATA[σ 10 ]]> <a,b,b,e,i,g,f,h> <![CDATA[σ5]]> <a,b,d,e,g,f,h> <![CDATA[σ 11 ]]> <a,d,e,c,f,g,h> <![CDATA[σ6]]> <a,d,b,e,g,f,h> <![CDATA[σ 12 ]]> <a,c,b,e,f,f,g,h>

[0057] Before building a decision tree, you first need to construct the attributes. The decision tree uses comparative attributes to select and divide the samples based on the optimal classification attributes. Therefore, the construction of attributes is the first step in building a decision tree.

[0058] First create n sets, set S i Store the activities in order i, |S i | represents the set S i For each trajectory in the fully fitted event log, the number of activities in the trajectory is accessed and the activities in the trajectory are saved in the above set in order, i = 1, 2, ..., n.

[0059] Let D = {D1, D2, ..., D n} is the attribute set, where D i It is represented as the i-th attribute. When applied to compliance checking, i is represented as the activity sequence. The set S i The middle element is D i The characteristic value of the attribute.

[0060] The property collection is constructed as follows:

[0061] Input: event log dataset; Output: attribute collection.

[0062] Step 1.1. Traverse the entire event log and repeat steps 1.2 to 1.3 below;

[0063] Step 1.2. Create a trajectory dictionary for the event that does not have a trajectory dictionary.

[0064] Step 1.3. Concatenate the activities in the event log into a trajectory.

[0065] Step 1.4. Traverse the trajectory generated in step 1.3 above and repeat steps 1.5 to 1.8 below;

[0066] Step 1.5. Determine whether the activity set for each position has been created; if not, proceed to step 1.6;

[0067] Step 1.6. Create an activity collection;

[0068] Step 1.7. Store the activities in the trace in the obtained set in order;

[0069] Step 1.8. Create an attribute set and assign the activity set to the attribute set;

[0070] Step 1.9. End and get the attribute set.

[0071] The time complexity of the attribute set construction method is analyzed. The method has two loop bodies. The first loop body includes a single loop, while the second loop body includes a double loop. The traversal object is the number of traces in the event log.

[0072] Therefore, the time complexity of the attribute construction method is O(l 2 ), where l is the number of trajectories in the event log.

[0073] Example 1: Using the plan log shown in Table 1 as an example, we access the earthquake emergency response event log and construct attributes for its processing. First, we create seven sets based on the activities in the trajectory and store the activities at each sequential position. Then, we construct an attribute set D. We run the attribute set construction method described above to obtain the attribute set. The result of constructing the attributes for this log is shown in Table 4.

[0074] Table 4 Decision tree attribute construction

[0075] property Property Value property Property Value <![CDATA[D1]]> {a} <![CDATA[D5]]> {f,g} <![CDATA[D2]]> {b,c,d} <![CDATA[D6]]> {f,g} <![CDATA[D3]]> {b,c,d} <![CDATA[D7]]> {h} <![CDATA[D4]]> {e}

[0076] Step 2. Perform attribute division based on the number of elements in each activity set as the attribute division standard. After the attribute set is constructed, define the only set of activities in the activity set as the key activity set. This activity is defined as the key activity, and define the priority of the key activity based on the activity position. The later the activity position, the higher the priority.

[0077] Since there is inconsistency between the behavior recorded in the enterprise information system and the process model or the fully fitted event log, there are behavioral deviations and abnormal activities. The definition of abnormal activities is given below.

[0078] There are four different types of abnormal activities in deviations, including missing activities, extra activities, misplaced activities, and repeated activities.

[0079] Lost activity: trace σ=a′da″, if Then the activities in a′ and d are lost activities.

[0080] Additional activities: The trajectory σ = a′da″, the activity set is denoted as A, if d∈A, then d is an additional activity.

[0081] Dislocation activity: trajectory σ=a′da″, if d∈A, but Then d is an error activity.

[0082] Repeated activity: trajectory σ = a′da″, if ·d∈a′ and ·d∈a″, then d is a repeated activity.

[0083] A missing activity is an activity that is not present in the trace but is present in the event log.

[0084] Extra activity refers to an activity that does not exist in the event log but exists in the trace.

[0085] Misplaced activities refer to the situation where the order in which a certain activity and its adjacent activities occur is different from the order in which the fitted trajectory is recorded in the log.

[0086] Repeating activities are activities that are executed repeatedly in the track.

[0087] Define key activities: After processing the fully fitted event log, the trajectory in the event log is divided according to the activity sequence to generate n sets. For each set, the activities that can be taken in the sequence are recorded.

[0088] According to the number of possible values ​​of each set, the n sets are classified, and the key activities are defined according to the number of possible activities. The set with 1 possible activity is defined as the key activity set, and the activity is defined as the key activity.

[0089] For all key activities in the entire event log, the priority of key activities can be defined according to their order in the event. The later the key activity is in the track, the higher the priority of the key activity set is, and the more preferred it is when selecting attributes.

[0090] The key activity extraction process is:

[0091] Input: attribute set; output: key activity set.

[0092] Step 2.1. Traverse the attribute set and repeat steps 2.2 to 2.4 below;

[0093] Step 2.2. If the number of elements in the active set is 1, proceed to step 2.3.

[0094] Step 2.3. Define the activity with activity number 1 in the key activity set;

[0095] Step 2.4. ends and the key activity set is obtained.

[0096] Analyzing the time complexity of the key activity extraction method, there is only one loop in the key activity extraction method to traverse the attribute set, so the time complexity of the method is O(n), where n is the size of the attribute set.

[0097] Example 2: Taking the above earthquake emergency plan record as an example, following Example 1, the set S constructed in Example 1 is i (i=1,2,…,n) performs key activity extraction and sets the key activity set S i (i=7,4,1) are extracted and sorted in descending order of priority.

[0098] Step 3. To more efficiently implement the compliance check of the activity sequence decision tree on the trajectory, the decision tree is pruned. This embodiment removes redundant judgments based on the Occam's razor principle (that is, converts all attributes into key activity attributes and implements the construction of the decision tree when iteratively selecting the optimal classification attributes) and then constructs the decision tree.

[0099] Decision trees produce many different algorithms based on the selection of optimal classification attributes during classification. ID3, C45, and CART are relatively classic and widely used in the early days. With numerous studies, the algorithms have also been improved. The SPRINT algorithm can quickly calculate the optimal classification attribute classification. The algorithm uses histograms and attribute lists to assist in the calculation.

[0100] The PUBLIC algorithm builds a decision tree similar to the SPRINT algorithm described above, but it addresses over-pruning during the pruning process, improving accuracy. The core of the PUBLIC algorithm's decision tree construction process is the selection and construction of the optimal classification attribute. After each node is partitioned, the optimal classification attribute selected in the previous step is removed from the attribute set. The previously partitioned features in the dataset are then removed to generate new data subsets. This process is repeated over and over again to create a classification decision tree.

[0101] In compliance checks, attributes are constructed based on specific circumstances. When selecting attributes, a set is created to store all attributes. When constructing a decision tree and making each decision—that is, selecting a categorical attribute—the highest-priority key activity is first selected. Then, key activities are selected as categorical attributes in order. Attributes with fewer possible values ​​are then prioritized. After each attribute is selected, it is removed from the set until the set is empty, completing the categorical selection and decision tree construction.

[0102] Example 3: Based on the above optimal classification attribute selection criteria, the constructed attribute analysis shows that the key activity set includes the seventh activity, the fourth activity, and the first activity, with the key activities prioritized in descending order. In addition, the order of non-key activities is the sixth activity, the fifth activity, the third activity, and the second activity.

[0103] After the optimal classification attribute is determined, the next step is to build a decision tree. The decision tree consists of a root node, internal nodes, and leaf nodes. The root node includes the entire sample data set, while the internal nodes correspond to attribute selection partitioning. Different data subsets are selected based on the attribute's characteristic value, and a decision tree is recursively generated step by step. The leaf nodes correspond to the classification condition, which in this example refers to the degree of trajectory fit. The decision tree construction stops when all attributes become internal nodes. In this example, this is when the second activity is selected as the optimal classification attribute.

[0104] For the above earthquake plan log, the following log can be generated: Figure 2 The above is the decision tree constructed after multiple selections of optimal classification attributes for the earthquake emergency plan. Since the length of the trajectory changes when abnormal activities occur, most of the fitting cases will be judged in the first judgment and fewer judgments will be used.

[0105] The prediction steps are as follows: First, analyze the number of judgments for each trajectory. For the fitting trajectory set {σ1,σ2,σ3,σ4,σ5,σ6,σ7,σ8}, it is necessary to determine whether the activity order of each position is compliant before the trajectory can be finally judged as a fitting trajectory. 10 ,σ 11 ,σ 12 When judging the trajectory sequence, the decision tree constructed in the offline stage is used to judge the node of the seventh activity. For trajectory σ9, σ 10 , σ 12 When position offset occurs, the fitting of the trajectory can be judged in the first judgment. 11Since only the sequence order is dislocated without a change in length, when judging according to the decision tree, no abnormal activity is found in the first activity judgment, so the second level of judgment is performed, and the offset activity is found. The comparison ends and the algorithm ends.

[0106] The above analysis shows that while there's no significant improvement in the efficiency of fitting trajectory judgment, the judgment process for abnormal trajectories can generally be completed and the result can be obtained within the first few comparisons. Compared to sequentially comparing the first activity to the one with the abnormality and determining compliance, this method significantly improves judgment efficiency. Theoretically, the business process compliance checking method based on the activity sequence decision tree performs both feature extraction and decision tree construction offline. This requires traversing the entire log to generate the decision tree, which takes a considerable amount of time, but this operation is performed only once. When using the decision tree for fitting judgment, the search and judgment speed is very high. One reason for this is that the decision tree is constructed offline, thus requiring only simple comparisons during the search and judgment process.

[0107] Therefore, the decision tree based on activity sequence can improve the efficiency of business process compliance inspection.

[0108] Analyzing the above data, while the decision tree based on this model achieves 100% accuracy, the depth of the resulting decision tree can be kept relatively low due to the small number of activities in the event logs. However, when there are too many activities, the decision tree trained on this data becomes complex, with increased depth and width. Experience has shown that increasing the size of the tree does not necessarily lead to more accurate classification. It not only makes the decision tree more difficult to interpret, but can even lead to overfitting of the training data. Therefore, pruning the decision tree is necessary. Pruning not only simplifies the tree structure but also reduces comparison time.

[0109] There are many ways to reduce the complexity of decision tree structures. The most effective and convenient method is to prune the decision tree. Pruning, as one of the methods for stopping branches, can not only prevent the occurrence of overfitting problems, but also simplify the tree structure. Fewer classification rules can reduce comparison time and improve the efficiency of the decision tree. Decision tree pruning is divided into pre-pruning and post-pruning. Pre-pruning, also known as pre-pruning, is to stop the growth of the decision tree in advance. Post-pruning is to prune the decision tree after it is constructed. First, during the training phase, the decision tree is allowed to grow fully to generate a complete decision tree. Then, the decision tree is pruned from the bottom up according to different standards. Although post-pruning can make the post-pruned decision tree grow more fully and its generalization ability will be higher than that of the pre-pruned decision tree, the complexity and time complexity of the algorithm for the post-pruned decision tree are greater than that of the pre-pruned decision tree because the pruning operation is performed after the decision tree training is completed.

[0110] Pre-pruning limits the full growth of the decision tree during its growth process, eliminating the extra overhead of post-pruning. This significantly improves algorithm efficiency and makes it suitable for solving large-scale problems. When performing large-scale event log compliance checks, pre-pruning can be used to suppress tree size growth and improve search and judgment efficiency.

[0111] When training a decision tree, it is generally necessary to perform pruning operations on it. Pruning is analyzed from the perspective of Occam's razor. The basic meaning of Occam's razor is "do not add entities unless necessary", and its core meaning is not to artificially complicate the problem.

[0112] Decision tree pruning is an application of this principle, removing redundant judgment rules to make the decision tree not only clearer but also more efficient. Because the decision tree in the above example has too many nodes and requires too much storage space, the decision tree used for fitting judgments is pruned. Therefore, a simplified decision tree based on key activities is introduced.

[0113] In the preceding text, we defined key activities and presented their relationship to trajectory fitting. In abnormal trajectories, the presence of abnormal activities can cause shifts in the order and length of activities within the trajectory, leading to misalignment of key activities. Therefore, we use key activities as attributes to construct a decision tree to analyze trajectories. Compared to the aforementioned decision tree, this method significantly reduces the tree's depth and width, thereby improving the efficiency of decision tree construction and detection.

[0114] The specific steps of step 3 are: changing the partitioning attributes from all attributes to only using the attributes corresponding to the key activities for pruning operations, and selecting the optimal classification attributes according to the priority of the key activities (the optimal classification attribute selection is to select an optimal classification attribute to generate a tree node each time when the decision tree is recursively constructed). Iteratively select the optimal attribute to generate a tree node, and remove the selected optimal classification attribute from the attribute set, and iterate continuously until the attribute set is empty, that is, all key activities have been traversed to realize a classification decision tree. The specific process of decision tree construction is: first, the decision tree is initialized to generate the initial root node, and the decision tree node insertion operation is executed. The decision tree node insertion function is a recursive function, and the decision tree node insertion is executed recursively until the attribute set is empty and the decision tree is built.

[0115] The initialization steps of the decision tree are as follows:

[0116] Input: key activity set; Output: tree node;

[0117] Step 3.1.1. Create a new tree node;

[0118] Step 3.1.2. Select the key activity attribute with the highest priority as the root node, set the root node name, and set its attribute characteristic value Liatts;

[0119] Step 3.1.3. Execute the insertion of decision tree nodes.

[0120] The steps for inserting a decision tree node are as follows:

[0121] Input: tree node; output: decision tree.

[0122] Step 3.2.1. Traverse the attribute feature values ​​and repeat the following steps 3.2.2 to 3.2.8;

[0123] Step 3.2.2. If the attribute set is not empty, execute steps 3.2.3 to 3.2.5.

[0124] Step 3.2.3. Create a new tree node;

[0125] Step 3.2.4. Set the node name, parent node, and attribute branch of the parent node, set its attribute feature value, and insert the node into the decision tree;

[0126] Step 3.2.5. Recursively insert nodes into the decision tree.

[0127] Step 3.2.6. If the attribute set is empty, execute steps 3.2.7 to 3.2.8;

[0128] Step 3.2.7. Create a new tree node;

[0129] Step 3.2.8. Set the node as a leaf node as a judgment node, name the node yes, and insert it into the decision tree;

[0130] Step 3.2.9. After traversing the attribute features, execute steps 3.2.10-3.2.11;

[0131] Step 3.2.10. Create a tree node;

[0132] Step 3.2.11. Set the node as a leaf node as a decision node, name it "no", and insert it into the decision tree;

[0133] Step 3.2.12. End and return to the decision tree.

[0134] The above two methods are the decision tree construction process, including the initialization of the decision tree and the insertion of its nodes, and finally outputting the decision tree. The decision tree initialization method calls the node insertion function. The node insertion method includes a layer of loop, and the number of loops is the possible characteristic value of the attribute. Therefore, the time complexity of the algorithm is O(n), where n represents the possible characteristic value of the attribute.

[0135] The present invention utilizes a decision tree to improve the efficiency of business process compliance checks. The specific analysis is as follows:

[0136] For abnormal trajectories, the sequence of activities and the length of the trajectory can vary due to the presence of various abnormal activities. Consequently, subsequent activities following an abnormal activity are misaligned in order. Key activities can be used to divide event activities into several parts. If an abnormal activity occurs between the mth key activity and its adjacent key activity, the activity at the position of the (m+1) key activity of the adjacent key activity will shift. Therefore, only a single determination of this key activity is required to identify an abnormal trajectory. For the entire event, when a shift occurs and an abnormal activity occurs, the last key activity in the event (i.e., the key activity near the end activity) will shift. First, determining this activity significantly reduces the number of determinations required during the fitting process. Since abnormal trajectories often exhibit changes in trajectory length, leveraging this characteristic can significantly improve the efficiency of business process determination.

[0137] Step 4. Replay the event logs with abnormal trajectories in the fitted event logs extracted from the actual business process on the decision tree, or replay the new event logs generated in the actual business process. For each trajectory in the event log, the decision tree internal decision nodes are used to continuously determine whether the activities at each position in the trajectory are compliant, and finally output the predicted label of the leaf node, that is, the business process compliance prediction result.

[0138] In addition, the present invention also conducted experiments to verify the effectiveness of the proposed method. In particular, the efficiency and accuracy of business process compliance checks were analyzed in the experiment. The comparison indicators in this experiment were the following two:

[0139] (1) Accuracy: Compare the judgment result with the label. If the judgment result is consistent with the label, the prediction is correct. Otherwise, the judgment is wrong. The accuracy rate is the ratio of the number of correct judgments to the total number of data. (2) Number of comparisons: The number of judgments between activities during compliance checks. The number of comparisons reflects the operating efficiency of the algorithm. The fewer the number of comparisons, the higher the efficiency of the algorithm.

[0140] I. Experimental setup.

[0141] Using earthquake emergency response process simulation data and manually generated abnormal trajectory data, the method of the present invention is compared with the existing business process compliance inspection method based on the activity sequence, and an experiment is conducted to compare the two indicators.

[0142] The data set in the experiment is the event log generated by the earthquake emergency response model. The training data is the fully fitted event log generated by the model, where the trajectory length is 5-8. The test set is to add abnormal trajectories on the basis of the fully fitted event log, and abnormal trajectories with missing, additional, misplaced, and repeated activities are added in proportion.

[0143] During the testing experiment, multiple test sets were generated based on the proportion of abnormal trajectories in the dataset, ranging from 5% to 30%, with a gradient of 5%. The abnormal trajectories were increased by 5% each time. Each test set was repeated 20 times, and the average was calculated. The accuracy and number of comparisons were also calculated. The proposed algorithm was compared with the traditional method.

[0144] In order to verify the effectiveness of the business process compliance checking method based on the activity sequence decision tree of the present invention and its effect on real data sets, experiments are conducted on real event logs again, and the experimental results are given.

[0145] II. Experimental environment.

[0146] This experiment first uses Python to process the data and Excel to summarize the data. Finally, the algorithms are compared in a Java environment, ultimately achieving experimental comparison. Table 5 shows the hardware and software environment configuration information used in the experiment.

[0147] Table 5 Experimental environment

[0148]

[0149]

[0150] III. Analysis of experimental results based on artificial data.

[0151] This experiment compares the business process compliance inspection method of the present invention with the traditional method, and compares the average number of comparisons and accuracy. The experimental results are as follows: Figure 3 and Figure 4 shown.

[0152] Depend on Figure 3 and Figure 4 The results show that the average number of comparisons is relatively stable. As the proportion of abnormal trajectories increases, the average number of comparisons stabilizes at a relatively low level. The traditional method uses 5-6 comparisons, while the method of the present invention maintains a stable number of 2-3, showing minimal impact from the proportion of abnormal trajectories. Furthermore, the accuracy of the traditional method remains high, exceeding 98%. The accuracy of the method of the present invention is 100% when no abnormal trajectories are added, i.e., during the training phase. With the addition of abnormal trajectories, the accuracy during the testing phase decreases significantly, but after this decrease, the accuracy stabilizes at a success rate of 92%-94%.

[0153] Although the accuracy of the proposed method shows a downward trend after the addition of abnormal trajectories, it still remains between 92% and 94%, which is still a high accuracy rate. Furthermore, by comparing the average number of comparisons, the proposed method has a significant advantage, thus greatly reducing the method's runtime. The above experimental results show that as the number of abnormal trajectories in the dataset increases, the accuracy and average number of comparisons remain relatively stable, essentially stabilizing at a certain relative level. This, to a certain extent, reflects the stability of the proposed method and its applicability to datasets of varying sizes and varying proportions of abnormal trajectories in the event log.

[0154] IV. Analysis based on real data.

[0155] In Part III, experiments were conducted on simulated data from earthquake emergency response processes, resulting in experimental data that was then visualized. The results demonstrate that the decision tree-based business process compliance check method significantly reduces the number of comparisons and shortens the method's runtime while maintaining high accuracy. To validate the effectiveness of this method on a real-world dataset, experiments were conducted on this dataset. Table 6 shows the data analysis of the real-world log data.

[0156] Table 6 Analysis of real event log data

[0157]

[0158] According to the real event log, different proportions of abnormal tracks are added to it respectively, and the compliance is checked using the method of the present invention and the traditional method. The comparison times and accuracy are compared. The average comparison times are as follows: Figure 5 and Figure 6 shown.

[0159] Figure 5 and Figure 6 The comparison of the average number of comparisons between the decision tree algorithm and the traditional algorithm when checking the compliance of business processes on two data sets was recorded. Figure 7 and Figure 8 The comparison of the accuracy of the two datasets is recorded.

[0160] The results of the two real case data sets show that the average number of comparisons of the decision tree method of the present invention is nearly half that of the traditional method. The average number of comparisons reflects the running time of the method. The experimental results show that the method of the present invention can complete the experiment in a relatively fast time and is therefore more efficient. In addition, through the comparison of the accuracy rate, the accuracy of the decision tree method of the present invention in the real case is slightly increased compared with the simulated data data set, and is stable at around 95%.

[0161] Through the above experiments, it is not difficult to find that the business process compliance check method based on the decision tree of the present invention has greatly improved operating efficiency, which shows that the method of the present invention can well complete the compliance check on real data. When the method of the present invention is applied to process the compliance check of complex real data and event logs, its operating efficiency is high and the accuracy is within an acceptable range.

[0162] The present invention also provides a computer device for implementing the aforementioned business process compliance checking method based on an activity sequence decision tree. The computer device includes a memory and one or more processors. The memory stores executable code, and when the processor executes the executable code, the computer device implements the aforementioned business process compliance checking method.

[0163] In this embodiment, the computer device is any device or apparatus with data processing capability, which will not be described in detail here.

[0164] In addition, an embodiment of the present invention further provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, it is used to implement the above-mentioned business process compliance checking method based on the activity sequence decision tree.

[0165] The computer-readable storage medium can be an internal storage unit of any device or apparatus with data processing capabilities, such as a hard disk or memory, or an external storage device of any device with data processing capabilities, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc. equipped on the device.

[0166] Of course, the above description is only a preferred embodiment of the present invention, and the present invention is not limited to the above-mentioned embodiments. It should be noted that all equivalent substitutions and obvious deformation forms made by any technician familiar with this field under the guidance of this specification fall within the substantive scope of this specification and should be protected by the present invention.

Claims

1. A business process compliance checking method based on an activity sequence decision tree, characterized in that: The steps include: Step 1. Preprocess the fully fitted event logs extracted from the actual business process; Extract fully fitting event logs and activities from the actual business process recorded in the information system, concatenate the extracted activities into a track according to the event ID, create an activity set according to the order of activities in the track, store the activities in the track in the order in the activity set, and then create an attribute set; Step 2. Perform attribute partitioning based on the number of elements in each activity set. After the attribute set is constructed, define the unique set of activities in the activity set as the key activity set. This set of activities is defined as the key activity, and the priority of the key activities is defined based on their position. The later the activity is positioned, the higher the priority. Step 3. Pre-prune the tree by removing redundant judgments according to Occam's razor principle, and then build the decision tree; This step is as follows: pruning is performed on the attributes corresponding to key activities instead of all attributes, the optimal classification attributes are selected based on the priority of key activities, the optimal classification attributes are iteratively selected to generate decision tree nodes, and the selected optimal classification attributes are removed from the attribute set; the iteration is continued until the attribute set is empty, that is, all key activities are traversed, and a classification decision tree is realized; the decision tree construction process is as follows: Initialize the decision tree to generate the initial root node, and execute the decision tree node insertion operation. The decision tree node insertion function is a recursive function, which executes the decision tree node insertion recursively until the attribute set is empty and the decision tree is built. In step 3, the decision tree node insertion steps are as follows: Input: tree node; Output: decision tree; Step 3.2.

1. Traverse the attribute feature values ​​and repeat steps 3.2.2 through 3.2.

8. Step 3.2.

2. If the attribute set is not empty, execute Step 3.2.3 to Step 3.2.

5. Step 3.2.

3. Create a new tree node; Step 3.2.

4. Set the node name, parent node, and attribute branch of the parent node, set its attribute feature value, and insert the node into the decision tree. Step 3.2.

5. Recursively insert nodes into the decision tree. Step 3.2.

6. If the attribute set is empty, execute steps 3.2.7 to 3.2.

8. Step 3.2.

7. Create a new tree node; Step 3.2.

8. Set the node as a leaf node, name it "yes", and insert it into the decision tree. Step 3.2.

9. After traversing the attribute features, execute steps 3.2.10-3.2.11; Step 3.2.

10. Create a tree node; Step 3.2.

11. Set the node as a leaf node, name it "no", and insert it into the decision tree. Step 3.2.

12. End and return to the decision tree; Step 4. Replay the fitted event logs extracted from the actual business process on the decision tree, adding event logs with abnormal trajectories or replaying new event logs generated in the actual business process. For each trajectory in the event log, the decision tree internal decision nodes continuously determine whether the activities at each location in the trajectory are compliant. Ultimately, the predicted label of the leaf node is output, which is the business process compliance prediction result.

2. The business process compliance checking method based on the activity sequence decision tree according to claim 1 is characterized in that: The step 1 is specifically as follows: Input: event log dataset; Output: attribute set; Step 1.

1. Traverse the entire event log and repeat steps 1.2-1.3 below. Step 1.

2. Create a trajectory dictionary for events that do not have a trajectory dictionary. Step 1.

3. Concatenate the activities in the event log into a trajectory. Step 1.

4. Traverse the trajectory generated in step 1.3 above and repeat steps 1.5 to 1.8 below. Step 1.

5. Determine whether the activity set for each position has been created; if not, proceed to step 1.6; Step 1.

6. Create an activity collection; Step 1.

7. Store the activities in the trace in the obtained set in order; Step 1.

8. Create an attribute set and assign the activity set to the attribute set. Step 1.

9. End, get the attribute set.

3. The business process compliance checking method based on the activity sequence decision tree according to claim 2 is characterized in that: In step 2, the key activity extraction process is as follows: Input: attribute set; Output: key activity set; Step 2.

1. Traverse the attribute set and repeat steps 2.2 to 2.4 below. Step 2.

2. If the number of elements in the active set is 1, proceed to step 2.

3. Step 2.

3. Define the activity with activity number 1 in the key activity set; Step 2.

4. End and obtain the key activity set.

4. The business process compliance checking method based on the activity sequence decision tree according to claim 3 is characterized in that: In step 3, the initialization steps of the decision tree are as follows: Input: key activity set; Output: tree node; Step 3.1.

1. Create a new tree node; Step 3.1.

2. Select the key activity attribute with the highest priority as the root node, set the root node name, and set its attribute characteristic value Liatts; Step 3.1.

3. Insert decision tree nodes.

5. A computer device comprising a memory and a processor, wherein the memory stores executable code, characterized in that: When the processor executes the executable code, it is used to implement the business process compliance checking method based on the activity sequence decision tree according to any one of claims 1 to 4.

6. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, it is used to implement the business process compliance checking method based on the activity sequence decision tree as described in any one of claims 1 to 4.

Images