An event log acquisition method based on task mining technology

Abstract

The application discloses an event log acquisition method based on task mining technology, relates to the field of task mining, and comprises the following steps: 1, when starting to implement a process mining project, business operations of the whole enterprise are stored by using task mining technology; 2, the format and attributes of event log data saved by task mining and event log data imported by a process mining tool are set to be consistent; 3, event log data stored by task mining technology is saved according to the data format required by the process mining tool for importing and cleaning data; 4, the stored event log data is imported into the process mining tool to perform format testing; and 5, after the testing is passed, the imported stored data can be used. The application standardizes and sorts event log data required by process mining, generates input data required by the process mining tool, can collect data across platforms and systems, and fundamentally solves the problems of difficulty in sorting and cleaning data of the process mining tool and few effective data.

Description

Technical Field

[0001] This invention relates to the fields of information technology and artificial intelligence, and is applicable to all projects and enterprises that use process mining technology, especially projects with complex preliminary event data that do not meet the conditions for process mining and cleaning. Background Technology

[0002] Process mining is a technology that collects and monitors event log information generated in the workplace to reconstruct real and detailed business processes, and then analyzes and optimizes the reconstructed workflows. Using this technology, one can gain a comprehensive understanding of business processes and greatly improve operational efficiency.

[0003] Task mining technology utilizes user interaction data (also known as desktop data) to record various tasks within a process. This process allows for the saving of data related to the process as needed. This type of data includes keystrokes, mouse clicks, and data inputs that occur during the completion of specified operations, and can be recorded across systems, terminals, and accounts. Task mining can use Optical Character Recognition (OCR), Natural Language Processing (NLP), and machine learning algorithms to interpret and record this data, and then use event log data in a format perfectly suited for process mining as the foundational event log. This allows analysts and other stakeholders to identify operational status, overall process performance, and more.

[0004] Event log information is a prerequisite for the efficiency of the entire technology and is of paramount importance in the entire process. However, the data retained from ERP systems, financial processing systems, and other systems used by enterprises in the early stages is mixed and has a chaotic data format, resulting in low availability and poor data quality of event logs, which fails to fully leverage the effectiveness of process mining technology. This invention will solve the problem of standardized and efficient event log data format by using a method for generating event logs based on task mining technology. Summary of the Invention

[0005] To address the aforementioned problems in existing technologies, the purpose of this invention is to standardize and organize the event log data required for process mining, generate input event log data that enterprises can use and that is efficient for process mining tools, and enable cross-platform and cross-system data collection, thereby fundamentally solving the problems of difficulty in organizing and cleaning data and the scarcity of effective data in process mining tools.

[0006] To achieve the above objectives, the main technical solution of this invention is: a method for obtaining event logs based on task mining technology, the core content of which includes setting up event log generation through task mining, specifically including:

[0007] Step 1: While launching the process mining project, task mining technology is used to retain all business operations across the enterprise.

[0008] Step 2: Ensure that the format and attributes of the event log data saved by the task mining tool are consistent with those of the event log data imported by the process mining tool;

[0009] Step 3: Import and clean the data according to the required data format using the data mining tool, and save the event log data retained by the task mining technology;

[0010] Step 4: Import the retained event log data into the process mining tool for format testing;

[0011] Step 5: Once the test is passed, the imported and retained data can be used on a large scale.

[0012] Furthermore, step 1 specifically involves: when an enterprise begins implementing a process mining project, it uses task mining technology to capture and extract relevant information about business operations to ensure the recording and retention of these operations. This relevant information includes user interactions, system logs, and event logs.

[0013] Furthermore, the task mining technology includes:

[0014] Step 1.1: Preprocess the initial text data by reverse translation. Translate the initial text to be used for task mining into a foreign language and then back into Chinese to maintain the balance of the data.

[0015] Step 1.2: Tag the preprocessed text for nouns and extract the nouns. Then find all their synonym sets on DANet, use glossbert's word sense disambiguation to determine the meaning of these nouns in the context, and thus obtain their correct synonym sets. Finally, replace these nouns with synonyms to create new input text.

[0016] Step 1.3: Encode the input text using the DeBERTa model; at the same time, use a knowledge graph generation system to generate a knowledge graph corresponding to the input text, then copy the generated knowledge graph so that it corresponds to the data-augmented input text, and finally input the knowledge graph into the knowledge graph encoder.

[0017] Step 1.4: After the input text is encoded, the input text embedding is obtained. The corresponding knowledge graph is encoded to obtain the knowledge graph embedding. Then, the input text embedding and the knowledge graph embedding are input into the fusion module.

[0018] Step 1.5: The merged result from the fusion module is fed into a linear layer, and the task information of the text is obtained after a softmax operation.

[0019] Furthermore, the fusion module in step 1.5 includes a self-attention module, a linear layer, a multi-head cross-attention module, and a merging layer connected in sequence. The text content is input from the self-attention module and output from the merging layer.

[0020] Furthermore, the processing procedure of the fusion module in step 1.5 is as follows: First, the input text embedding and knowledge graph embedding are processed through a self-attention module, and then they are input into a linear layer to obtain the input of the multi-head cross-attention layer. The input text embedding is used as the query input of the multi-head cross-attention layer, and the knowledge graph embedding is used as the two support inputs of the multi-head cross-attention layer. Then, the output of the multi-head cross-attention layer is merged with the input text embedding, so that the information of the original input text embedding is preserved without loss.

[0021] Furthermore, the attributes in step 2 include: field naming, data type, and data structure.

[0022] Furthermore, the data cleaning in step 3 is performed using Large Language Models (LLMs) and iterative processing.

[0023] Furthermore, the specific steps include:

[0024] Step 3.1: Input the data to be cleaned and the specified format into the Large Language Model (LLMs). Then, the Large Language Model (LLMs) cleans the data according to the provided format to obtain Data 1.

[0025] Step 3.2: Perform iterative cleaning: Input the obtained data 1 and the specified format back into the large language model LLMs to obtain data 2; then input data 2 and the specified format back into the large language model LLMs to obtain data 3;

[0026] Step 3.3: Perform cross-entropy loss on data 1, data 2, data 3 and the original data to control the validity of the data obtained in each iteration.

[0027] Compared with the prior art, the beneficial effects of this application are as follows:

[0028] [1] To address the data imbalance problem in task mining, we use reverse translation to enrich the data in the less common classes.

[0029] [2] In terms of task mining data augmentation, we used DANet and glosbert to create additional data to train our task mining network by replacing noun synonyms in the original text.

[0030] [3] In order to make up for the information that is easily missing in task mining, we propose to use knowledge graphs to jointly carry out task mining, which greatly improves the performance of task mining.

[0031] [4] In the data cleaning process, we use an iterative processing method to clean the data, which not only ensures the validity of the data, but also makes the cleaned data more applicable to the subsequent processes. Attached Figure Description

[0032] Figure 1 This is a schematic diagram illustrating the data flow direction in the event logs of existing mining tools.

[0033] Figure 2 This is a flowchart of the log acquisition method based on task mining technology in this invention.

[0034] Figure 3 This is a schematic diagram of the overall process of task mining technology in this invention.

[0035] Figure 4 This is a schematic diagram of the fusion module structure and workflow in this invention.

[0036] Figure 5 This is a flowchart illustrating the cross-entropy loss step in this invention. Detailed Implementation

[0037] The technical solution of the present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0038] Figure 1 This diagram illustrates the data flow direction in event logs from existing data mining tools. Data from previously used ERP and financial systems is often mixed and poorly formatted, resulting in low availability and poor data quality in event logs, thus failing to fully leverage the capabilities of process mining techniques.

[0039] Figure 2 This is a flowchart of the log acquisition method based on task mining technology in this invention. It standardizes and organizes the event log data required for process mining, generating input event log data that is usable and efficient for enterprise process mining tools. It can collect data across platforms and systems, fundamentally solving problems such as the difficulty in organizing and cleaning data for process mining tools and the scarcity of effective data.

[0040] This embodiment employs an event log acquisition method based on task mining technology, including the following steps:

[0041] Step 1: While launching the process mining project, task mining technology is used to retain data across all business operations within the enterprise. The specific steps are as follows:

[0042] Enterprises should ensure the recording and retention of business operations when initiating a process mining project. Task mining techniques can be used to capture and extract relevant information from business operations, such as user interactions, system logs, and event records. Retaining this data is essential for subsequent process mining and analysis.

[0043] Task mining is an emerging technology. It focuses on tasks—smaller components of processes or subprocesses that consist of multiple steps, typically performed manually by employees at their workstations. It enables businesses to better understand how users perform tasks by tracking user activity and collecting user interaction information. Businesses can use the collected information to review how they manage operations, identify the most common errors in task execution, and identify tasks that can be automated.

[0044] The specific process of the task mining method used in this embodiment is as follows:

[0045] Reverse translation technology is employed to address the data imbalance problem. Then, a data augmentation network, DANet, is used to augment the text data. Simultaneously, to address issues such as inaccuracies in task mining caused by direct task mining, a knowledge graph is introduced to assist in task mining. This embodiment uses the latest DeBERTA as the text encoder and the KGE module as the knowledge graph encoder. The overall process of the task mining method is as follows: Figure 3 As shown.

[0046] Many texts used in task mining suffer from problems such as uneven keyword distribution and unclear mining information. To address these issues, preprocessing of the text data is necessary. This embodiment addresses the uneven data distribution problem through reverse translation. The text to be mined is first translated into English or another language, and then translated back into Chinese. This preprocessing maintains data balance.

[0047] More data often leads to better model performance because more data is available for training. This embodiment first tags the text to be mined for nouns, then extracts the nouns. Next, we find all their synonym sets on DANet and use glosbert's word sense disambiguation to determine the meaning of these nouns in context, thus obtaining their correct synonym sets. Finally, we replace these nouns with synonyms to create new input text.

[0048] To encode the input, this embodiment uses DeBERTa, which has demonstrated good performance across various natural language processing tasks. To better mine task information, a knowledge graph is preferably introduced to supplement key information and assist task mining. A knowledge graph generation system is used to generate a knowledge graph corresponding to the input text. The generated knowledge graph is then copied to correspond with the data-augmented input text. The knowledge graph is then input into the knowledge graph encoder (KGE module), which consists of a state-of-the-art large language model.

[0049] The input text is encoded to obtain the input text embedding. The corresponding knowledge graph is encoded to obtain the knowledge graph embedding. Then, the input text embedding and the knowledge graph embedding are input to the fusion module.

[0050] Fusion modules such as Figure 4 As shown, this includes a self-attention module and a multi-head cross-attention module. First, the input text embedding and knowledge graph embedding are passed through a self-attention module, and then input into a linear layer to obtain the input for the multi-head cross-attention layer. The input text embedding serves as the query input for the multi-head cross-attention layer, and the knowledge graph embedding serves as two support inputs. Then, the output of the multi-head cross-attention layer is merged with the input text embedding, thus preserving the original information from the input text embedding.

[0051] Finally, the merged result is passed through a linear layer, and then a softmax operation is performed to obtain the task information of the text.

[0052] Step 2: Set the event log data saved by the task mining tool to be consistent with the event log data imported by the process mining tool in terms of format and attributes.

[0053] To ensure that the event log data saved by task mining can be successfully imported and processed by process mining tools, it is necessary to ensure the consistency of data format and attributes. This includes matching in terms of field naming, data type, and data structure.

[0054] Step 3: Import and clean the data according to the required data format using the data mining tool, and save the event log data retained by the task mining technology.

[0055] According to the requirements of process mining tools, the event log data retained by task mining techniques is imported and cleaned. This involves steps such as data format conversion, data cleaning, and preprocessing to make the data suitable for the subsequent requirements of process mining tools.

[0056] This method utilizes Large Language Models (LLMs) and iterative processing to clean the data. The specific implementation process is as follows:

[0057] First, the data to be cleaned and the specified format are input into a large language model (LLM). Then, the LLM cleans the data according to the provided format to obtain Data 1. The format of the data in Data 1 already initially meets the requirements of the subsequent process. However, in order to obtain data that better meets the requirements, we need to perform iterative cleaning.

[0058] Then, the obtained data 1 and the specified format are input again into the Large Language Model (LLMs) to obtain data 2. Then, data 2 and the specified format are input again into the Large Language Model (LLMs) to obtain data 3, and so on, iterating. In our method, the iteration is performed three times.

[0059] After three iterations, the cleaned data was finally obtained, which basically met the prescribed format and could be used for subsequent operations.

[0060] To control the validity of the data obtained in each iteration—that is, to ensure that the data after each iteration does not change its original meaning—we apply cross-entropy loss to data 1, data 2, and the original data. The specific process is as follows: Figure 5 As shown.

[0061] Step 4: Import the retained event log data into the process mining tool for format testing;

[0062] The cleaned and transformed event log data is imported into the process mining tool and its format is tested. This is to ensure that the data can be correctly imported into the tool and that the tool can correctly parse and process the data.

[0063] Step 5: Once the test is passed, the imported and retained data can be used on a large scale.

[0064] The above solution designs a task mining tool that can set relevant software and keywords in the workflow. When related operations are triggered, the task mining log generation tool automatically starts. All work operations and business processes of the enterprise are recorded through task mining technology, generating and saving formatted event logs. These formatted event logs generated by task mining technology are then imported into subsequent process mining systems and tools for data verification and other follow-up operations. This solution primarily addresses low-quality time log data that is difficult to process and format using process mining technology, using task mining technology to rewrite the formatted data.

[0065] While the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Those skilled in the art can make various modifications and refinements without departing from the spirit and scope of the invention. Therefore, the scope of protection of the present invention shall be determined by the claims.

Claims

1. A method for obtaining event logs based on task mining technology, characterized in that, Includes the following steps: Step 1: While launching the process mining project, task mining technology is used to retain all business operations across the enterprise. The task mining technology includes: Step 1.1: Preprocess the initial text data by reverse translation. Translate the initial text to be used for task mining into a foreign language and then back into Chinese to maintain the balance of the data. Step 1.2: Tag the preprocessed text for nouns and extract the nouns. Then find all their synonym sets on DANet, use Glossbert's word sense disambiguation to determine the meaning of these nouns in the context, and thus obtain their correct synonym sets. Finally, replace these nouns with synonyms to create new input text. Step 1.3: Encode the input text using the DeBERTa model; at the same time, use a knowledge graph generation system to generate a knowledge graph corresponding to the input text, then copy the generated knowledge graph so that it corresponds to the data-augmented input text, and finally input the knowledge graph into the knowledge graph encoder. Step 1.4: After the input text is encoded, the input text embedding is obtained. After the corresponding knowledge graph is encoded, the knowledge graph embedding is obtained. Then, the input text embedding and the knowledge graph embedding are input into the fusion module. Step 1.5: The merged result from the fusion module is fed into a linear layer, and after a softmax operation, the task information of the text is obtained; Step 2: Ensure that the format and attributes of the event log data saved by the task mining tool are consistent with those of the event log data imported by the process mining tool; Step 3: Import and clean the data according to the required data format using the data mining tool, and save the event log data retained by the task mining technology; Step 4: Import the retained event log data into the process mining tool for format testing; Step 5: Once the test is passed, you can import and retain the data.

2. The event log acquisition method based on task mining technology according to claim 1, characterized in that, Step 1 specifically involves the enterprise using task mining technology to capture and extract relevant information about business operations when it begins implementing the process mining project, ensuring that the business operations are recorded and retained. This relevant information includes user interactions, system logs, and event records.

3. The event log acquisition method based on task mining technology according to claim 1, characterized in that, The fusion module in step 1.5 includes a self-attention module, a linear layer, a multi-head cross-attention module, and a merging layer connected in sequence. The text content is input from the self-attention module and output from the merging layer.

4. The event log acquisition method based on task mining technology according to claim 3, characterized in that, The processing procedure of the fusion module in step 1.5 is as follows: First, the input text embedding and knowledge graph embedding are processed through a self-attention module, and then they are input into a linear layer to obtain the input of the multi-head cross-attention layer. The input text embedding is used as the query input of the multi-head cross-attention layer, and the knowledge graph embedding is used as the two support inputs of the multi-head cross-attention layer. Then, the output of the multi-head cross-attention layer is merged with the input text embedding, so that the information of the original input text embedding is preserved without loss.

5. The event log acquisition method based on task mining technology according to claim 1, characterized in that, The attributes in step 2 include: field naming, data type, and data structure.

6. The event log acquisition method based on task mining technology according to claim 1, characterized in that: The data cleaning in step 3 is performed using large language models (LLMs) and iterative processing.

7. The event log acquisition method based on task mining technology according to claim 6, characterized in that, Specifically, the following steps are included: Step 3.1: Input the data to be cleaned and the specified format into the Large Language Model (LLMs). Then, the Large Language Model (LLMs) cleans the data according to the provided format to obtain Data 1. Step 3.2: Perform iterative cleaning: Input the obtained data 1 and the specified format back into the large language model LLMs to obtain data 2; then input data 2 and the specified format back into the large language model LLMs to obtain data 3; Step 3.3: Perform cross-entropy loss on data 1, data 2, data 3 and the original data to control the validity of the data obtained in each iteration.

Images