Automatic discovery and modeling method for textual business process
By introducing technologies such as bidirectional ON-LSTM, GNN, CNN and XGBoost in traditional modeling methods, the accuracy and flexibility of traditional modeling methods in complex business process processing are solved, and a more efficient and flexible automated modeling method is achieved.
Patent Information
- Application Number
- CN202411712015.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-27
- Publication Date
- 2025-03-14
AI Technical Summary
Traditional manual modeling methods have consistency and accuracy problems when dealing with complex business processes, and rule-based modeling methods have insufficient flexibility and maintenance costs, making it difficult to adapt to rapidly changing business environments.
Based on the Bi-LSTM+Attention+CRF model, Bi-LSTM is replaced as a bidirectional ON-LSTM, and graph neural network (GNN) is introduced, combining CNN and XGBoost to improve the recognition accuracy of task attribute elements and the accuracy and robustness of task relationship recognition.
Through automated modeling methods, the identification accuracy and integrity of business process models are improved, the dependence on labor is reduced, efficiency is improved, and the adaptability and flexibility are shown when dealing with complex and diverse business processes.
Smart Images
Figure CN119624274A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of artificial intelligence and modeling technology, and in particular to a method for automatic modeling from a textual business process description. Background Art
[0002] In the information age, business process management (BPM) has become one of the key factors for enterprises to enhance their competitiveness. By effectively implementing BPM, enterprises can not only improve work efficiency, but also significantly improve service quality and customer satisfaction. As a core component of BPM, business process modeling plays a vital role in this process. Business process modeling represents the business activities, tasks and decision logic of an enterprise in a graphical or structured way, which is easy to understand and optimize, thereby achieving efficient process management. However, the traditional manual modeling method relies on personal experience, and the designed process varies from person to person, resulting in low consistency and accuracy of the model; in addition, manual flow chart drawing and document writing are time-consuming and laborious, and are prone to omissions or errors, especially when dealing with large or complex projects. At the same time, frequent adjustments to the process also bring a huge burden to manual updates. With the increase in business complexity, these challenges have prompted enterprises to turn to rule-based modeling methods to seek higher levels of standardization and automation. However, rule-based modeling methods also have problems such as insufficient flexibility, high maintenance costs, increased complexity and poor adaptability. When the business environment changes rapidly or encounters special circumstances, fixed rules may be difficult to adjust automatically, resulting in rigid processes; at the same time, as the number of rules increases, the difficulty and cost of system maintenance will also increase significantly, and too many rules may make the system complex and difficult to understand, affecting efficiency. Therefore, although rule-based modeling has solved the problem of manual modeling to a certain extent, further technological innovation and optimization are still needed. In order to better meet the needs of modern enterprises, enterprises have begun to use artificial intelligence technology to analyze unstructured texts and automatically model from texts. Compared with previous modeling methods, this method can greatly improve work efficiency and flexibility, but the model's shortcomings in dealing with long-distance dependencies and integrating contextual information, as well as its limitations in capturing complex relationships between tasks (such as selection relationships and jump relationships), will affect the accuracy and completeness of building BPMN models.
[0003] Therefore, the present application aims to provide a method for automatic discovery and modeling of textual business processes. Summary of the invention
[0004] The purpose of this application is to provide an automatic discovery and modeling method for textual business processes, and to improve the recognition accuracy of task attribute elements and the accuracy and robustness of task relationship recognition through the following improvements. Specifically, based on the prior art Bi-LSTM+Attention+CRF model, this application replaces Bi-LSTM with bidirectional ON-LSTM, and enhances the model's ability to capture long-term dependencies through ordered neurons, better integrates contextual information, and improves the recognition accuracy of task attribute elements; at the same time, in terms of process element relationship extraction, XGBoost is improved and a graph neural network (GNN) is introduced. GNN can extract the graph structure features of task relationships and capture the complex relationships between tasks. When combined with XGBoost, the model can further improve the accuracy and robustness of task relationship recognition, thereby building a more accurate BPMN model. Compared with the prior art, the beneficial effects of this application are: compared with the traditional LSTM, ON_LSTM (Ordered Neurons LSTM) can better handle dependencies in long sequences by optimizing the gating mechanism of LSTM, and has a better ability to capture long-distance information in the text; ON_LSTM can better handle hierarchical texts, especially for those with obvious hierarchical texts, such as business process descriptions, and can more effectively capture hierarchical information in the text and improve the accuracy of element extraction; the introduction of CNN can effectively capture local features in the text, which is very useful for identifying the relationship between tasks, because the task relationship is often reflected in the local structure of the text. CNN can capture these static text features, and XGBoost can further utilize these features, combined with the dynamic information of the context, for a deeper analysis. Therefore, by capturing local features through CNN and then using XGBoost for classification, the accuracy of task relationship recognition can be effectively improved, especially when dealing with complex task relationships; compared with rule-based methods, automated modeling methods are more robust and flexible when dealing with flexible and diverse unstructured texts, and can better adapt to different text content and forms. It does not need to rely on a fixed text structure, and also avoids the problem of frequent adjustment of rules due to text changes, thereby showing higher adaptability and efficiency when dealing with diverse and complex information; through automated model extraction methods, the dependence on manual labor is greatly reduced and efficiency is improved.
[0005] In order to achieve the above purpose, this application adopts the following technical solutions:
[0006] The present application embodiment provides a method for automatically discovering and modeling a textual business process. The specific steps of the modeling method are as follows:
[0007] S1: Obtain complete and accurate unstructured business process description for modeling;
[0008] S2: Use natural language processing technology to analyze the business process description and identify the process elements in the description;
[0009] S3: Then identify the logical relationship between process elements and construct intermediate representation;
[0010] S4: Map the process into a standardized BPMN model based on the intermediate representation.
[0011] As a further solution of this solution, the specific steps of using natural language processing technology to identify process elements in the unstructured business process description in step S2 are as follows:
[0012] First, the business process description text is preprocessed and annotated, including data cleaning (noise removal, missing value processing, etc.), sentence segmentation, word segmentation, and part-of-speech tagging. Then, according to the task attributes to be extracted, the word segmentation results are annotated with elements (using the designed labeling rules) and relationship labels are set for the task pairs.
[0013] Convert the words in the text into vector form for deep learning model processing. The word sequence is encoded using the FastText algorithm, and the part of speech is encoded using One-Hot encoding. The two are concatenated to construct a feature vector.
[0014] Construct input feature vectors and feed them into the ON_LSTM network. In addition, introduce the attention mechanism to focus on key parts to improve recognition accuracy. Use conditional random fields (CRF) for sequence labeling. Combine the labeled data to train the ON_LSTM-Attention-CRF network to identify task attribute elements.
[0015] By comprehensively analyzing the attribute elements identified in the short sentences, these elements are integrated into a specific task description based on the combination relationship between them.
[0016] As a further solution of this solution, the specific steps of identifying the logical relationship between the process elements in step S3 and constructing the intermediate representation are as follows:
[0017] Analyze text to generate task relationship feature vector;
[0018] Build a graph based on tasks and their relationships, with each task as a node and the relationships between tasks as edges;
[0019] The GNN model is used to learn the constructed graph to capture deeper interaction patterns between tasks. The GNN model can update the features of nodes so that each node not only contains its own information but also integrates the information of neighboring nodes.
[0020] The node features output by GNN are combined with the original task relationship feature vector to form a new feature vector. The XGboost model is trained using the pre-annotated relationship element labels and the fused feature set. Five logical relationships between elements are identified: Parallel, Synchronous, Exclusive, Sequential, and Jump.
[0021] As a further solution of this solution, the specific steps of mapping the process into a standardized BPMN model according to the intermediate representation in step S4 are as follows:
[0022] Use the Camunda workflow engine to build a BPMN file and integrate all the elements obtained in the above steps (activities, roles, events, gateways, pools, sequence flows, etc.) into the BPMN model. Specifically, based on the extracted tasks, roles, events, gateways and other elements, call the API provided by Camunda to create corresponding BPMN model elements, define sequence flows, and establish connections between activities and between activities and gateways.
[0023] Compared with the prior art, the beneficial effects of this application are:
[0024] 1.ON_LSTM can better handle dependencies in long sequences by optimizing the gating mechanism of LSTM, and has a better ability to capture long-distance information in text;
[0025] 2. Compared with traditional LSTM, ON_LSTM (Ordered Neurons LSTM) can better process hierarchical texts, especially for those with obvious hierarchical structures, such as business process descriptions. It can more effectively capture the hierarchical information in the text and improve the accuracy of element extraction;
[0026] 3. The introduction of CNN can effectively capture local features in the text, which is very useful for identifying the relationship between tasks, because the task relationship is often reflected in the local structure of the text. CNN can capture these static text features, and XGBoost can further use these features and combine them with the dynamic information of the context for a deeper analysis. Therefore, by capturing local features through CNN and then using XGBoost for classification, the accuracy of task relationship recognition can be effectively improved, especially when dealing with complex task relationships;
[0027] 4. Compared with rule-based methods, automated modeling methods are more robust and flexible when dealing with flexible and diverse unstructured texts. They can better adapt to different text contents and forms. They do not need to rely on fixed text structures and avoid the problem of frequent rule adjustments due to text changes. They can thus show higher adaptability and efficiency when dealing with diverse and complex information.
[0028] 5. Through the automated model extraction method, the reliance on manual labor is greatly reduced and the efficiency is improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] The accompanying drawings are used to provide further understanding of the present application and constitute a part of the specification. They are used to explain the present application together with the embodiments of the present application and do not constitute a limitation of the present application.
[0030] Figure 1 A modeling flowchart of the business process modeling method for this application;
[0031] Figure 2 This is an example diagram of the BPMN model for this application;
[0032] Figure 3 This is the ON_LSTM model architecture diagram in the embodiment of this application;
[0033] Figure 4 This is a model architecture diagram of entity recognition in an embodiment of the present application;
[0034] Figure 5 This is a diagram of the relationship extraction model architecture in an embodiment of the present application. DETAILED DESCRIPTION
[0035] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments.
[0036] Example 1
[0037] Reference Figure 1-Figure 5 The present application embodiment provides a method for automatically discovering and modeling textual business processes, and the specific steps of the method are as follows:
[0038] S1: Obtain complete and accurate unstructured business process description for modeling;
[0039] S2: Use natural language processing technology to analyze the business process description and identify the process elements in the description. The specific steps are as follows:
[0040] S21: First, the business process description text is preprocessed and annotated, including data cleaning (noise removal, missing value processing, etc.), sentence segmentation, word segmentation, and part-of-speech tagging. Then, according to the task attributes to be extracted, the word segmentation results are annotated with elements (using the designed labeling rules), and relationship labels are set for the task pairs.
[0041] S22: Convert the words in the text into vector form for deep learning model processing, where the word sequence is encoded using the FastText algorithm, and the part of speech is encoded using One-Hot, and the two are concatenated to construct a feature vector. The constructed feature vector is sent to the bidirectional ON_LSTM network to capture the context information of the word, generate a hidden layer representation, input it into the attention layer to calculate the attention score, and weighted sum the hidden layer representation and the attention score to remove unimportant word information, highlight and amplify key information, improve recognition accuracy, use conditional random fields (CRF) for sequence labeling, capture the dependency between labels, improve the performance of sequence labeling tasks, and identify process elements;
[0042] S23: Through comprehensive analysis of the attribute elements identified in the short sentences, these elements are integrated into a specific task description based on the combination relationship between them.
[0043] S3: Then identify the logical relationship between process elements and construct an intermediate representation. The specific steps are as follows:
[0044] S31: Analyze text to generate task relationship feature vector;
[0045] S32: Build a graph based on tasks and their relationships, with each task as a node and the relationships between tasks as edges;
[0046] S33: Use the GNN model to learn the constructed graph and capture the deeper interaction patterns between tasks. The GNN model can update the features of the nodes so that each node not only contains its own information, but also integrates the information of neighboring nodes.
[0047] S34: Combine the node features output by GNN with the original task relationship feature vector to form a new feature vector, and use the pre-annotated relationship element labels and the fused feature set to train the XGboost model; identify five logical relationships between elements: parallel, synchronous, exclusive, sequential, and jump.
[0048] S4: Map the process into a standardized BPMN model based on the intermediate representation. The specific steps are as follows:
[0049] S41: Use the Camunda workflow engine to build a BPMN file, and integrate all the elements obtained in the above steps (activities, roles, events, gateways, pools, sequence flows, etc.) into the BPMN model. Specifically, based on the extracted tasks, roles, events, gateways and other elements, call the API provided by Camunda to create corresponding BPMN model elements, define sequence flows, and establish connections between activities and between activities and gateways.
[0050] In summary, this example is an automatic discovery and modeling method for textual business processes. Through artificial intelligence technology, unstructured text is analyzed to identify process elements and extract relationships, and a BPMN model is automatically constructed from the text.
[0051] Note that the above content only describes the preferred implementation of the embodiment of the present application and the technical principles of its application. The embodiment of the present application is not limited to the specific examples described in detail here. For those skilled in the art, a variety of obvious modifications, reconfigurations and replacements can be made without departing from the core idea of the embodiment of the present application. Although the embodiment of the present application has been described in more detail through the above-mentioned specific embodiments, the embodiment of the present application is far more than this. Without departing from the basic concept of the embodiment of the present application, more other equivalent implementation methods can also be covered, and the protection scope of the embodiment of the present application is defined by the attached claims.
[0052] The above is a detailed description of an embodiment of the present application, but the content is only a preferred embodiment of the present application and cannot be considered to limit the scope of implementation of the present application. All equivalent changes and improvements made within the scope of the present application should still fall within the scope of the patent coverage of the present application.
Claims
1. A method for automatically discovering and modeling textual business processes, characterized in that: The specific steps of this method are as follows: S1: Obtain complete and accurate unstructured business process description for modeling; S2: Use natural language processing technology to analyze the business process description and identify the process elements in the description; S3: Then identify the logical relationship between process elements and construct intermediate representation; S4: Map the process into a standardized BPMN model based on the intermediate representation.
2. The method for automatically discovering and modeling textual business processes according to claim 1, characterized in that: In using natural language processing technology to identify process elements in unstructured business process descriptions, the specific steps for using natural language processing to preprocess and annotate data are as follows: Perform data cleaning on the business process description text: remove noise in the text, process missing values, etc.; Sentence and word segmentation: divide the text into sentences and segment sentences into words or phrases; Part of speech and element labeling: Label the part of speech for each word, and label the elements of the word segmentation results according to the attributes of the task to be extracted, and set relationship labels for the task pairs.
3. The method for automatically discovering and modeling textual business processes according to claim 1, characterized in that: When using natural language processing technology to identify process elements in unstructured business process descriptions, the specific steps of model training and element identification are as follows: Convert the words in the text into vector form for deep learning model processing. The word sequence is encoded using the FastText algorithm, and the part of speech is encoded using One-Hot encoding. The two are concatenated to construct a feature vector. Construct a combined feature vector of word order and part of speech, send it to the ON_LSTM network, and introduce the attention mechanism to focus on the key parts to improve recognition accuracy. Use conditional random fields (CRF) for sequence labeling, and combine the labeled data to train the ON_LSTM-Attention-CRF network to identify task attribute elements. By comprehensively analyzing the attribute elements identified in the short sentences, these elements are integrated into a specific task description based on the combination relationship between them.
4. The method for automatically discovering and modeling textual business processes according to claim 1, characterized in that: The specific steps to identify the logical relationship between process elements are as follows: Analyze text to generate task relationship feature vector; Build a graph based on tasks and their relationships, with each task as a node and the relationships between tasks as edges; Use the GNN model to learn the constructed graph and capture deeper interaction patterns between tasks; the GNN model can update the features of nodes so that each node not only contains its own information, but also integrates the information of neighboring nodes; The node features output by GNN are combined with the original task relationship feature vector to form a new feature vector. The XGboost model is trained using the pre-annotated relationship element labels and the fused feature set. Five logical relationships between elements are identified: Parallel, Synchronous, Exclusive, Sequential, and Jump.
5. The method for automatically discovering and modeling textual business processes according to claim 1, characterized in that: The specific steps to map the process into a standardized BPMN model based on the intermediate representation are as follows: Use the Camunda workflow engine to build a BPMN file and integrate all the elements obtained in the above steps (activities, roles, events, gateways, pools, sequence flows, etc.) into the BPMN model. Specifically, based on the extracted tasks, roles, events, gateways and other elements, call the API provided by Camunda to create corresponding BPMN model elements, define sequence flows, and establish connections between activities and between activities and gateways.