Automatic generation method of process model

Abstract

The invention discloses an automatic generation method of a process model, which belongs to the technical field of computer application, and comprises the following steps: S1, preprocessing an original business text based on a large language model to obtain a standard business text; s2, extracting elements in the standard business text and dividing the elements into classification sets corresponding to the elements; s3, simplifying a classification set corresponding to the element, and identifying an end node in the standard business text; s4, performing visualization processing on the simplified classification set based on an end node in the standard business text to obtain a business flow chart; according to the method, the problems of low efficiency and high error rate caused by manual intervention in the process modeling process are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of computer application, and in particular to an automatic generation method of a process model. Background Art

[0002] A process model is a graphical or textual expression method that describes in detail and clearly each link in a business flow or workflow, including steps, activities, decision points, branch paths, loop mechanisms, and the logical sequence and interrelationships between them. The purpose of building this model is to assist people in deeply understanding and thoroughly analyzing complex business flows, thereby achieving process optimization, automation, real-time monitoring and efficient management.

[0003] Existing process modeling technologies mainly include: manual modeling, rule-driven modeling and semi-automatic modeling; among them, manual modeling manually draws each step and element of the business flow into a model step by step. This method is time-consuming, labor-intensive and error-prone; rule-driven modeling uses predefined rules and templates to generate business process models in combination with input structured data, but this method does not support unstructured natural language descriptions sufficiently and has high requirements for users; semi-automatic modeling combines manual and automated tools for modeling. Although it improves efficiency to a certain extent, it still requires a lot of manual intervention and is difficult to meet the needs of modern enterprises for efficient and accurate business process modeling; in summary, when using existing technologies to establish process models, there are obvious problems of complex use, time-consuming, labor-intensive and error-prone.

[0004] Chinese patent, publication number: CN 118133233 A, publication date: June 4, 2024, discloses a business process mining method, device, electronic device and medium, including obtaining original session data of a customer service system and the intent scenario category corresponding to the session data; performing data cleaning processing on the original session data to obtain key data in the original session data; splicing the key data in the original session data, the intent scenario category corresponding to the original session data and the initial prompt word to obtain a process mining prompt word; inputting the process mining prompt word as input data into a large language model to obtain a first business process corresponding to the intent scenario category output by the large language model; and the invention needs to match the original session data with the preset intent scenario when performing business process modeling, so the invention needs manual intervention to set the intent scenario when performing process modeling, which leads to the process and results of the modeling being mixed with the user's subjective factors, affecting the efficiency and accuracy of the modeling. Summary of the invention

[0005] The purpose of the present invention is to address the problems of low efficiency and easy errors caused by human intervention in the process of process modeling in the prior art; a method for automatically generating a process model is proposed, which processes the original business text through a large language model to extract elements in the original business text and divide the elements into classified sets, and finally visualizes the classified sets of elements to obtain a business process diagram. The present invention overcomes the problems of low efficiency and easy errors caused by human intervention in the process of process modeling.

[0006] In a first aspect, a technical solution provided in an embodiment of the present invention is a method for automatically generating a process model, comprising the following steps: S1. Preprocess the original business text based on the large language model to obtain the standardized business text; S2, extracting elements from the standard business text and dividing the classification sets corresponding to the elements; S3, simplifying the classification set corresponding to the element and identifying the end node in the standard business text; S4. Visualize the simplified classification set based on the end node in the standard business text to obtain a business process diagram.

[0007] In this solution, the original business text is input into the large language model for preprocessing to generate a standardized business text, so that different languages ​​in the text are converted into the language required for process modeling, and the time for processing the original business text in the process modeling process is reduced, which effectively improves efficiency and provides strong support for the subsequent extraction of elements in the standardized business text; then the elements in the standardized business text are extracted and divided into classification sets corresponding to the elements, and the elements in the standardized business text are classified and counted, so that the large language model can understand what the elements in the standardized business text do respectively, and the classification sets corresponding to the elements are simplified, and the end nodes in the standardized business text are identified, so that the standardized business text is more concise and the end points of the events in the text are clarified, which effectively improves the accuracy and efficiency of process modeling; finally, the simplified classification sets are visualized based on the end nodes in the standardized business text to generate a business process diagram that is easy to understand and can be adjusted and optimized according to different needs.

[0008] Preferably, in S1, the specific process of preprocessing the original business text based on the large language model to obtain the standardized business text is: S11, using a large language model to filter redundant information in the original business text to obtain a streamlined business text; S12. Identify and mark conditional statements in the simplified business text and improve the simplified business text; S13. Based on the improved and streamlined business text, implicit information is made explicit to obtain a standardized business text.

[0009] In this solution, a large language model is used to filter redundant information in the original business text, and non-essential elements are systematically eliminated to obtain a streamlined business text, which significantly improves the clarity, accuracy and consistency of the process; then, the conditional statements in the streamlined business text are identified and marked, and the streamlined business text is improved, and the implicit information of the improved streamlined business text is made explicit to obtain a standardized business text, which effectively enhances the logical connection between text sentences and eliminates ambiguous descriptions, significantly improves the parsing efficiency of the overall process, and thus improves the efficiency of process modeling, while providing strong support for the subsequent generation of a more complete and accurate business process model.

[0010] Preferably, in S2, the specific process of extracting elements in the standard business text and dividing the classification sets corresponding to the elements is: S21. Use the natural language processing library to parse the standard business text and generate structured clauses; S22. Extract identity features and behavior features in the standardized business text based on structured clauses, and divide the classification set according to the identity features and behavior features.

[0011] In this solution, the natural language processing library is used to parse the standardized business text to accurately identify each clause, and the clauses are structured to generate structured clauses. Each clause is an independent text unit, and the clauses are independent of each other and do not affect each other. Secondly, the identity characteristics and behavioral characteristics in the standardized business text are extracted based on the clauses, and the classification sets are divided according to the identity characteristics and behavioral characteristics, so that the logic between the clauses is clearer, and at the same time, it provides strong support for subsequent steps.

[0012] Preferably, in S21, the specific process of using the natural language processing library to parse the standard business text and generate structured clauses is: S211, pre-processing the standard business text through the pre-trained Chinese model in the natural language processing library to obtain a document object; S212, performing syntactic analysis on the document object based on the dependency parser of the natural language processing library to obtain a dependency relationship tree; S213, decomposing the document object based on the dependency tree combined with the short sentence recognition function of the natural language processing library to obtain independent short sentences and subordinate short sentences; S214. Generate structured clauses based on independent clauses and dependent clauses.

[0013] In this solution, the natural language processing library has powerful text parsing and language analysis capabilities. The natural language processing library is used to parse standard business texts to obtain structured clauses, which can efficiently parse complex sentences and clearly find the structural components in the sentences, providing strong support for the subsequent classification and aggregation of divided elements.

[0014] Preferably, in S22, the specific process of extracting identity features and behavior features in the standard business text based on the structured clauses and dividing the classification set according to the identity features and behavior features is as follows: S221, extract identity features and behavior features from standardized business texts based on structured clauses combined with large language models; S222, establishing a list based on the structured clause in combination with the document object, wherein the list is divided into a process tuple and an activity tuple, and storing the identity features and behavior features in the structured clause in the process tuple; At the same time, the identity features and the behavior features are combined to obtain activities, and the activities are stored in activity tuples.

[0015] In this scheme, a list is established based on structured clauses combined with document objects, and the list is divided into process tuples and activity tuples, wherein the process tuples correspond to the identity characteristics and behavior characteristics in the structured clause components, and the activity tuples correspond to the activities. By classifying and storing the structured clause components, the efficiency of text processing in subsequent steps is effectively improved.

[0016] Preferably, in S3, the specific process of simplifying the classification set corresponding to the element is: S311, using a large language model to identify identity features in the elements, and summarizing the identity features in the same list; S312, calculating the similarity of different identity features based on the semantic similarity algorithm of identity features combined with vocabulary, merging the identity features when the identity features involved in the calculation are determined to be similar, and updating the element with the merged identity features.

[0017] In this solution, the identity features in the elements are identified, similar identity features are merged, and the elements are updated with the merged identity features, making the elements more concise and clear, achieving the effect of simplifying the element set and providing strong support for subsequent visualization processing.

[0018] Preferably, in S3, the specific process of identifying the end node in the standard business text is: The standardized business text and the simplified classification set are used as the input of the large language model for recognition and analysis. If the Boolean value of the feedback at the end of the large language model analysis is true, the behavior feature corresponding to the Boolean value is determined to be the end node. If the Boolean value of the feedback at the end of the large language model analysis is false, the behavior feature corresponding to the Boolean value is determined not to be the end node.

[0019] In this solution, the standard business text and the simplified classification set are combined with a large language model to identify the end nodes in the standard business text, clarify the end points of the events contained in the standard business text, and determine each end node during process modeling.

[0020] Preferably, in S4, the simplified classification set is visualized based on the end node in the standard business text to obtain the specific process of the business process diagram as follows: S41, generating PiperFlow syntax text based on the simplified classification set and the end node in the standard business text; S42. Use a large language model combined with standardized business text to optimize task labels in PiperFlow grammar text; S43. Generate a visual business process diagram based on the optimized PiperFlow syntax text.

[0021] In this solution, the efficiency and accuracy of generating visual business process diagrams are effectively improved by converting the simplified classification set into the PiperFlow syntax text required in the visualization processing process, and converting the simplified classification set into PiperFlow syntax text while using standardized business text for support.

[0022] Preferably, in S41, the specific process of generating the PiperFlow syntax text based on the simplified classification set and the end node in the standard business text is: S411, inputting the standardized business text and the simplified classification set into the large language model, and extracting identity features, behavior features and activities based on the simplified classification set; S412, based on the large language model combined with the standard business text, and the end node in the standard business text, a circular judgment is performed to make the behavior characteristics and activities correspond to the identity characteristics; S413. Based on the matching results of the behavioral features, activities and identity features, and in combination with the writing format of the PiperFlow grammatical text, generate the PiperFlow grammatical text.

[0023] Preferably, in S42, the specific process of optimizing the task labels in the PiperFlow grammar text using the large language model is as follows: The standard business text and PiperFlow grammar text are input into the large language model. When the large language model finds that the descriptions of multiple task labels in the input PiperFlow grammar text are for the same activity and the objects of the activities are the same, the large language model will merge the multiple task labels corresponding to the situation according to the standard business text.

[0024] The present invention has at least the following substantial effects: First, the original business text is input into the large language model for preprocessing to generate standardized business text, so that the different languages ​​in the text are converted into the language required for process modeling, and the time for processing the original business text in the process modeling is reduced, which effectively improves the efficiency and provides strong support for the subsequent extraction of elements in the standardized business text; then, the elements in the standardized business text are extracted and divided into the classification sets corresponding to the elements, and the elements in the standardized business text are classified and counted, so that the large language model can understand what the elements in the standardized business text do respectively, and simplify the classification sets corresponding to the elements, and identify the end nodes in the standardized business text, so that the standardized business text is more concise and the end points of the events in the text are clarified, which effectively improves the process The accuracy and efficiency of modeling; finally, based on the end node in the standard business text, the simplified classification set is visualized to generate a business process diagram that is easy to understand and can be adjusted and optimized according to different needs; in summary, the automatic generation method of a process model provided by the present invention does not require a lot of time and human resources, and the understanding of the original business text by the large language model is an objective understanding without subjective factors. Therefore, the present invention significantly improves the efficiency and accuracy of process modeling. At the same time, the present invention has a high degree of flexibility and adaptability in the processing of original business texts and the standard requirements of process modeling, and can be adjusted and optimized according to the specific needs of different enterprises and industries to meet the diverse business process modeling needs.

[0025] The above invention content is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Other features, objects and advantages of the present invention will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings. The drawings are only for the purpose of illustrating preferred embodiments and are not to be considered as limiting the present invention. Also, the same reference symbols are used throughout the drawings to represent the same parts.

[0027] Figure 1 A schematic diagram of a process flow of a method for automatically generating a process model according to the present invention; Figure 2 Business process diagram generated for the present invention. DETAILED DESCRIPTION

[0028] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation method described herein is only an optimal embodiment of the present invention, which is only used to explain the present invention and does not limit the scope of protection of the present invention. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0029] Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flow charts. Although the flow charts describe the operations (or steps) as sequential processes, many of the operations (or steps) therein can be implemented in parallel, concurrently, or simultaneously. In addition, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but can also have additional steps not included in the drawings; the process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0030] Embodiment 1: like Figure 1 As shown, this embodiment provides a method for automatically generating a process model, comprising the following steps: S1. Preprocess the original business text based on the large language model to obtain the standardized business text.

[0031] In one embodiment, in order to simplify the original business text, this embodiment adopts a large language model solution, which specifically includes the following steps: S11, using a large language model to filter redundant information in the original business text to obtain a streamlined business text; S12. Identify and mark conditional statements in the simplified business text and improve the simplified business text; S13. Based on the improved and streamlined business text, implicit information is made explicit to obtain a standardized business text.

[0032] In this embodiment, the original business text is input into the large language model for filtering, so that a large amount of information irrelevant to the actual process contained in the original business text can be deleted, so that the basic steps and connections required for accurately representing the process are exposed, and the simplified business text is obtained after being sorted by the large language model; secondly, since the Chinese language is complex, rich in expression, and emphasizes conciseness, the conditional relationship can usually be clearly inferred from the context, and keywords such as "if" that express conditional judgments are often omitted, so it is necessary to identify and mark the conditional statements in the simplified business text to facilitate the understanding of the logical relationship between the basic steps of the process. Specifically, A large language model is used to identify conditional statements in the text, and the word "if" is inserted at the appropriate position in the sentence to make the conditional relationship clearer, effectively enhancing the clarity of the text and providing strong support for the subsequent text parsing steps. Finally, since the original business process text usually hides the action subject, inferred actions or unexplained assumptions, which may lead to ambiguity in understanding, it is necessary to combine the large language model to carefully analyze the sentences in the text, capture explicit and implicit operations, accurately parse and understand the implicit information in the natural language text, and complete this implicit information to eliminate ambiguity, so as to facilitate the subsequent generation of a more complete and accurate business process model. The large language model described in this embodiment can select different large language models according to different actual needs. In this embodiment, the large language model specifically selected is ChatGPT.

[0033] S2. Extract elements from the standard business text and divide them into classification sets corresponding to the elements.

[0034] In one embodiment, in order to extract elements from the standardized business text and divide the elements into classification sets, this embodiment adopts a clause scheme, which specifically includes the following steps: S21. Use the natural language processing library to parse standard business texts and generate structured clauses.

[0035] In one embodiment, in order to generate structured clauses, this embodiment adopts a natural language processing library solution, which specifically includes the following steps: S211, pre-processing the standard business text through the pre-trained Chinese model in the natural language processing library to obtain a document object; S212, performing syntactic analysis on the document object based on the dependency parser of the natural language processing library to obtain a dependency relationship tree; S213, decomposing the document object based on the dependency tree combined with the short sentence recognition function of the natural language processing library to obtain independent short sentences and subordinate short sentences; S214. Generate structured clauses based on independent clauses and dependent clauses.

[0036] In this embodiment, the natural language processing library is an open source advanced NLP library, the spaCy library, which has powerful text parsing and language analysis capabilities and is widely used in various NLP tasks; first, the pre-trained Chinese language model is loaded through the spaCy library to pre-process the standard business text, including segmenting the standard business text into words and punctuation marks, and marking the part of speech of each word. After the above steps are completed, a document object is generated, wherein the Chinese language model is the zh_core_web_trf framework, and the document object is the Doc object; the Doc object is one of the core data structures of the spaCy library, which is a container of text and contains all processed information such as every word, sentence boundary, dependency, named entity, etc. of the text; then, the dependency parser of the spaCy library is used to perform syntactic analysis on the pre-processed Doc object, and the dependency parser can generate a dependency tree between each word , clearly point out the syntactic components such as the subject, predicate, and object in the sentence; after the dependency tree is established, the short sentence recognition function of the spaCy library is used to decompose the independent short sentences and subordinate short sentences in the sentences contained in the Doc object, where an independent short sentence is a sentence part that can express a complete idea or meaning. It can exist independently and can form a complete sentence without relying on other short sentences. The subordinate short sentence is a sentence part that cannot form an independent sentence. It relies on the main sentence to provide context or complete the expression of meaning; In summary, based on the dependency tree combined with the short sentence recognition function of the spaCy library, and by analyzing the dependency relationship and conjunctions, the sentences in the Doc object can be decomposed to obtain independent short sentences and subordinate short sentences, and finally the independent short sentences and subordinate short sentences are extracted to generate structured clauses; the structured clauses exist in the form of a list, and each clause contained therein is an independent text unit, which can be further processed separately.

[0037] S22. Extract identity features and behavior features in the standardized business text based on structured clauses, and divide the classification set according to the identity features and behavior features.

[0038] In one embodiment, in order to divide the classification set according to participants and actions, this embodiment adopts a list solution, which specifically includes the following steps: S221, extract identity features and behavior features from standardized business texts based on structured clauses combined with large language models; S222, establishing a list based on the structured clause in combination with the document object, wherein the list is divided into a process tuple and an activity tuple, and storing the identity features and behavior features in the structured clause in the process tuple; At the same time, the identity features and the behavior features are combined to obtain activities, and the activities are stored in activity tuples.

[0039] In this embodiment, based on the structured clause combined with the large language model, elements in the standard business text are extracted, specifically identity features and behavior features, and the identity features and behavior features are combined to obtain an activity, wherein the identity feature is a participant, and the behavior feature is an action; at the same time, a Doc object with vocabulary is input to generate a list with various tags, the list contains a process tuple and an activity tuple, and the identity features and behavior features in the structured clause are stored in the process tuple, and the activity is stored in the activity tuple; corresponding to the various tags in the list, the vocabulary is Token, the list is a container list, the process tuple is a process class, the activity tuple is an activity class, the participant corresponds to the actor class, the action corresponds to the action class, and the activity is the activity class; specifically, a container class tag is created for each sentence in the list, and each clause is stored in a process class, and the identity features and behavior features in the clause correspond to the actor class and the action class stored in the process class respectively; after the above steps are completed, an activity structure is defined, and an activity structure corresponds to a process class, and the participants and actions in the actor class and the action class are combined to form an activity.

[0040] S3. Simplify the classification set corresponding to the element and identify the end node in the standard business text.

[0041] In one embodiment, in order to simplify the classification set of elements, this embodiment adopts a similarity algorithm solution, which specifically includes the following steps: S311, using a large language model to identify identity features in the elements, and summarizing the identity features in the same list; S312, calculating the similarity of different identity features based on the semantic similarity algorithm of identity features combined with vocabulary, merging the identity features when the identity features involved in the calculation are determined to be similar, and updating the element with the merged identity features.

[0042] In this embodiment, since the same object may have different names in the standard business text, resulting in multiple identity features in the element being actually the same object, a semantic similarity evaluation algorithm based on vocabulary tokens is used to analyze and compare the similarities of different identity features, and when identity features are determined to be similar, they are merged; specifically, the similarity of the identity features is evaluated, including: first, using a large language model to identify the identity features in the element, and summarizing the identity features in the same list to determine the actual action issuer in the element, and then randomly selecting two identity feature strings to input into a function, which is processed through a predefined natural language processing pipeline to generate each identity feature. The Doc object from which the function and the predefined natural language processing pipeline can be adjusted according to different actual situations; then, the function compares the standard form lemmas of the vocabulary tokens derived from the two identity features in pairs, specifically, first calculates the number of identical elements in the lemmas, and then divides it by the average number of lemmas to obtain the similarity of the two identity features, and the similarity is a number from 0 to 1. When the calculated similarity is greater than or equal to 0.95, it is determined that the corresponding two real participants are similar. At this time, the large language model combines the two identity features with the standard business text, and updates the element with the merged identity feature.

[0043] In one embodiment, in order to identify the ending action in the standard business text, this embodiment adopts a large language model analysis solution, which specifically includes: The standardized business text and the simplified classification set are used as the input of the large language model for recognition analysis. If the Boolean value of the feedback at the end of the large language model analysis is true, the behavior feature corresponding to the Boolean value is determined to be the end node. If the Boolean value of the feedback at the end of the large language model analysis is false, the behavior feature corresponding to the Boolean value is determined to be the end node.

[0044] In this solution, the standardized business text is input into the large language model for understanding to identify potential final activities; at the same time, the simplified classification set is input into the large language model for identification and analysis. The main thing is to input the activities in the simplified classification set. When the large language model completes the analysis, it will feedback a Boolean value. If the Boolean value is true, the large language model determines that the behavior feature corresponding to the Boolean value marks the end of the process. If the Boolean value is false, the large language model determines that the behavior feature corresponding to the Boolean value is an action in the process.

[0045] S4. Visualize the simplified classification set based on the end node in the standard business text to obtain a business process diagram.

[0046] In one embodiment, in order to visualize the simplified classification set, this embodiment adopts the PiperFlow grammar text solution, which specifically includes the following steps: S41. Generate PiperFlow syntax text based on the simplified classification set and the end node in the standard business text.

[0047] In a real-time example, in order to convert the simplified classification set into PiperFlow grammar text, this embodiment adopts a loop judgment scheme, which specifically includes the following steps: S411, inputting the standardized business text and the simplified classification set into the large language model, and extracting identity features, behavior features and activities based on the simplified classification set; S412, based on the large language model combined with the standard business text, and the end node in the standard business text, a circular judgment is performed to make the behavior characteristics and activities correspond to the identity characteristics; S413. Based on the matching results of the behavioral features and the activities and identity features, and in combination with the writing format of the PiperFlow grammatical text, generate the PiperFlow grammatical text.

[0048] In this embodiment, a large language model is used to perform cyclic judgment on the identity features, behavioral features and activities in the simplified classification set according to the logic between the contexts in the standard business text, and the identity features are used as the channels in the PiperFlow grammar text to combine activities and activity structures to obtain the matching results of the behavioral features and the channel-identity features, as well as the connection order and connection method of the identity features and the behavioral features. After the judgment is completed, the PiperFlow grammar text is generated based on the judgment result combined with the writing format of the PiperFlow grammar text.

[0049] S42. Use a large language model combined with standardized business text to optimize task labels in PiperFlow grammatical text.

[0050] In one embodiment, in order to generate a clearer flowchart, this embodiment adopts a task label optimization solution, which specifically includes: The standard business text and PiperFlow grammar text are input into the large language model. When the large language model finds that the descriptions of multiple task labels in the input PiperFlow grammar text are for the same activity and the objects of the activities are the same, the large language model will merge the multiple task labels corresponding to the situation according to the standard business text.

[0051] In this embodiment, when the large language model determines that the activities described by at least two task tags are the same activity, and the activities described by the two task tags involve the same object, the large language model will optimize and merge the two task tags according to the logic in the standard business text.

[0052] S43. Generate a visual business process diagram based on the optimized PiperFlow syntax text.

[0053] In this embodiment, the rendering function of an open source Python library, Processpiper library, is used to render the PiperFlow syntax text, and a clear and accurate business process diagram is generated after the rendering is completed.

[0054] Embodiment 2: In this embodiment, an original business text is input to verify the automatic generation method of a process model proposed by the present invention, which specifically includes: A1. Preprocess the original business text based on the large language model to obtain the standardized business text.

[0055] In this embodiment, the input original business text is: "Each unit initiates a service request. In an emergency, each unit can directly contact the third-line operation and maintenance manufacturer. The third-line operation and maintenance manufacturer directly provides operation and maintenance services. Each unit and the third-line operation and maintenance manufacturer need to report to the third-line operator; in a non-emergency situation, each unit initiates a service request to the third-line technical support center operator, and the third-line operator accepts the service and dispatches a work order; the third-line engineer accepts the dispatch response request, understands the situation and determines the support type. If the support type is determined to be telephone support, the third-line engineer provides solutions to each unit via telephone and email; each unit handles the problem with reference to the solution and checks whether the problem has been solved. If the problem has been solved, each unit confirms that the support service has been completed; the third-line engineer closes the work order, and the third-line operator conducts a satisfaction survey; if the support type is determined to be a remote access solution, the third-line engineer remotely accesses the solution to solve the problem, and each unit cooperates with and monitors the remote access solution. and check whether the problem has been solved. If it has been solved, the support service is confirmed to be completed. If on-site support or remote access does not solve the problem, the third-line engineer coordinates the manufacturer to provide on-site support, and each unit cooperates and monitors the on-site processing to confirm that the support service is completed. When the above original business text is input into the big language model, the big language model recognizes the conditional sentences in the text, including "Is it an emergency situation", "Support type", and "Is the problem solved". Among them, the conditional sentence "Is it an emergency situation" is not obvious. The big language model marks it by adding the word "if". The marked sentence is "If it is an emergency situation, each unit can directly contact the third-line operation and maintenance manufacturer, and the third-line operation and maintenance manufacturer directly provides operation and maintenance services. Each unit and the third-line operation and maintenance manufacturer need to report to the third-line operator; if it is a non-emergency situation, each unit initiates a service request to the operator of the third-line technical support center, and the third-line operator accepts the service and dispatches a work order."

[0056] A2. Use NLP to extract elements from standard business texts and divide them into classification sets corresponding to the elements.

[0057] In this embodiment, NLP first divides the entire text into 6 clauses, namely: "Each unit initiates a service request. If it is an emergency situation, each unit can directly contact the third-line operation and maintenance manufacturer. The third-line operation and maintenance manufacturer directly provides operation and maintenance services. Each unit and the third-line operation and maintenance manufacturer report to the third-line operator;" "If it is a non-emergency situation, each unit initiates a service request to the third-line technical support center operator. The third-line operator accepts the service and dispatches a work order; the third-line engineer accepts the dispatch response request, understands the situation and determines the support type;" "If the support type is determined to be telephone support, the third-line engineer will provide solutions to each unit via telephone and email; each unit will refer to the solution to handle the problem and check whether the problem has been solved;" "If the problem has been solved, each unit confirms the support service Completed; the third-line engineer closes the work order, and the third-line operator conducts a satisfaction survey;", "If the support type is determined to be a remote access solution, the third-line engineer remotely accesses to solve the problem, and each unit cooperates and monitors the remote access processing, and checks whether the problem has been solved. If it has been solved, it is confirmed that the support service is completed;", "If the support type is determined to be on-site support, or remote access does not solve the problem, the third-line engineer coordinates the manufacturer to provide on-site support, and each unit cooperates and monitors the on-site processing to confirm that the support service is completed;" Then extract the elements in the text based on the divided clauses, taking the participants as examples, including "each unit", "third-line operation and maintenance manufacturer", "third-line operator", "third-line technical support center operator" and "third-line engineer"; and divide the corresponding classification set based on this.

[0058] A3. Simplify the classification set corresponding to the element and identify the end action in the standard business text.

[0059] In this embodiment, the specific operation of simplifying the classification set can be understood as: the large language model determines that the "third-line operator" and the "third-line technical support center operator" in the participant element are the same entity, but the "third-line operator" appears more times in the text, so the above participants are merged into "third-line operator"; after the simplification of the classification set is completed, the large language model identifies the end action through context understanding in the standardized business text. For example, if the large language model determines that "each unit confirms that the support service has been completed; the third-line engineer closes the work order, and the third-line operator conducts a satisfaction survey;" is an end event, then the three actions of "confirming that the support service has been completed", "closing the work order", and "conducting a satisfaction survey" are further marked as end actions.

[0060] A4. Visualize the simplified classification set based on the ending action in the standard business text to obtain a business process diagram.

[0061] In this embodiment, before visualizing the simplified classification set, the simplified classification set needs to be converted into a PiperFlow grammar text as shown below: title: Flowchart of processing of third-line service request; width:10000; colourtheme:BLUEMOUNTAIN; Lane: each unit; (start)as start; [Initiate service request] as activity_1; <Is it an emergency situation? >as gateway_1; [Contact third-line operation and maintenance vendors]as activity_2; [Initiate service request to the third line operator] as activity_5; [Refer to the solution to solve the problem] as activity_10; <Has the problem been solved? >as gateway_3; [Confirm that support services have been completed] as activity_11; [Cooperation and monitoring of remote access processing] as activity_15; [Cooperation and supervision on-site processing] as activity_17; lane: a third-tier operation and maintenance vendor; [Provide operation and maintenance services] as activity_3; [Report to the third line operator]as activity_4; lane: third line operator; [Service acceptance and dispatch of work orders] as activity_6; [Conduct a satisfaction survey]as activity_13; (end)as end; lane: third-line engineer; [Accept work assignment response request] as activity_7; [Understand the situation and determine the type of support] as activity_8; <Support type?> as gateway_2; [Provide solutions] as activity_9; [Remote access to solve problems] as activity_14; [Coordinate with manufacturers to provide on-site support] as activity_16; [Close the work order] as activity_12; start->activity_1->gateway_1; gateway_1-"is"->activity_2->activity_3->activity_4->end; gateway_1-"No"->activity_5->activity_6->activity_7->activity_8->gateway_2; gateway_2-"Phone support"->activity_9->activity_10->gateway_3; gateway_2-"remote access"->activity_14->activity_15->gateway_3; gateway_2-"Field support"->activity_16->activity_17->activity_11; gateway_3-"is"->activity_11->activity_12->activity_13->end; gateway_3-"No"->activity_16; Then use the rendering function of the Processpiper library to render the PiperFlow syntax text, generating Figure 2 The business process diagram shown can clearly and intuitively understand the process.

[0062] The embodiments of the present invention have at least the following technical effects: The embodiment of the present invention provides an automatic generation method of a process model, which is mainly divided into three modules; Module 1 is to pre-process the original business text described in natural language through a large language model to filter out unnecessary information in the original business text, expose the elements that are really needed, and improve the efficiency of process modeling while improving the accuracy of process modeling by enhancing the logic between text sentences; Module 2 is to divide clauses through the open source advanced NLP library-spaCy library to extract identity features, behavioral features, activities and conditional statements in the text description process, and then identify and merge real participation and end nodes through regularized NLP steps to make the output element classification set reach the optimal and simplest state, and further improve the efficiency and accuracy of process modeling with the help of processing text elements; In module 3, the simplified element classification set is first converted into PiperFlow syntax text, and then the PiperFlow syntax text is rendered using the rendering function of the Processpiper library to generate a simple, clear and intuitive business process diagram that reflects the business process described by the original business text.

[0063] The above specific embodiments are preferred embodiments of the present invention, and are not intended to limit the specific implementation scope of the present invention. The scope of the present invention includes but is not limited to the present specific embodiments. All equivalent changes made in accordance with the shape, structure, and method of the present invention are within the protection scope of the present invention.

Claims

1. A method for automatically generating a process model, characterized in that: The method comprises the following steps: S1. Preprocess the original business text based on the large language model to obtain the standardized business text; S2, extracting elements from the standard business text and dividing the classification sets corresponding to the elements; S3, simplifying the classification set corresponding to the element and identifying the end node in the standard business text; S4. Visualize the simplified classification set based on the end node in the standard business text to obtain a business process diagram.

2. The method for automatically generating a process model according to claim 1, characterized in that: In S1, the specific process of preprocessing the original business text based on the large language model to obtain the standardized business text is as follows: S11, using a large language model to filter redundant information in the original business text to obtain a streamlined business text; S12. Identify and mark conditional statements in the simplified business text and improve the simplified business text; S13. Based on the improved and streamlined business text, implicit information is made explicit to obtain a standardized business text.

3. The method for automatically generating a process model according to claim 1, characterized in that: In S2, the specific process of extracting elements in the standard business text and dividing the classification sets corresponding to the elements is: S21. Use the natural language processing library to parse the standard business text and generate structured clauses; S22. Extract identity features and behavior features in the standardized business text based on structured clauses, and divide the classification set according to the identity features and behavior features.

4. The method for automatically generating a process model according to claim 3, characterized in that: In S21, the specific process of using the natural language processing library to parse the standard business text and generate structured clauses is as follows: S211, preprocessing the standard business text through the pre-trained Chinese model in the natural language processing library to obtain a document object; S212, performing syntactic analysis on the document object based on a dependency parser of a natural language processing library to obtain a dependency tree; S213, decomposing the document object based on the dependency tree combined with the short sentence recognition function of the natural language processing library to obtain independent short sentences and subordinate short sentences; S214. Generate structured clauses based on independent clauses and dependent clauses.

5. The method for automatically generating a process model according to claim 4, characterized in that: In S22, the specific process of extracting identity features and behavior features in the standard business text based on the structured clauses and dividing the classification set according to the identity features and behavior features is as follows: S221, extract identity features and behavior features from standardized business texts based on structured clauses combined with large language models; S222, establishing a list based on the structured clause in combination with the document object, wherein the list is divided into a process tuple and an activity tuple, and storing the identity features and behavior features in the structured clause in the process tuple; At the same time, the identity features and the behavior features are combined to obtain activities, and the activities are stored in activity tuples.

6. The method for automatically generating a process model according to claim 1, characterized in that: In S3, the specific process of simplifying the classification set corresponding to the element is: S311, using a large language model to identify identity features in the elements, and summarizing the identity features in the same list; S312, calculating the similarity of different identity features based on the semantic similarity algorithm of identity features combined with vocabulary, merging the identity features when the identity features involved in the calculation are determined to be similar, and updating the element with the merged identity features.

7. The method for automatically generating a process model according to claim 1, characterized in that: In S3, the specific process of identifying the end node in the standard business text is: The standardized business text and the simplified classification set are used as the input of the large language model for recognition and analysis. If the Boolean value of the feedback at the end of the large language model analysis is true, the behavior feature corresponding to the Boolean value is determined to be the end node. If the Boolean value of the feedback at the end of the large language model analysis is false, the behavior feature corresponding to the Boolean value is determined not to be the end node.

8. The method for automatically generating a process model according to claim 1, characterized in that: In S4, the simplified classification set is visualized based on the end node in the standard business text to obtain the specific process of the business process diagram as follows: S41, generating PiperFlow syntax text based on the simplified classification set and the end node in the standard business text; S42. Use a large language model combined with standardized business text to optimize PiperFlow grammar text; S43. Generate a visual business process diagram based on the optimized PiperFlow syntax text.

9. The method for automatically generating a process model according to claim 8, characterized in that: In S41, the specific process of generating the PiperFlow syntax text based on the simplified classification set and the end node in the standard business text is: S411, inputting the standardized business text and the simplified classification set into the large language model, and extracting identity features, behavior features and activities based on the simplified classification set; S412, based on the large language model combined with the standard business text, and the end node in the standard business text, a circular judgment is performed to make the behavior characteristics and activities correspond to the identity characteristics; S413. Based on the matching results of the behavioral features, activities and identity features, and in combination with the writing format of the PiperFlow grammatical text, generate the PiperFlow grammatical text.

10. The method for automatically generating a process model according to claim 8, characterized in that: In S42, the specific process of optimizing the task labels in the PiperFlow grammar text using the large language model is as follows: The standard business text and PiperFlow grammar text are input into the large language model. When the large language model finds that the descriptions of multiple task labels in the input PiperFlow grammar text are for the same activity and the objects of the activities are the same, the large language model will merge the multiple task labels corresponding to the situation according to the standard business text.

Images