Standard operation process extraction method and equipment based on large language model

Abstract

The invention discloses a standard operation process extraction method and equipment based on a large language model, and relates to the technical field of computers. The method comprises the following steps: analyzing operation documents corresponding to a standard operation process extraction task, segmenting each operation document into at least one fragment, and determining fragment index information corresponding to each fragment; for each fragment, determining a structuring step corresponding to the fragment based on an extraction result of structuring extraction of the fragment by a large language model; and according to the fragment index information corresponding to each fragment, combining the structured steps corresponding to each fragment to obtain a structured process corresponding to the standard operation process extraction task. According to the technical scheme, through the technologies of text fragmentation, structured extraction, structured aggregation and the like, the problems of low extraction efficiency and low accuracy of the structured process are effectively solved, and automatic and intelligent extraction of the structured process is realized.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method and device for extracting standard operating procedures based on a large language model. Background Technology

[0002] In the field of equipment maintenance, Standard Operating Procedures (SOPs) are crucial for ensuring maintenance quality and safety. Therefore, extracting corresponding SOPs from equipment maintenance manuals has become an urgent issue.

[0003] Traditional information processing methods cannot automatically extract directly usable Standard Operating Procedures (SOPs) from equipment maintenance manuals. Usually, it is necessary to manually read the equipment maintenance manuals and manually compile the SOPs. This method is not only time-consuming and labor-intensive, but also highly dependent on personnel experience, resulting in poor SOP extraction efficiency and low accuracy of the obtained SOPs. Summary of the Invention

[0004] This invention provides a method and device for extracting standard operating procedures based on a large language model, in order to solve the problems of poor efficiency and inaccuracy in extracting structured procedures.

[0005] According to one aspect of the present invention, a method for extracting standard operation procedures based on a large language model is provided, comprising: By parsing the operation documents corresponding to the task extracted from the standard operation process, each operation document is divided into at least one segment, and the segment index information corresponding to each segment is determined. For each of the aforementioned segments, the structured extraction results based on the large language model are used to determine the structured steps corresponding to the segment. According to the shard index information corresponding to each shard, the structured steps corresponding to each shard are merged to obtain the structured process corresponding to the standard operation process extraction task.

[0006] Furthermore, by parsing the operation documents corresponding to the tasks extracted from the standard operating procedures, each operation document is divided into at least one fragment, and the fragment index information corresponding to each fragment is determined, including: For each of the operation documents, the operation document is divided into segments according to natural paragraphs to obtain the segments corresponding to the operation document; For each of the aforementioned shards, the shard index information corresponding to the shard is determined based on the operation document to which the shard belongs and its order within the operation document.

[0007] Further, the operation document is segmented according to natural paragraphs to obtain the segment corresponding to the operation document, including: Based on a preset context window threshold, the operation document is segmented according to natural paragraphs to obtain the segment corresponding to the operation document.

[0008] Furthermore, before performing structured extraction of the fragment according to the extraction rules corresponding to the operation document to which the fragment belongs, the method further includes: For each of the operation documents, the extraction rule corresponding to the operation document is determined based on the selection request triggered by the user for the operation document, and / or the custom rule input by the user for the operation document.

[0009] Furthermore, the structure extraction result based on the large language model to determine the structure corresponding to the segment includes: Extraction prompts are constructed based on the fragments, the fragment index information, structural constraints, and the extraction rules corresponding to the operation documents to which the fragments belong. The structuring step is determined by inputting the extracted prompt words into a large language model and then analyzing the extraction results output by the large language model.

[0010] Further, the structuring step is determined based on the extraction results output by the large language model, including: Based on the judgment criteria of the standard structured steps and the extraction results, judgment prompt words are constructed; The validity of the extraction result is determined by inputting the judgment prompt words into a large language model and judging the judgment result output by the large language model. When the extraction result is determined to be invalid, the extraction status corresponding to the segment is set to re-extraction, and the segment with the extraction status set to re-extraction is re-extracted in a structured manner until the structured step corresponding to the segment is obtained.

[0011] Further, according to the shard index information corresponding to each shard, the structured steps corresponding to each shard are merged to obtain the structured process corresponding to the standard operation process extraction task, including: Based on the fragment index information corresponding to each fragment, determine the connection relationship between the structured steps corresponding to each fragment, wherein the connection relationship is a combination or nesting; Based on the connection relationship between the structured steps corresponding to each segment, the structured steps corresponding to each segment are combined or nested to obtain the structured process corresponding to the standard operation process extraction task.

[0012] Furthermore, it also includes: The extraction progress of the operation document to which each segment belongs is determined based on the extraction progress of each segment; The standard operating procedure extraction progress is determined based on the extraction progress of each of the aforementioned operation documents. A progress report is constructed based on the extraction progress of each segment, the extraction progress of each operation document, and the extraction progress of the standard operation process.

[0013] Furthermore, after obtaining the structured process corresponding to the standard operating procedure extraction task, the method further includes: The process text corresponding to the structured process is vectorized to obtain a vectorized process; The vectorized process is written into the process vector library according to the classification method of the structured process, wherein the process vector library is used to search for standard operation processes.

[0014] According to another aspect of the present invention, a standard operating procedure extraction apparatus is provided, the apparatus comprising: The parsing module is used to parse the operation documents corresponding to the tasks extracted from the standard operation process, divide each operation document into at least one segment, and determine the segment index information corresponding to each segment. The extraction module is used to determine the structured steps corresponding to each segment based on the extraction results of the structured extraction of the segment using a large language model. The aggregation module is used to merge the structured steps corresponding to each of the shards according to the shard index information corresponding to each shard, so as to obtain the structured process corresponding to the standard operation process extraction task.

[0015] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and a memory communicatively connected to said at least one processor; The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform the standard operating procedure extraction method described in any one of the first aspects.

[0016] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the standard operation flow extraction method described in any embodiment of the present invention.

[0017] The technical solution of this invention firstly parses the operation documents corresponding to the standard operation process extraction task to divide each operation document into at least one segment, thereby achieving the segmentation of the operation document. At the same time, it determines the segment index information corresponding to each segment, and achieves accurate positioning of each segment through the segment index information corresponding to each segment. Then, it can perform structured extraction on each segment corresponding to each operation document to determine the structured steps corresponding to each segment. Furthermore, it can merge the structured steps corresponding to each segment with reference to the segment index information corresponding to each segment to obtain a structured process with reasonable sequence and nesting relationship, thereby realizing automatic extraction of the structured process from the operation document.

[0018] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of a standard operating procedure extraction method provided in Embodiment 1; Figure 2 This is a flowchart of a standard operating procedure extraction method provided in Embodiment 2; Figure 3 This is a schematic diagram of the structure of a standard operating procedure extraction device provided in Embodiment 3; Figure 4 This is a schematic diagram of a standard operating procedure extraction system provided in Embodiment 4. Figure 5a Create an interface for the task. Figure 5b Configure the rules interface. Figure 5c For task management interface, Figure 5d Provides a structured workflow display interface; Figure 6 This is a sequence diagram for extracting standard operating procedures (SOPs) based on the SOP extraction system. Figure 7 This is a schematic diagram of the structure of an electronic device provided in Embodiment 5 of the present invention. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0022] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0023] In existing technologies, due to misunderstandings among different personnel, the procedures compiled from equipment maintenance manuals vary, and problems such as missing steps, disordered sequences, or inconsistencies with the equipment maintenance manual are prone to occur. Once the equipment manufacturer updates the equipment maintenance manual, the Standard Operating Procedures (SOPs) need to be manually revised again to maintain their accuracy and consistency. This results in frequent repetitive work and high maintenance costs.

[0024] Therefore, this application proposes a standard operating procedure extraction method that automatically extracts SOPs from equipment operation manuals based on explicit, standardized, and automated steps.

[0025] The standard operating procedure extraction method proposed in this application will be described in detail below with reference to the illustrations and embodiments.

[0026] Example 1 Figure 1 This is a flowchart illustrating a standard operating procedure extraction method provided in Embodiment 1. This embodiment is applicable to situations requiring automatic generation of standard operating procedures. The method can be executed by a standard operating procedure extraction device, which can be implemented in hardware and / or software and can be configured in a computer device. Figure 1 As shown, the method includes: Step 110: Parse the operation documents corresponding to the tasks extracted from the standard operation process, divide each operation document into at least one segment, and determine the segment index information corresponding to each segment.

[0027] The standard operating procedure (SOP) extraction task is triggered by the user based on actual needs. For example, when it is necessary to determine the standard operating procedure required for refrigerator malfunction repair, a SOP extraction task for refrigerator malfunction repair can be triggered. When triggering the SOP extraction task, the user needs to upload at least one required operating document. For example, when the SOP extraction task is for refrigerator malfunction repair, the operating document can be a refrigerator repair manual.

[0028] Specifically, after receiving the user-triggered standard operation process extraction task and task start request, the corresponding operation document for the standard operation process extraction task can be determined. That is, the document uploaded synchronously when the user triggers the standard operation process extraction task can be identified as the operation document. Then, each operation document can be parsed. Since operation documents typically contain a large amount of text information, to improve data processing efficiency, each operation document can be pre-sharded, i.e., each operation document can be divided into at least one shard. Specifically, each operation document can be divided into a corresponding number of shards based on the pre-set total number of shards for each operation document. Since the standard operation process extraction task corresponds to at least one operation document, to ensure accurate shard location, the operation document to which the shard belongs and its order within the belonging operation document can be determined as the shard index information corresponding to the shard.

[0029] It should be noted that various operation documents can be parsed simultaneously, improving parsing efficiency.

[0030] In this embodiment of the invention, the operation documents corresponding to the standard operation process extraction task are parsed and each operation document is divided into at least one segment to achieve the segmentation of the operation documents. At the same time, the segment index information corresponding to each segment is determined to achieve the accurate positioning of each segment.

[0031] Step 120: For each of the segments, determine the structured steps corresponding to the segment based on the extraction results of the structured extraction of the segment using the large language model.

[0032] For each operation document, when uploading the operation document, the user needs to determine the extraction rules corresponding to the operation document. The extraction rules corresponding to the operation document can be determined based on the selection request triggered by the user, such as global extraction rules, or they can be determined based on the rule information entered by the user, such as special extraction rules for the operation document.

[0033] Specifically, the segments can be extracted in a structured manner based on the extraction rules corresponding to the operation document to which the segments belong. More specifically, the segments can be input into a large language model, and the model can be used to extract the structured steps corresponding to each segment. For example, the large language model can extract the structured steps corresponding to each segment based on the extraction prompts derived from the segments.

[0034] It should be noted that structured extraction can be performed on each segment simultaneously, improving extraction efficiency.

[0035] In this embodiment of the invention, the structured steps corresponding to each segment are determined by extracting the structured data from each segment corresponding to each operation document.

[0036] Step 130: According to the shard index information corresponding to each shard, merge the structured steps corresponding to each shard to obtain the structured process corresponding to the standard operation process extraction task.

[0037] The shard index information may include the operation document to which the shard belongs and its order within the operation document.

[0038] Specifically, after determining the structured steps corresponding to each segment, the segment index information corresponding to each segment can be determined, which can determine the operation document to which each segment belongs and its order in the operation document. Then, the structured steps corresponding to each segment can be merged according to the operation document to which each segment belongs and its order in the operation document to obtain at least one structured process. The at least one structured process obtained by merging is the structured process corresponding to the standard operation process extraction task.

[0039] In this embodiment of the invention, the structured steps corresponding to each segment are merged with reference to the segment index information corresponding to each segment to obtain a structured process with reasonable sequence and nesting relationship, thereby realizing automatic extraction of the structured process of the operation document.

[0040] The standard operation process extraction method provided in this embodiment includes: parsing the operation documents corresponding to the standard operation process extraction task, dividing each operation document into at least one segment, and determining the segment index information corresponding to each segment; for each segment, determining the structured steps corresponding to the segment based on the extraction result of the structured extraction of the segment using a large language model; and merging the structured steps corresponding to each segment according to the segment index information to obtain the structured process corresponding to the standard operation process extraction task. The above technical solution first parses the operation documents corresponding to the standard operation process extraction task to divide each operation document into at least one segment, thereby achieving document segmentation. Simultaneously, it determines the segment index information corresponding to each segment, enabling precise location of each segment. Then, it performs structured extraction on each segment corresponding to each operation document to determine the structured steps for each segment. Furthermore, it merges the structured steps corresponding to each segment with reference to the segment index information, resulting in a structured process with a reasonable sequence and nesting relationship. This achieves highly accurate automatic extraction of structured processes corresponding to operation documents, improving the efficiency and accuracy of structured process extraction.

[0041] Example 2 Figure 2 This is a flowchart illustrating a standard operating procedure extraction method provided in Embodiment 2. This embodiment is a specific implementation based on the above embodiments. Figure 2 As shown, the method includes: Step 210: For each operation document corresponding to the standard operation process extraction task, the operation document is segmented according to natural paragraphs to obtain the segment corresponding to the operation document.

[0042] The operation document can be in PDF or Word format. When the operation document is in PDF format, it first needs to be recognized based on Optical Character Recognition (OCR). By extracting the text in the operation document into a machine-processable string, the operation document is converted into editable text.

[0043] Of course, OCR has difficulty restoring the original structure of text, such as headings, paragraphs, lists, and hierarchical relationships. Therefore, after converting the document into editable text using OCR, the layout structure of the editable text can be restored based on the layout analysis results of the document.

[0044] Specifically, after receiving the user-triggered standard operation process extraction task and task start request, the document synchronously uploaded when the user triggers the standard operation process extraction task can be identified as the operation document. Then, each operation document can be segmented into at least one segment.

[0045] In addition, the page where users trigger the standard operation process to extract tasks includes multiple file upload controls. Next to each file upload control is a selection control for choosing extraction rules. Users can select the appropriate extraction rules for the corresponding operation document based on the selection control. Each file upload control also has an input box next to it, allowing users to enter personalized extraction rules specifically for the operation document corresponding to that file upload control.

[0046] Therefore, after determining the operation document, the method further includes: for each operation document, determining the extraction rule corresponding to the operation document based on the selection request triggered by the user for the operation document, and / or the custom rule input by the user for the operation document.

[0047] Specifically, for each operation document, it can first be determined whether the operation document corresponds to a selection request and a custom rule. When it is determined that the operation document only corresponds to a selection request, the selected rule contained in the selection request is determined as the extraction rule corresponding to the operation document. For example, the selection rule can be a global rule. When it is determined that the operation document only corresponds to a custom rule, the custom rule is usually a feature extraction rule. For example, the custom rule can be SOP (Simple and Professional). Therefore, the custom rule can be determined as the extraction rule corresponding to the operation document. When it is determined that the operation document corresponds to both a selection request and a custom rule, the custom rule is usually a supplementary rule. Therefore, the extraction rule corresponding to the operation document can be determined according to the selected rule and the custom rule corresponding to the selection request, thereby realizing the automated determination of the extraction rule corresponding to the operation document.

[0048] In one embodiment, step 210 may specifically include: Based on a preset context window threshold, the operation document is segmented according to natural paragraphs to obtain the segment corresponding to the operation document.

[0049] Specifically, in order to preserve the semantics of each segment, a general parser can be used to segment the operation document according to natural paragraphs based on a preset context window threshold. That is, while ensuring that the number of texts in a segment does not exceed the preset context window threshold, the operation document is segmented according to natural paragraphs to obtain the segment corresponding to the operation document.

[0050] When the text length of a segment reaches a preset context window threshold, it is not truncated immediately. Instead, it searches forward or backward for the nearest natural paragraph end character (e.g., period, newline character) to ensure that a maintenance step is not cut off in two segments.

[0051] In this embodiment of the invention, the operation documents corresponding to the standard operation process extraction task are parsed, and each operation document is divided into at least one segment, thereby achieving the segmentation of the operation documents. Furthermore, based on a preset context window threshold and natural paragraphs, the segmentation of each operation document is achieved while ensuring contextual continuity, enabling smooth segmentation of wide-format, multi-page operation documents.

[0052] Step 220: For each of the fragments, determine the fragment index information corresponding to the fragment based on the operation document to which the fragment belongs and its order in the operation document.

[0053] Specifically, while dividing each operation document into at least one fragment, the fragment index information of each fragment can be determined. That is, the operation document to which the fragment belongs and its order in the operation document can be determined as the fragment index information corresponding to the fragment.

[0054] In this embodiment of the invention, precise positioning of each fragment is achieved by determining the fragment index information corresponding to each fragment.

[0055] Step 230: For each of the segments, determine the structured steps corresponding to the segment based on the extraction results of the structured extraction of the segment using the large language model.

[0056] In one implementation, step 230 may specifically include: Extraction prompts are constructed based on the fragments, fragment index information, structural constraints, and extraction rules corresponding to the operation documents to which the fragments belong; the extraction prompts are input into a large language model, and the structuring step is determined based on the extraction results output by the large language model.

[0057] Specifically, structured extraction can be performed on each segment. First, extraction prompts can be constructed based on the segment, segment index information, structural constraints, and extraction rules. The role definition of the extraction prompts can be "workstation maintenance expert," and the structural constraints can be that the output must conform to the schema definition in JSON format, including faultName and stepList (nested is supported). Extraction prompts can also include contextual information, which can be reflected through the segment content and segment index information. The segment index information can include the operation document to which the segment belongs and the order of the segment in the operation document. The order of the segment in the operation document can be reflected by {chunkIndex} / {totalChunks}, where chunkIndex represents the segment identifier and totalChunks represents the total number of segments. Then, the extraction prompts can be input into a large language model, which can output the corresponding extraction results. The structured steps can then be determined based on the extraction results output by the large language model.

[0058] Further, the structuring step is determined based on the extraction results output by the large language model, including: Based on the judgment criteria of the standard structured steps and the extraction results, judgment prompts are constructed; by inputting the judgment prompts into the large language model, the validity of the extraction results is determined based on the judgment results output by the large language model; if the extraction results are determined to be valid, the extraction results are determined as the structured steps; if the extraction results are determined to be invalid, the extraction status corresponding to the segment is set to re-extraction, and the segment with the extraction status set to re-extraction is re-extracted in a structured manner until the structured steps are obtained.

[0059] Specifically, after the large language model outputs the extraction results corresponding to the segments, the structuring step only accepts valid JSON that starts with an array / object. Therefore, the Markdown code blocks and abnormal formats in the extraction results can be stripped first. Specifically, regular expressions can be used to remove them from the extraction results.

[0060] Next, the validity of the extracted results needs to be verified. Furthermore, the validity of the extracted results also needs to be verified by a large language model. Therefore, judgment prompts can be constructed based on the judgment criteria of the standard structured steps and the extracted results. The judgment criteria here are the validity judgment criteria. The role of the judgment prompts can be defined as an SOP auditor in the equipment operation and maintenance system, with the task of judging whether the extracted results belong to equipment maintenance SOPs. Judgment criteria can include: 1. Whether it directly describes technical operations such as equipment maintenance, repair, fault handling, component replacement, and debugging; 2. Whether it focuses on "maintenance behavior" rather than management, approval, or production processes. The judgment results can include validity, confidence level, and reason.

[0061] After constructing the judgment prompts, the judgment prompts can be input into the large language model. The large language model can output the validity judgment result of the extraction result. When the extraction result is determined to be valid, the extraction result can be determined as a structured step. When the extraction result is determined to be invalid, the extraction status corresponding to the segment can be set to re-extraction, and the segment with the extraction status set to re-extraction can be re-extracted in a structured manner until the structured step corresponding to the segment is obtained.

[0062] It should be noted that invalid extraction results of a fragment do not block the execution of the task. The process of resetting the extraction status of the fragment to re-extract and performing structured extraction on the fragment can be carried out in parallel with the process of performing structured extraction on other fragments.

[0063] Of course, you can also compare the extracted structured steps and their contents, and merge duplicate structured steps.

[0064] In this embodiment of the invention, a large language model is used to automatically identify the operation steps, nested structures and logical relationships corresponding to each segment by leveraging its deep semantic reasoning capabilities. This enables accurate extraction of complex structured steps within each segment. Furthermore, by constraining the structured output method through JSON Schema, the structured steps can be generated in a format that can be directly used for calling, searching, building maintenance libraries and intelligent question answering, thereby improving the extraction efficiency of structured steps.

[0065] Step 240: According to the shard index information corresponding to each shard, merge the structured steps corresponding to each shard to obtain the structured process corresponding to the standard operation process extraction task.

[0066] In one implementation, step 240 may specifically include: Based on the shard index information corresponding to each shard, the connection relationship between the structured steps corresponding to each shard is determined, wherein the connection relationship is a combination or nesting; the structured steps corresponding to each shard are combined or nested according to the connection relationship between the structured steps corresponding to each shard to obtain the structured process corresponding to the standard operation process extraction task.

[0067] Specifically, the shard index information can include the order of the shards in their respective operation documents. Based on this order, the connection order between each shard can be determined, which means that the connection relationship between the structured steps corresponding to each shard can be determined as a combination or nesting. Then, based on the connection relationship between the structured steps corresponding to each shard, the structured steps corresponding to each shard can be combined or nested, and finally, the structured process corresponding to the standard operation process extraction task can be obtained.

[0068] In this embodiment of the invention, the structured steps corresponding to each segment are merged with reference to the segment index information corresponding to each segment to obtain a structured process with reasonable sequence and nesting relationship, thereby realizing automatic extraction of the structured process of the operation document.

[0069] In one embodiment, during the execution of steps 210-240, the method further includes: The extraction progress of the operation document to which each of the aforementioned segments belongs is determined based on the extraction progress of each segment; the extraction progress of the standard operation process is determined based on the extraction progress of each of the aforementioned operation documents; and a progress report is constructed based on the extraction progress of each segment, the extraction progress of each of the aforementioned operation documents, and the extraction progress of the standard operation process.

[0070] Specifically, during the standard operation process extraction based on steps 210-240, the extraction progress of each segment can be monitored in real time. The extraction progress can be any value between 0-100%. The extraction progress of the operation document to which each segment belongs can also be determined based on the extraction progress of each segment. For each operation document, the extraction progress can be determined based on the extraction progress of its corresponding segments and the total number of segments. The standard operation process extraction progress can also be determined based on the extraction progress of each operation document and the number of operation documents. Furthermore, a progress report can be constructed based on the extraction progress of each segment, each operation document, and the standard operation process. In practical applications, to facilitate user monitoring of the standard operation process extraction progress, the progress report can be sent to the user terminal in real time, and the user terminal can display the progress report on the monitoring page.

[0071] In this embodiment of the invention, the extraction progress of each operation document and the extraction progress of the standard operation process are determined based on the extraction progress of each segment during the standard operation process extraction. A progress report constructed from the extraction progress of each segment, the extraction progress of each operation document, and the extraction progress of the standard operation process is sent to the user terminal so that the user can control the extraction progress of the standard operation process.

[0072] Step 250: Vectorize the process text corresponding to the structured process to obtain a vectorized process; write the vectorized process into the process vector library according to the classification method of the structured process.

[0073] The process vector library is used to look up standard operating procedures.

[0074] Specifically, after determining the structured process, the corresponding process text can be identified, and then the process text can be vectorized to obtain the vectorized process. The categorization method of the structured process can also be determined, such as equipment type or fault type. Based on the categorization method, the vectorized process is written into the corresponding category's process vector library. For example, if the categorization method is equipment type, the vectorized process can be written into the process vector library corresponding to the equipment type in the operation document.

[0075] It should be noted that the process vector library supports vectorized lookup, which means that the corresponding vectorized process can be found in the process vector library based on the description vector corresponding to the fault description. Then, the vectorized process can be converted into a structured process to provide a reference for fault repair.

[0076] In this embodiment of the invention, the vectorized process corresponding to the structured process is determined by vectorizing the process text corresponding to the structured process. The vectorized process is written into the process vector library according to the classification method of the structured process, so as to realize the storage of vectorized processes by category and provide a data foundation for process retrieval.

[0077] The standard operation process extraction method provided in this embodiment includes: for each operation document corresponding to the standard operation process extraction task, segmenting the operation document according to natural paragraphs to obtain the segment corresponding to the operation document; for each segment, determining the segment index information corresponding to the segment according to the operation document to which the segment belongs and its order in the operation document; for each operation document, determining the extraction rule corresponding to the operation document according to the selection request triggered by the user for the operation document and / or the custom rule input by the user for the operation document; for each segment, determining the structured steps corresponding to the segment based on the extraction result of the structured extraction of the segment based on a large language model; merging the structured steps corresponding to each segment according to the segment index information corresponding to each segment to obtain the structured process corresponding to the standard operation process extraction task; vectorizing the process text corresponding to the structured process to obtain the vectorized process; and writing the vectorized process into a process vector library according to the classification method of the structured process. The above technical solution parses the operation documents corresponding to the standard operation process extraction task, dividing each operation document into at least one segment. Specifically, based on a preset context window threshold and natural paragraphs, the segmentation of each operation document is achieved while ensuring contextual continuity, enabling smooth segmentation of wide-format, multi-page operation documents. Simultaneously, the segment index information corresponding to each segment can be determined, achieving precise location of each segment. Then, according to the extraction rules corresponding to the operation document to which each segment belongs, the segment is structurally extracted to obtain the structured steps corresponding to each segment. Through a large language model, its deep semantic reasoning capabilities are used to automatically identify the operation steps, nested structures, and logical relationships corresponding to each segment, achieving accurate extraction of complex structured steps within each segment. Furthermore, JSON Schema is used... This approach constrains the structured output method, enabling the formatted generation of structured steps that can be directly used for calling, searching, repair library construction, and intelligent question answering, thus improving the extraction efficiency of structured steps. By merging the structured steps corresponding to each segment based on the segment index information, a structured process with a reasonable sequence and nesting relationship is obtained. This achieves highly accurate automatic extraction of structured processes corresponding to operation documents, improving the efficiency and accuracy of structured process extraction. Through vectorization of the process text corresponding to the structured process, the corresponding vectorized process is determined. Based on the classification method of the structured process, the vectorized process is written into the process vector library, enabling categorized storage of vectorized processes and providing a data foundation for process retrieval. Through technologies such as large language models, text segmentation, structured extraction, and structured aggregation, the approach effectively solves problems such as non-standard, non-automatic, inefficient, and error-prone structured process extraction, achieving automated and intelligent extraction of structured processes.

[0078] Furthermore, during the extraction of structured processes, the extraction progress of each operation document and the standard operation process are determined based on the extraction progress of each segment during the extraction of standard operation processes. A progress report, which is constructed by the extraction progress of each segment, the extraction progress of each operation document, and the extraction progress of standard operation processes, is sent to the user terminal so that the user can control the extraction progress of standard operation processes.

[0079] Example 3 Figure 3 This is a schematic diagram of a standard operating procedure extraction device provided in Embodiment 3. Figure 3 As shown, the device includes: The parsing module 310 is used to parse the operation documents corresponding to the task extracted from the standard operation process, divide each operation document into at least one piece, and determine the piece index information corresponding to each piece. Extraction module 320 is used to determine the structured steps corresponding to each segment based on the extraction results of the structured extraction of the segment using a large language model. The aggregation module 330 is used to merge the structured steps corresponding to each of the shards according to the shard index information corresponding to each shard, so as to obtain the structured process corresponding to the standard operation process extraction task.

[0080] The standard operation process extraction device provided in this embodiment parses the operation documents corresponding to the standard operation process extraction task, divides each operation document into at least one segment, and determines the segment index information corresponding to each segment; for each segment, the structured steps corresponding to the segment are determined based on the extraction result of the structured extraction of the segment using a large language model; according to the segment index information corresponding to each segment, the structured steps corresponding to each segment are merged to obtain the structured process corresponding to the standard operation process extraction task. The above technical solution first parses the operation documents corresponding to the standard operation process extraction task to divide each operation document into at least one segment, thereby achieving document segmentation. Simultaneously, it determines the segment index information corresponding to each segment, enabling precise location of each segment. Then, it performs structured extraction on each segment corresponding to each operation document to determine the structured steps for each segment. Furthermore, it merges the structured steps corresponding to each segment with reference to the segment index information, resulting in a structured process with a reasonable sequence and nesting relationship. This achieves highly accurate automatic extraction of structured processes corresponding to operation documents, improving the efficiency and accuracy of structured process extraction.

[0081] Based on the above embodiments, the parsing module 310 is specifically used for: For each of the operation documents, the operation document is divided into segments according to natural paragraphs to obtain the segments corresponding to the operation document; for each segment, the segment index information corresponding to the segment is determined according to the operation document to which the segment belongs and its order in the operation document.

[0082] In one implementation, the operation document is segmented according to natural paragraphs to obtain the segment corresponding to the operation document, including: Based on a preset context window threshold, the operation document is segmented according to natural paragraphs to obtain the segment corresponding to the operation document.

[0083] Based on the above embodiments, the device further includes: The determination module is used to determine the extraction rule corresponding to each operation document before performing structured extraction of the segment according to the extraction rule corresponding to the operation document to which the segment belongs, based on the selection request triggered by the user for the operation document and / or the custom rule input by the user for the operation document.

[0084] Based on the above embodiments, the extraction module 320 is specifically used for: Extraction prompts are constructed based on the fragmentation, fragmentation index information, structural constraints, and extraction rules; the extraction prompts are input into a large language model, and the structuring step is determined based on the extraction results output by the large language model.

[0085] In one implementation, determining the structuring step based on the extraction results output by the large language model includes: Based on the judgment criteria of the standard structured steps and the extraction results, judgment prompts are constructed; by inputting the judgment prompts into the large language model, the validity of the extraction results is determined based on the judgment results output by the large language model; if the extraction results are determined to be valid, the extraction results are determined as the structured steps; if the extraction results are determined to be invalid, the extraction status corresponding to the segment is set to re-extraction, and the segment with the extraction status set to re-extraction is re-extracted in a structured manner until the structured steps are obtained.

[0086] Based on the above embodiments, the aggregation module 330 is specifically used for: Based on the shard index information corresponding to each shard, the connection relationship between the structured steps corresponding to each shard is determined, wherein the connection relationship is a combination or nesting; the structured steps corresponding to each shard are combined or nested according to the connection relationship between the structured steps corresponding to each shard to obtain the structured process corresponding to the standard operation process extraction task.

[0087] Based on the above embodiments, the device further includes: The monitoring module is used to determine the extraction progress of the operation document to which each of the segments belongs based on the extraction progress of each segment; determine the extraction progress of the standard operation process based on the extraction progress of each operation document; and construct a progress report based on the extraction progress of each segment, the extraction progress of each operation document, and the extraction progress of the standard operation process.

[0088] Based on the above embodiments, the device further includes: The execution module is used to vectorize the process text corresponding to the structured process to obtain a vectorized process; and to write the vectorized process into a process vector library according to the classification method of the structured process, wherein the process vector library is used to search for standard operation processes.

[0089] The standard operating procedure extraction device provided in this embodiment can execute the standard operating procedure extraction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of executing the standard operating procedure extraction method.

[0090] Example 4 Figure 4 This is a schematic diagram of a standard operating procedure extraction system provided in Embodiment 4. Figure 4 As shown, the system includes a terminal device 410 and a server 420, which interact with each other through an interface layer.

[0091] Terminal device 410 is used to provide an interactive interface, through which users can create, start and cancel standard operation process extraction tasks, subscribe to the progress of standard operation process extraction tasks, preview and export the results of standard operation process extraction, etc. Figure 5a Create an interface for the task. Figure 5b For the rule configuration interface, users can base it on... Figure 5a The task creation page shown allows users to create tasks and upload corresponding operation documents. After uploading the operation documents, users can be redirected to... Figure 5b The rule configuration interface shown allows users to configure rules based on... Figure 5b The rule configuration interface shown specifies the extraction rules corresponding to the operation document. Figure 5cThe task management interface displays information such as task identifier, task name, number of documents to be processed, progress, task status, synchronization status, number of documents extracted, number of documents synchronized, and creation time for each task. It also includes various operation buttons, such as: details, start, retry, cancel, synchronize all, download, and delete. Figure 5d This is a structured workflow display interface. After determining the structured workflow corresponding to a task, the structured workflow can be displayed on this interface. Users can also review and modify the structured workflow based on this interface.

[0092] Server 420 may include a task management server, a document parsing server, an extraction server, a progress management server, a data storage server, a result processing server, and a process synchronization server.

[0093] The task management server is used for full lifecycle management of tasks, including creation, startup, cancellation, retry, and statistics. For example, when a task is created, a task identifier is generated, and the number of operation documents and rule templates are registered. The task status is set to pending. When a task is started, the task status is changed to processing, and the cumulative success and failure counts of the task's segment extraction are counted. Finally, it is determined whether the extraction result of the corresponding segment of the task is all successful, partially successful, or all failed. When a task is canceled, task-level cancellation and document-level cancellation are supported, and the corresponding task or document is marked as aborted. When a task is retried, task-level retry and document-level retry are supported, and the corresponding task or document is marked as retried. When performing task statistics, the progress percentage, number of completed tasks and failures, number of segment processed, etc. are updated in real time, and the progress report is sent to the terminal device.

[0094] The document parsing server converts operation documents into processable fragments while preserving sequential information. The extraction server performs structured extraction on the fragments, obtaining hierarchical fault descriptions and step descriptions. The progress management server determines progress at the fragment, document, and task levels and generates progress reports that are sent to terminal devices. Specifically, it establishes an SSE connection (a one-way server push technology based on HTTP long connections) with critical devices, pushes progress reports to terminal devices via the SSE connection, and automatically removes the connection if the push fails. It can also clear the cache after task completion or failure to ensure continuous usability. The data storage server persistently stores all intermediate and final state data during task execution, ensuring traceability and auditability. This includes task and document status, progress, total number of fragments, model output results, structured workflows before and after cleaning, error messages, timestamps, synchronization status, and counts. The results processing server aggregates the extraction results from multiple fragments into a single, valid structured workflow, preparing for previewing, exporting, and synchronization. It can also skip illegal fragments and log data to ensure the overall results are usable and formatted stably. The process synchronization server is used to transform structured processes into directly implementable maintenance process entities and synchronize them to the maintenance process library. It also constructs vectorized processes and writes them into a searchable process vector library. Furthermore, the process synchronization server can synchronously record the reasons for failed processes during task execution; a single failure does not affect other processes, and the recorded results are traceable and retryable.

[0095] Figure 6 This is a sequence diagram of standard operating procedure (SOP) extraction based on the SOP extraction system, such as... Figure 6As shown, the standard operation process extraction method includes: the terminal device creates a task, uploads the operation document, and selects rules, setting the initial task status to PENDING. Subsequently, based on the received task start request, the task status is set to PROCESSING, and the task cancellation flag (cancelFlag) is cleared. The document parsing server parses each operation document according to the selected rules, obtaining the corresponding fragments and fragment retrieval information. The extraction server performs structured extraction on each fragment, obtaining the structured steps corresponding to each fragment. The extraction server sends a progress report to the terminal device via the interface layer. The terminal device displays the progress report through the interactive interface after the user triggers the progress button. Furthermore, before each extraction, the cancellation manager checks the cancelFlag; if the user initiates cancellation, the task is terminated, ensuring immediate cancellation when a user cancels any task to prevent wasting background computing resources. The result processing server aggregates the structured steps corresponding to each fragment to obtain a structured process. It can also mark the document status as SUCCESS or FAILED based on the aggregation results and record statistical information such as the number of fragments and the number of processed fragments. The process synchronization server sends the final task progress status (fully completed, partially completed, or failed) to the terminal device.

[0096] The standard operating procedure extraction exercise provided in this embodiment can execute the standard operating procedure extraction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of executing the standard operating procedure extraction method.

[0097] Example 5 Figure 7 This is a schematic diagram of the structure of an electronic device provided in Embodiment 5 of the present invention. Figure 7 A block diagram of an exemplary electronic device 7 suitable for implementing embodiments of the present invention is shown. Figure 7 The electronic device 7 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0098] like Figure 7 As shown, the electronic device 7 is represented in the form of a general-purpose computing electronic device. The components of the electronic device 7 may include, but are not limited to: one or more processors or processing units 16, system memory 28, and bus 18 connecting different system components (including system memory 28 and processing unit 16).

[0099] Bus 18 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. For example, these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0100] Electronic device 7 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by electronic device 7, including volatile and non-volatile media, removable and non-removable media.

[0101] System memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. Electronic device 7 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 34 may be used to read and write non-removable, non-volatile magnetic media (… Figure 7 Not shown; usually referred to as a "hard drive"). Although Figure 7 As not shown, disk drives for reading and writing to removable non-volatile disks (e.g., "floppy disks") and optical disc drives for reading and writing to removable non-volatile optical discs (e.g., CD-ROMs, DVD-ROMs, or other optical media) may be provided. In these cases, each drive may be connected to bus 18 via one or more data media interfaces. System memory 28 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of the present invention.

[0102] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in system memory 28. Such program modules 42 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 42 typically perform the functions and / or methods described in the embodiments of the present invention.

[0103] Electronic device 7 can also communicate with one or more external devices 14 (e.g., keyboard, pointing device, display 24, etc.), and with one or more devices that enable a user to interact with electronic device 7, and / or with any device that enables electronic device 7 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed through input / output (I / O) interface 22. Furthermore, electronic device 7 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) through network adapter 20. Figure 7 As shown, network adapter 20 communicates with other modules of electronic device 7 via bus 18. It should be understood that, although... Figure 7 Not shown, it can be combined with electronic device 7 to use other hardware and / or software modules, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0104] Processing unit 16 executes various functional applications and page displays by running programs stored in system memory 28, such as implementing the standard operation flow extraction method provided in this embodiment of the invention, which includes: By parsing the operation documents corresponding to the task extracted from the standard operation process, each operation document is divided into at least one segment, and the segment index information corresponding to each segment is determined. For each of the aforementioned segments, the structured extraction results based on the large language model are used to determine the structured steps corresponding to the segment. According to the shard index information corresponding to each shard, the structured steps corresponding to each shard are merged to obtain the structured process corresponding to the standard operation process extraction task.

[0105] Of course, those skilled in the art will understand that the processor can also implement the technical solution of the standard operation process extraction method provided in any embodiment of the present invention.

[0106] Example 6 Embodiment 6 of the present invention provides a computer-readable storage medium having a computer program stored thereon. When executed by a processor, the program implements, for example, the standard operation flow extraction method provided in the embodiments of the present invention, the method comprising: By parsing the operation documents corresponding to the task extracted from the standard operation process, each operation document is divided into at least one segment, and the segment index information corresponding to each segment is determined. For each of the aforementioned segments, the structured extraction results based on the large language model are used to determine the structured steps corresponding to the segment. According to the shard index information corresponding to each shard, the structured steps corresponding to each shard are merged to obtain the structured process corresponding to the standard operation process extraction task.

[0107] The computer storage medium of this invention can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0108] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0109] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0110] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0111] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computing device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0112] Furthermore, the acquisition, storage, use, and processing of data in the technical solution of this invention all comply with relevant laws and regulations.

[0113] Note that the above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention, the scope of which is determined by the scope of the appended claims.

Claims

1. A method for extracting a standard operation procedure based on a large language model, characterized in that, The method comprises the following steps: parsing operation documents corresponding to a standard operation process extraction task by extracting tasks, dividing each operation document into at least one segment, and determining segment index information corresponding to each segment; for each segment, determining the structured step corresponding to the segment based on the extraction result of the structured extraction of the segment by the large language model; combining the structured steps corresponding to each segment according to the segment index information corresponding to each segment to obtain the structured process corresponding to the standard operation process extraction task.

2. The standard operation flow extraction method based on a large language model according to claim 1, characterized in that, The method comprises the following steps: for each operation document, the operation document is divided into segments according to natural paragraphs to obtain the segments corresponding to the operation document; for each segment, the segment index information corresponding to the segment is determined according to the operation document to which the segment belongs and the order in the operation document. 3.The standard operation flow extraction method based on a large language model according to claim 2, wherein, The method comprises the following steps: On the basis of the preset context window threshold, the operation document is divided into segments according to natural paragraphs to obtain the segments corresponding to the operation document. 4.The standard operation flow extraction method based on a large language model according to claim 1, wherein, Before structurally extracting the segment according to the extraction rule corresponding to the operation document to which the segment belongs, the method further comprises the following steps: for each operation document, the extraction rule corresponding to the operation document is determined according to the selection request triggered by the user for the operation document, and / or the custom rule input by the user for the operation document. 5.The standard operation flow extraction method based on a large language model according to claim 1, wherein, The method comprises the following steps: According to the segment, the segment index information, the structure constraint condition and the extraction rule corresponding to the operation document to which the segment belongs, an extraction prompt word is constructed; by inputting the extraction prompt word into the large language model, the structured step is determined according to the extraction result output by the large language model. 6.The standard operation flow extraction method based on a large language model according to claim 5, wherein, The method comprises the following steps: According to the judgment standard of the standard structured step and the extraction result, a judgment prompt word is constructed; by inputting the judgment prompt word into the large language model, whether the extraction result is valid is determined according to the judgment result output by the large language model; when it is determined that the extraction result is valid, the extraction result is determined as the structured step; when it is determined that the extraction result is invalid, the extraction state corresponding to the segment is set to re-extraction, and the segment with the re-extraction state is re-extracted until the structured step is obtained.

7. The big language model-based standard operation flow extraction method according to claim 1, wherein, The method comprises the following steps: According to the slice index information corresponding to each of the slices, a connection relationship between the structured steps corresponding to each of the slices is determined, wherein the connection relationship is combination or nesting; According to the connection relationship between the structured steps corresponding to each of the slices, the structured steps corresponding to each of the slices are combined or nested to obtain a structured process corresponding to the standard operation process extraction task. 8.The method of claim 7, wherein, Further comprising: According to the extraction progress of each of the slices, the extraction progress of the operation document to which each of the slices belongs is determined; According to the extraction progress of each of the operation documents, the standard operation process extraction progress is determined; According to the extraction progress of each of the slices, the extraction progress of each of the operation documents, and the standard operation process extraction progress, a progress report is constructed. 9.The standard operation flow extraction method based on a large language model according to claim 1, wherein, After obtaining the structured process corresponding to the standard operation process extraction task, further comprising: The process text corresponding to the structured process is vectorized to obtain a vectorized process; According to the category division mode of the structured process, the vectorized process is written into a process vector library, wherein the process vector library is used for standard operation process searching.

10. An electronic device, comprising: The electronic device comprises: At least one processor; and a memory connected in communication with the at least one processor; Wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute any one of the standard operation process extraction methods based on a large language model in claims 1-9.

Images