A multi-granularity semantic coverage detection and step deviation diagnosis method based on knowledge graph ontology constraint

Abstract

The application discloses a kind of multi-granularity semantic coverage detection and step deviation diagnosis method based on knowledge graph ontology constraint.The method inquires standard step sequence and each standard step pre-associated check item, atomic semantic slot, weight and order constraint from knowledge graph according to current job task, and obtains to be audited step sequence;According to knowledge graph ontology, the semantic analysis and standard concept mapping of to-be-audited step are carried out, and the filling step semantic slot set is obtained;Combining vector similarity generates candidate to-be-audited step set for each standard step;In the range of candidate set, check item coverage audit and atomic semantic slot coverage audit are carried out respectively, and check item coverage score and atomic semantic coverage score are obtained;Based on the score, order constraint penalty term and forced atomic slot missing penalty term, fusion coverage score is calculated, and coverage state is determined according to fusion coverage score and forced atomic slot coverage condition;According to the preset priority of missing step or pre-step missing, order error, over-generalized merging, split expression, effective merging, redundant step from high to low, determine main diagnosis type;Finally, through knowledge graph ontology, step deviation type is associated to corresponding harm, consequence and control measure and output.The method can improve the accuracy, explainability and traceability of process step audit.

Description

Technical Field

[0001] This invention belongs to the field of process step review and production safety, and particularly relates to a method for multi-granular semantic coverage detection and step deviation diagnosis based on knowledge graph ontology constraints. Background Technology

[0002] In scenarios such as work safety analysis, work permit review, and standard operating procedure verification, it is usually necessary to compare the sequence of steps filled in on-site with the preset standard sequence of steps to determine whether the on-site content is complete, accurate, and meets the sequence requirements.

[0003] Existing methods for reviewing steps often employ keyword matching, rule matching, or vector similarity matching. While these methods can identify some steps that are superficially similar, their ability to identify steps with inconsistent granularity, multiple steps broken down into smaller parts, multiple standard steps combined into a single, general expression, incorrect step order, or superficial similarity without crucial security semantics is limited.

[0004] Especially in operational safety analysis scenarios, whether a step meets the review requirements depends not only on text similarity but also on whether it covers key safety semantic units such as actions, objects, tools, locations, conditions, and control measures. Judging solely based on keyword or vector similarity can easily lead to misjudging vague expressions that lack key safety requirements as if they have been covered.

[0005] In addition, existing methods often lack expression of the relationship between step deviations and hazards, consequences, and control measures, resulting in insufficient interpretability of audit results and difficulty in providing direct basis for subsequent risk review and rectification closure.

[0006] Therefore, it is necessary to provide a step-by-step review method that can combine knowledge graph ontology, fine-grained semantic slots, and sequence constraints for comprehensive judgment, so as to improve the accuracy, interpretability, and traceability of coverage detection and deviation diagnosis. Summary of the Invention

[0007] The purpose of this invention is to provide a multi-granularity semantic coverage detection and step deviation diagnosis method based on knowledge graph ontology constraints, so as to solve the problem that existing technologies rely solely on keyword matching or vector similarity, which makes it difficult to accurately identify true coverage, partial coverage, over-generalized merging, split expression, and sequence errors.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] A multi-granularity semantic coverage detection and step deviation diagnosis method based on knowledge graph ontology constraints includes:

[0010] (1) Query the standard step sequence from the knowledge graph based on the current task and obtain the step sequence to be reviewed;

[0011] (2) Read the semantic requirements associated with each standard step from the knowledge graph, and perform semantic parsing only on the step to be reviewed based on the knowledge graph ontology to obtain the set of semantic slots for the filling step;

[0012] (3) Obtain the pre-associated mandatory atomic slots, ordinary atomic slots, check items and corresponding weights for each standard step from the knowledge graph;

[0013] (4) Combine vector similarity to generate a set of candidate steps to be reviewed for each standard step;

[0014] (5) Within the limited scope of the candidate set, perform check item coverage audit and atomic semantic slot coverage audit respectively to obtain check item coverage score and atomic semantic coverage score;

[0015] (6) Calculate the fusion coverage score by combining the sequence constraint penalty and the forced atomic slot missing penalty;

[0016] (7) Determine the coverage status based on the fusion coverage score and the forced atomic slot coverage;

[0017] (8) Based on the coverage status, mapping relationship and positional relationship, diagnose the deviation type of the steps according to the preset priority and determine the main diagnosis type;

[0018] (9) Associate the main diagnostic type and auxiliary tags with the corresponding hazards, consequences and control measures through the knowledge graph ontology and output them. Attached Figure Description

[0019] Figure 1 The flowchart for multi-granularity semantic coverage detection and step deviation diagnosis provided in the embodiments of the present invention is shown.

[0020] Figure 2 This is a knowledge graph ontology constraint structure diagram provided in an embodiment of the present invention. Detailed Implementation

[0021] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that the following embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention.

[0022] like Figure 1 As shown, this embodiment uses the review of a work safety analysis form as the application scenario. The system first queries the corresponding standard step sequence from the knowledge graph based on the current work task identifier, and simultaneously reads the pre-associated inspection items, atomic semantic slots, weights, and sequence constraints of each standard step; the step sequence to be reviewed comes from the work tickets filled out by on-site personnel, electronic forms, or mobile input.

[0023] The system retrieves the standard step sequence S={S1, S2, ..., S...} from the knowledge graph. m}, and obtain the sequence of steps to be reviewed W = {W1, W2, ..., W}. n}. Among them, S i W represents the i-th standard step. j This represents the j-th step to be reviewed, where m and n represent the number of steps in the two sequences, respectively. The system retains the identifier and sequential position of the standard steps in the knowledge graph, as well as the number and position of the steps to be reviewed in the original input sequence.

[0024] The system only performs word segmentation, terminology normalization, dependency relationship analysis, or semantic role annotation on the text to be reviewed, and maps colloquialisms, abbreviations, or non-standard expressions to standardized concepts in the knowledge graph ontology, resulting in a set of semantic slots for the filling steps. These semantic slots include at least one of action, object, tool, location, condition, and temporal prerequisite. Standard steps and their semantic requirements are directly read from the knowledge graph, eliminating the need for further ontology semantic parsing.

[0025] like Figure 2 As shown, the knowledge graph ontology includes at least the following concepts: task, standard steps, filling steps, check items, atomic semantic slots, mandatory atomic slots, ordinary atomic slots, sequence constraints, step mapping relationships, step deviation types, hazards, consequences, and control measures. These concepts are connected through relationships such as querying, receiving and filling, associating check items, associating semantic slots, being constrained by, forming mappings, classifying as, providing coverage evidence, providing missing evidence, providing sequence evidence, providing quantity and location evidence, associating hazards, associating consequences, and associating control measures.

[0026] For each standard step S obtained from the knowledge graph query i The system directly reads the pre-associated set of atomic semantic slots and distinguishes between mandatory atomic slots and ordinary atomic slots. Mandatory atomic slots represent semantic requirements that cannot be omitted and must be explicitly filled in; ordinary atomic slots represent semantic requirements used to supplement, refine, or improve the completeness of expression. This reading process does not re-execute word segmentation, ontology mapping, or semantic slot parsing as standard steps.

[0027] The system obtains standard steps S from the knowledge graph. i The corresponding set of inspection items Q i ={q i1 q i2 , ..., q ik} and its weight u ik Each inspection item q ik Used to determine whether a candidate step explicitly answers the standard step S. iOne of the audit questions; when given a clear answer, the check item coverage value c ik Select 1 if the value is 1, otherwise select 0.

[0028] Standard Step S i The inspection item coverage score is C. i Calculate using the following formula: C i =Σ(u ik ·c ik ) / Σu ik Among them, u ik c represents the weight of the k-th inspection item. ik This represents the coverage value of the k-th check item.

[0029] The system obtains standard steps S from the knowledge graph. i The corresponding set of atomic semantic slots G i ={g i1 g i2 , ..., g il} and its weight w il And the semantic slots of the candidate filling steps are linked to G. i Perform alignment. Explicitly cover the atomic semantic slot g. il At that time, the coverage value r il Take 1; when coverage is formed only by generalized concepts, superordinate concepts, or incomplete expressions, r il Take partial coverage value p a , of which 0 <p a <1; when not covered, r il Take 0.

[0030] Standard Step S i Atomic semantic coverage score A i Calculate according to the following formula: A i =Σ(w il ·r il ) / Σw il Among them, w il r represents the weight of the l-th atomic semantic slot. il This represents the coverage value of the l-th atomic semantic slot.

[0031] To limit the comparison scope of fine-grained review, the system vectorizes the standard step specification text and the text of the step to be reviewed read from the knowledge graph, and calculates the standard step S. i With the pending review step W j Vector similarity V between ij and V ij Steps that reach the candidate threshold are added to the initial candidate set. The system will also work with S... i Pre-configured semantic requirements specify that steps awaiting review that have synonymous, subordinate overriding, or adjacent window relationships should be added to the candidate step set H.i Candidate step set H i It is only used to limit the scope of subsequent comparisons and is not directly used as the basis for determining the final coverage status.

[0032] For the candidate step set H i For one or more steps to be reviewed, the system performs check item coverage review and atomic semantic slot coverage review respectively. When multiple steps to be reviewed jointly express the same standard step, the system takes the joint coverage result of their semantic slots and then calculates C based on this. i and A i When the same semantic slot is expressed repeatedly, it is only counted as one coverage.

[0033] The system directly reads pre-configured pre-, post-, adjacent, or non-reversible sequence constraints between standard steps from the knowledge graph and establishes a positional mapping from standard steps to valid steps awaiting review. For constraint S... i Prepositioned in S k If S i The corresponding step to be reviewed is later than S. k The sequence violates the marker e at the location of the step to be reviewed. ik Take 1; otherwise e ik Take 0.

[0034] Sequence constraint penalty term R i According to R i =Σ(ζ ik ·e ik ) / Σζ ik Determined. Wherein, ζ ik S represents i With S k The importance weight of the order constraints between them, e ik This indicates a violation of the corresponding sequence constraint. When there is no sequence constraint violation, R... i =0; when a normal order violation exists, R = 0; i Increase the weight accordingly; when there is a forced inversion of the preceding step, the adjacent step is interrupted, or the order of critical safety steps is incorrect, use a higher weight.

[0035] The system obtains standard steps S from the knowledge graph. i Forced atomic slot set F i ={f i1 f i2 , ..., f it For each forced atomic slot f it If the candidate step set H i If the pending review steps in the process form explicit coverage, then the missing marker m it Take 0; if not covered, or if coverage is only formed by generalized concepts, superordinate concepts, or incomplete expressions, then mit Take 1.

[0036] Forced atomic slot missing penalty item M i According to M i =Σ(v it ·m it ) / Σv it Calculate. Where, v it m represents the missing penalty weight for the t-th forced atomic slot. it This indicates the missing flag for the t-th forced atomic slot. When no forced atomic slot is missing, M... i =0; The more missing slots or the higher the safety importance of the missing slots, the lower the M value. i The larger.

[0037] The system scores C based on the coverage of the inspection items. i Atomic semantic coverage score A i Sequence constraint penalty term R i And the forced atomic slot missing penalty item M i Calculate S according to the following standard steps. i The integrated coverage segment: Coverage i =λC i +μA i −ρR i −ηM i Wherein, λ, μ, ρ, and η are preset weighting coefficients. Each weighting coefficient can be determined based on the importance of the safety procedure, historical audit samples, or expert annotation results.

[0038] The coverage state is determined jointly by the fused coverage and the forced atomic slot coverage. When Coverage i If the value is not lower than the first threshold and there are no forced missing atomic slots, it is considered a complete cover; when Coverage i If the value is below the first threshold but not below the second threshold, it is considered partial coverage; when Coverage i If the value is below the second threshold but not below the third threshold, it is considered weak coverage; when H... i Empty or Coverage i If the value is below the third threshold, it is considered missing.

[0039] To diagnose deviations in the process, the system establishes a reverse correspondence between the steps to be reviewed and the standard steps, and presets diagnostic priorities from high to low: the first priority is missing steps or missing prerequisite steps, the second priority is incorrect order, the third priority is overly generalized merging, the fourth priority is splitting the expression, the fifth priority is effective merging, and the sixth priority is redundant steps.

[0040] When the candidate step set H iEmpty or Coverage i When the value is below the third threshold, it is diagnosed as a missing step of the first priority; when S i To enforce the prerequisite step and H i When S is empty, the diagnosis is a missing first-priority prerequisite step. i It should be placed before S k However, its corresponding step is later than S. k When the corresponding step is located, it is diagnosed as a second-priority sequence error. This occurs when a step to be reviewed corresponds to multiple standard steps, but at least one standard step has a missing mandatory atomic slot or coverage. i If the value is below the first threshold, it is diagnosed as over-merging with the third priority.

[0041] When a standard step is jointly covered by multiple steps pending review, and the combined Coverage i If the value is not lower than the first threshold and there are no missing mandatory atomic slots, it is diagnosed as a fourth-priority split expression. When a step to be reviewed corresponds to multiple standard steps, and the mandatory atomic slots of each standard step are effectively covered, it is diagnosed as a fifth-priority effective merge.

[0042] When awaiting review step W j If a step does not correspond to any standard step, it is diagnosed as a redundant step with the sixth priority. Necessary risk descriptions, control measures, or reasonable supplements to the standard steps can be configured as allowed supplementary types in the knowledge graph. When the same diagnostic object simultaneously meets multiple type conditions, the system will determine the highest priority type as the primary diagnostic type and output the remaining non-conflicting types as auxiliary labels.

[0043] Once the primary diagnostic type is determined, the system starts with the standard steps, inspection items, or atomic semantic slots where deviations occur, and searches along the relationships of associated hazards, associated consequences, and associated control measures in the knowledge graph to obtain the corresponding hazard set, consequence set, and control measure set. This is then output together with the primary diagnostic type, auxiliary tags, coverage status, missing slots, and step location evidence.

[0044] For example, the system retrieves the standard step S3 "Gas detection before operation" from the knowledge graph based on the current task, and directly reads its pre-associated check items and mandatory atomic slots. The check items include whether it is specified whether pre-operation detection is required, whether oxygen concentration is detected, whether combustible gas is detected, whether toxic and harmful gas is detected, and whether it is specified that entry is only allowed after passing the detection. The corresponding mandatory atomic slots include at least the detection action, oxygen concentration, combustible gas, toxic and harmful gas, and pre-operation timing prerequisites.

[0045] If the step W2 to be reviewed is "Inspect the site environment", and W2 passes the candidate screening and enters H3, the system will only parse W2 and align it with the pre-configured semantic requirements of S3 in a fine-grained manner. Since W2 only provides partial coverage of general inspection actions and does not explicitly cover oxygen concentration, combustible gases, toxic and harmful gases, and pre-operation timing prerequisites, related c 3k and r 3l Take 0 or p a multiple forced atomic slots m 3t Set it to 1, which decreases A3 and increases M3. When Coverage3 is below the third threshold, the system determines the first priority missing step as the primary diagnostic type.

[0046] For example, when the standard step "perform welding and arrange for dedicated personnel to monitor" is expressed by both the steps to be reviewed, "perform welding" and "arrange for dedicated personnel to monitor throughout the process," the system combines the semantic slots of the two steps to be reviewed. If the combined result covers all forced atomic slots and the fusion coverage score reaches the first threshold, and no higher priority type is hit, then the fourth priority split expression is determined as the main diagnostic type.

[0047] When the pending step "Safety check before operation" corresponds to three standard steps—disclosure zone setting, gas detection, and ventilation—but does not explicitly cover the mandatory atomic slots of at least one of the standard steps, the system will determine the overgeneralization of the third priority as the main diagnostic type; only when it effectively covers the mandatory atomic slots of all three standard steps and does not hit a higher priority type will the effective merging of the fifth priority be determined as the main diagnostic type.

[0048] This invention can be deployed in job safety analysis systems, job ticket review systems, enterprise safety procedure review platforms, or mobile job reporting systems. After the processor executes the program instructions in memory, it can automatically output the candidate step number for each standard step, C. i A i R i M i Coverage i Coverage status, missing forced atomic slots, diagnostic priority, primary diagnostic type, auxiliary labels, and associated hazards, consequences, and control measures.

[0049] In one application embodiment based on a JSA review system, the JSA review system includes a JSA form management module, a knowledge graph query module, a vector retrieval module, the method execution module of this invention, a large model review module, and a review history management module. The client submits tree-structured JSA data containing the job name, job location, tools and equipment, job steps, hazards, consequences, and control measures, along with its version identifier. The server locks the form version corresponding to this review, identifies the job task type and job scenario, queries the knowledge graph for standard step sequences and their pre-associated check items, atomic semantic slots, forced atomic slots, and sequence constraints, and inputs the user-filled job steps as the step sequence W to be reviewed into the method execution module of this invention.

[0050] The method execution module of the present invention is for each standard step S i The structured process audit result is generated, which includes at least the JSA form identifier, version identifier, standard process identifier, and candidate process set H. i And the corresponding pending audit step number, inspection item coverage score C i Atomic semantic coverage score A i Sequence constraint penalty term R i Forced atomic slot missing penalty item M i Integrated Coverage i The structured review process includes information such as coverage status, missing mandatory atomic slots, diagnostic priority, primary diagnostic type, auxiliary tags, and associated hazards, consequences, and control measures. The results can also record the location of the text to be reviewed, the hit knowledge graph nodes, and relationship paths, enabling each review conclusion to be traced back to the input steps and knowledge graph basis.

[0051] To provide enhanced retrieval and generation of RAG knowledge for large models, the system converts the structured step review results into review evidence knowledge units. Each review evidence knowledge unit is granular, consisting of an evidence summary for semantic retrieval and structured metadata for precise filtering. The structured metadata includes at least one of the following: job type, job scenario, standard step identifier, step to be reviewed identifier, coverage status, diagnosis type, diagnosis priority, knowledge graph version, and review time. The system can generate vector representations of the evidence summaries and write them into the vector index of the knowledge graph or independent vector storage. During retrieval, it first performs structured filtering by job type, version, or diagnosis type, and then selects review evidence knowledge units related to the current JSA review object based on vector similarity.

[0052] When constructing the audit prompt context, the large model audit module injects the current JSA tree data, audit rule documents, retrieved historical reference cases, and audit evidence knowledge units output by this invention into the prompt context. The coverage status, missing mandatory atomic slots, main diagnosis type, and diagnosis priority output by this invention serve as constrained audit facts. The large model does not re-replace the coverage calculation and priority determination of this invention; instead, it generates problem merging, duplicate elimination, terminology standardization, cause summarization, and rectification suggestions based on these audit facts. When the content generated by the large model conflicts with the audit evidence knowledge units or the evidence is insufficient, the system retains the results from the method execution module and marks it as awaiting manual review.

[0053] When summarizing and standardizing the results of the step-by-step audit, the large model can generate audit conclusions, main problem analyses, key step missing reminders, and improvement suggestions according to preset templates. For each issue, the standardized output should include at least one of the following: the step or hazard involved, the issue category, the diagnostic priority, the issue description, the evidence, the missing action, object, tool, location, condition or time sequence prerequisites, associated hazards and consequences, control measures, and modification suggestions. The system can simultaneously output machine-readable structured data and report text for auditors, and display the audit process through streaming responses, saving the complete report, structured issue items, and their corresponding audit evidence knowledge units to the audit history.

[0054] For the initial audit, the system executes the method of this invention on all pending steps of the current JSA form and binds the audit evidence knowledge unit to the locked form version. For incremental audits, the system compares the steps, hazards, consequences, and control measures of the previous and subsequent versions, and only re-executes coverage detection and deviation diagnosis on the steps affected by additions, deletions, or modifications. For unchanged steps, the system reuses the saved audit evidence knowledge units. At the same time, it provides the large model with the previous round of audit conclusions, issues, audit notes, and version differences, so that the standardized report can distinguish between rectified issues, new issues, and unresolved issues.

[0055] Auditors can confirm, reject, modify, or annotate issues in the standardized report. The system will associate and save the manual processing results with the JSA form version, knowledge graph version, and original audit evidence knowledge units. Historical evidence, after audit confirmation and version verification, can serve as a reference sample for subsequent RAG retrieval, but does not directly cover the standard steps and ontology constraints in the knowledge graph. Therefore, the combination of calculable and traceable audit results with the language inductive ability of the large model provided by this invention reduces the risk of the large model generating conclusions deviating from the audit basis, and improves the consistency, interpretability, and reusability of the JSA audit report.

[0056] Those skilled in the art should understand that various equivalent substitutions and modifications can be made to this invention without departing from its spirit and essence, and all such substitutions and modifications should fall within the protection scope of this invention. Those skilled in the art should understand that embodiments of this invention can be provided as methods, systems, or computer program products. Therefore, this invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this invention can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0057] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0058] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0059] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment, whereby a series of operational steps are performed to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0060] The above description is merely a preferred embodiment of this disclosure and is not intended to limit this disclosure. Various modifications and variations can be made to this disclosure by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for multi-granularity semantic coverage detection and step deviation diagnosis based on knowledge graph ontology constraints, characterized in that, Includes the following steps: (1) Query the standard step sequence S={S1,...,S1} corresponding to the current task from the knowledge graph. m }, and obtain the sequence of steps to be reviewed W={W1,...,W n }; (2) Obtain the semantic slot requirements pre-associated for each standard step from the knowledge graph, and parse each filling step into a set of semantic slots for filling steps based on the knowledge graph ontology. The semantic slots include at least one of the following: action, object, tool, location, condition, and temporal premise. (3) Obtain the pre-associated forced atomic slots, ordinary atomic slots, check items and corresponding weights for each standard step from the knowledge graph; (4) Vectorize the standard step sequence and the step sequence to be reviewed obtained from the knowledge graph query, construct the vector similarity matrix V, and generate each standard step S. i The corresponding candidate step set H i H i It is only used to limit the comparison scope for subsequent fine-grained reviews and is not directly used as the basis for determining the final coverage status; (5) In the candidate step set H i Within the defined scope, for standard step S i Perform inspection item coverage audit and atomic semantic slot coverage audit separately to obtain inspection item coverage score C. i Atomic semantic coverage score A i ; (6) Based on C i A i Standard steps for calculating sequence constraint penalty and forced atomic slot missing penalty S i The integration of coverage i ; (7) Determine whether the coverage is complete, partial, weak or missing based on the fusion coverage score and the forced atomic slot coverage. (8) Based on the coverage status, the quantitative relationship of the candidate filling step set, the reverse correspondence between the filling steps and the standard steps, and the step position mapping relationship, diagnose at least one step deviation type in the following order of priority from high to low: missing steps or missing preceding steps, incorrect order, over-generalization, split expression, effective merging, and redundant steps, and determine the highest priority as the main diagnostic type. (9) The main diagnostic type and auxiliary labels are associated with the corresponding hazards, consequences and control measures through a knowledge graph ontology; The implementation of semantic slot parsing in step (2) includes: a. Perform word segmentation, terminology unification, dependency analysis, or semantic role labeling on the filling steps; b. Map the colloquialisms, abbreviations, or non-standard expressions used in the filling steps to the standardized concepts in the knowledge graph ontology; c. When there is a synonym or hierarchical relationship between the filled step slot and the standard step semantic slot requirement obtained from the knowledge graph, an ontology relation score is generated based on the relation type. The candidate step set H mentioned in step (4) i The implementation includes: a. Calculation standard steps S i With filling in steps W j Vector similarity V between ij Select V ij The filling steps that reach the candidate threshold form an initial candidate set; b. Compare with standard step S i Filling steps involving synonyms, subordinate overlays, or adjacent window relationships are added to the candidate set H. i ; c. When the initial candidate set is empty or the number of candidates is less than the preset number, add the filling steps within the adjacent position window that have reached the supplementary threshold to the candidate set H. i ; The implementation of the inspection item coverage audit and atomic semantic slot coverage audit in step (5) includes: a. Regarding standard procedure S i Obtain the standard step S from the knowledge graph. i The corresponding set of inspection items Q i ={q i1 ,q i2 ,...,q ik } and the corresponding weight u ik Each inspection item q ik Used to determine the candidate step set H i Does the filling step in the form answer the standard step S? i One of the review issues; b. When the candidate step set H i The steps for filling out the form clearly answer the questions in the checklist. ik At that time, c ik Take the full coverage value 1; otherwise c ik Set the full coverage value to 0; c. Calculate the standard step S according to the following formula. i The inspection item coverage score is C. i : C i =Σ(u ik ·c ik ) / Σu ik Among them, C i Indicates standard step S i The degree of coverage at the granularity of the inspection items, u ik c represents the weight of the k-th inspection item. ik This represents the coverage value of the k-th inspection item; d. For standard procedure S i Obtain the standard step S from the knowledge graph. i The corresponding set of atomic semantic slots G i ={g i1 ,g i2 ,...,g il }, where each atomic semantic slot g il Indicates standard step S i The minimum security semantic requirements in the atomic semantic slots include at least one of action, object, tool, position, condition, and timing prerequisite; e. Set the candidate steps H i The filling steps are parsed into a set of semantic slots for filling steps, and based on synonym relationships, hierarchical relationships, inclusion relationships, and forced slot tags in the knowledge graph, the semantic slots for filling steps are matched with the standard steps S obtained from the knowledge graph. i atomic semantic slot g il Alignment is performed to obtain the atomic semantic slot coverage value r. il , where the candidate step set H i The filling steps in the text explicitly cover the atomic semantic slot g. il At that time, r il Take the full coverage value of 1; when the candidate step set H i The filling steps in the text only cover the atomic semantic slot g with generalized concepts, higher-level concepts, or incomplete expressions. il At that time, r il Take partial coverage value p a , of which 0 <p a <1; when the candidate step set H i The filling steps in the code do not cover the atomic semantic slot g. il At that time, r il Take the uncovered value 0; f. Calculate standard step S according to the following formula. i Atomic semantic coverage score A i : A i =Σ(w il ·r il ) / Σw il Among them, A i Indicates standard step S i The degree of coverage at the atomic semantic slot granularity, w il r represents the weight of the l-th atomic semantic slot. il This represents the coverage value of the l-th atomic semantic slot; g. Based on the inspection item coverage score C i Atomic semantic coverage score A i Sequence constraint penalty term R i And the forced atomic slot missing penalty term M i Calculate the standard step S according to the following formula. i The integration of coverage i : Coverage i =λC i +μA i −ρR i −ηM i Where λ, μ, ρ, and η are weighting coefficients; R i Indicates standard step S i Penalties for violating sequence constraints; M i Indicates standard step S i The penalty term resulting from the forced absence of atomic slots; The sequence constraint penalty term R i The methods for determining include: a. Obtain the pre-, post-, adjacent, or non-reversible order constraints between standard steps from the knowledge graph; b. According to standard procedure S i Establish a positional mapping relationship between standard steps and filling steps based on the position of the corresponding candidate filling steps in the sequence of steps to be reviewed W; c. When standard steps S exist in the knowledge graph i Preceded by standard step S k The order constraint, and the filling steps S i The corresponding step is later than S. k When the corresponding step is in position, determine S. i With S k There is a violation of sequence constraints; d. Determine R based on the importance, number of violations, or type of violation of the sequence constraints. i The value of R; where, when there is no order constraint violation, R i =0; when there is a violation of ordinary order constraints, R i Take the first penalty value; when there is a forced inversion of the preceding step, the adjacent step is isolated, or the critical safety step sequence is incorrect, R i Take a second penalty value that is greater than the first penalty value; The forced atomic slot missing penalty term M i The methods for determining include: a. Obtaining standard steps S from the knowledge graph i Forced atomic slot set F i ={f i1 ,f i2 ,...,f it }, where the forced atomic slot represents the minimum security semantic requirement that cannot be omitted and must be explicitly filled in; b. Set the candidate steps H i The filling steps in the semantic slots and the forced atomic slot set F i Perform alignment and determine the position f of each forced atomic slot. it Whether it is explicitly covered; c. When the forced atomic slot f it When explicitly covered, its missing marker m it =0; when the forced atomic slot f it When it is not covered, or is only covered by a superordinate concept, a generalized concept, or an incomplete expression, its missing marker m it =1; d. Calculate the forced atomic slot missing penalty term M according to the following formula. i : M i =Σ(v it ·m it ) / Sv it Among them, v it m represents the missing penalty weight of the t-th forced atomic slot. it This indicates the missing marker for the t-th forced atomic slot; when standard step S... i When there is no forced atomic slot missing, M i =0; The coverage status determination and step deviation priority diagnosis in steps (7) and (8) include: a. For each standard step S i Obtain its candidate step set H i Integrated Coverage i Coverage state, set of missing forced atomic slots, and set of candidate steps H i The steps to fill in are numbered; the preset deviation diagnosis priority is from high to low as follows: missing steps or missing preceding steps, incorrect order, overly generalized merging, split expression, effective merging, and redundant steps; b. When the candidate step set H i Empty, or merged coverage i When the value is below the third threshold, the standard step S will be executed. i Diagnosed as a missing step; when S i For mandatory prerequisite steps and candidate step set H i When the value is empty, it is diagnosed as a missing prerequisite step, and the missing step or the missing prerequisite step is of the first priority. c. When standard steps S exist in the knowledge graph i Preceded by standard step S k The order constraints, and fill in the steps S i The corresponding step is later than S. k When the corresponding step is in position, the standard step S will be... i and standard steps S k The relationship between them is diagnosed as an order error, and the order error is of second priority. d. When a filling step W j If multiple standard steps are involved, but at least one corresponding standard step has a missing forced atomic slot or the fusion coverage is below the first threshold, then the filling step W will be cancelled. j The diagnosis is overgeneralization, which is classified as the third priority. e. When a standard step S i The semantic slots are jointly covered by multiple filling steps, and the fused coverage after the joint process is divided into Coverage. i Not lower than the first threshold, and standard step S i When there is no forced atomic slot missing, the standard step S will be performed. i The diagnosis is a split expression, which is of fourth priority. f. When a filling step W j There are multiple standard steps, and this filling step W j When the forced atomic slots of each of the multiple standard steps are effectively covered, the filling step W is performed. j The diagnosis is a valid confluence, which is classified as the fifth priority. g. When filling in step W j If it does not correspond to any standard step, then fill in step W. j The diagnosis is a redundant step, which has the sixth priority. When the same diagnostic object meets multiple step deviation type conditions at the same time, the one with the highest priority is determined as the main diagnostic type, and the remaining types are output as auxiliary labels.

Images