Inspection work order intelligent quality inspection method and system based on semantic features

Abstract

The invention discloses an intelligent inspection work order quality inspection method and system based on semantic features, and relates to the technical field of inspection work order quality inspection.The method comprises the steps that historical inspection work order data are read, and an identification noun candidate set is recognized; calculating a synonymous expression updating frequency set of the identification noun candidate set; dividing the identification noun candidate set into multiple types of identification noun candidate sets according to the size of the synonymous expression updating frequency set; and connecting an inspection work order quality inspection system, performing priority rule quality inspection on the identification noun set of the current inspection work order, and outputting an identification noun quality inspection return result. The technical problems that in the prior art, due to the fact that a single quality inspection rule is adopted for inspection work order identification nouns, expression changes of nouns of different active degrees cannot be adapted, the quality inspection efficiency is low, and the accuracy is insufficient are solved, and the technical effects that the nouns of different active degrees are accurately adapted, and the quality inspection efficiency and accuracy are improved are achieved.

Description

Technical Field

[0001] This invention relates to the field of audit work order quality inspection technology, specifically to an intelligent quality inspection method and system for audit work orders based on semantic features. Background Technology

[0002] In the field of audit work order processing, work order quality inspection is a crucial link in ensuring the standardized execution of business procedures and improving service quality. Its core requires verifying the standardization of the terminology used in the work orders. Currently, traditional quality inspection methods largely rely on manual verification, which not only consumes significant manpower but also suffers from low efficiency and errors when dealing with massive volumes of work orders due to subjective judgment differences and fatigue. While some automated quality inspection tools have been introduced, they often employ single keyword matching rules, failing to consider the varying activity levels of terminology in actual business operations. A single rule cannot adapt to the verification needs of terms with different activity levels; either rigid rules lead to misjudgments of highly active terms, or insufficient coverage results in missed detections of inactive terms. Ultimately, it is difficult to balance efficiency and accuracy in quality inspection, failing to meet the large-scale and refined quality inspection needs of audit work orders.

[0003] The existing technology uses a single quality inspection rule for the identification terms of inspection work orders, which cannot adapt to the changes in the expression of terms with different activity levels, resulting in technical problems such as low quality inspection efficiency and insufficient accuracy. Summary of the Invention

[0004] This application provides an intelligent quality inspection method and system for audit work orders based on semantic features, which addresses the technical problem that the existing technology uses a single quality inspection rule for audit work order identifiers, which cannot adapt to the changes in the expression of terms with different activity levels, resulting in low quality inspection efficiency and insufficient accuracy.

[0005] In view of the above problems, this application provides an intelligent quality inspection method and system for audit work orders based on semantic features.

[0006] The first aspect of this application provides an intelligent quality inspection method for audit work orders based on semantic features, the method comprising: Read historical audit work order data, identify the candidate set of identifier terms in the historical audit work order data, where identifier terms are non-technical terms in the work order text; calculate the synonym expression update frequency set of the candidate set of identifier terms based on semantic features; divide the candidate set of identifier terms into multiple categories according to the size of the synonym expression update frequency set, where the multiple categories of candidate sets of identifier terms include at least a stable category, a low-activity category, and a high-activity category; connect to the audit work order quality inspection system, where the audit work order quality inspection system performs priority rule quality inspection on the identifier term set of the current audit work order based on the multiple categories of candidate sets of identifier terms, and outputs the identifier term quality inspection return result, where the rule quality inspection includes static keyword matching rules, dynamic keyword matching rules, and semantic model matching rules.

[0007] A second aspect of this application provides an intelligent quality inspection system for audit work orders based on semantic features, the system comprising: The candidate set identification module is used to read historical audit work order data and identify the candidate set of identifier terms in the historical audit work order data. The identifier terms are non-technical terms in the work order text. The update frequency set calculation module is used to calculate the update frequency set of synonym expressions of the candidate set of identifier terms based on semantic features. The candidate set partitioning module is used to partition the candidate set of identifier terms into multiple categories of identifier term candidate sets according to the size of the synonym expression update frequency set. The multiple categories of identifier term candidate sets include at least stable identifier term candidate sets, low-activity identifier term candidate sets, and high-activity identifier term candidate sets. The rule quality inspection module is used to connect to the audit work order quality inspection system. The audit work order quality inspection system performs priority rule quality inspection on the identifier term set of the current audit work order based on the multiple categories of identifier term candidate sets and outputs the identifier term quality inspection return results. The rule quality inspection includes static keyword matching rules, dynamic keyword matching rules, and semantic model matching rules.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: The system reads historical audit work order data, identifies a candidate set of identifier terms from the historical audit work order data (identifier terms are non-technical terms in the work order text), calculates the synonym expression update frequency set of the candidate identifier terms based on semantic features, divides the candidate identifier terms into multiple categories according to the size of the synonym expression update frequency set, and connects to the audit work order quality inspection system. The audit work order quality inspection system performs priority rule quality inspection on the identifier term set of the current audit work order based on the multiple categories of identifier term candidate sets, and outputs the identifier term quality inspection return results. This achieves the technical effect of accurately adapting to terms with different activity levels, improving the efficiency and accuracy of quality inspection. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A schematic diagram of the intelligent quality inspection method for audit work orders based on semantic features provided in this application embodiment; Figure 2 This is a schematic diagram of the structure of an intelligent quality inspection system for audit work orders based on semantic features, provided in an embodiment of this application.

[0011] Figure labeling: Candidate set identification module 10, update frequency set calculation module 20, candidate set partitioning module 30, rule quality inspection module 40. Detailed Implementation

[0012] This application provides an intelligent quality inspection method and system for audit work orders based on semantic features, which addresses the technical problem that existing technologies use a single quality inspection rule for audit work order identifiers, which cannot adapt to the changing expressions of terms with different activity levels, resulting in low quality inspection efficiency and insufficient accuracy.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0014] Example 1, as Figure 1 As shown, this application provides an intelligent quality inspection method for audit work orders based on semantic features, the method comprising: Step S100: Read historical audit work order data and identify the candidate set of identifier terms in the historical audit work order data. The identifier terms are non-technical terms in the work order text.

[0015] Specifically, firstly, historical audit work order data stored in the database is retrieved through a data interface. This data covers work order text information generated in past audit operations, including work order descriptions, processing object information, and feedback records. Next, the retrieved historical audit work order text is preprocessed using natural language processing techniques such as text cleaning, word segmentation, and part-of-speech tagging to break down and analyze the text, filtering out nouns from the segmentation results. Then, based on the definition that the identifying nouns are non-technical terms in the work order text, technical terms related to the audit business, such as "audit process code" and "violation judgment standard," are excluded. Nouns used to describe general information and are not business-specific, such as "customer name," "feedback date," "contact address," and "submission channel," are retained. Finally, these filtered non-technical terms are integrated and summarized to form a candidate set of identifying nouns corresponding to the historical audit work order data, providing basic data support for subsequent steps such as calculating the frequency of synonym expressions based on semantic features.

[0016] Step S200: Calculate the synonym expression update frequency set of the candidate set of identified nouns based on semantic features.

[0017] Specifically, the obtained candidate set of identifiers is sliced ​​into time-series data segments according to the time dimension. Based on the generation cycle of inspection work orders, such as by month or by quarter, multiple consecutive time-series segments are divided. Each segment corresponds to a subset of candidate identifiers within a specific time period, ensuring that the temporal variation patterns of synonym expressions can be captured. Next, the candidate subset of identifiers in each time-series segment is semantically vectorized. Using a pre-trained semantic model, such as a word embedding model, each identifier is transformed into a high-dimensional semantic feature vector, forming a semantic feature vector set. This set numerically represents the semantic connotation of the term, providing a basis for synonym judgment. Subsequently, based on the semantic feature vector set, the semantic similarity of the same identifier in different time-series segments is calculated. Terms with similarity reaching a preset threshold are grouped into the same set of synonym expressions, determining the composition of synonym expressions corresponding to each time-series segment. Finally, the frequency of each synonym expression in each time slice is counted, and its frequency proportion is calculated to determine the dominant synonym expression in each slice. Then, by comparing whether the dominant synonym expression in adjacent time slices changes, and if the frequency proportion of the changed dominant synonym expression exceeds a preset threshold, the expression update event is recorded. Finally, the update frequency of each identified noun synonym expression is calculated based on the number of update events, and the results are integrated to form a synonym expression update frequency set.

[0018] Step S300: Divide the candidate set of identifiers into multiple candidate sets of identifiers according to the size of the synonym expression update frequency set. The multiple candidate sets of identifiers include at least a stable candidate set of identifiers, a low-activity candidate set of identifiers, and a high-activity candidate set of identifiers.

[0019] Specifically, considering the business scenario characteristics of inspection work orders, the changing patterns of historical synonym expressions, and the requirements for quality inspection accuracy, two frequency thresholds are pre-set: a first threshold and a second threshold. For example, based on past data, the first threshold is set to once per month, and the second threshold is set to three times per month. The specific values ​​can be adjusted according to actual business needs. Next, the calculated synonym expression update frequency set is retrieved, and the synonym expression update frequency value corresponding to each candidate synonym is extracted one by one. This value is then compared sequentially with the pre-set first and second thresholds. Subsequently, the candidate set of identifiers is categorized according to the comparison results: If the update frequency of a synonym expression for an identifier is less than the first threshold, it indicates that its synonym expression is stable in the long term and rarely changes, and it is classified into the stable identifier candidate set, such as "customer name," which is only used as "customer name" for a long time without any other synonym updates; if the update frequency is greater than or equal to the first threshold but less than the second threshold, it indicates that its synonym expression is occasionally adjusted and changes at a low frequency, and it is classified into the low-activity identifier candidate set, such as "business number," which occasionally undergoes a brief replacement with synonym expressions like "work order number"; if the update frequency is greater than or equal to the second threshold, it means that its synonym expression changes frequently due to business optimization, expression adjustments, etc., and it is classified into the high-activity identifier candidate set, such as "feedback type," which frequently undergoes synonym updates such as "problem feedback" and "request feedback" due to the addition or merging of business categories. This classification process provides a clear classification basis for triggering differentiated quality inspection rules for different types of identifiers in subsequent steps, ensuring the targeting and efficiency of the quality inspection process.

[0020] Step S400: Connect to the audit work order quality inspection system. The audit work order quality inspection system performs priority rule quality inspection on the set of identifiers of the current audit work order based on the candidate set of multiple identifiers, and outputs the identifier quality inspection return result. The rule quality inspection includes static keyword matching rules, dynamic keyword matching rules, and semantic model matching rules.

[0021] Specifically, a connection is established with the audit work order quality inspection system, synchronizing the three categories of candidate identifiers—stable, inactive, and highly active—to this system. This provides a classification basis for subsequent priority rule matching. Next, the audit work order quality inspection system processes the current audit work order text to be inspected, identifying the set of identifiers—the non-technical terms in the work order. It then iterates through each identifier in this set, comparing it with the three candidate identifier sets to determine the type of the current term, such as stable, inactive, or highly active. If the current noun belongs to the stable noun candidate set, the static keyword matching rule is triggered. A preset static keyword library, containing standard expressions of stable nouns, is used to precisely match the current noun with the content of the keyword library, completing the quality inspection. If the current noun belongs to the low-activity noun candidate set, the dynamic keyword matching rule is triggered. A dynamic keyword library, containing the static keyword library and synonymous extended keywords within a preset time window, is used to perform a matching quality inspection on the current noun, covering occasional synonymous expressions of low-activity nouns. If the current noun belongs to the high-activity noun candidate set, the semantic model matching rule is triggered. A semantic model pre-trained based on preset semantic samples is used to calculate the similarity between the current noun and the preset semantic samples, completing the quality inspection through flexible semantic matching. After all nouns have been traversed and matched, the audit work order quality inspection system integrates the quality inspection results for each type of noun, such as successful matching, unsuccessful matching, and the reason for the unsuccessful match, and finally outputs a complete noun quality inspection return result, realizing intelligent and differentiated quality inspection of the nouns in the current audit work order.

[0022] In one possible implementation, step S200 further includes: Step S210: Divide the candidate set of identifiers into time-series data slices to obtain multiple candidate subsets of identifiers corresponding to multiple time-series slices.

[0023] Step S220: Perform semantic vectorization processing on the multiple candidate sets of identifier nouns to obtain a set of semantic feature vectors.

[0024] Step S230: Calculate the similarity of the candidate subset of identifiers corresponding to each time slice according to the semantic feature vector set, and obtain the set of synonym expressions for each time slice based on the similarity calculation results.

[0025] Step S240: Calculate the synonym update frequency based on the synonym expression set of each time slice, and output the synonym expression update frequency set.

[0026] Specifically, by combining the generation cycle of historical audit work orders and the characteristics of business data, the time division standard for time-series slices is determined. For example, based on the work order generation density, the slices can be divided by natural month, quarter, or fixed time window to ensure that each slice can cover a sufficient amount of time-representative work order data. Next, the candidate set of identified historical audit work order identifiers, which consists of non-technical terms in the work order text, is retrieved. According to the preset time division standard, this candidate set is split into multiple continuous and non-overlapping time-series slices, each corresponding to a specific time period. Finally, the identifiers appearing in all historical audit work orders within that time period are extracted from each time-series slice to form a candidate subset of identifiers that corresponds one-to-one with each time-series slice.

[0027] First, a semantic encoding model, such as a word embedding model or a contextual semantic model, pre-trained on a large-scale text corpus is invoked. This model can capture the semantic associations and contextual features of words. Next, for the multiple candidate subsets of identifier nouns obtained, all identifier nouns in each subset are traversed, and each noun is input into the semantic encoding model for processing. The model outputs a corresponding high-dimensional numerical vector. These vectors represent the semantic connotations of different identifier nouns and their semantic distances through differences and associations in numerical dimensions. Finally, the high-dimensional vectors corresponding to all identifier nouns are associated with and stored with the nouns themselves according to their respective temporal slices, and integrated to form a semantic feature vector set covering all identifier nouns, realizing the transformation from textual identifier nouns to numerical semantic feature vectors.

[0028] For each time-series slice, a subset of candidate identifiers is created. Semantic feature vectors corresponding to all identifiers within this subset are extracted from the resulting semantic feature vector set, forming a vector subset that matches the time-series slice. Next, a vector similarity calculation algorithm, such as cosine similarity, is used to perform pairwise calculations on the semantic feature vectors of every two identifiers in the vector subset, obtaining a similarity value representing the degree of semantic association between them. Subsequently, based on the definition of "synonym" in the audit work order business scenario, a preset similarity threshold, such as 0.85, is set. This threshold can be adjusted according to actual semantic matching needs. Identifiers with similarity values ​​greater than or equal to this threshold are judged as synonymous expressions. Finally, after completing pairwise similarity calculations and synonym determination for all identifiers within each time-series slice, the synonymous identifiers are grouped together. All groups are integrated to form a set of synonymous expressions for that time-series slice, achieving clustering and integration of identifiers with the same or similar semantics within the same time period.

[0029] For each time-series slice's set of synonyms, the frequency of each group of synonyms in the corresponding historical audit work orders is counted. The frequency percentage of each group of synonyms is calculated, i.e., the proportion of the group's frequency to the total frequency of all identified terms in the slice. Based on the frequency percentage, the dominant synonym with the highest frequency percentage for each group is determined. If there are cases with the same percentage, common business expressions are considered for further determination. Next, the dominant synonyms of the same type of identified term in two adjacent time-series slices are compared in chronological order: if the dominant synonym of an adjacent slice changes, and the frequency percentage of the new dominant synonym exceeds a preset threshold (e.g., 0.5), ensuring that the change is valid and not a random fluctuation, an expression update event for that identified term is recorded. Subsequently, all time-series slices are traversed, and the total number of expression update events for each identified term throughout the entire time range is counted. The total number is divided by the total number of time-series slices or the total time span to obtain the synonym update frequency for that identified term. Finally, the synonym update frequencies corresponding to all identified terms are integrated in a unified format to form and output a synonym update frequency set.

[0030] In one possible implementation, step S240 further includes: Step S241: Calculate the frequency proportion of each synonym expression based on the synonym expression set of each time slice, and determine the dominant synonym expression based on the frequency proportion of each synonym expression.

[0031] Step S242: Compare the dominant synonym representations of adjacent time slices in the plurality of time slices and record the representation update events.

[0032] Step S243: Calculate the synonym update frequency of each synonym expression based on the number of expression update events, and output the synonym update frequency set.

[0033] Specifically, for each time-series slice, the set of synonym expressions is traversed, and the frequency of each synonym expression in the historical audit work orders covered by the time-series slice is counted. That is, the number of times the expression is used in the work order text. Next, the frequency ratio of each synonym expression is calculated, that is, the ratio of the frequency of the synonym expression to the total frequency of all synonym expressions in the group. This quantifies the usage ratio of the expression in the same group of synonym expressions. Finally, the expression with the highest frequency ratio in each group of synonym expressions is determined as the dominant synonym expression of the group. If multiple expressions have the same frequency ratio and are all the highest, a secondary judgment is made in combination with the standard or commonly used expressions in the audit business scenario. This clarifies the main expression forms of various identifiers in each time-series slice, laying the foundation for comparing expression changes between adjacent slices.

[0034] By capturing the dynamic changes of synonym expressions through temporal comparison, multiple time-series slices are first sorted chronologically to form a continuous time-series sequence. Then, for each group of synonym expressions corresponding to a identified noun, the dominant synonym expression of that group is extracted from two adjacent time-series slices, i.e., the expression with the highest frequency. Subsequently, through string exact matching or semantic similarity verification, for cases where the expression forms are different but the semantics are consistent, it is determined whether there is a difference in the dominant synonym expression of adjacent time-series slices. If the two are inconsistent, and the frequency of the dominant synonym expression in the new time-series slice exceeds a preset validity threshold, such as 50%, to ensure that the change is stable rather than random fluctuations, it is determined that the synonym expression of the identified noun has been updated, an expression update event is recorded, and the corresponding time-series slice time information is associated, thereby accurately capturing the nodes of synonym expression change over time and providing event basis for subsequent update frequency calculation.

[0035] The algorithm iterates through all identified nouns and records the total number of expression update events associated with each noun. This number directly reflects the frequency of changes in its synonyms over the entire time series. Next, it combines the total number of time series slices or the total time span, such as the total number of months, to calculate the synonym update frequency for each noun, which is the ratio of the total number of expression update events to the total number of time series slices or the total time span. This quantifies the rate of change of its synonyms over time. Finally, the synonym update frequencies corresponding to all identified nouns are integrated in a unified data format to form and output a synonym update frequency set. This set clearly presents the dynamic change characteristics of the synonyms for each noun.

[0036] In one possible implementation, step S240 further includes: Compare the dominant synonyms of adjacent time slices among the multiple time slices and record the expression update events.

[0037] The expression update event is an event in which the dominant synonym expression of an adjacent time slice changes and the frequency of the updated dominant synonym expression exceeds a preset threshold.

[0038] Specifically, in the intelligent quality inspection process of audit work orders, the precise capture of synonym expression changes is achieved through dynamic tracking of time-series slices. First, multiple time-series slices are arranged chronologically to form a continuous time series. Then, for each group of synonym expressions corresponding to an identified noun, the dominant synonym expression of that group—that is, the expression with the highest frequency—is extracted from adjacent time-series slices. String comparison or semantic similarity calculation is used to determine if the two are consistent. If the dominant synonym expressions of adjacent slices differ, and the frequency of the updated dominant synonym expression in the new slice exceeds a preset threshold (e.g., 50%) to ensure the stability of the change, an expression update event is recorded and associated with the corresponding identified noun and time-series information. This captures the substantial changes in synonym expressions over time, providing a reliable event basis for subsequent update frequency calculations.

[0039] The definition of expression update events must simultaneously meet two core conditions to ensure the recorded changes have business-level validity and stability. The first condition is a change in the dominant synonym in adjacent time slices. Specifically, for a group of synonyms for the same identifier, the dominant synonym identified in the previous time slice differs substantially in expression form or semantic connotation from the dominant synonym in the next adjacent time slice. Non-substantial changes with different expressions but consistent semantics must be excluded through string matching or semantic similarity verification. The second condition is that the frequency percentage of the updated dominant synonym exceeds a preset threshold. Here, the updated dominant synonym refers to the new dominant synonym identified in the next time slice. Its frequency percentage data within that slice must be retrieved, i.e., the proportion of its frequency to the total frequency of synonyms in the same group, and compared with a preset threshold, such as 50%. The specific value can be adjusted and compared according to the audit business's requirements for expression stability. Only when this frequency percentage exceeds the threshold can it be concluded that the new dominant synonym has formed a stable usage trend within that time period, rather than being a few occasional expressions. Only when both of the above conditions are met can it be judged as a valid expression update event, avoiding misjudging occasional expression fluctuations as valid updates and providing a reliable basis for the accurate calculation of subsequent synonym expression update frequencies.

[0040] In one possible implementation, step S300 further includes: Step S310: Set the first threshold and the second threshold.

[0041] Step S320: According to the size of the synonym expression update frequency set, divide the candidate set of identifiers less than the first threshold into a stable candidate set of identifiers, divide the candidate set of identifiers greater than or equal to the first threshold and less than the second threshold into a low-activity candidate set of identifiers, and divide the candidate set of identifiers greater than or equal to the second threshold into a high-activity candidate set of identifiers.

[0042] Specifically, focusing on threshold settings, and combining the update patterns of synonyms in historical audit work orders, the audit business's need for stable expression (e.g., stable terms need to remain unchanged for a long time, while highly active terms are mostly related to business adjustments), and the numerical distribution characteristics of the synonym update frequency set (e.g., the median and quartiles of statistical frequency data), two key dividing thresholds are determined: a first threshold and a second threshold. For example, if most of the identified terms in historical data are updated less than twice a year, the first threshold can be set to "2 times / year"; if frequently updated terms are mostly updated more than 5 times a year, the second threshold can be set to "5 times / year". The specific values ​​can be dynamically adjusted according to the business scenario to ensure that the thresholds can effectively distinguish identified terms with different update activity levels.

[0043] The classification operation is performed based on the set thresholds and the synonym expression update frequency set. First, the output synonym expression update frequency set is retrieved, and the update frequency value corresponding to each candidate identifier is extracted. Then, the update frequency of each identifier is compared with the first threshold and the second threshold in turn: if the synonym expression update frequency of an identifier is less than the first threshold, it means that its synonym expression has remained stable for a long time and rarely changes, and has existed with only a single expression for a long time, so it is classified into the stable identifier candidate set; if the update frequency is greater than or equal to the first threshold and less than the second threshold, it means that its synonym expression has been updated a little, but the frequency is low, and the synonym expression changes occasionally due to minor adjustments to the recording specifications, so it is classified into the low-activity identifier candidate set; if the update frequency is greater than or equal to the second threshold, it means that its synonym expression changes frequently due to business optimization, classification adjustment, etc., and the synonym expression is frequently updated due to new feedback scenarios, so it is classified into the high-activity identifier candidate set.

[0044] In one possible implementation, step S400 further includes: Step S410: The inspection work order quality inspection system identifies the set of identifiers for the current inspection work order, traverses the set of identifiers for the current inspection work order, and determines the type of the current term.

[0045] Step S420: If the current noun belongs to the stable noun candidate set, trigger the static keyword matching rule for quality inspection; if the current noun belongs to the low-activity noun candidate set, trigger the dynamic keyword matching rule for quality inspection; if the current noun belongs to the high-activity noun candidate set, trigger the semantic model matching rule for quality inspection.

[0046] Step S430: Until the traversal is complete, output the identifier noun quality inspection results and return them.

[0047] Specifically, the current audit work order text awaiting quality inspection is processed using a noun identification logic. Based on text segmentation and non-technical terminology filtering, all nouns are extracted from the work order content to form a set of nouns for the current audit work order. Subsequently, each noun in this set is traversed sequentially and compared with the divided stable noun candidate set, low-activity noun candidate set, and high-activity noun candidate set. Through precise name matching or semantic similarity secondary verification, for possible differences in synonymous expressions, it is determined which candidate set the currently traversed noun belongs to, thus clarifying its type.

[0048] Differentiated quality inspection rules are implemented for different types of identifiers: If the current term belongs to the stable identifier candidate set, the static keyword matching rule is triggered, and the term is compared with the preset static keyword library, which includes standard expressions and fixed synonyms that have been verified over a long period of time, to complete the quality inspection; if it belongs to the low-activity identifier candidate set, the dynamic keyword matching rule is triggered, and a dynamic library that integrates static keywords and recently added synonyms is called to cover a small number of expression changes through flexible matching; if it belongs to the high-activity identifier candidate set, the system triggers the semantic model matching rule. This model is built on the Transformer architecture, uses multiple versions of synonyms of historically high-activity terms as training samples, learns semantic association features through pre-training, and then adapts to the audit work order business scenario through transfer learning. Finally, it can convert the term into a semantic vector and calculate the similarity, thereby realizing flexible matching and quality inspection of frequently changing synonyms.

[0049] After completing quality inspection of all identifiers in the current audit work order and outputting the results, the process continuously iterates through the set of identifiers for the current work order. For each identifier, it sequentially performs type determination and corresponding rule-based quality inspection operations, simultaneously recording the quality inspection status of each identifier, such as qualified, unqualified, its type, and specific anomaly information, such as static matching mismatch or insufficient semantic similarity. Once all identifiers in the set have been processed, the scattered quality inspection records are integrated into structured data, forming complete quality inspection details for each identifier. Finally, the quality inspection return result for that identifier is output, providing specific and traceable evidence for the overall quality assessment of the audit work order.

[0050] In one possible implementation, step S420 further includes: Step S421: If the current noun belongs to the stable identifier noun candidate set, trigger the static keyword matching rule, and use the static keyword library of the static keyword matching rule to read the stable identifier noun candidate set for matching quality inspection.

[0051] Specifically, when a noun is determined to belong to the stable identifier noun candidate set, a static keyword matching rule is immediately triggered. This rule calls a specially constructed static keyword library, which stores the standard expressions of stable identifier nouns confirmed by business specifications and long-standing synonyms. The content is only updated when there are significant adjustments to the underlying business specifications, maintaining high stability. All keyword entries corresponding to the stable identifier noun candidate set are extracted from the static keyword library through a data interaction interface. Then, a string exact match method is used to compare the current noun to be inspected with each entry in the library. If a completely identical match is found, the noun is deemed to have passed the quality inspection; if no match is found, it is marked as unqualified, and specific information regarding discrepancies with the static keyword library's expression is recorded simultaneously, ensuring that the standardization of stable noun expressions is strictly verified.

[0052] In one possible implementation, step S420 further includes: Step S422: Wherein, the dynamic keyword matching rule includes a dynamic keyword library, which includes a static keyword library and a synonym expansion keyword library within a preset time window.

[0053] Step S423: Use the dynamic keyword library of the dynamic keyword matching rules to read the candidate set of low-activity identifiers and perform matching quality checks.

[0054] Specifically, the core of the dynamic keyword matching rules is defined as the dynamic keyword library, which adopts a two-layer integrated architecture: on the one hand, it fully incorporates the static keyword library to ensure stable coverage of the basic standard expressions of inactive identifiers; on the other hand, it includes a synonym expansion keyword library within a preset time window. This expansion library dynamically introduces high-frequency synonyms that have recently appeared in audit work orders, such as in the past three months, through an automatic system mechanism. These synonyms must meet the requirement that their frequency of occurrence exceeds a preset threshold and be confirmed as valid synonym expressions through semantic verification, thus forming a dynamically updated expansion word set. Through the joint use of the static basic word library and the dynamic expansion word library, this dynamic keyword library can both ensure the basic standardization of the expressions of inactive identifiers and cover reasonable changes in their expressions within a preset time range, providing comprehensive and timely source support for subsequent matching quality inspection.

[0055] When a noun is determined to belong to the low-activity candidate noun set, a dynamic keyword matching rule is activated, calling its core component, the dynamic keyword library. This library integrates the basic standard expressions of the static keyword library with synonymous extended keywords within a preset time window, including recently frequently occurring effective synonyms. All keyword entries corresponding to the low-activity candidate noun set are read from the dynamic keyword library via a data interface, including static basic words and dynamic extended words, forming a complete matching word pool. Subsequently, a multi-pattern matching algorithm compares the current noun to be inspected with all entries in the word pool one by one. If the current noun successfully matches any keyword in the word pool, whether it's a static standard expression or a recently frequently occurring synonym, the noun is deemed to have passed the quality inspection. If no match is found, it is marked as unqualified, and the specific information of "no match with static basic words or recently occurring extended words in the dynamic keyword library" is recorded. This ensures both the standardization of the expression of low-activity nouns and flexibly adapts to their limited expression update characteristics.

[0056] In one possible implementation, step S420 further includes: Step S424: The semantic model matching rules include a semantic model pre-trained based on preset semantic samples.

[0057] Step S425: Read the candidate set of highly active identifiers and perform similarity matching quality checks on the preset semantic samples according to the pre-trained semantic model.

[0058] Specifically, the semantic model matching rule, at its core, employs a Transformer architecture and is pre-trained on a semantic model based on preset semantic samples. This model is trained using preset semantic samples of highly active identifying nouns, encompassing their standard expressions and various synonymous variants generated at different times, all of which have been verified for semantic association through business validation. Leveraging the multi-head self-attention mechanism of the Transformer architecture, the model can simultaneously capture the bidirectional dependencies between words and the semantic relationships within the context. Through the pre-training process, it transforms textual expressions into high-dimensional vectors containing deep semantic information. During training, basic language rules are first learned on a general text corpus, and then fine-tuned using synonymous expression samples from the audit work order domain. This allows the model to accurately adapt to the expression characteristics of highly active nouns in business scenarios, ultimately enabling it to determine whether expressions are synonymous through semantic vector similarity calculation. This provides flexible and accurate algorithmic support for the dynamic quality inspection of highly active identifying nouns.

[0059] When it is determined that the current noun belongs to the candidate set of highly active identification nouns, the pre-trained semantic model that has been completed will be called. First, the highly active identification noun to be quality inspected currently and the preset semantic samples, including the standard expressions of this type of noun, historical synonymous variants, and recently added valid expressions, will be input into the model respectively. Through the internal multi-head self-attention mechanism and fully connected layer of the model, the two are transformed into high-dimensional vectors containing deep semantic information, and then the cosine similarity algorithm is used to calculate the similarity value between the two vectors. Subsequently, the calculated similarity value is compared with the preset threshold, which is determined through cross-validation of the business scenario, such as 0.85: If the similarity ≥ threshold, it is determined that the current highly active identification noun and the preset semantic sample are semantically consistent and the quality inspection is qualified; if the similarity < threshold, it is marked as unqualified, and the specific similarity value and the corresponding preset semantic sample are recorded synchronously, which not only adapts to the frequent expression changes of highly active nouns but also ensures the accuracy and traceability of the quality inspection results.

[0060] Embodiment 2, based on the same inventive concept as the semantic feature-based intelligent quality inspection method for inspection work orders in the foregoing embodiment, as Figure 2 shown, this application provides a semantic feature-based intelligent quality inspection system for inspection work orders. The system in the embodiments of this application and the method embodiments are based on the same inventive concept. Among them, the system includes: Candidate set recognition module 10, which is used to read historical inspection work order data and recognize the candidate set of identification nouns in the historical inspection work order data. The identification noun is a non-professional term noun in the work order text.

[0061] Update frequency set calculation module 20, which is used to calculate the synonym expression update frequency set of the candidate set of identification nouns according to semantic features.

[0062] Candidate set division module 30, which is used to divide the candidate set of identification nouns into multiple types of candidate sets of identification nouns according to the size of the synonym expression update frequency set. The multiple types of candidate sets of identification nouns at least include a stable type candidate set of identification nouns, a low-active candidate set of identification nouns, and a highly active candidate set of identification nouns.

[0063] Rule quality inspection module 40, which is used to connect to the inspection work order quality inspection system. The inspection work order quality inspection system performs priority rule quality inspection on the set of identification nouns of the current inspection work order according to the multiple types of candidate sets of identification nouns and outputs the identification noun quality inspection return result. The rule quality inspection includes static keyword matching rules, dynamic keyword matching rules, and semantic model matching rules.

[0064] Furthermore, the system is also used to implement the following functions: The candidate set of identifiers is divided into time-series data slices to obtain multiple candidate subsets of identifiers corresponding to multiple time-series slices; the multiple candidate sets of identifiers are semantically vectorized to obtain a set of semantic feature vectors; similarity is calculated for the candidate subsets of identifiers corresponding to each time-series slice according to the set of semantic feature vectors, and a set of synonym expressions for each time-series slice is obtained based on the similarity calculation results; the synonym expression update frequency is calculated based on the set of synonym expressions for each time-series slice, and a set of synonym expression update frequencies is output.

[0065] Furthermore, the system is also used to implement the following functions: The frequency proportion of each synonym expression is calculated based on the synonym expression set of each time slice, and the dominant synonym expression is determined based on the frequency proportion of each synonym expression; the dominant synonym expressions of adjacent time slices in the multiple time slices are compared, and expression update events are recorded; the synonym expression update frequency of each synonym expression is calculated based on the number of expression update events, and the synonym expression update frequency set is output.

[0066] Furthermore, the system is also used to implement the following functions: Compare the dominant synonyms of adjacent time slices among the multiple time slices and record the expression update event; wherein, the expression update event is an event in which the dominant synonym of adjacent time slices changes and the frequency ratio of the updated dominant synonym exceeds a preset threshold.

[0067] Furthermore, the system is also used to implement the following functions: Set a first threshold and a second threshold; according to the size of the synonym expression update frequency set, divide the candidate set of identifiers less than the first threshold into a stable candidate set of identifiers, divide the candidate set of identifiers greater than or equal to the first threshold and less than the second threshold into a low-activity candidate set of identifiers, and divide the candidate set of identifiers greater than or equal to the second threshold into a high-activity candidate set of identifiers.

[0068] Furthermore, the system is also used to implement the following functions: The audit work order quality inspection system identifies the set of identifier terms for the current audit work order, traverses the set of identifier terms for the current audit work order, and determines the type of the current term. If the current term belongs to a stable set of identifier term candidates, static keyword matching rules are triggered for quality inspection; if the current term belongs to a low-activity set of identifier term candidates, dynamic keyword matching rules are triggered for quality inspection; if the current term belongs to a high-activity set of identifier term candidates, semantic model matching rules are triggered for quality inspection; until the traversal is completed, the identifier term quality inspection results are returned.

[0069] Furthermore, the system is also used to implement the following functions: If the current noun belongs to the stable identifier noun candidate set, the static keyword matching rule is triggered, and the stable identifier noun candidate set is read using the static keyword library of the static keyword matching rule for matching quality inspection.

[0070] Furthermore, the system is also used to implement the following functions: The dynamic keyword matching rules include a dynamic keyword library, which includes a static keyword library and a synonym expansion keyword library within a preset time window; the dynamic keyword library of the dynamic keyword matching rules is used to read the candidate set of low-activity identifiers for matching quality inspection.

[0071] Furthermore, the system is also used to implement the following functions: The semantic model matching rules include a semantic model pre-trained based on preset semantic samples; and a similarity matching quality check is performed between the candidate set of highly active identifiers and the preset semantic samples according to the pre-trained semantic model.

[0072] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0073] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0074] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for intelligent quality inspection of an inspection work order based on semantic features, characterized in that, The method comprises: reading historical inspection work order data, identifying a set of identified noun candidates of the historical inspection work order data, the identified noun being a non-professional term noun of the work order text; calculating a set of synonym expression update frequencies of the set of identified noun candidates according to semantic features; dividing the set of identified noun candidates into a plurality of sets of identified noun candidates according to the sizes of the set of synonym expression update frequencies, the plurality of sets of identified noun candidates including at least a stable set of identified noun candidates, a low-active set of identified noun candidates, and a high-active set of identified noun candidates; connecting an inspection work order quality inspection system, the inspection work order quality inspection system performing priority rule quality inspection on a set of identified nouns of a current inspection work order according to the plurality of sets of identified noun candidates, outputting an identified noun quality inspection return result, and the rule quality inspection including static keyword matching rules, dynamic keyword matching rules, and semantic model matching rules. 2.The method of claim 1, wherein, The method for calculating a set of synonym expression update frequencies of the set of identified noun candidates according to semantic features comprises: performing time series data slicing on the set of identified noun candidates to obtain a plurality of sets of identified noun candidate subsets corresponding to a plurality of time series slices; performing semantic vectorization processing on the plurality of sets of identified noun candidates to obtain a set of semantic feature vectors; performing similarity calculation on the set of identified noun candidate subsets corresponding to each time series slice according to the set of semantic feature vectors, and obtaining a set of synonym expressions of each time series slice according to the similarity calculation result; performing synonym expression update frequency calculation according to the set of synonym expressions of each time series slice, and outputting a set of synonym expression update frequencies. 3.The method of claim 2, wherein, The method for performing synonym expression update frequency calculation according to the set of synonym expressions of each time series slice and outputting a set of synonym expression update frequencies comprises: calculating a frequency proportion of each synonym expression according to the set of synonym expressions of each time series slice, determining a dominant synonym expression based on the frequency proportion of each synonym expression; comparing the dominant synonym expressions of adjacent time series slices in the plurality of time series slices, and recording expression update events; calculating a synonym expression update frequency of each synonym expression according to the number of expression update events, and outputting a set of synonym expression update frequencies. 4.The method of claim 3, wherein, comparing the dominant synonym expressions of adjacent time series slices in the plurality of time series slices, and recording expression update events; The expression update event is a change in the dominant synonym expression of adjacent time series slices, and the frequency proportion of the updated dominant synonym expression exceeds a preset threshold.

5. The method of claim 1, wherein the method further comprises: The method for dividing the set of identified noun candidates into a plurality of sets of identified noun candidates according to the sizes of the set of synonym expression update frequencies comprises: setting a first threshold and a second threshold; dividing the set of identified noun candidates into a stable set of identified noun candidates when the size is less than the first threshold, into a low-active set of identified noun candidates when the size is greater than or equal to the first threshold and less than the second threshold, and into a high-active set of identified noun candidates when the size is greater than or equal to the second threshold. 6.The method of claim 1, wherein the method further comprises: The method for performing priority rule quality inspection on a set of identified nouns of a current inspection work order according to the plurality of sets of identified noun candidates comprises: The inspection work order quality inspection system identifies the identification noun set of the current inspection work order, traverses the identification noun set of the current inspection work order, and judges the type to which the current noun belongs; If the type to which the current noun belongs is a stable identification noun candidate set, a static keyword matching rule is triggered for quality inspection, if the type to which the current noun belongs is a low-active identification noun candidate set, a dynamic keyword matching rule is triggered for quality inspection, and if the type to which the current noun belongs is a high-active identification noun candidate set, a semantic model matching rule is triggered for quality inspection; Until the traversal is completed, the identification noun quality inspection return result is output.

7. The method of claim 6, wherein the semantic feature-based intelligent quality inspection of the inspection ticket is performed by a computer system. If the type to which the current noun belongs is a stable identification noun candidate set, a static keyword matching rule is triggered, and the static keyword library of the static keyword matching rule is used to read the stable identification noun candidate set for matching quality inspection. 8.The method of claim 6, wherein, If the type to which the current noun belongs is a low-active identification noun candidate set, a dynamic keyword matching rule is triggered for quality inspection, the method The dynamic keyword matching rule includes a dynamic keyword library, and the dynamic keyword library includes a static keyword library and a synonym expansion keyword library within a preset time window. The dynamic keyword library of the dynamic keyword matching rule is used to read the low-active identification noun candidate set for matching quality inspection. If the type to which the current noun belongs is a high-active identification noun candidate set, a semantic model matching rule is triggered for quality inspection, the method includes: 9.The method of claim 6, wherein, The semantic model matching rule includes a semantic model pre-trained based on a preset semantic sample; The high-active identification noun candidate set is read according to the pre-trained semantic model to perform similarity matching quality inspection with the preset semantic sample. The system is used to implement the semantic feature-based intelligent inspection work order quality inspection method of any one of claims 1-9, and the system includes:

10. The intelligent quality inspection system for inspection work order based on semantic features, characterized in that, A candidate set identification module is configured to read historical inspection work order data, identify an identification noun candidate set of the historical inspection work order data, and identify a non-professional term noun of a work order text as an identification noun; An update frequency set calculation module is configured to calculate a synonym expression update frequency set of the identification noun candidate set according to semantic features; A candidate set division module is configured to divide the identification noun candidate set into multiple types of identification noun candidate sets according to the size of the synonym expression update frequency set, and the multiple types of identification noun candidate sets at least include a stable identification noun candidate set, a low-active identification noun candidate set, and a high-active identification noun candidate set; A rule quality inspection module is configured to be connected to an inspection work order quality inspection system, the inspection work order quality inspection system performs priority rule quality inspection on an identification noun set of a current inspection work order according to the multiple types of identification noun candidate sets, and outputs an identification noun quality inspection return result, and the rule quality inspection includes a static keyword matching rule, a dynamic keyword matching rule, and a semantic model matching rule. ​

Images