Intelligent Inspection Method and System for Engineering Indicators Based on Retrieval Generation

Abstract

This invention discloses an intelligent verification method and system for engineering indicators based on retrieval, comprising: acquiring engineering text to be verified; retrieving engineering indicators to be verified from the engineering text; performing a first validity check and a second validity check on the engineering indicators to be verified; if both the first validity check and the second validity check using the context distribution vector pass, then the engineering indicator to be verified is determined to be a valid indicator; this invention improves the level of judgment on the standardization of engineering indicators by constructing a structural expression vector, and introduces field distribution features based on the distribution of condition fields, solving the problem of lack of condition integrity judgment basis in the prior art by calculating the positional relationship between condition fields in statement number and the character length between them; finally, by combining the dual verification of structural expression vector and context distribution vector, the level of validity judgment of engineering indicators in complex document environments is improved.

Description

Technical Field

[0001] This invention relates to the field of engineering indicator verification, specifically to an intelligent verification method for engineering indicators generated through retrieval. Background Technology

[0002] During the engineering design and review process, engineering documents such as design specifications, construction drawings, and foundation calculation sheets typically contain a large amount of critical engineering indicator information, such as "pile foundation depth is 25 meters" and "concrete strength grade is C30." These indicators are important reflections of engineering quality and design intent, and need to be accurately extracted and automatically verified. Existing engineering indicator verification technologies mainly identify and judge indicators in documents through keyword extraction and field matching. These methods usually only focus on whether the numerical value and unit of the indicator are reasonable and consistent with the standard value, but they have the following shortcomings:

[0003] Engineering specifications are often not presented in isolation in documents; their reasonableness frequently depends on the geological conditions, structural type, and construction methods described in the text. For example, whether "the pile foundation depth is 25 meters" is appropriate needs to be judged in conjunction with contextual information such as "the stratum is soft soil" or "the foundation type is friction pile." However, most existing methods do not extract these conditional fields, which can easily lead to a lack of basis for judgment.

[0004] Therefore, existing technologies lack contextual recognition of indicator conditions when verifying engineering indicators, and also lack means to determine whether indicator conditions are fully expressed or close to the indicator fields, resulting in incomplete verification results that are difficult to meet the actual needs of engineering review. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent verification method and system for engineering indicators based on retrieval and generation. This method solves the technical problems mentioned in the background by introducing engineering condition fields of engineering indicators and context distribution vectors of engineering condition fields.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] The intelligent verification method for engineering indicators generated by retrieval includes the following steps:

[0008] S1. Obtain the project text to be verified;

[0009] S2. Retrieve the project indicators to be verified from the project text to be verified;

[0010] S3. Perform the first validity check and the second validity check on the indicators of the project to be verified.

[0011] If the first validity check passes, then the condition fields are searched in the N standard statements to extract T engineering condition fields; otherwise, the engineering indicators to be checked are determined to be illegal indicators.

[0012] Based on T project condition fields, construct the context distribution vector of the project indicators to be verified;

[0013] The context distribution vector is generated by concatenating the field distribution features of T engineering condition fields based on the numerical order of their condition statement numbers, and is used to characterize the text distribution status of all engineering condition fields.

[0014] The field distribution characteristics are represented by the character length of the standard statement for each of the Q nonconditional statement numbers;

[0015] The non-conditional statement number is used to measure the degree of separation between engineering condition fields, and the character length is used to characterize the degree of text redundancy between engineering condition fields.

[0016] S4. If both the first validity check and the second validity check of the context distribution vector pass, the engineering indicator to be checked is determined to be a valid indicator.

[0017] In some embodiments, retrieving project metrics from the project text to be verified includes:

[0018] S2-1. Perform text cleaning on the engineering text to be verified to generate standardized engineering text; wherein, the standardized engineering text contains N standard statements that have been cleaned by text cleaning.

[0019] S2-2. Perform position identification on N standard statements that have undergone text cleaning, and generate N statement numbers corresponding to the N standard statements;

[0020] S2-3. Based on the numbering order of N statement numbers, perform main field retrieval on N standard statements and extract the main fields of the indicators;

[0021] S2-4. For any indicator subject field, extract the numerical field and unit field that are successively adjacent to it in its standard statement;

[0022] S2-5. Concatenate the main field, numerical field and unit field of the indicator in sequence to generate the engineering indicator to be verified.

[0023] In some embodiments, position identification is performed on N text-cleaned standard statements to generate N statement numbers corresponding to the N standard statements, including:

[0024] S2-2-1. Obtain the order of appearance of N standard statements in the standardized engineering text;

[0025] S2-2-2. Based on the order of appearance, assign unique and incremental statement numbers to the N standard statements from top to bottom.

[0026] In some embodiments, the engineering metrics to be validated undergo a first validity check and a second validity check, including:

[0027] S3-1. Perform field featureization on the engineering indicators to be verified, and construct the structural expression vector of the engineering indicators to be verified.

[0028] S3-2. Based on the structural representation vector, perform the first validity check on the engineering indicators to be verified.

[0029] S3-3. Based on the context distribution vector, perform a second validity check on the engineering indicators to be verified.

[0030] In some of these embodiments, a first validity check is performed on the engineering metrics to be validated based on the structural representation vector, including:

[0031] S3-2-1. Construct a structural representation vector library containing multiple verified engineering indicators;

[0032] S3-2-2. In the structural expression vector library, calculate the vector similarity between the engineering indicator to be verified and multiple verified engineering indicators;

[0033] S3-2-3. If the vector similarity meets the first effective threshold, the structural representation vector is determined to be a valid representation, and the first validity check is passed.

[0034] In some embodiments, a second validity check is performed on the engineering metrics to be verified based on the context distribution vector, including:

[0035] S3-3-1. Construct a template library containing multiple context distribution vector templates;

[0036] S3-3-2. Calculate the similarity between the context distribution vector of the engineering indicator to be verified and multiple context distribution vector templates in the context distribution vector template library;

[0037] S3-3-3, Determine whether the similarity of the context distribution vectors meets the second validity threshold;

[0038] S3-3-4. If the similarity meets the second validity threshold, the second validity check passes; otherwise, the engineering indicator to be checked is determined to be an illegal indicator.

[0039] In some embodiments, the context distribution vector of the engineering indicator to be verified is retrieved based on T engineering condition fields, including:

[0040] A1. Anchor the standard statements containing T engineering condition fields, and define the statement number corresponding to the standard statement as the condition statement number;

[0041] A2. Among the T conditional statement numbers, anchor the target statement number corresponding to a project condition field;

[0042] A3. Based on the anchored target statement number, select the condition statement number that is adjacent to it from among N statement numbers;

[0043] A4. Select Q non-conditional statement numbers between the anchored target statement number and its adjacent conditional statement number.

[0044] A5. Obtain the Q standard statements corresponding to the unconditional statement numbers, and calculate the character length of the Q standard statements;

[0045] A6. Characterize the anchored condition statement number and the character length of Q standard statements to generate the field distribution characteristics of the anchored engineering condition field.

[0046] A7. Traverse the T project condition fields and repeatedly generate field distribution features until the field distribution features of the T project condition fields are obtained;

[0047] A8. The field distribution characteristics of the T engineering condition fields are concatenated sequentially based on the numbering order of their condition statement numbers to generate the context distribution vector of the engineering indicators to be verified.

[0048] This invention provides an intelligent verification method for engineering indicators based on retrieval, which has the following beneficial effects:

[0049] The intelligent verification method for engineering indicators based on retrieval proposed in this invention constructs a structural expression vector, which combines and encodes the main field, numerical field, and unit field of the engineering indicator in a preset order to form a standardized expression. This method can effectively detect structural anomalies in engineering text and improve the verification capability of engineering indicators at the structural level.

[0050] Furthermore, by retrieving condition fields related to the indicators from the engineering text, the contextual conditions affecting indicator judgment were completed, thus solving the problem of indicators lacking supporting conditions in traditional methods.

[0051] Furthermore, by introducing field distribution features, calculating the spacing between these fields in the statement number and the character length of the non-conditional statements between them, we can determine whether the conditional information is expressed in a concentrated manner, thus solving the problem that existing methods cannot identify whether the conditional statements are coherent and focused.

[0052] Ultimately, this invention encodes the condition fields and their distribution information to form a context distribution vector that reflects the expression of engineering indicators in the text. It then combines this vector with the structural expression vector for double verification, simultaneously verifying the structural integrity of the indicators and the completeness of the condition descriptions, thereby improving the level of verification of the validity of indicators in complex engineering documents.

[0053] Secondly, this invention provides an intelligent inspection system for engineering indicators based on retrieval enhancement, comprising:

[0054] The project text acquisition unit is used to acquire the project text to be verified.

[0055] The indicator generation unit is used to retrieve the indicators of the project to be verified from the project text to be verified.

[0056] The validity verification unit is used to perform the first validity verification and the second validity verification on the engineering indicators to be verified.

[0057] If the first validity check passes, then the condition fields are searched in the N standard statements to extract T engineering condition fields; otherwise, the engineering indicators to be checked are determined to be illegal indicators.

[0058] Based on T project condition fields, construct the context distribution vector of the project indicators to be verified;

[0059] The context distribution vector is generated by concatenating the field distribution features of T engineering condition fields based on the numerical order of their condition statement numbers, and is used to characterize the text distribution status of all engineering condition fields.

[0060] The field distribution characteristics are represented by the character length of the standard statement for each of the Q nonconditional statement numbers;

[0061] The non-conditional statement number is used to measure the degree of separation between engineering condition fields, and the character length is used to characterize the degree of text redundancy between engineering condition fields.

[0062] The indicator determination unit is used to determine the indicator to be verified as a valid indicator if both the first validity check and the second validity check of the context distribution vector pass.

[0063] Compared with the prior art, the beneficial effects of the intelligent inspection system for engineering indicators based on retrieval enhancement of the present invention are the same as those of the intelligent inspection method for engineering indicators based on retrieval, so they will not be repeated here. Attached Figure Description

[0064] Figure 1 This is a flowchart illustrating the intelligent inspection method for engineering indicators based on retrieval generated according to the present invention.

[0065] Figure 2This is a schematic diagram of the process for generating the engineering indicators to be verified according to the present invention;

[0066] Figure 3 This is a schematic diagram of the validity verification process described in this invention;

[0067] Figure 4 This is a schematic diagram of the context distribution vector generation process described in this invention;

[0068] Figure 5 This is a structural block diagram of the intelligent inspection system for engineering indicators based on retrieval, as described in this invention. Detailed Implementation

[0069] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] Please see Figures 1 to 4 This invention provides an intelligent verification method for engineering indicators based on retrieval, comprising the following steps:

[0071] S1. Obtain the project text to be verified;

[0072] For example, the engineering text refers to natural language text extracted from engineering project-related documents, including structural design specifications, construction drawing specifications, foundation design calculation sheets, etc., and the file format can be PDF, Word, etc.

[0073] S2. Retrieve the project indicators to be verified from the project text to be verified;

[0074] S3. Perform the first validity check and the second validity check on the indicators of the project to be verified.

[0075] S4. If both the first validity check and the second validity check of the context distribution vector pass, the engineering indicator to be checked is determined to be a valid indicator.

[0076] In this embodiment, the structural representation vector focuses on the structural composition and semantic rationality of the indicator field itself, while the context distribution vector extracts semantic association features from the context of the statement in which the indicator is located, helping to determine whether the indicator's expression environment in the engineering text is compliant. Through collaborative verification at the structural and contextual levels, the validity determination of the engineering indicator to be verified is achieved.

[0077] For example, in this embodiment, step S2 further includes:

[0078] S2-1. Perform text cleaning on the engineering text to be verified to generate standardized engineering text; wherein, the standardized engineering text contains N standard statements that have been cleaned by text cleaning.

[0079] For example, the text cleaning includes: removing invalid information from the engineering text, such as headers, footers, table titles, etc.; and segmenting and dividing the text into sentences, removing formatting symbols, and converting special symbols and Chinese units (such as "kilogram" and "ton") into standard formats.

[0080] S2-2. Perform position identification on N standard statements that have undergone text cleaning, and generate N statement numbers corresponding to the N standard statements;

[0081] Specifically, the statement number represents a unique statement number that is sequentially increased from top to bottom, indicating that each standard statement has a unique and sequentially arranged statement number;

[0082] S2-3. Based on the numbering order of N statement numbers, perform main field retrieval on N standard statements and extract the main fields of the indicators;

[0083] For example, the main index fields include "pile foundation depth", "concrete strength", "reinforcing bar diameter", "story height", etc.; the field retrieval adopts a string matching method based on the index subject vocabulary to retrieve and extract the index subject fields in the text that are consistent with the vocabulary.

[0084] In practical engineering systems, to improve retrieval accuracy, a hybrid retrieval strategy is often employed, combining multiple retrieval methods to process text. For example:

[0085] Retrieve the indicator "pile foundation depth" from standardized engineering texts; assuming the text is: "Based on the soil structure and engineering requirements, the foundation design adopts a pile foundation depth of 25 meters, which is suitable for the load standard of soft soil layers."

[0086] First, a match is made from a predefined index subject vocabulary (such as "pile foundation depth"), and "pile foundation depth" is successfully retrieved. If "foundation depth" appears in the text, but the vocabulary only contains "pile foundation depth", then by expanding the vocabulary, "foundation depth" can be retrieved as a variant of "pile foundation depth" and correctly identified as an index subject field.

[0087] S2-4. For any indicator subject field, extract the numerical field and unit field that are successively adjacent to it in its standard statement;

[0088] S2-5. Concatenate the main field, numerical field and unit field of the indicator in sequence to generate the engineering indicator to be verified.

[0089] Specifically, the engineering indicator to be verified represents pairing the indicator's main field with its corresponding numerical and unit fields in a predetermined order to ensure their associated combination in the structural representation vector. In this embodiment, the structural representation vector is represented by fields in a triplet structure, specifically:

[0090] For example, (pile foundation depth, 25, meters) constitutes a complete structural expression vector, where "pile foundation depth" is the header field, "25" is the numerical field, and "meters" is the unit field.

[0091] Furthermore, step S2-2 also includes:

[0092] S2-2-1. Obtain the order of appearance of N standard statements in the standardized engineering text;

[0093] S2-2-2. Based on the order of appearance, assign unique and incremental statement numbers to the N standard statements from top to bottom.

[0094] It should be noted that the location identification does not rely on the original paragraph numbering or chapter structure of the engineering text, but is based on the order of the sentences after text cleaning, and is numbered according to their physical position from top to bottom in the standardized engineering text.

[0095] For example, in this embodiment, step S3 further includes:

[0096] S3-1. Perform field featureization on the engineering indicators to be verified, and construct the structural expression vector of the engineering indicators to be verified.

[0097] Specifically, the field characterization refers to encoding and combining the indicator subject field, numerical field, and unit field according to a preset field structure order, and generating a vector representation that characterizes the semantic features of the indicator structure, thus forming a structural expression vector.

[0098] S3-2. Based on the structural representation vector, perform the first validity check on the engineering indicators to be verified.

[0099] Furthermore, step S3-2 also includes:

[0100] S3-2-1. Construct a structural representation vector library containing multiple verified engineering indicators;

[0101] S3-2-2. In the structure expression vector library, calculate the vector similarity between the engineering indicator to be verified and multiple verified engineering indicators; for example, the similarity calculation method is preferably vector Euclidean distance, but other alternative vector distance measurement methods such as cosine similarity can also be used.

[0102] S3-2-3. If the vector similarity meets the first effective threshold, the structural representation vector is determined to be a valid representation, and the first validity check is passed.

[0103] For example, the first effective threshold may be set to a similarity score of not less than 0.85 (when using cosine similarity) or an Euclidean distance of not greater than ε (where ε is an empirically set error range threshold).

[0104] Specifically, in this embodiment, if the first validity check passes, the condition fields are searched in N standard statements to extract T engineering condition fields; otherwise, the engineering indicator to be checked is determined to be an illegal indicator.

[0105] For example, the engineering condition field includes condition information related to the main index field, such as geological type (e.g., soft soil, clay, rock strata), structural type (e.g., frame structure, steel structure), design load (e.g., heavy load, light load), construction method (e.g., pile foundation construction, underground construction), project area (e.g., urban, mountainous area, etc.).

[0106] The condition field retrieval still uses a string matching method; in this embodiment, for the index subject field in each structural expression vector, matching is performed based on a predefined condition field vocabulary to retrieve engineering condition fields related to the structural expression vector.

[0107] Furthermore, based on the T engineering condition fields, a context distribution vector of the engineering indicators to be verified is constructed;

[0108] The steps for constructing the context distribution vector include:

[0109] A1. Anchor the standard statements containing T engineering condition fields, and define the statement number corresponding to the standard statement as the condition statement number;

[0110] The condition statement number is used to identify the specific location of the standard statement containing the engineering condition field in the standardized engineering text.

[0111] A2. Among the T conditional statement numbers, anchor the target statement number corresponding to a project condition field;

[0112] In this step, a target statement number is selected from all condition statement numbers as the central anchor point to construct the upstream and downstream windows of the project's condition field.

[0113] A3. Based on the anchored target statement number, select the condition statement number that is adjacent to it from among N statement numbers;

[0114] It is important to note that the adjacent conditional statement numbers do not refer to visually adjacent statements or sentences that are close together in paragraphs within the standard text. Instead, they refer to the statement numbers that appear most recently before or after the target statement number out of all N statement numbers, and whose corresponding statements contain the engineering condition field. This is used to identify the positions of upstream and downstream conditional fields that are associated with the currently anchored engineering condition field, thereby capturing the contextual structure of the engineering condition field in the standardized engineering text.

[0115] A4. Select Q non-conditional statement numbers between the anchored target statement number and its adjacent conditional statement number.

[0116] The non-conditional statement number represents the statement number that does not contain any engineering condition fields. It is used to measure the degree of separation between engineering condition fields and is a manifestation of context sparsity.

[0117] A5. Obtain the Q standard statements corresponding to the unconditional statement numbers, and calculate the character length of the Q standard statements;

[0118] Character length reflects the degree of text redundancy between condition fields and can be used to help determine the discreteness of the expression of engineering indicators.

[0119] A6. Characterize the anchored condition statement number and the character length of Q standard statements to generate the field distribution characteristics of the anchored engineering condition field.

[0120] Specifically, the characterization refers to quantifying the character distribution of non-conditional statements between the anchored conditional statement number and its adjacent conditional statements. This field distribution characteristic reflects the contextual tightness of the engineering condition field in the standardized engineering text, and thus characterizes whether the indicator to be verified has a structured and clear conditional expression.

[0121] A7. Traverse the T project condition fields and repeatedly generate field distribution features until the field distribution features of the T project condition fields are obtained;

[0122] A8. The field distribution characteristics of the T engineering condition fields are concatenated sequentially based on the numbering order of their condition statement numbers to generate the context distribution vector of the engineering indicators to be verified.

[0123] Ultimately, the context distribution vector comprehensively reflects the text distribution status of all condition fields related to the engineering metrics to be validated. This vector preserves semantic positional relationships and contextual tightness, facilitating the judgment of contextual validity and the assessment of the semantic integrity of the metrics.

[0124] S3-3. Based on the context distribution vector, perform a second validity check on the engineering indicators to be verified.

[0125] In this embodiment, the first validity check is used to determine the structural integrity of the engineering indicators in terms of field combination based on the structural expression vector, ensuring that the engineering indicators to be checked meet the engineering expression specifications in terms of format; while the second validity check constructs a context distribution vector based on the engineering condition fields obtained from the standard text, thereby determining the contextual consistency of the indicator in the current engineering text.

[0126] Furthermore, step S3-3 also includes:

[0127] S3-3-1. Construct a template library containing multiple context distribution vector templates;

[0128] Specifically, the context distribution vector templates of the template library can be constructed by engineering experts based on typical engineering cases, specification requirements, and text distribution experience to create condition field distribution templates that various engineering indicators should meet, and represented in a vector manner, thereby forming a context distribution vector template library with engineering rationality.

[0129] S3-3-2. Calculate the similarity between the context distribution vector of the engineering indicator to be verified and multiple context distribution vector templates in the context distribution vector template library;

[0130] S3-3-3, Determine whether the similarity of the context distribution vectors meets the second validity threshold;

[0131] S3-3-4. If the similarity meets the second validity threshold, the second validity check passes; otherwise, the engineering indicator to be checked is determined to be an illegal indicator.

[0132] When the similarity between the context distribution vector of the engineering indicator to be verified and the verified vector is higher than the preset second validity threshold, it indicates that the distribution of its conditional context has a reasonable expression, and its context structure expression is deemed valid.

[0133] The intelligent verification method for engineering indicators based on retrieval proposed in this invention constructs a structural expression vector, which combines and encodes the main field, numerical field, and unit field of the engineering indicator according to a preset structural order to generate a standardized indicator expression. This method can effectively identify structural standardization issues in engineering texts, such as missing fields, inconsistent units, and abnormal values, thereby improving the accuracy of engineering indicator judgment in terms of structural expression.

[0134] Furthermore, based on a predefined condition field vocabulary and combined with the numbering order of standard statements, this invention automatically retrieves engineering condition fields related to indicators in engineering text, including but not limited to geological type, structural form, load type, construction method, project area, etc., thereby enhancing the conditional semantic coverage capability during engineering indicator verification and solving the problem of ignoring contextual condition constraints in the existing technology for engineering indicator verification.

[0135] Furthermore, to comprehensively measure the distribution of conditional fields in the text, this invention introduces field distribution features. By calculating the interval between statement numbers containing conditional fields and the character length of non-conditional statements in between, it determines whether the conditional information is expressed in a concentrated and logically aggregated manner in the text, thus overcoming the problem that traditional methods cannot perceive the incompleteness of conditional descriptions.

[0136] Finally, the condition fields and their distribution characteristics are vector-encoded to form a context distribution vector representing the contextual state of engineering indicators in the document. This vector is then combined with the structural expression vector to construct a dual verification system. This dual verification system balances the structural standardization and contextual rationality of engineering indicators, improving the effectiveness judgment of indicators in complex engineering documents.

[0137] This invention also provides an intelligent inspection system for engineering indicators based on retrieval enhancement, which is used to implement the above-described method embodiments. Details already described will not be repeated. The terms "module," "unit," and "subunit," etc., used below refer to combinations of software and / or hardware that perform predetermined functions. Although the system described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0138] like Figure 5 As shown, Figure 5 This is a structural block diagram of the intelligent inspection system for engineering indicators based on retrieval enhancement of the present invention. The system includes:

[0139] The project text acquisition unit is used to acquire the project text to be verified.

[0140] The indicator generation unit is used to retrieve the indicators of the project to be verified from the project text to be verified.

[0141] The validity verification unit is used to perform the first validity verification and the second validity verification on the engineering indicators to be verified.

[0142] The indicator determination unit is used to determine the indicator to be verified as a valid indicator if both the first validity check and the second validity check of the context distribution vector pass.

[0143] In the above system, the engineering text to be verified is obtained through the engineering text acquisition unit; the engineering indicators to be verified are obtained through the indicator generation unit; the validity verification unit performs the first validity verification and the second validity verification; and the indicator determination unit determines whether the engineering indicators to be verified are valid indicators. This solves the problem that the existing engineering text verification lacks context recognition of indicator conditions and determines whether the indicator conditions are fully expressed.

[0144] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means.

[0145] The computer-readable storage medium can be any available medium that a computer can access, or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g.,...), etc. DVD ( ), or semiconductor media. Semiconductor media can be solid-state drives (SSDs).

[0146] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interface; the indirect coupling or communication connection of apparatuses or units may be electrical, mechanical, or other forms.

[0147] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. An intelligent inspection method based on search-generated engineering indicators, characterized in that, The method comprises the following steps: S1, obtaining an engineering text to be verified; S2, searching for an engineering index to be verified in the engineering text to be verified; S3, performing first validity verification and second validity verification on the engineering index to be verified; If the first validity verification passes, condition field searching is performed in N standard sentences to extract T engineering condition fields; otherwise, the engineering index to be verified is determined as an illegal index; According to the T engineering condition fields, a context distribution vector of the engineering index to be verified is constructed; The context distribution vector is generated by splicing field distribution features of the T engineering condition fields based on the numbering order of their condition sentence numbers, and is used to represent the text distribution state of all the engineering condition fields; The field distribution features are represented by character lengths of standard sentences of Q non-condition sentence numbers; The non-condition sentence numbers are used to measure the separation degree between the engineering condition fields, and the character lengths are used to represent the text redundancy degree between the engineering condition fields; S4, if the first validity verification and the second validity verification of the context distribution vector both pass, the engineering index to be verified is determined as a valid index.

2. The method of claim 1, wherein, In the engineering text to be verified, the engineering index to be verified is searched, comprising: S2-1, performing text cleaning on the engineering text to be verified to generate a standardized engineering text; wherein the standardized engineering text comprises N standard sentences cleaned by text; S2-2, performing position recognition on the N standard sentences cleaned by text to generate N sentence numbers corresponding to the N standard sentences; S2-3, based on the numbering order of the N sentence numbers, performing subject field searching on the N standard sentences to extract index subject fields; S2-4, for any index subject field, extracting numerical fields and unit fields successively adjacent to it in its standard sentence; S2-5, sequentially splicing the index subject field, the numerical field and the unit field to generate the engineering index to be verified. 3.The method of claim 2, wherein, The position recognition on the N standard sentences cleaned by text generates N sentence numbers corresponding to the N standard sentences, comprising: S2-2-1, obtaining the appearance order of the N standard sentences in the standardized engineering text; S2-2-2, according to the appearance order, assigning unique and increasing sentence numbers to the N standard sentences from top to bottom.

4. The method of claim 1, wherein the method is based on a search-generated engineering index intelligent inspection method. The first validity verification and the second validity verification on the engineering index to be verified comprise: S3-1, field featureizing the engineering index to be verified to construct a structure expression vector of the engineering index to be verified; S3-2, based on the structure expression vector, performing first validity verification on the engineering index to be verified; S3-3, based on the context distribution vector, performing second validity verification on the engineering index to be verified.

5. The method of claim 4, wherein, The first validity verification on the engineering index to be verified based on the structure expression vector comprises: S3-2-1, constructing a structure expression vector library comprising a plurality of verified engineering indexes; S3-2-2, in the structure expression vector library, calculating the vector similarity between the engineering index to be verified and the plurality of verified engineering indexes; S3-2-3, if the vector similarity meets the first validity threshold, it is determined that the structure expression vector is a valid expression, and the first validity check is passed.

6. The method of claim 5, wherein the engineering index intelligent inspection method based on retrieval generation is characterized in that, Based on the context distribution vector, the second validity check of the to-be-checked engineering index is performed, including: S3-3-1, a template library containing a plurality of context distribution vector templates is constructed; S3-3-2, similarity calculation is performed between the context distribution vector of the to-be-checked engineering index and the plurality of context distribution vector templates in the context distribution vector template library; S3-3-3, it is judged whether the similarity of the context distribution vector meets the second validity threshold; S3-3-4, if the similarity meets the second validity threshold, the second validity check is passed; otherwise, the to-be-checked engineering index is determined to be an illegal index.

7. The search-based engineered indicator intelligent inspection method of claim 1, wherein, According to the T engineering condition fields, the context distribution vector of the to-be-checked engineering index is retrieved, including: A1, anchor the standard sentence where the T engineering condition fields are located, and define the sentence number corresponding to the standard sentence as the condition sentence number; A2, in the T condition sentence numbers, anchor the target sentence number corresponding to an engineering condition field; A3, based on the anchored target sentence number, select the condition sentence number adjacent to it in the N sentence numbers; A4, select Q non-condition sentence numbers between the anchored target sentence number and the adjacent condition sentence numbers; A5, obtain Q standard sentences corresponding to the non-condition sentence numbers, and calculate the character length of the Q standard sentences; A6, the anchored condition sentence number and the character length of the Q standard sentences are characterized to generate the field distribution feature of the anchored engineering condition field; A7, traverse the T engineering condition fields, and repeatedly generate the field distribution feature until the field distribution features of the T engineering condition fields are obtained; A8, the field distribution features of the T engineering condition fields are sequentially concatenated based on the number order of their condition sentence numbers to generate the context distribution vector of the to-be-checked engineering index.

8. An intelligent inspection system for engineering indicators based on search augmentation generation, characterized in that, It includes: An engineering text acquisition unit for acquiring a to-be-checked engineering text; An index generation unit for retrieving a to-be-checked engineering index in the to-be-checked engineering text; An effectiveness checking unit for performing first and second validity checks on the to-be-checked engineering index; If the first validity check is passed, condition field retrieval is performed in the N standard sentences to extract T engineering condition fields; otherwise, the to-be-checked engineering index is determined to be an illegal index; According to the T engineering condition fields, the context distribution vector of the to-be-checked engineering index is constructed; The context distribution vector is generated by concatenating the field distribution features of the T engineering condition fields based on the number order of their condition sentence numbers, and is used to represent the text distribution state of all engineering condition fields; The field distribution feature is represented by the character length of the standard sentence of each non-condition sentence number; The non-condition sentence number is used to measure the separation degree between the engineering condition fields, and the character length is used to represent the text redundancy degree between the engineering condition fields; The index determination unit is configured to determine the to-be-verified engineering index as a valid index if both the first validity verification and the second validity verification of the context distribution vector pass.

Images