Bridge disease naming entity normalization processing method and device based on large language model and storage medium

Abstract

The application provides a bridge disease naming entity normalization processing method and device based on a large language model and a storage medium, which comprises the following steps: establishing a bridge disease structured knowledge base; extracting potential bridge disease names and corresponding disease locations and disease description information from a target bridge regular inspection report; traversing a standard disease name set for the potential bridge disease names to obtain a first normalization processing result; calculating the string similarity of each potential bridge disease name and each standard disease name in the standard disease name set; extracting potential bridge disease names and corresponding standard disease names with a string similarity greater than a set threshold; constructing a prompt word for a normalization task for the potential bridge disease names in the to-be-normalized relation group; and inputting the prompt word into a large language model to obtain a second normalization processing result. The processing method provided by the application can quickly normalize bridge disease naming entities without the need for special model training.

Description

Technical Field

[0001] This invention relates to the field of bridge defect entity processing technology, and in particular to a method, apparatus and storage medium for bridge defect named entity standardization processing based on a large language model. Background Technology

[0002] Bridge maintenance units need to standardize the descriptions of bridge defects in periodic inspection reports. This is a prerequisite for effective data statistical analysis and scientific maintenance decision-making. Maintenance units need to integrate periodic inspection reports from multiple bridges over several years for maintenance decision analysis. These reports are often written by inspection engineers from different units and with different backgrounds. Due to differences in the professional backgrounds of inspection engineers, their personal habits, and the dynamic evolution of industry standards, non-standardized descriptions of defects frequently appear in bridge periodic inspection reports. This lack of standardization in bridge defect naming leads to data redundancy and disorder, affecting not only the accuracy of basic mathematical statistics but also hindering the establishment of reliable bridge performance prediction models to support maintenance decisions.

[0003] In the process of realizing this invention, the inventors discovered at least the following problems in the prior art: Existing named entity normalization methods heavily rely on high-quality word embedding models. However, the bridge detection field lacks large-scale text corpora suitable for training high-quality word embedding models, hindering the effectiveness of existing methods in this specialized area. Currently, named entity normalization methods have shifted from traditional dictionary- and rule-based approaches to embedding representation learning-based methods. The core of embedding representation learning-based methods lies in relying on word embedding models trained on large-scale text corpora—only well-trained models can capture the rich semantic information of entities. Taking general domains as an example, corpora such as certain encyclopedias are massive (4.1G in size, containing 745 million characters and 5.42 million unique words), providing ample data support for word vector learning. However, the bridge detection field presents a significant challenge: firstly, it lacks open-source text corpora suitable for word vector training, with existing documents mostly in PDF format and containing a large amount of non-textual content such as charts and formulas; secondly, due to limitations in data format and domain characteristics, it is difficult to construct a bridge maintenance corpus of comparable size to general corpora. In this case, word embedding models trained on small-scale domain corpora often fail to accurately learn the vector representations of entities (names of bridge defects). Consequently, when the vector representations of bridge defect names are difficult to obtain accurately, the normalized methods based on embedding representation learning are ineffective.

[0004] Therefore, a method, device, and storage medium for the normalization of bridge defect named entities based on a large language model are needed to at least partially solve the above-mentioned technical problems. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a method, apparatus and storage medium for bridge defect naming entity standardization processing based on a large language model, so as to at least solve one of the problems in the prior art.

[0006] In a first aspect, embodiments of the present invention provide a method for normalizing bridge defect named entities based on a large language model, the method comprising: A structured knowledge base for bridge defects will be established based on the current standards for highway bridges. This base includes standard defect names and defect reference information consisting of the corresponding bridge type, bridge defect components, and qualitative and quantitative descriptions of defects at different scales. A set of standard defect names will also be established. Extract the names of potential bridge defects and their corresponding locations and descriptions from the periodic inspection reports of the target bridges. For each potential bridge defect name, the standard defect name set is traversed. If there is a standard defect name that is the same as the potential bridge defect name, the corresponding potential bridge defect name is mapped to a normalized defect name as the first normalization result. Otherwise, the string similarity between the potential bridge defect name and each standard defect name in the standard defect name set is calculated. Extract the names of potential bridge defects with a string similarity greater than a set threshold and one or more corresponding standard defect names to form a group of relationships to be normalized. For each potential bridge defect name in the relation group to be normalized, a prompt word is constructed to instruct the large language model to perform the bridge defect named entity normalization task. The prompt word consists of a task instruction, domain reference information, bridge defect information to be normalized, and an output instruction. The domain reference information is generated based on the defect reference information corresponding to all standard defect names in the relation group to be normalized. Input the prompt word into the large language model to obtain the second normalization result, including the matching standard disease name or the standard disease name not matched.

[0007] Secondly, embodiments of the present invention also provide a bridge defect naming entity standardization processing device based on a large language model, the processing device comprising: Memory is used to store executable instructions for a computer; A processor, configured to execute computer-executable instructions stored in the memory, and to implement the above-described technical solution.

[0008] Thirdly, embodiments of the present invention also provide a storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute the processing method of the above-described technical solution.

[0009] According to the processing method of this invention, a structured knowledge base for bridge defects, including standard defect names and corresponding defect reference information, is established. For each potential bridge defect name, the set of standard defect names is traversed to determine whether there is a standard defect name identical to the potential bridge defect name. For those that do not exist, the string similarity between the potential bridge defect name and each standard defect name in the set of standard defect names is calculated. Potential bridge defect names with string similarity greater than a set threshold and their corresponding standard defect names are extracted. For each potential bridge defect name in the relation group to be normalized, a prompt word is constructed to instruct the large language model to perform the bridge defect named entity normalization task. The prompt word is input into the large language model to obtain the second normalization processing result. The first normalization processing result is combined to form the final normalization result. According to the processing method of this invention, bridge defect named entity normalization can be performed quickly without the need for special model training.

[0010] Additional advantages, objects, and features of the invention will be set forth in part in the description which follows, and will also become apparent in part to those skilled in the art upon studying the description, or may be learned by practice of the invention. The objects and other advantages of the invention can be realized and obtained by means of the structures specifically pointed out in the description and drawings.

[0011] Those skilled in the art will understand that the objectives and advantages achievable with the present invention are not limited to those specifically described above, and that the above and other objectives achievable with the present invention will become clearer from the following detailed description. Attached Figure Description

[0012] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, are not intended to limit the scope of the invention. The components in the drawings are not drawn to scale but are merely illustrative of the principles of the invention. For ease of illustration and description of certain parts of the invention, corresponding portions in the drawings may be enlarged, i.e., may appear larger relative to other components in an exemplary device actually manufactured according to the invention. In the drawings: Figure 1 This is a flowchart of a processing method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a processing apparatus according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a processing system according to an embodiment of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.

[0014] It should also be noted that, in order to avoid obscuring the invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the invention are shown in the accompanying drawings, while other details that are not closely related to the invention are omitted.

[0015] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0016] It should also be noted that, unless otherwise specified, the term "connection" in this article can refer not only to a direct connection, but also to an indirect connection involving an intermediary.

[0017] In the following description, embodiments of the invention will be illustrated with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar parts, or the same or similar steps.

[0018] First, refer to Figure 1 This application describes a method 100 for bridge defect named entity normalization based on a large language model, according to an embodiment of this application. For example... Figure 1 As shown, the processing method 100 may include steps S110 and S160. Specifically: In step S110, a structured knowledge base for bridge defects is established based on the current standards for highway bridges. This base includes standard defect names and defect reference information consisting of the corresponding bridge type, bridge defect components, and qualitative and quantitative descriptions of defects at different scales. A set of standard defect names is also established.

[0019] In step S120, the names of potential bridge defects and their corresponding locations and descriptions are extracted from the periodic inspection reports of the target bridge.

[0020] In step S130, for each potential bridge defect name, the standard defect name set is traversed. If there is a standard defect name that is the same as the potential bridge defect name, the corresponding potential bridge defect name is mapped to a normalized defect name as the first normalization processing result. Otherwise, the string similarity between the potential bridge defect name and each standard defect name in the standard defect name set is calculated.

[0021] In step S140, potential bridge defect names and one or more corresponding standard defect names with string similarity greater than a set threshold are extracted to form a relationship group to be normalized.

[0022] In step S150, for each potential bridge defect name in the relation group to be normalized, a prompt word is constructed to instruct the large language model to perform the bridge defect named entity normalization task. The prompt word consists of task instructions, domain reference information, bridge defect information to be normalized, and output instructions. The domain reference information is generated based on the defect reference information corresponding to all standard defect names in the relation group to be normalized.

[0023] In step S160, the prompt word is input into the large language model to obtain the second normalization processing result, including the matching standard disease name or the absence of a matching standard disease name.

[0024] In the embodiments of this application, firstly, a structured knowledge base of bridge defects is established based on current highway bridge standards, and a set of standard defect names is established; potential bridge defect names and corresponding defect locations and descriptions are extracted from the periodic inspection reports of the target bridge; for each potential bridge defect name, the set of standard defect names is traversed; if a standard defect name with the same name exists, the corresponding potential bridge defect name is mapped to a standardized defect name as the first standardization result; otherwise, the string similarity between the potential bridge defect name and each standard defect name in the set of standard defect names is calculated; the string similarity is then extracted. Potential bridge defect names exceeding a set threshold and their corresponding one or more standard defect names form a group of relationships to be normalized. For each potential bridge defect name in the group to be normalized, a prompt word is constructed to instruct the large language model to perform the bridge defect named entity normalization task. The prompt word consists of a task instruction, domain reference information, bridge defect information to be normalized, and an output instruction. The domain reference information is generated based on the defect reference information corresponding to all standard defect names in the group to be normalized. The prompt word is input into the large language model to obtain the second normalization processing result, including a matched standard defect name or no matched standard defect name.

[0025] As can be seen from the above description, the processing method according to the embodiments of this application utilizes the defect reference information corresponding to the standard defect name to construct prompt words for potential bridge defect names that meet the requirements, and obtains the corresponding standardized processing results through a general large language model. This method can quickly standardize bridge defect naming entities without the need for special model training.

[0026] The following will combine Figure 1 The above-described steps of the processing method 100 according to the embodiments of this application are described in detail.

[0027] In the embodiments of this application, step S110 establishes a structured knowledge base of bridge defects based on the current standards for highway bridges, including standard defect names and defect reference information consisting of corresponding bridge types, bridge defect components, and qualitative and quantitative descriptions of defects corresponding to different scales, and establishes a set of standard defect names.

[0028] Specifically, the "Standard for Technical Condition Assessment of Highway Bridges" (JTG / T H21—2011) is the current standard for the periodic inspection and technical condition assessment of bridges. This standard includes the standard names of bridge defects and qualitative and quantitative descriptions of defects corresponding to different scales. In this embodiment, the current standard for highway bridges can be the "Standard for Technical Condition Assessment of Highway Bridges" (JTG / T H21—2011).

[0029] Then, a structured knowledge base for bridge defects is established based on current highway bridge standards. This knowledge base includes a large number of standard defect names and defect reference information consisting of corresponding bridge types, defective bridge components, and qualitative and quantitative descriptions of defects at different scales. A set of standard defect names is also established based on this large number of standard defect names.

[0030] For example, .

[0031] In the formula, Indicates the first A standard disease name This represents the total number of standard disease names. This is a set of standard disease names.

[0032] In the embodiments of this application, step S120 involves extracting the names of potential bridge defects and their corresponding locations and descriptions from the periodic inspection reports of the target bridge.

[0033] Specifically, since the defect information in bridge periodic inspection reports is mostly in tabular form, knowledge extraction methods for structured data (such as automatic mapping) can be used to extract potential bridge defect names from the bridge periodic inspection reports. (No. Information such as the name of each potential bridge defect, the location of the defect, and a detailed description of the defect.

[0034] In the embodiments of this application, in step S130, for each potential bridge defect name, the standard defect name set is traversed. If there is a standard defect name that is the same as the potential bridge defect name, the corresponding potential bridge defect name is mapped to a normalized defect name as the first normalization processing result. Otherwise, the string similarity between the potential bridge defect name and each standard defect name in the standard defect name set is calculated.

[0035] Specifically, for each potential bridge defect name, the standard defect name set is traversed. If there is a standard defect name that is the same as the potential bridge defect name, the corresponding potential bridge defect name is mapped to a normalized defect name, which is used as the first normalization processing result.

[0036] When a potential bridge defect name is found to be identical to any standard defect name in the set of standard defect names after traversing the entire set, the string similarity between the potential bridge defect name and each standard defect name in the set is calculated. Assume the set of standard defect names contains... A standard disease name can be obtained through calculation. String similarity.

[0037] Among these methods, the Jaccard similarity coefficient (J) or other similarity metrics can be used for string similarity. The Jaccard similarity coefficient is calculated as follows: in, This refers to the set of characters contained in the name of a potential bridge defect in the periodic inspection report of the target bridge. This represents the set of characters contained in a specific standard disease name within the set of standard disease names. Represents a set , Number of elements in the intersection Represents a set , Number of elements in the union set.

[0038] In the embodiments of this application, in step S140, potential bridge defect names and one or more corresponding standard defect names with string similarity greater than a set threshold are extracted to form a relationship group to be normalized.

[0039] Specifically, continuing from the previous text, when the name of a potential bridge defect is obtained through calculation... String similarity, from this From a set of string similarity scores, strings with similarity scores greater than a predetermined threshold are extracted. For example, one or more strings with similarity scores greater than the predetermined threshold are obtained, and one or more standard defect names corresponding to the potential bridge defect name are obtained. One or more sets of relationships to be normalized can be formed between the potential bridge defect name and the corresponding one or more standard defect names.

[0040] In the embodiments of this application, in step S150, for each potential bridge defect name in the relation group to be normalized, a prompt word is constructed to instruct the large language model to perform the bridge defect named entity normalization task. The prompt word consists of a task instruction, domain reference information, bridge defect information to be normalized, and an output instruction. The domain reference information is generated based on the defect reference information corresponding to all standard defect names in the relation group to be normalized.

[0041] Specifically, the structured knowledge base of bridge defects established in step S110 (containing structured knowledge such as standard defect names and qualitative / quantitative descriptions) is integrated with the specific defect information (including contextual background information such as defect names, defect locations, and detailed descriptions of defects) extracted from the periodic inspection reports of the target bridges in step S120, to design input prompts for the large language model. These prompts are designed to explicitly instruct the large language model to perform the bridge defect named entity normalization task.

[0042] The prompt word has the following four parts: 1) Task instructions: Clearly require the large language model to map the input non-standard defect names (potential bridge defect names) to the standard bridge defect names in the "Technical Condition Assessment Standard for Highway Bridges" (JTG / T H21—2011).

[0043] 2) Domain knowledge injection: Embed standard disease names and their reference information descriptions (such as qualitative and quantitative descriptions) extracted from the structured knowledge base of bridge diseases as standardized domain reference information.

[0044] 3) Contextual information provision: Integrate the names of bridge defects to be standardized, their corresponding locations, and descriptions to provide sufficient basis for the model to understand the semantics of specific instances.

[0045] 4) Output format constraints: The output of the large language model is strictly limited to the corresponding standard disease name. The analysis basis (reason) can also be given for subsequent manual review; or a preset unrecognized label, such as "not matched".

[0046] When designing prompts, the knowledge related to bridge defects in the "Standards for the Assessment of Technical Condition of Highway Bridges" and the contextual information of bridge defects extracted from periodic bridge inspection reports are mostly structured data. Preferably, considering that large language models understand information through text pattern recognition rather than structured parsing, this invention first converts the structured data into text and then embeds it into the prompts. Table 1 below provides an example of a prompt template for a bridge defect named entity normalization task.

[0047] Table 1 By incorporating professional reference information on bridge defects from the "Standards for the Assessment of Technical Condition of Highway Bridges" and contextual information on bridge defects from periodic bridge inspection reports into the prompts, the large language model can understand bridge professional knowledge. Based on its understanding of bridge professional reference information, it can identify the standard names of bridge defects. This eliminates the need for pre-training of a large language model specifically for the bridge field, reducing time and cost costs, and significantly reducing the model's computational requirements.

[0048] For example, here is a specific implementation case: (1) Based on the "Standard for Technical Condition Assessment of Highway Bridges" (JTG / T H21—2011), establish a structured knowledge base covering bridge types, bridge components, bridge defects (i.e. standard defect names), qualitative descriptions of defects, and quantitative descriptions of defects, and store it using a relational database.

[0049] (2) An automatic mapping method was used to extract a defect information from the bridge periodic inspection report: the defect name is "rust swelling", the defect location is "2-3-1# diaphragm", and the detailed description is as follows: .

[0050] (3) Based on the prompt word requirements, prompt words for the bridge defect naming entity standardization task were designed, as shown in Table 2. Due to the large amount of content in the domain knowledge base, "..." is used to represent other defect information.

[0051] Table 2 In an embodiment of this application, step S160 involves inputting the prompt word into the large language model to obtain a second normalization processing result, including a matched standard disease name or no matched standard disease name.

[0052] Specifically, continuing from the previous text, we invoked the mainstream large language model and input the prompts from Table 2 into the model (with the temperature coefficient set to 0.6). The large language model correctly identified the standard name of the "rust expansion" defect in the bridge's periodic inspection report as "steel reinforcement corrosion".

[0053] Furthermore, the analysis is based on the following: According to the "Standard for Technical Condition Assessment of Highway Bridges" (JTG / TH21—2011), the term "rust expansion" is a non-standard term. It describes the expansion caused by steel reinforcement corrosion (such as expansion or cracks on the concrete surface), directly corresponding to the "steel reinforcement corrosion" defect in the standard. This defect applies to the general superstructure components (including diaphragms) of concrete beam bridges. The standard explicitly describes "significant surface expansion rust layer" at scale 4, which aligns with the input specific manifestation (area of ​​rust expansion). They are consistent in their phenomena.

[0054] Based on the above description, the processing method 100 according to the embodiments of this application effectively overcomes the bottleneck of scarce corpus in the bridge entity domain and significantly improves the accuracy of bridge defect entity standardization. Traditional methods rely on large-scale domain corpora to train high-quality word embedding models, but the lack of structured text resources in the bridge detection domain makes it difficult to learn entity vector representations. This invention transforms the structured knowledge (such as standard defect names and qualitative / quantitative descriptions) in the "Highway Bridge Technical Condition Assessment Standard" into text instructions that can be understood by a large language model through prompt word engineering, and integrates the contextual information (such as defect location and specific defect information) in the bridge inspection report, so that the large language model can accurately understand professional semantics without special training, thereby achieving reliable mapping of non-standard defect names. This invention also reduces the technical implementation threshold and resource dependence, and improves the universality and feasibility of the method. Existing embedding learning-based solutions require a large investment in building domain corpora and training dedicated models, while this invention directly utilizes the almost zero-sample reasoning ability of large language models, and only requires the design of structured prompt words to complete the task. By dynamically embedding prompt words into the professional knowledge base of bridge defects and the contextual information of bridge defects, the reliance on domain-annotated data and model training process is avoided, which greatly reduces technical complexity and resource consumption, and is more suitable for the challenges of scattered data and inconsistent formats in actual engineering scenarios.

[0055] Furthermore, this invention optimizes computational efficiency and cost control, balancing processing speed and accuracy. By introducing a pre-screening mechanism (such as a string similarity matching method), simple cases that perfectly match the standard disease name are prioritized for filtering, and large language models are only invoked for deep semantic analysis when necessary. This strategy significantly reduces the number of costly calls to large models, improving overall processing efficiency while ensuring the accuracy of core tasks.

[0056] refer to Figure 2 A processing apparatus 200 for implementing the processing method according to embodiments of this application includes a processor 210 and a memory 220. The processing apparatus 200 may include one or more processors 210 and one or more memories 220. The memory 220 stores an executable program executed by the processor 210, which, when executed by the processor 210, causes the processor 210 to perform the processing method 100 described above according to embodiments of this application.

[0057] The processor 210 may be a central processing unit (CPU) or other processing units with data processing capabilities and / or instruction execution capabilities.

[0058] The memory 220 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 210 may execute the program instructions to implement the client functions (implemented by the processor) in the embodiments of this application described herein, and / or other desired functions. Various applications and various data may also be stored in the computer-readable storage medium, such as various data used and / or generated by the applications.

[0059] The processing device 200 may also include input devices and output devices, these components being interconnected via a bus system and / or other forms of connection mechanisms. It should be noted that... Figure 2 The components and structure of the processing device 200 shown are merely exemplary and not limiting; the processing device 200 may also have other components and structures as needed.

[0060] The input device can be a device used by a user to input commands, and can include one or more of a keyboard, mouse, microphone, and touchscreen. Furthermore, the input device can also be any interface for receiving information.

[0061] The output device can output various information (e.g., images or sounds) to the outside (e.g., a user), and may include one or more of a display, speaker, etc. Furthermore, the output device can also be any other device with output functionality.

[0062] For example, the example processing apparatus 200 for implementing the processing method 100 according to the embodiments of this application can be applied to terminal devices (such as mobile phones), tablet computers, laptop computers, ultra-mobile personal computers (UMPCs), handheld computers, netbooks, personal digital assistants (PDAs), wearable devices (such as smartwatches, smart glasses, or smart helmets), augmented reality (AR), virtual reality (VR) devices, smart home devices, in-vehicle computers, and other electronic devices. The embodiments of this application do not impose any limitations on this.

[0063] Those skilled in the art can understand the specific operation of the processing device 200 for implementing the processing method 100 according to the embodiments of this application in conjunction with the content described above. For the sake of brevity, the specific details will not be repeated here, but only some main operations of the processor 210 will be described.

[0064] In one embodiment of this application, when the executable program is run by the processor 210, the processor 210 performs the following steps: A structured knowledge base for bridge defects will be established based on the current standards for highway bridges. This base includes standard defect names and defect reference information consisting of the corresponding bridge type, bridge defect components, and qualitative and quantitative descriptions of defects at different scales. A set of standard defect names will also be established. Extract the names of potential bridge defects and their corresponding locations and descriptions from the periodic inspection reports of the target bridges. For each potential bridge defect name, the standard defect name set is traversed. If there is a standard defect name that is the same as the potential bridge defect name, the corresponding potential bridge defect name is mapped to a normalized defect name as the first normalization result. Otherwise, the string similarity between the potential bridge defect name and each standard defect name in the standard defect name set is calculated. Extract the names of potential bridge defects with a string similarity greater than a set threshold and one or more corresponding standard defect names to form a group of relationships to be normalized. For each potential bridge defect name in the relation group to be normalized, a prompt word is constructed to instruct the large language model to perform the bridge defect named entity normalization task. The prompt word consists of a task instruction, domain reference information, bridge defect information to be normalized, and an output instruction. The domain reference information is generated based on the defect reference information corresponding to all standard defect names in the relation group to be normalized. Input the prompt word into the large language model to obtain the second normalization result, including the matching standard disease name or the standard disease name not matched.

[0065] The above exemplarily illustrates a processing method 100 according to an embodiment of this application. The following, in conjunction with... Figure 3 The present application describes a processing system 300 provided in another aspect of the embodiments.

[0066] Reference Figure 3 This describes an example processing system 300 for implementing the processing methods of embodiments of this application. The processing system 300 may include a knowledge base building module 310, a report extraction module 320, a traversal and calculation module 330, a relation group formation module 340, a prompt word construction module 350, and a model input module 360. Wherein: The knowledge base establishment module 310 is used to: establish a structured knowledge base for bridge defects based on the current standards for highway bridges, including standard defect names and defect reference information consisting of corresponding bridge types, bridge defect components, and qualitative and quantitative descriptions of defects at different scales, and to establish a set of standard defect names.

[0067] The report extraction module 320 is used to extract the names of potential bridge defects and their corresponding locations and descriptions from the periodic inspection reports of the target bridge.

[0068] The traversal and calculation module 330 is used to: traverse the set of standard defect names for each potential bridge defect name; if there is a standard defect name that is the same as the potential bridge defect name, the corresponding potential bridge defect name is mapped to a normalized defect name as the first normalization processing result; otherwise, the string similarity between the potential bridge defect name and each standard defect name in the set of standard defect names is calculated.

[0069] The relation group forming module 340 is used to extract potential bridge defect names and one or more corresponding standard defect names with string similarity greater than a set threshold, and form relation groups to be normalized.

[0070] The prompt word construction module 350 is used to: construct prompt words for each potential bridge defect name in the relation group to be normalized, instructing the large language model to perform the bridge defect named entity normalization task. The prompt words consist of task instructions, domain reference information, bridge defect information to be normalized, and output instructions. The domain reference information is generated based on the defect reference information corresponding to all standard defect names in the relation group to be normalized.

[0071] The model input module 360 ​​is used to: input the prompt words into the large language model and obtain the second normalization processing result, including the matching standard disease name or the standard disease name that is not matched.

[0072] The processing system 300 proposed in this embodiment of the invention can quickly perform bridge defect name entity standardization without the need for special model training.

[0073] Furthermore, according to embodiments of this application, this application also provides a storage medium on which a computer program is stored. When the computer program is run by a processor, it is used to execute corresponding steps of the processing method 100 of this application. The storage medium may, for example, include a memory card of a smartphone, a storage component of a tablet computer, a hard disk of a personal computer, a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a portable compact disc read-only memory (CD-ROM), a USB memory, or any combination of the above storage media. The computer-readable storage medium may be any combination of one or more computer-readable storage media.

[0074] Furthermore, according to embodiments of this application, this application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the processing method of embodiments of this application.

[0075] Although exemplary embodiments have been described herein with reference to the accompanying drawings, it should be understood that the above exemplary embodiments are merely illustrative and are not intended to limit the scope of this application. Various changes and modifications can be made therein by those skilled in the art without departing from the scope and spirit of this application. All such changes and modifications are intended to be included within the scope of this application as claimed in the appended claims.

[0076] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0077] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed.

[0078] Furthermore, those skilled in the art will understand that although some embodiments described herein include certain features but not others included in other embodiments, combinations of features from different embodiments are intended to be within the scope of this application and form different embodiments. For example, in the claims, any one of the claimed embodiments can be used in any combination.

[0079] It should be noted that the above embodiments are illustrative of this application and not restrictive, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. This application can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.

[0080] The above description is merely a specific embodiment or illustration of the embodiments of this application. The scope of protection of this application is not limited thereto. Any variations 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. The scope of protection of this application shall be determined by the scope of the claims.

Claims

1. A method for normalizing bridge defect named entities based on a large language model, characterized in that, The processing method includes: A structured knowledge base for bridge defects will be established based on the current standards for highway bridges. This base includes standard defect names and defect reference information consisting of the corresponding bridge type, bridge defect components, and qualitative and quantitative descriptions of defects at different scales. A set of standard defect names will also be established. Extract the names of potential bridge defects and their corresponding locations and descriptions from the periodic inspection reports of the target bridges. For each potential bridge defect name, the standard defect name set is traversed. If there is a standard defect name that is the same as the potential bridge defect name, the corresponding potential bridge defect name is mapped to a normalized defect name as the first normalization result. Otherwise, the string similarity between the potential bridge defect name and each standard defect name in the standard defect name set is calculated. Specifically, calculating the string similarity between the potential bridge defect name and each standard defect name in the standard defect name set refers to: The Jaccard similarity coefficient is used to calculate string similarity. in, This refers to the set of characters contained in the name of a potential bridge defect in the periodic inspection report of the target bridge. This represents the set of characters contained in a specific standard disease name within the set of standard disease names. Represents a set , Number of elements in the intersection Represents a set , Number of elements in the union; Extract the names of potential bridge defects with a string similarity greater than a set threshold and one or more corresponding standard defect names to form a group of relationships to be normalized. For each potential bridge defect name in the relation group to be normalized, a prompt word is constructed to instruct the large language model to perform the bridge defect named entity normalization task. The prompt word consists of a task instruction, domain reference information, bridge defect information to be normalized, and an output instruction. The domain reference information is generated based on the defect reference information corresponding to all standard defect names in the relation group to be normalized. Among them, the domain reference information generated from the disease reference information corresponding to a certain standard disease name is specifically manifested as follows: A certain type of bridge with a certain bridge defect component may have a standard defect name. When a qualitative or quantitative description appears, the scale of the defect is a scale. Input the prompt word into the large language model to obtain the second normalization result, including the matching standard disease name or the standard disease name not matched.

2. The processing method according to claim 1, characterized in that, The bridge defect information to be standardized is generated based on the potential bridge defect name, the corresponding defect location, and defect description information.

3. The processing method according to claim 1, characterized in that, By employing a knowledge extraction method oriented towards structured data, a structured knowledge base for bridge defects is established. Potential bridge defect names, corresponding defect locations, and defect descriptions are extracted from periodic inspection reports of target bridges and pre-converted into text data to construct the prompt words through embedding.

4. The processing method according to claim 2, characterized in that, The bridge defect information to be standardized is generated based on the potential bridge defect names, corresponding defect locations, and defect descriptions, specifically as follows: The {potential bridge defects} extracted from the periodic inspection reports of the target bridge occur at {defect locations}, and are specifically described as {defect description information}.

5. The processing method according to claim 1, characterized in that, The established set of standard disease names is represented as follows: In the formula, Indicates the first A standard disease name This represents the total number of standard disease names. This is a set of standard disease names.

6. The processing method according to claim 1, characterized in that, The threshold value for string similarity is set to 0.8, and the temperature coefficient of the large language model is set to 0.

6.

7. A bridge defect naming entity standardization processing device based on a large language model, characterized in that, The processing device includes: Memory is used to store executable instructions for a computer; A processor, when executing computer-executable instructions stored in the memory, implements the processing method according to any one of claims 1 to 6.

8. A storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the processing method according to any one of claims 1 to 6.

Images