Material information matching method based on attention mechanism
Through the material information matching method based on attention mechanism, using technologies such as Transformer framework and mask language model, the matching difficulties caused by the difference in material description information in the design institute and enterprise databases are solved, and efficient and accurate material information matching is achieved.
Patent Information
- Application Number
- CN202510010586.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-03
- Publication Date
- 2025-05-06
AI Technical Summary
In the early stage of the construction of the industrial park, due to the large difference between the material description information provided by the design institute and the material description in the enterprise material database, the material information matching is time-consuming and labor-intensive, and human errors are prone to occur, which affects the progress and quality of procurement construction.
The material information matching method based on attention mechanism is adopted, and the material description information in the design institute and enterprise databases is collected and marked, data cleaning and feature extraction are carried out, and the information matching model of the Transformer framework is established, and the mask language model and the prediction algorithm below are combined to achieve intelligent matching of material information.
The labor cost and time cost of material information matching are reduced, the accuracy and speed of material information matching is significantly improved, and the accuracy of material information matching can be achieved in 300ms.
Smart Images
Figure CN119939268A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of material information matching, and in particular to a material information matching method based on an attention mechanism. Background Art
[0002] my country's industry is in a stage of rapid development towards digitalization and intelligence, and various emerging technologies and equipment are being used in large quantities in various industrial parks. In the early stages of the construction of large industrial parks, in order to ensure that the factories in the park can be put into production quickly and efficiently, the design institute needs to issue detailed construction drawings, and the enterprise is responsible for the purchase and construction of the required materials. In the early stages of park planning, the material description information provided by the design institute for the same material may differ to varying degrees from the material description in the enterprise's material database. In order to unify material procurement and ensure smooth production and construction, it is necessary to compare the material information provided by the design institute with the material information in the enterprise's material database before purchasing.
[0003] There are significant differences between the material information in the design drawings provided by the design institute and the material information maintained in the enterprise database, mainly reflected in the existence of some redundant information, different expressions of the same specification model, abbreviations and substitutions of some indicators, etc. For example, the same material is expressed as "welded steel pipe SAWH SH / T3405 BE THK=8mm DN400 L245N / BN-GB / T9711" in the design institute, and "carbon steel straight seam welded pipe\Φ406×8L245N GB / T9711" in the enterprise material database, which means that the pipe diameter is expressed differently and some information is redundant.
[0004] However, it is very time-consuming and labor-intensive for business personnel to match material information one by one, and a large number of professional business personnel are required to work together to carry out the matching work. Skilled business personnel can only match about 100 pieces of data a day, and there are matching errors caused by human factors, which lead to problems in the procurement and construction stage, resulting in economic and property losses and affecting the construction period. Summary of the invention
[0005] The present invention provides a material information matching method based on an attention mechanism, which involves text semantic comparison, and intelligently matches the material description information of a design institute with the material description information in an enterprise database, thereby reducing the labor cost and time cost of material information matching.
[0006] The present invention provides a material information matching method based on an attention mechanism, comprising:
[0007] Collect the material description information provided by each design institute and the material description information maintained in the enterprise material database as original data, and perform data matching and labeling on the original data;
[0008] Cleaning the annotated raw data and creating a basic data set for training a preset information matching model, dividing the basic data set into a training set, a validation set and a test set;
[0009] The training set is used to perform preliminary training on a preset information matching model to obtain a basic model, the verification set is used to adjust the basic model, and the test set is used to test the accuracy of the model to establish a target model;
[0010] The material description information provided by the design institute is input into the target model, and the existing material information in the material database with matching semantics is output.
[0011] Furthermore, after the step of performing data cleaning on the annotated raw data and preparing a basic data set for training a preset information matching model and dividing the basic data set into a training set, a validation set and a test set, the method further includes:
[0012] Features that are helpful for learning the preset information matching model are extracted from the original data and screened, and the contribution of the features to the matching results of the preset information matching model is analyzed by combining structured information in the chemical industry and engineering construction fields.
[0013] Furthermore, before the step of using the training set to preliminarily train the preset information matching model to obtain a basic model, using the validation set to adjust the basic model, and using the test set to perform an accuracy test on the model to establish the target model, the step further includes:
[0014] The differences between the standards and description methods in the original data are determined, and the data are organized into structured data in the form of bias items and incorporated into the training process of the preset information matching model.
[0015] Furthermore, in the step of using the training set to preliminarily train the preset information matching model to obtain a basic model, using the validation set to adjust the basic model, and using the test set to test the accuracy of the model to establish the target model, the preset information matching model is built using the Transformer framework and combined with the masked language model and the following prediction algorithm;
[0016] The masked language model assumes that there is an input sequence X, and the sequence after random masking is X'. The input sequence is represented by X = {x1, x2, ..., x n}, randomly mask a set proportion of the marks, and obtain the masked sequence X', which needs to maximize the following formula:
[0017] L MLM =∑xP(x|X';θ)
[0018] Among them, θ represents the model parameter. The larger the value is, the more likely it is to predict the original masked mark and assist in semantic matching.
[0019] The following prediction algorithm model predicts the next sentence of the text. Given two text paragraphs A and B, the following formula is maximized to predict whether B is really the next sentence of A:
[0020] L LSB =∑YP(Y|A,B;θ)
[0021] Among them, Y is a binary classification label, indicating whether B is the next sentence of A.
[0022] Furthermore, in the preset information matching model, the encoding layer of the Transformer framework includes a self-attention layer and a feedforward neural network layer, and the decoding layer includes a self-attention layer, an encoding-decoding attention layer, and a feedforward neural network layer; the self-attention layer in the encoding layer enables the encoder to always pay attention to the impact of each parameter indicator, and the self-attention layer in the decoding layer focuses on the impact of indicators related to the decoding input parameters.
[0023] Furthermore, a multi-head attention layer is introduced in the encoding layer and the decoding layer, and the output of the multi-head attention layer is:
[0024] MultiHead(Q,K,V)=Concat(head1,head2,...,head n )
[0025]
[0026] Among them, Q, K and V represent query, key and value respectively, and is the weight matrix of the i-th attention head, and the concatenation function Concat combines the outputs of different attention heads.
[0027] After the multi-head attention layer, each encoder and decoder layer contains a feedforward neural network, which adds nonlinear capabilities to the model through nonlinear activation functions; a residual connection layer and a normalization layer are placed after each multi-head attention layer and feedforward network layer.
[0028] Furthermore, after the step of inputting the material description information provided by the design institute into the target model and outputting the existing material information in the material database with semantic matching, the step further includes:
[0029] The input data and output data of the target model are manually reviewed and processed for a second time, and the target model performs self-learning optimization based on the processed data.
[0030] The present invention also provides a material information matching device based on an attention mechanism, comprising:
[0031] The collection module is used to collect the material description information provided by each design institute and the material description information maintained in the enterprise material database as original data, and perform data matching and labeling on the original data;
[0032] A partitioning module is used to clean the annotated raw data and prepare a basic data set for training a preset information matching model, and to divide the basic data set into a training set, a validation set and a test set;
[0033] A training module, used to use the training set to perform preliminary training on a preset information matching model to obtain a basic model, use the validation set to adjust the basic model, and use the test set to perform an accuracy test on the model to establish a target model;
[0034] The matching module is used to input the material description information provided by the design institute into the target model and output the existing material information in the material database with semantic matching.
[0035] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.
[0036] The present invention also provides a computer-readable storage medium on which a computer program is stored, and the computer program implements the steps of the above method when executed by a processor.
[0037] The beneficial effects of the present invention are:
[0038] The present invention collects material description information provided by each design institute and material description information maintained in the enterprise material database as original data, and performs data matching, labeling, and data cleaning, and divides the data into a training set, a verification set, and a test set; the training set is used to perform preliminary training on a preset information matching model to obtain a basic model, and the verification set and the test set are used to optimize the model to establish a target model; the material description information provided by the design institute is input into the target model, and the existing material information in the material database with semantic matching is output, thereby reducing the labor cost and time cost of material information matching; the material information matching method based on the attention mechanism proposed in the present invention integrates professional knowledge in the chemical industry and construction production fields, unifies material information of different standards and in different forms, and can significantly improve the accuracy and matching speed of material information matching. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1 The figure is a schematic diagram of a method flow according to an embodiment of the present invention.
[0040] Figure 2 It is a schematic diagram of the structure of the coding layer in the present invention.
[0041] Figure 3 It is a schematic diagram of the structure of the decoding layer in the present invention.
[0042] Figure 4 FIG. 1 is a schematic diagram of a device structure according to an embodiment of the present invention.
[0043] Figure 5 The figure is a schematic diagram of the internal structure of a computer device according to an embodiment of the present invention.
[0044] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION
[0045] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.
[0046] There are many mature algorithms in the field of text semantic comparison, which mainly use OCR to identify the differences in the text in the document. However, they only focus on text similarity and do not consider the text differences caused by different standard expressions, which leads to poor algorithm performance. At the same time, they focus on the field of computer vision, which is different from semantic recognition.
[0047] The present invention proposes a material information matching algorithm based on the attention mechanism, which can unify the description information of the same material from different design institutes, so as to achieve the purpose of quickly comparing the material description information from different sources with the existing material information in the enterprise material database.
[0048] like Figure 1 As shown, the present invention provides a material information matching method based on an attention mechanism, comprising:
[0049] S1. Collect the material description information provided by each design institute and the material description information maintained in the enterprise material database as original data, and have business personnel match and annotate part of the original data.
[0050] The original data mainly comes from a large listed company. When a new factory was built, the material information was provided by various design institutes and maintained in the company's material database. Business personnel also did some data matching and labeling.
[0051] S2. Clean the annotated original data and create a basic data set for training a preset information matching model, dividing the basic data set into a training set, a validation set and a test set.
[0052] Clean the raw data and remove irrelevant and abnormal data. Ensure the uniformity of data format, unit and scope to prevent analysis errors caused by data inconsistency. After completing data preparation, divide the data set into training set, validation set and test set for subsequent model training and evaluation.
[0053] In one embodiment, step S2 also includes feature extraction and evaluation: extracting features that are helpful for model learning from the original data, screening out features that have a greater impact on model performance, and combining with the structured information in the chemical industry and engineering construction fields provided by business personnel to analyze the contribution of the features to the model matching results.
[0054] S3. During the model training process, the business personnel will sort out the differences between some standards and description methods, organize them into structured data, and integrate them into the training process in the form of bias items. The training set is used to perform preliminary training on the preset information matching model to obtain a basic model, and then the validation set is used to evaluate and fine-tune the basic model, and the test set is used to test the accuracy of the model. Finally, the business personnel will check and correct the deviation to establish the target model.
[0055] The preset information matching model is built using the Transformer framework, and combines the masked language model (Masked LM) with the next sentence prediction algorithm (Next Sentence Prediction).
[0056] The masked language model assumes that there is an input sequence X, and the sequence after random masking is X'. The input sequence is represented by X = {x1, x2, ..., x n}, randomly mask a set proportion of the marks, and obtain the masked sequence X', which needs to maximize the following formula:
[0057] L MLM =∑x P(x|X';θ) (1)
[0058] Here, θ represents the model parameter. The larger the value in (1), the better it can predict the original masked mark and assist in semantic matching.
[0059] The following prediction algorithm model predicts the next sentence of the text. Given two text paragraphs A and B, maximize the following formula to predict whether B is really the next sentence of A:
[0060] L LSB =∑YP(Y|A,B;θ) (2)
[0061] Among them, Y is a binary classification label, indicating whether B is the next sentence of A.
[0062] The construction of the encoding layer of the Transformer framework includes a self-attention layer and a feedforward neural network layer, and the decoding layer includes a self-attention layer, an encoding-decoding attention layer, and a feedforward neural network layer, such as Figure 2 The structure of the encoding layer is shown in Figure 1. The self-attention layer in the encoding layer is to make the encoder pay attention to the impact of each parameter indicator at all times, and the self-attention layer in the decoding layer is to pay attention to the impact of the indicators related to the decoding input parameters. The multi-head attention mechanism is introduced to expand the model's ability to focus on different positions and fully reflect the impact of various indicators on the results. The specific structure of the decoding layer is shown in Figure 1. Figure 3 shown.
[0063] Among them, the output of the multi-head attention layer is:
[0064] MultiHead(Q,K,V)=Concat(head1,head2,...,head n )
[0065]
[0066] Among them, Q, K and V represent query, key and value respectively, and is the weight matrix of the i-th attention head, and the concatenation function Concat combines the outputs of different attention heads.
[0067] After the multi-head attention layer, each encoder and decoder layer contains a feedforward neural network, which has the same application at each position, but is independent for different positions. The feedforward neural network adds nonlinear capabilities to the model through nonlinear activation functions. And increases the number of parameters of the model, thereby enhancing the expressive power of the model, allowing the model to learn more complex patterns and features of the input data.
[0068] A residual connection layer and a normalization layer are placed after each self-attention layer and feed-forward network layer. The residual connection layer helps to avoid the gradient vanishing problem in deep networks, while the normalization layer helps to stabilize the training process.
[0069] After establishing the algorithm framework and auxiliary algorithms, it is necessary to adjust the model's hyperparameters, including selecting a suitable activation function, setting an appropriate learning rate, determining the loss function, etc. At the same time, the business personnel's professional knowledge of chemical engineering and engineering construction should be integrated into the model bias item for training. This can ensure that the professional knowledge of the business personnel is effectively utilized, thereby improving the reliability and accuracy of the model.
[0070] S4. After the target model is established, the material description information provided by the design institute is input into the target model, and the existing material information in the material database with semantic matching is output, and the model is submitted to the business personnel for secondary review to increase the reliability of the model. The model is self-learning and optimized according to the review of the business personnel to achieve the matching of all types of material information.
[0071] The present invention uses Transformer to develop a material information matching algorithm model. The model is trained based on the matched design institute material description information and the existing information in the material database, and incorporates the knowledge of business experts in the chemical industry and construction production fields. It has continuous learning capabilities and can achieve self-optimization based on matching feedback information, covering material knowledge not involved.
[0072] The present invention utilizes a semantic matching algorithm model to realize rapid matching of the design institute's material description to the material information in the material database, and the comprehensive time consumption can be shortened to 300ms with an accuracy rate of 98%.
[0073] like Figure 4 As shown, the present invention also provides a material information matching device based on an attention mechanism, comprising:
[0074] Collection module 1, used to collect material description information provided by each design institute and material description information maintained in the enterprise material database as original data, and perform data matching and labeling on the original data;
[0075] A division module 2 is used to clean the annotated raw data and prepare a basic data set for training a preset information matching model, and divide the basic data set into a training set, a validation set and a test set;
[0076] The training module 3 is used to perform preliminary training on the preset information matching model using the training set to obtain a basic model, adjust the basic model using the validation set, and perform accuracy test on the model using the test set to establish a target model;
[0077] The matching module 4 is used to input the material description information provided by the design institute into the target model, and output the existing material information in the material database with semantic matching.
[0078] In one embodiment, it further includes:
[0079] The screening module is used to extract and screen features that are helpful for learning the preset information matching model from the original data, and analyze the contribution of the features to the matching results of the preset information matching model in combination with the structured information in the chemical industry and engineering construction fields.
[0080] In one embodiment, it further includes:
[0081] The determination module is used to determine the differences between the standard and the description method in the original data, and organize them into structured data in the form of bias items to be integrated into the training process of the preset information matching model.
[0082] In one embodiment, in the training module 3, the preset information matching model is constructed using the Transformer framework and combined with the masked language model and the following prediction algorithm;
[0083] The masked language model assumes that there is an input sequence X, and the sequence after random masking is X'. The input sequence is represented by X = {x1, x2, ..., x n}, randomly mask a set proportion of the marks, and obtain the masked sequence X', which needs to maximize the following formula:
[0084] L MLM =∑xP(x|X';θ)
[0085] Among them, θ represents the model parameter. The larger the value is, the more likely it is to predict the original masked mark and assist in semantic matching.
[0086] The following prediction algorithm model predicts the next sentence of the text. Given two text paragraphs A and B, the following formula is maximized to predict whether B is really the next sentence of A:
[0087] L LSB =∑YP(Y|A,B;θ)
[0088] Among them, Y is a binary classification label, indicating whether B is the next sentence of A.
[0089] In one embodiment, in the preset information matching model of the training module 3, the encoding layer of the Transformer framework includes a self-attention layer and a feedforward neural network layer, and the decoding layer includes a self-attention layer, an encoding-decoding attention layer, and a feedforward neural network layer; the self-attention layer in the encoding layer enables the encoder to always pay attention to the impact of each parameter indicator, and the self-attention layer in the decoding layer focuses on the impact of indicators related to the decoding input parameters.
[0090] In one embodiment, in the training module 3, a multi-head attention layer is introduced in the encoding layer and the decoding layer, and the output of the multi-head attention layer is:
[0091] MultiHead(Q,K,V)=Concat(head1,head2,...,head n )
[0092]
[0093] Among them, Q, K and V represent query, key and value respectively, and is the weight matrix of the i-th attention head, and the concatenation function Concat combines the outputs of different attention heads.
[0094] After the multi-head attention layer, each encoder and decoder layer contains a feedforward neural network, which adds nonlinear capabilities to the model through nonlinear activation functions; a residual connection layer and a normalization layer are placed after each multi-head attention layer and feedforward network layer.
[0095] In one embodiment, it further includes:
[0096] The self-learning module is used to perform manual secondary review processing on the input data and output data of the target model, and the target model performs self-learning optimization based on the processed data.
[0097] Each of the above modules is used to execute the corresponding steps in the above-mentioned material information matching method based on the attention mechanism. The specific implementation method thereof is described in the above-mentioned method embodiment and will not be repeated here.
[0098] like Figure 5 As shown, the present invention also provides a computer device, which can be a server, and its internal structure can be as shown in Figure 5 As shown. The computer device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store all data required for the process of the material information matching method based on the attention mechanism. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the material information matching method based on the attention mechanism is implemented.
[0099] Those skilled in the art will understand that Figure 5 The structure shown in is merely a block diagram of a portion of the structure related to the present application solution and does not constitute a limitation on the computer device to which the present application solution is applied.
[0100] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, any one of the above-mentioned material information matching methods based on the attention mechanism is implemented.
[0101] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing related hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided in this application and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
[0102] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.
[0103] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.
Claims
1. A material information matching method based on attention mechanism, characterized in that: include: Collect the material description information provided by each design institute and the material description information maintained in the enterprise material database as original data, and perform data matching and labeling on the original data; Cleaning the annotated raw data and creating a basic data set for training a preset information matching model, dividing the basic data set into a training set, a validation set and a test set; The training set is used to perform preliminary training on a preset information matching model to obtain a basic model, the verification set is used to adjust the basic model, and the test set is used to test the accuracy of the model to establish a target model; The material description information provided by the design institute is input into the target model, and the existing material information in the material database with matching semantics is output.
2. The material information matching method based on the attention mechanism according to claim 1 is characterized in that: After the steps of cleaning the labeled original data and preparing a basic data set for training a preset information matching model, and dividing the basic data set into a training set, a validation set, and a test set, the method further includes: Features that are helpful for learning the preset information matching model are extracted from the original data and screened, and the contribution of the features to the matching results of the preset information matching model is analyzed by combining structured information in the chemical industry and engineering construction fields.
3. The material information matching method based on the attention mechanism according to claim 1 is characterized in that: Before the step of using the training set to preliminarily train the preset information matching model to obtain a basic model, using the validation set to adjust the basic model, and using the test set to perform an accuracy test on the model to establish the target model, the step further includes: The differences between the standards and description methods in the original data are determined, and the data are organized into structured data in the form of bias items and incorporated into the training process of the preset information matching model.
4. The material information matching method based on the attention mechanism according to claim 1 is characterized in that: In the step of using the training set to preliminarily train the preset information matching model to obtain a basic model, using the validation set to adjust the basic model, and using the test set to test the accuracy of the model to establish the target model, the preset information matching model is built using the Transformer framework and combined with the masked language model and the following prediction algorithm; The masked language model assumes that there is an input sequence X, and the sequence after random masking is X'. The input sequence is represented by X = {x1, x2, ..., x n }, randomly mask a set proportion of the marks, and obtain the masked sequence X', which needs to maximize the following formula: L MLM = ∑ x P(x|X';θ) Among them, θ represents the model parameter. The larger the value is, the more likely it is to predict the original masked mark and assist in semantic matching. The following prediction algorithm model predicts the next sentence of the text. Given two text paragraphs A and B, the following formula is maximized to predict whether B is really the next sentence of A: L LSB =∑YP(Y|A,B;θ) Among them, Y is a binary classification label, indicating whether B is the next sentence of A.
5. The material information matching method based on the attention mechanism according to claim 4 is characterized in that: In the preset information matching model, the encoding layer of the Transformer framework includes a self-attention layer and a feedforward neural network layer, and the decoding layer includes a self-attention layer, an encoding-decoding attention layer, and a feedforward neural network layer; the self-attention layer in the encoding layer enables the encoder to always pay attention to the impact of each parameter indicator, and the self-attention layer in the decoding layer focuses on the impact of indicators related to the decoding input parameters.
6. The material information matching method based on the attention mechanism according to claim 5 is characterized in that: A multi-head attention layer is introduced in the encoding layer and the decoding layer, and the output of the multi-head attention layer is: MultiHead(Q,K,V)=Concat(head1,head2,...,head n ) Among them, Q, K and V represent query, key and value respectively, and is the weight matrix of the i-th attention head, and the concatenation function Concat combines the outputs of different attention heads; After the multi-head attention layer, each encoder and decoder layer contains a feedforward neural network, which adds nonlinear capabilities to the model through nonlinear activation functions; a residual connection layer and a normalization layer are placed after each multi-head attention layer and feedforward network layer.
7. The material information matching method based on the attention mechanism according to claim 1 is characterized in that: After the step of inputting the material description information provided by the design institute into the target model and outputting the existing material information in the material database with matching semantics, the method further includes: The input data and output data of the target model are manually reviewed and processed for a second time, and the target model performs self-learning optimization based on the processed data.
8. A material information matching device based on attention mechanism, characterized in that: include: The collection module is used to collect the material description information provided by each design institute and the material description information maintained in the enterprise material database as original data, and perform data matching and labeling on the original data; A partitioning module is used to clean the annotated raw data and prepare a basic data set for training a preset information matching model, and to divide the basic data set into a training set, a validation set and a test set; A training module, used to use the training set to perform preliminary training on a preset information matching model to obtain a basic model, use the validation set to adjust the basic model, and use the test set to perform an accuracy test on the model to establish a target model; The matching module is used to input the material description information provided by the design institute into the target model and output the existing material information in the material database with semantic matching.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.