Material master data semantic matching method and system based on word vector similarity

Abstract

The invention is suitable for the technical field of data management, and provides a material master data semantic matching method and system based on word vector similarity, and the method comprises the following steps: S1, constructing a standard library; S2, preprocessing data: carrying out word segmentation and preprocessing on material master data and text data in the standard library; s3, word vector generation: utilizing a natural language processing technology to generate word vectors for the material main data and words or phrases in the standard library; and S4, similarity calculation: calculating the similarity between the word vector of each material description and characteristic in the material main data and the word vector of the text data in the standard library. According to the word vector similarity-based material master data semantic matching method and system provided by the scheme, the matching accuracy and flexibility are improved, material descriptions with similar semantics but different expressions can be processed, and the matching precision and efficiency are further improved by optimizing the matching result.

Description

Technical Field

[0001] The present invention relates to the field of data management technology, and in particular to a material master data semantic matching method and system based on word vector similarity. Background Art

[0002] In today's digital age, enterprises are increasingly demanding the management of material master data. As one of the company's core data assets, material master data permeates multiple business processes, including procurement, production, and sales. Its accuracy and completeness are crucial to operational efficiency and decision-making quality. However, with the continuous expansion of business operations and the rapid growth of data volumes, material master data management faces numerous challenges.

[0003] Traditional material master data management methods primarily rely on manual matching and simple text matching algorithms. Manual matching is not only inefficient but also susceptible to human error, leading to inaccurate and inconsistent matching results. Simple text matching algorithms, such as those based on keywords, can only compare the surface form of the text and lack the ability to understand the underlying semantic information. Consequently, when processing material master data with complex semantics, numerous missed matches and incorrect matches often occur.

[0004] For example, in some large manufacturing companies, material master data contains detailed information on various raw materials, parts, and finished products. This information may come from different suppliers, departments, and even different countries and regions. Due to factors such as language usage and terminology differences, the same or similar materials may be presented in different text forms. Traditional matching methods struggle to accurately identify the semantic connections between materials in such situations, leading to redundancy and duplication in material master data, increasing management costs and operational risks for the company.

[0005] Furthermore, with the continuous updating of industry standards and the optimization of enterprise business processes, the requirements for semantic matching accuracy of material master data are becoming increasingly stringent. For example, in the chemical industry, different types of chemicals may have similar properties and uses, but their specific composition and indicators may vary. Failure to accurately semantically match these materials can lead to serious consequences such as procurement errors and production accidents.

[0006] Although several semantic-based material master data matching methods have been proposed, most of these approaches have limitations. For one thing, some rely on manually constructed semantic dictionaries or ontologies, which are expensive to build and maintain and struggle to cover all material domains and business scenarios. Other methods, on the other hand, employ complex machine learning algorithms but require large amounts of labeled data for training. However, obtaining labeled data is often difficult in practice.

[0007] Furthermore, some existing semantic matching methods face computational efficiency and performance bottlenecks when processing large amounts of material master data. Because material master data volumes are typically very large, traditional semantic matching methods often require extensive matrix operations and complex logical reasoning when calculating similarities, resulting in excessive computation time and difficulty meeting the real-time or near-real-time data processing needs of enterprises. Furthermore, as data volumes continue to increase, system scalability becomes a significant issue, making existing methods often unable to process and match large amounts of data within a reasonable time and cost.

[0008] Furthermore, existing semantic matching methods also have shortcomings in terms of user interaction and result visualization. Most methods only provide simple matching result outputs, making it difficult for users to intuitively understand and analyze the matching results, and unable to effectively utilize the matching results for subsequent applications such as data integration and knowledge discovery. Furthermore, these methods often lack flexible user interaction features, making it impossible for users to customize the matching process and results according to their actual needs, reducing the system's ease of use and practicality.

[0009] In summary, traditional methods have many problems in semantic matching of material master data, including low matching accuracy, low efficiency, difficulty in adapting to multi-domain scenarios, insufficient ability to process large-scale data, and poor user interaction experience.

[0010] Therefore, developing an efficient, accurate, and adaptable material master data semantic matching method and system that can overcome these shortcomings has become an urgent problem to be solved in the field of enterprise data management.

[0011] To this end, the present invention proposes a material master data semantic matching method and system based on word vector similarity to solve the above technical problems, providing enterprises with a new material master data semantic matching solution to meet the growing data management needs of enterprises. Summary of the Invention

[0012] The present invention provides a material master data semantic matching method and system based on word vector similarity, aiming to solve the technical problem that traditional matching methods often rely on string matching or keyword matching and are not effective when processing material descriptions that are semantically similar but have different expressions.

[0013] The present invention is implemented as follows: a material master data semantic matching method based on word vector similarity includes the following steps: S1. Build a standard library: Collect national and industry standard documents, scan paper documents, and perform OCR character recognition on PDF / JPG format documents; Extract technical parameters from the standard documents, including standard numbers, material names, and technical attributes; Store the extracted data in a database, and the database table contains fields: leaf class code, material name, model specification, and long description; S2. Data preprocessing: Perform word segmentation and preprocessing on the text data in the material master data and the standard library. The preprocessing includes: Splitting the text data into multiple word segments according to semantics; Removing stop words, removing words such as "de", "di", "de" that do not have actual semantic information; And词性标注, tagging the remaining words with词性 to screen out the parts with semantic value; S3. Word vector generation: Use natural language processing technology to input the words or phrases in the material master data and the standard library after word segmentation into a pre-trained word vector model and convert them into corresponding word vectors; The training process of the word vector model is: Collect a large amount of text corpora related to the material master data, including standard documents, technical materials, product descriptions, etc.; Perform preprocessing on the collected corpora, including cleaning, normalization, etc.; Input the preprocessed corpora into the word vector model for training, and adjust the parameters of the model, such as vector dimension, window size, number of iterations, etc., to obtain the best word vector representation; S4. Similarity calculation: Calculate the similarity between the word vectors of each material description and characteristic in the material master data and the word vectors of the text data in the standard library; S5. Matching result screening: Set a similarity threshold. When the similarity exceeds this threshold, it is considered that the text data matches a certain material description or characteristic; This threshold is set manually according to the actual application scenario and business requirements or automatically optimized through machine learning algorithms. Compare the similarity between the calculated standard library word vector and the material master data word vector with this threshold. When the similarity is greater than or equal to the threshold, it is determined that the standard library text data matches the material master data successfully; S6. Matching result optimization: According to business requirements, further optimize the matching results, such as considering the natural attributes of materials and the data controlled by the SAP system, that is, group and sort the matching results by material type, then display the standard document association information and differential attributes through a visualization interface, and then generate a list of material master data revision suggestions and synchronize it to the SAP system; S7. Result display and application: The matching results are displayed to the user in a visual manner and applied to the revision and optimization of the material master data. The steps for visually displaying the matching results are to build an intuitive visual interface to present the similarity values, matching relationships, semantic associations, etc. between the standard library text data and the material master data in the form of charts, graphics, lists, etc., so that users can quickly view, understand and analyze the matching results, improve users' trust in the matching process and results, and support users to perform interactive operations in the visual interface, such as filtering, sorting, querying, etc., so that users can obtain the required information more flexibly, further improving the usability and practicality of the material master data semantic matching system.

[0014] Preferably, the data preprocessing in step S1 also includes noise removal, text cleaning, and text sentence segmentation.

[0015] Preferably, the word vectors in step S2 are generated using Word2Vec or GloVe natural language processing technology.

[0016] Preferably, the similarity calculation in step S3 adopts the cosine similarity and Euclidean distance methods. If cosine similarity is adopted, the ratio of the dot product of the two word vectors to the product of the module lengths of the two word vectors is calculated; if Euclidean distance is adopted, the square root of the sum of the squares of the numerical differences of the corresponding dimensions of the two word vectors is calculated.

[0017] Preferably, the matching result optimization in step S4 further includes limiting the matching range by grouping material types to improve matching accuracy.

[0018] The present invention also proposes a material master data semantic matching system based on word vector similarity, the system comprising: Data preprocessing module, used to preprocess the text data in the standard library and the description and characteristics of the material master data; The word vector generation module is used to generate word vectors for words or phrases in the material master data and standard library. This module uses word vector models from natural language processing technologies such as word2vec or glove to train the standard library text data processed by the data preprocessing module and generate a corresponding set of standard library word vectors. Each piece of standard library text data corresponds to a word vector. Similarity calculation module, used to calculate the similarity between the word vectors of each material description and characteristic in the material master data and the word vectors of the text data in the standard library; The matching result screening module is used to set a similarity threshold based on actual application scenarios and business needs, and compare the similarity obtained by the similarity calculation module with the threshold. When the similarity is greater than or equal to the threshold, it is determined that the standard library text data and the material master data are successfully matched; The matching result optimization module is used to collect user feedback information and matching result evaluation data, and optimize the threshold using machine learning algorithms or manual adjustment to improve the accuracy and adaptability of matching results; The result display and application module is used to display the matching results to users in a visual manner and apply them to the revision and optimization of material master data.

[0019] Preferably, the data preprocessing module also includes a word segmentation optimization submodule, which adopts a word segmentation algorithm based on a combination of dictionary matching and statistical language model, wherein the dictionary contains professional vocabulary, industry terms and common abbreviations related to materials, and the statistical language model is used to identify unregistered words to improve the recognition ability of professional terms and special expressions in material master data, and ensure the accuracy and completeness of the word segmentation results.

[0020] Preferably, the word vector generation module also includes a corpus training submodule, which is used to train the word vector model using multi-level corpus data. First, large-scale general domain text corpus is used for pre-training to obtain basic word vector representations, and then professional corpus data related to the material master data field is used for fine-tuning training to enhance the word vector model's ability to capture the semantic features of material domain text.

[0021] Preferably, the similarity calculation module also includes a vector normalization submodule, which normalizes the standard library word vector set and the material master data word vector set before calculating the similarity, limiting the word vector modulus to unit length to eliminate modulus length differences and improve the fairness and accuracy of the similarity calculation.

[0022] Preferably, the matching result optimization module further includes a material type grouping unit for limiting the matching range and improving the matching accuracy.

[0023] Compared with related technologies, the material master data semantic matching method and system based on word vector similarity provided by the present invention has the following beneficial effects: This paper proposes a method and system for semantic matching of material master data based on word vector similarity. This method generates word vectors using natural language processing techniques and calculates the word vector similarity between the material master data and text data in a standard library, achieving semantic matching of material master data. This method not only improves matching accuracy and flexibility but also can handle material descriptions that are semantically similar but have different representations.

[0024] In addition, the present invention further improves the accuracy and efficiency of matching by optimizing the matching results by setting similarity thresholds and grouping material types. Therefore, the present invention has broad application prospects and significant economic benefits in the field of material management. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 Flowchart of a semantic matching method for material master data based on word vector similarity provided by the present invention; Figure 2 Example diagram from input word to word vector and then to output word; Figure 3 Example diagram of the operation interface for result display. Specific implementation manner

[0026] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0027] The present invention proposes a semantic matching method for material master data based on word vector similarity, as Figure 1 shown, including the following steps: S1. Construct a standard library: Collect national / industry standard documents, scan paper documents, and perform OCR character recognition on PDF / JPG format documents; Extract technical parameters from the standard documents, including standard numbers, material names, and technical attributes; Store the extracted data in a database, and the database table contains fields: leaf class code, material name, model specification, long description; S2. Data preprocessing: Perform word segmentation and preprocessing on the text data in the material master data and the standard library. The preprocessing includes: Segmenting the text data into multiple word segments according to semantics; Removing stop words, removing words such as "de", "di", "de" that do not have actual semantic information; And词性标注 (词性标注 is not a recognized English term. It should be something like "pos tagging"), performing pos tagging on the remaining words to filter out the parts with semantic value; S3. Word vector generation: Use natural language processing technology to input the words or phrases in the material master data and the standard library after word segmentation into a pre-trained word vector model, and convert them into corresponding word vectors; The training process of the word vector model is: Collect a large amount of text corpora related to the material master data, including standard documents, technical materials, product descriptions, etc.; Perform preprocessing on the collected corpora, including operations such as cleaning and normalization; Input the preprocessed corpora into the word vector model for training, and adjust the parameters of the model, such as vector dimension, window size, iteration times, etc., to obtain the best word vector representation; S4. Similarity calculation: Calculate the similarity between the word vectors of each material description and characteristics in the material master data and the word vectors of the text data in the standard library; S5. Matching result screening: A similarity threshold is set. When the similarity exceeds the threshold, the text data is considered to match a material description or characteristic. The threshold is set manually based on the actual application scenario and business needs or automatically optimized through a machine learning algorithm. The similarity between the calculated standard library word vector and the material master data word vector is compared with the threshold. When the similarity is greater than or equal to the threshold, the standard library text data is considered to be a match with the material master data. S6. Matching result optimization: Based on business needs, the matching results are further optimized. For example, the matching results are grouped and sorted by material type, taking into account the natural properties of the material and the data controlled by the SAP system. The standard document association information and difference attributes are then displayed through a visual interface. A list of material master data revision suggestions is then generated and synchronized to the SAP system. S7. Result display and application: The matching results are displayed to the user in a visual manner and applied to the revision and optimization of the material master data. The steps for visually displaying the matching results are to build an intuitive visual interface to present the similarity values, matching relationships, semantic associations, etc. between the standard library text data and the material master data in the form of charts, graphics, lists, etc., so that users can quickly view, understand and analyze the matching results, improve users' trust in the matching process and results, and support users to perform interactive operations in the visual interface, such as filtering, sorting, querying, etc., so that users can obtain the required information more flexibly, further improving the usability and practicality of the material master data semantic matching system.

[0028] Furthermore, the data preprocessing in step S1 also includes noise removal, text cleaning, and text sentence segmentation.

[0029] Furthermore, the word vectors in step S2 are generated using Word2Vec and GloVe natural language processing technologies.

[0030] Furthermore, the similarity calculation in step S3 adopts the cosine similarity and Euclidean distance methods. If cosine similarity is adopted, the ratio of the dot product of the two word vectors to the product of the module lengths of the two word vectors is calculated; if Euclidean distance is adopted, the square root of the sum of the squares of the numerical differences of the corresponding dimensions of the two word vectors is calculated.

[0031] Furthermore, the matching result optimization in step S4 also includes limiting the matching range by grouping material types to improve matching accuracy.

[0032] The present invention also proposes a material master data semantic matching system based on word vector similarity, the system comprising: Data preprocessing module, used to preprocess the text data in the standard library and the description and characteristics of the material master data; The word vector generation module is used to generate word vectors for words or phrases in the material master data and standard library. This module uses word vector models from natural language processing technologies such as word2vec or glove to train the standard library text data processed by the data preprocessing module and generate a corresponding set of standard library word vectors. Each piece of standard library text data corresponds to a word vector. Similarity calculation module, used to calculate the similarity between the word vectors of each material description and characteristic in the material master data and the word vectors of the text data in the standard library; The matching result screening module is used to set a similarity threshold based on actual application scenarios and business needs, and compare the similarity obtained by the similarity calculation module with the threshold. When the similarity is greater than or equal to the threshold, it is determined that the standard library text data and the material master data are successfully matched; The matching result optimization module is used to collect user feedback information and matching result evaluation data, and optimize the threshold using machine learning algorithms or manual adjustment to improve the accuracy and adaptability of matching results; The result display and application module is used to display the matching results to users in a visual manner and apply them to the revision and optimization of material master data.

[0033] Furthermore, the data preprocessing module also includes a word segmentation optimization submodule, which adopts a word segmentation algorithm based on a combination of dictionary matching and statistical language model, wherein the dictionary contains professional vocabulary, industry terms and common abbreviations related to materials, and the statistical language model is used to identify unregistered words to improve the recognition ability of professional terms and special expressions in material master data, and ensure the accuracy and completeness of the word segmentation results.

[0034] Furthermore, the word vector generation module also includes a corpus training sub-module, which is used to train the word vector model using multi-level corpus data. First, large-scale general domain text corpus is used for pre-training to obtain basic word vector representation, and then professional corpus data related to the material master data field is used for fine-tuning training to enhance the word vector model's ability to capture the semantic features of material domain text.

[0035] Furthermore, the similarity calculation module also includes a vector normalization submodule, which normalizes the standard library word vector set and the material master data word vector set before calculating the similarity, limiting the word vector modulus to unit length to eliminate modulus length differences and improve the fairness and accuracy of the similarity calculation.

[0036] Furthermore, the matching result optimization module also includes a material type grouping unit for limiting the matching range and improving the matching accuracy.

[0037] Next, the specific implementation of the material master data semantic matching method based on word vector similarity proposed in the present invention is explained.

[0038] 1. Build a standard library First, collect all relevant national standards (GB) and industry standards (IX) documents. Paper documents need to be scanned and converted to electronic formats to ensure that all documents are in electronic format for easy processing. For standard documents sourced from official internet portals, crawler technology can be used to obtain the latest data, updated in real time. During the collection process, ensure the completeness and accuracy of the documents, verifying the validity of the versions and avoiding missing or unclear pages.

[0039] Next, the collected electronic documents are parsed to identify their structure, such as chapters, titles, and paragraphs, and extract the required data elements. These key data elements include the standard number, name, scope of application, and technical requirements. Of particular note, if the document is in an image format (such as JPG) or a non-text format like PDF, optical character recognition (OCR) is used to identify the document's content as electronic text data, enabling subsequent parsing and extraction.

[0040] The extracted data is then cleaned and standardized. Cleaning primarily involves removing duplicate data, correcting erroneous data, and deleting irrelevant information. Standardization involves standardizing data formats, units, and terminology to ensure consistency. Data is classified and organized according to the specific requirements of national and industry standards. For example, materials are filed by major, minor, and subcategories, using different fields for differentiation and classification, forming a hierarchical data structure with a logically inclusive relationship.

[0041] After organizing the data, begin designing the database structure. Based on the structure and data requirements of national and industry standards, determine which database tables need to be created, which fields each table should contain, and the relationships between fields. Also, define each field's data type, constraints (such as primary keys and unique keys), and indexing strategies. The goal of this design is to establish a reasonable database table structure and relationships to facilitate efficient data storage and support query operations.

[0042] After designing the database structure, import the cleaned and standardized data into the database. Store the data in the corresponding tables and fields according to the designed table structure. During the import process, ensure the accuracy and completeness of the data and perform necessary verification and testing, such as checking whether the data was successfully written, whether the field values ​​meet expectations, and whether the constraints are effective.

[0043] After data is imported, ensure database query efficiency. Develop appropriate query statements based on business needs to facilitate user access to the required information from the database. Optimize database performance, such as by creating indexes on key fields and optimizing complex query statements, to improve query speed and accuracy. Regular database backup and maintenance are also essential to ensure data security and recoverability in the event of an outage.

[0044] In order to support other systems or applications to use the data in the standard library, corresponding interfaces need to be developed. For example, a standard API interface (such as ` / api / material-master`) is provided, and the HTTP POST method is used. The request header must include `Content-Type: application / json` and `Authorization: Bearer`. <token>`, which can return structured JSON data containing standard document information such as leaf code, material name, model specification, long description, and other fields. These interfaces are fully tested and verified to ensure their stability and reliability, and can correctly respond to requests and return data.

[0045] Finally, since national and industry standards are constantly updated and changing, a continuous updating and maintenance mechanism is necessary. New national and industry standard documents should be regularly collected and processed according to the aforementioned steps (scanning, OCR, parsing, extraction, cleaning, and standardization). The updated data should be synchronized with the database. At the same time, existing documents in the database should be reviewed and cleaned up, identifying and removing any obsolete or outdated standards data to ensure that the standards information stored in the database is always accurate, complete, and up-to-date.

[0046] 2. Word vector generation First, the text data in the standard library is segmented and preprocessed.

[0047] Generate word vectors for each word or phrase using natural language processing techniques (such as Word2Vec, GloVe, etc.). These word vectors map words or phrases to vectors in a high-dimensional space, so that words with similar semantics are closer in the space.

[0048] Similar word segmentation and preprocessing are performed on the material master data, especially the description and characteristics of the material. The text of these descriptions and characteristics is converted into word vectors to capture their semantic information.

[0049] 3. Similarity calculation For each text data in the standard library, calculate the similarity between its word vector and the word vector of each material description and characteristic in the material master data. This can be achieved through methods such as cosine similarity and Euclidean distance. An example diagram from input word to word vector and then to output word is shown below. Figure 2 shown.

[0050] Set a similarity threshold. When the similarity exceeds the threshold, the text data is considered to match a material description or characteristic.

[0051] Matching results can be further optimized based on business needs. For example, the material's natural attributes (such as name, size, weight, etc.) and data controlled by the SAP system (such as MRP-related data) can be considered to further filter and confirm matching results.

[0052] You can also use material type grouping (such as raw materials, semi-finished products, finished products, etc.) to limit the matching scope and improve matching accuracy.

[0053] 4. Results presentation and application: Matching results are presented to users visually, processing the data into an easily understandable and interactive format so that users can intuitively review and analyze the data. A detailed description of the matching results is presented to users in the form of a list, table, or graphical interface. Lists are a simple and intuitive display method, particularly suitable for presenting short or structured data. In the matching results, lists can be used to display brief information about each standard document.

[0054] Matching results are presented to users in a visual format. Displaying structured data clearly presents multiple dimensions and attributes of the data. Within the matching results, tables can be used to display detailed information about the standard document. Figure 3 shows an example of the result display interface.

[0055] Users can clearly see the detailed information of each standard document through the table. They can use the table's sorting and filtering functions to quickly locate specific standard documents according to their needs.

[0056] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the scope of protection of the invention. Obviously, the embodiments described are only some embodiments of the present invention, rather than all embodiments. Based on these embodiments, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in this field can still combine, add, delete or make other adjustments to the features in the various embodiments of the present invention according to the circumstances without conflict, without making creative work, so as to obtain different other technical solutions that do not deviate from the concept of the present invention in essence, and these technical solutions also fall within the scope of protection of the present invention.< / token>

Claims

1. A semantic matching method for material master data based on word vector similarity, characterized in that: The following steps are involved: S1. Build a standard library: Collect national standard / industry standard documents, scan paper documents, and perform OCR text recognition on PDF / JPG format documents; extract technical parameters from standard documents, including standard number, material name, and technical attributes; store the extracted data in a database. The database table contains the following fields: leaf code, material name, model specification, and long description; S2. Data preprocessing: Segment and preprocess the text data in the material master data and standard library. The preprocessing includes: segmenting the text data into multiple word segments according to semantics; removing stop words and words without actual semantic information; and part-of-speech tagging, marking the remaining words with parts of speech to filter out those with semantic value. S3. Word vector generation: Using natural language processing technology, the word-segmented material master data and words or phrases in the standard library are input into a pre-trained word vector model to convert them into corresponding word vectors. The training process of the word vector model is as follows: a large amount of text corpus related to the material master data, including standard documents, technical information, and product descriptions, is collected; the collected corpus is preprocessed, including cleaning and normalization operations; the preprocessed corpus is input into the word vector model for training, and the model parameters, such as vector dimension, window size, and number of iterations, are adjusted to obtain the optimal word vector representation. S4. Similarity calculation: Calculate the similarity between the word vectors of each material description and characteristic in the material master data and the word vectors of the text data in the standard library; S5. Matching result screening: A similarity threshold is set. When the similarity exceeds the threshold, the text data is considered to match a material description or characteristic. The threshold is set manually based on the actual application scenario and business needs or automatically optimized through a machine learning algorithm. The similarity between the calculated standard library word vector and the material master data word vector is compared with the threshold. When the similarity is greater than or equal to the threshold, the standard library text data is considered to be a match with the material master data. S6. Matching result optimization: Based on business needs, the matching results are further optimized by grouping and sorting them by material type. The standard document association information and difference attributes are then displayed through a visual interface. A list of suggested revisions to the material master data is then generated and synchronized to the SAP system. S7. Result display and application: The matching results are displayed to the user in a visual manner and applied to the revision and optimization of the material master data. The steps for visually displaying the matching results are to build an intuitive visual interface to present the similarity values, matching relationships, semantic associations, etc. between the standard library text data and the material master data in the form of charts, graphics, and lists, so that users can quickly view, understand and analyze the matching results, improve users' trust in the matching process and results, and support users to perform interactive operations in the visual interface, such as filtering, sorting, and querying, so that users can obtain the required information more flexibly, further improving the usability and practicality of the material master data semantic matching system.

2. The material master data semantic matching method based on word vector similarity according to claim 1 is characterized in that: The data preprocessing in step S1 also includes noise removal, text cleaning, and text sentence segmentation.

3. The material master data semantic matching method based on word vector similarity according to claim 1 is characterized in that: The word vectors in step S2 are generated using Word2Vec and GloVe natural language processing technologies.

4. The material master data semantic matching method based on word vector similarity according to claim 1 is characterized in that: The similarity calculation in step S3 adopts the cosine similarity and Euclidean distance methods. If cosine similarity is adopted, the ratio of the dot product of the two word vectors to the product of the module lengths of the two word vectors is calculated; if Euclidean distance is adopted, the square root of the sum of the squares of the numerical differences of the corresponding dimensions of the two word vectors is calculated.

5. The material master data semantic matching method based on word vector similarity according to claim 1 is characterized in that: The matching result optimization in step S4 also includes limiting the matching range by grouping material types to improve matching accuracy.

6. A material master data semantic matching system based on word vector similarity, characterized by: The system is applied to the material master data semantic matching method based on word vector similarity according to any one of claims 1 to 5, and the system includes: Data preprocessing module, used to preprocess the text data in the standard library and the description and characteristics of the material master data; The word vector generation module is used to generate word vectors for words or phrases in the material master data and standard library. This module uses the word vector model in word2vec or glove natural language processing technology to train the standard library text data processed by the data preprocessing module to generate a corresponding set of standard library word vectors. Each standard library text data corresponds to a word vector. Similarity calculation module, used to calculate the similarity between the word vectors of each material description and characteristic in the material master data and the word vectors of the text data in the standard library; The matching result screening module is used to set a similarity threshold based on actual application scenarios and business needs, and compare the similarity obtained by the similarity calculation module with the threshold. When the similarity is greater than or equal to the threshold, it is determined that the standard library text data and the material master data are successfully matched; The matching result optimization module is used to optimize the threshold by collecting user feedback information and matching result evaluation data, using machine learning algorithms or manual adjustment; The result display and application module is used to display the matching results to users in a visual manner and apply them to the revision and optimization of material master data.

7. The material master data semantic matching system based on word vector similarity according to claim 6 is characterized in that: The data preprocessing module also includes a word segmentation optimization submodule, which adopts a word segmentation algorithm based on a combination of dictionary matching and statistical language model. The dictionary contains professional vocabulary, industry terms and common abbreviations related to materials, and the statistical language model is used to identify unregistered words to improve the recognition ability of professional terms and special expressions in material master data, thereby ensuring the accuracy and completeness of the word segmentation results.

8. The material master data semantic matching system based on word vector similarity according to claim 6 is characterized in that: The word vector generation module also includes a corpus training submodule, which is used to train the word vector model using multi-level corpus data. First, large-scale general domain text corpus is used for pre-training to obtain basic word vector representations, and then professional corpus data related to the material master data field is used for fine-tuning training to enhance the word vector model's ability to capture the semantic features of material domain text.

9. The material master data semantic matching system based on word vector similarity according to claim 6 is characterized in that: The similarity calculation module also includes a vector normalization submodule, which normalizes the standard library word vector set and the material master data word vector set before calculating the similarity, limiting the word vector modulus to unit length to eliminate modulus length differences and improve the fairness and accuracy of the similarity calculation.

10. The material master data semantic matching system based on word vector similarity according to claim 6 is characterized in that: The matching result optimization module also includes a material type grouping unit for limiting the matching range and improving the matching accuracy.

Images