Supply chain material data management method and system

Abstract

The present invention discloses a supply chain material data management method and system, the method comprising receiving material data of upstream and downstream enterprises in the supply chain; performing named entity recognition for each material data, and analyzing the semantic dependency of attributes using dependency analysis technology; mapping different entities to corresponding concept nodes in an ontology, and establishing associations between entities and concepts; using a constructed material ontology, performing semantic standardized representation of entities, mapping entities to corresponding concept nodes in the ontology, and replacing them with standardized concept names or identifiers to form a standard material attribute feature library. The present invention realizes the collection and standardized processing of material data of upstream and downstream enterprises in the supply chain and the precise association of upstream and downstream material data.

Description

Technical Field

[0001] The present invention relates to the technical field of material data processing, and in particular to a supply chain material data management method and system. Background Art

[0002] The existing supplier information of enterprises is scattered and difficult to fully understand. At the same time, there are multiple supplier data sources inside and outside the enterprise, such as ERP, CRM, supplier websites, etc. This information is scattered in different departments and systems, making it difficult to manage and analyze in a unified manner; at the same time, supplier information not only contains structured data, but also a large amount of unstructured data, such as text introductions, product introductions, comments and feedback on supplier websites, etc. This information is difficult to quickly extract and analyze; among the massive supplier resources, it is difficult for enterprises to actively discover the high-quality suppliers that best suit their needs, and a lot of manpower is needed for manual screening. Summary of the invention

[0003] In view of the problems existing in the above-mentioned prior art, the present invention provides a supply chain material data management method and system, which effectively improves the efficiency of supply chain data management. The technical solution is as follows:

[0004] In a first aspect, a supply chain material data management method is provided, comprising the following steps:

[0005] Receive material data from upstream and downstream companies in the supply chain;

[0006] Perform named entity recognition on each material data to identify the entity type in the material data, where the entity type includes the name of the material subject and the attributes of the material subject, where the attributes include attribute categories and attribute values, and where the attribute categories include specification model, material, and purpose;

[0007] Analyzing the semantic dependency of attributes using dependency analysis technology, wherein the categories of the semantic dependency include: synonymy, hyponymy, part-whole, and belonging;

[0008] Map different entities to corresponding concept nodes in the ontology and establish associations between entities and concepts. The concept nodes include material names, attributes, and attribute values.

[0009] Using the constructed material ontology, the entities are semantically standardized and mapped to the corresponding concept nodes in the ontology, replaced with standardized concept names or identifiers to form a standard material attribute feature library.

[0010] In some implementations, performing named entity recognition on each material data to identify the entity type in the material data includes:

[0011] Perform word segmentation and part-of-speech tagging for each material data;

[0012] Using dependency syntactic analysis technology, obtain syntactic dependency and semantic dependency in material data, wherein the syntactic dependency is used to characterize the relationship between each word and predicate, and the semantic dependency is used to characterize the semantic association relationship;

[0013] Based on part-of-speech tagging, syntactic dependency, and semantic dependency, the entity types in the material data are obtained.

[0014] In some implementations, the word segmentation and part-of-speech tagging for each material data further includes: analyzing the word segmentation results as candidate word segmentation results, wherein the analysis process is:

[0015] Get the current word segmentation result and record it as the first word segmentation result;

[0016] Get the current word in the word segmentation result and record it as the first word;

[0017] Get the words in the adjacent positions before and after the current word position, and record them as the second word;

[0018] Based on at least one character in the second word that is close to the first word and at least one character in the first word that is close to the second word, a new word is formed, which is recorded as a third word;

[0019] When the frequency of the third word in the vocabulary is greater than 0, the third word is used as a candidate word;

[0020] Based on the situation that the third word is a candidate word, the word segmentation results before and after the third word are re-determined and recorded as the second word segmentation results.

[0021] In some embodiments, the forming of a new word based on at least one character in the second word close to the first word and at least one character in the first word close to the second word, recorded as a third word, includes:

[0022] For the first word, the word library is searched for matching words with the first character of the first word as a non-first position and matching words with the last character of the first word as a non-last position; when a matching word is found, the matching search word is used as the third word.

[0023] In some implementations, the word segmentation and part-of-speech tagging for each material data further includes: performing credibility analysis on the word segmentation results, wherein the credibility analysis includes the following steps:

[0024] Obtain the semantic features of each word in the word segmentation results;

[0025] The credibility of the word segmentation result is analyzed based on the gradient of the semantic features of the current word in the word segmentation result and the semantic features of the second word in the adjacent position and the semantic adaptability of all words in the word segmentation result, wherein the larger the gradient data corresponding to all words in the word segmentation result and the greater the semantic adaptability of all words, the greater the credibility of the word segmentation result.

[0026] In some implementations, the semantic adaptability of all words in the word segmentation result is obtained by:

[0027] The word embeddings of all words in the word segmentation results are sequentially composed into the first vector.

[0028] Delete the word embedding of the current word in the word segmentation result, and compose the second vector with the word embeddings of the remaining words in order;

[0029] The first vector and the second vector are input into a semantic comparison analysis model, and semantic similarity between the first vector and the second vector is obtained through semantic coding as the semantic adaptability.

[0030] In some embodiments, the supply chain material data management method further includes: after forming the standard material attribute feature library, performing correlation matching analysis on the material data of upstream and downstream enterprises in the supply chain;

[0031] The matching analysis includes:

[0032] For the first material data and the second material data, a matching search is performed in the standard material attribute feature library to extract the corresponding key attributes;

[0033] For each attribute of the first material data and the second material data, determine a corresponding preset similarity calculation method according to a data type of the attribute, and obtain the similarity of the attribute of the first material data and the second material data;

[0034] The similarity calculation results based on the multiple attributes are integrated to determine the similarity between the first material data and the second material data.

[0035] In some implementations, for each attribute of the first material data and the second material data, determining a corresponding preset similarity calculation method according to the attribute data type includes:

[0036] For the similarity of text attributes, based on the edit distance similarity as the first similarity parameter and the Jaccard similarity as the second similarity parameter, different weights are assigned to the first similarity parameter and the second similarity parameter, and the comprehensive similarity result of the text attributes is obtained based on the weighted fusion of the first similarity parameter and the second similarity parameter;

[0037] For numerical attributes, the Euclidean distance is used to calculate the similarity;

[0038] For category attributes, the semantic similarity calculation method is used.

[0039] In some implementations, the similarity based on edit distance is used as the first similarity parameter, the Jaccard similarity is used as the second similarity parameter, and different weights are assigned to the first similarity parameter and the second similarity parameter, respectively. The weight assignment method includes the following steps:

[0040] w1 and w2 represent the weights of edit distance similarity and Jaccard similarity respectively, and w1 + w2 = 1;

[0041] The known matching pairs and non-matching pairs are used as training data, and the supervised learning algorithm is used to learn the optimal values ​​of w1 and w2; the difference between the comprehensive similarity and the actual matching situation is used as the loss function, and w1 and w2 are continuously adjusted through the optimization algorithm until the loss function is minimized and the optimal value is determined.

[0042] In a second aspect, a supply chain material data management system is provided, the system comprising:

[0043] A data acquisition unit, used to receive material data from upstream and downstream enterprises in the supply chain;

[0044] An entity recognition unit is used to perform named entity recognition on each material data to identify the entity type in the material data, wherein the entity type includes the name of the material subject and the attributes of the material subject, wherein the attributes include attribute categories and attribute values, and wherein the attribute categories include specification model, material, and purpose;

[0045] An entity relationship recognition unit, used to analyze the semantic dependency of attributes using dependency analysis technology, wherein the categories of the semantic dependency include: synonymy, hyponymy, part-whole, and belonging;

[0046] The standard material attribute feature library unit is used to map different entities to corresponding concept nodes in the ontology and establish associations between entities and concepts. The concept nodes include material names, attributes, and attribute values. The constructed material ontology is used to perform semantic standardization on entities, map entities to corresponding concept nodes in the ontology, and replace them with standardized concept names or identifiers to form a standard material attribute feature library.

[0047] According to a third aspect, an electronic device is provided, the electronic device comprising:

[0048] A memory for storing executable instructions;

[0049] The processor is used to implement the supply chain material data management method as described in the first aspect above when running the executable instructions stored in the memory.

[0050] In a fourth aspect, a computer-readable storage medium is provided, storing executable instructions, which, when executed by a processor, implement the supply chain material data management method as described in the first aspect above.

[0051] A supply chain material data management method and system of the present invention has the following beneficial effects: a standard material attribute feature library is established, material data of upstream and downstream enterprises in the supply chain are collected, cleaned and standardized, various internal and external structured and unstructured supplier data are integrated, and the unstructured material data provided by the supplier is structured and decomposed using named entity recognition technology to form specific material keywords, brands, models, materials, sales specifications, units, etc.; for different enterprises and different suppliers, the naming of materials is not unified, and the accurate association of upstream and downstream material data is found by using fuzzy matching of key material attributes. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a flow chart of a supply chain material data management method in an embodiment of the present application;

[0053] Figure 2 It is a flow chart of the material data entity identification method in the embodiment of the present application;

[0054] Figure 3 It is a flow chart of a method for performing association matching analysis on material data of upstream and downstream enterprises in the supply chain in an embodiment of the present application;

[0055] Figure 4 It is a structural diagram of a supply chain material data management system in an embodiment of the present application. DETAILED DESCRIPTION

[0056] 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.

[0057] See also Figure 1 The present application embodiment provides a supply chain material data management method, comprising the following steps:

[0058] Step 1: Receive material data from upstream and downstream enterprises in the supply chain;

[0059] Step 2: Perform named entity recognition on each material data to identify the entity type in the material data. The entity type includes the name of the material subject and the attributes of the material subject. The attributes include attribute categories and attribute values. The attribute categories include specifications, models, materials, and uses. The entity recognition results include material keywords, brands, models, materials, sales specifications, units, etc.

[0060] Step 3, using dependency analysis technology to analyze the semantic dependency of the attributes, the categories of the semantic dependency include: synonymy, hyponymy, part-whole, and belonging;

[0061] Step 4: Map different entities to corresponding concept nodes in the ontology and establish associations between entities and concepts. The concept nodes include material names, attributes, and attribute values.

[0062] Step 5: Use the constructed material ontology to perform semantic standardization on the entities, map the entities to the corresponding concept nodes in the ontology, replace them with standardized concept names or identifiers, and form a standard material attribute feature library.

[0063] In the embodiment of the present application, based on the platform transaction data and combined with the standard data platform, a unified data access and sharing platform is provided for upstream and downstream participants in the supply chain. In view of the problem that different enterprises and suppliers do not name materials uniformly and material data is not standardized, a standard material attribute feature library is constructed based on named entity recognition technology and dependency analysis technology in the embodiment of the present application, which is convenient for subsequent real-time analysis of supply chain data. Based on the standard material attribute feature library and the standard data platform, the accurate association of upstream and downstream materials in the supply chain is realized.

[0064] In the categories of semantic relations in step 3 above,

[0065] Synonymous relationship: indicates that two attribute values ​​express the same or similar concepts, such as "color" and "hue".

[0066] Hyponym-hypernym relationship: indicates that one attribute value is a superordinate concept or subordinate concept of another attribute value, such as "plastic" is a subordinate concept of "material";

[0067] Part-whole relationship: indicates that one attribute value is a component of another attribute value, such as "diameter" is a part of "specification"

[0068] Attribute value relationship: indicates that one attribute value is the value of another attribute, such as "red" is a value of "color".

[0069] In the embodiments of the present application, when extracting semantic relations, training data with semantic relationship annotations can be used to train a classifier or a sequence annotation model, and a small amount of annotated data and a large amount of unlabeled data can be combined to utilize guided learning, self-training and other methods to improve the generalization ability of the model.

[0070] In step 4 above, the attribute values ​​are mapped to the corresponding concept nodes in the ontology, including:

[0071] For synonymous attribute values, they are mapped to the same concept node to achieve semantic normalization;

[0072] For hierarchical and part-whole relationships, attribute values ​​are organized into a hierarchical structure to represent the semantic relationship between concepts.

[0073] See also Figure 2 In one implementation, in the above step 2, named entity recognition is performed for each material data to identify the entity type in the material data, including:

[0074] Step 21, perform word segmentation and part-of-speech tagging for each material data;

[0075] Step 22, using dependency syntactic analysis technology to obtain syntactic dependency and semantic dependency in the material data, wherein the syntactic dependency is used to characterize the relationship between each word and the predicate, and the semantic dependency is used to characterize the semantic association relationship;

[0076] Step 23, based on part-of-speech tagging, syntactic dependency, and semantic dependency, obtain the entity type in the material data.

[0077] In the embodiment of the present application, entity types are analyzed based on multiple annotations, and part-of-speech tagging, syntactic dependencies, and semantic dependencies are integrated in the entity recognition process to improve the accuracy of entity recognition.

[0078] Specifically, the above step 23 obtains entity types in the material data based on part-of-speech tagging, syntactic dependency, and semantic dependency, including:

[0079] Step 231, based on the training sample set of material data, annotate each sample with word segmentation tagging, part of speech tagging, syntactic dependency, and semantic dependency;

[0080] Step 232, training a conditional random field model based on the training sample set to obtain a material data entity recognition model;

[0081] Step 233: perform entity recognition on the material data to be analyzed based on the trained conditional random field model.

[0082] In one implementation, after the above step 21, the method further includes: analyzing the candidate word segmentation results on the word segmentation results, wherein the analysis process is as follows:

[0083] Step 21, obtaining the current word segmentation result and recording it as the first word segmentation result;

[0084] Step 22, obtaining the current word in the word segmentation result, recorded as the first word;

[0085] Step 23, obtaining the words at the positions before and after the current word position, and recording them as the second words;

[0086] Step 24, based on at least one character in the second word close to the first word and at least one character in the first word close to the second word, a new word is formed, which is recorded as a third word;

[0087] Step 25, when the frequency of the third word appearing in the word library is greater than 0, the third word is used as a candidate word;

[0088] Step 26, based on the situation that the third word is a candidate word, re-determine the word segmentation results before and after the third word, and record them as the second word segmentation results.

[0089] In the embodiment of the present application, for the preliminary word segmentation results, it is necessary to further analyze whether there are other word segmentation results. There may be multiple word segmentation results. For example, "Nanjing Yangtze River Bridge" can be segmented into the first word segmentation result "Nanjing City / Yangtze River / Bridge", or it can also be segmented into "Nanjing / Mayor / Jiang Bridge", where "Jiang Bridge" is a person's name, "Nanjing / Mayor / Jiang / Bridge", and there is no "Jiang Bridge" in the vocabulary.

[0090] In one embodiment, the above step 24, based on at least one character in the second word close to the first word and at least one character in the first word close to the second word, forms a new word, recorded as a third word, including:

[0091] For the first word, the word library is searched for matching words with the first character of the first word as a non-first position and matching words with the last character of the first word as a non-last position; when a matching word is found, the matching search word is used as the third word.

[0092] It is understandable that after the third word is determined, re-segmenting the characters before and after the third word may result in some characters not being matched. In this case, there is no second segmentation result.

[0093] In one implementation, after the above step 21, the method further includes: performing credibility analysis on the word segmentation results, wherein the credibility analysis includes the following steps:

[0094] Step 27, obtaining the semantic features of each word in the word segmentation result;

[0095] Step 28, performing a credibility analysis on the word segmentation result based on the gradient of the semantic features of the current word in the word segmentation result and the semantic features of the second word in the adjacent position and the semantic adaptability of all words in the word segmentation result, wherein the larger the gradient data corresponding to all words in the word segmentation result and the greater the semantic adaptability of all words, the greater the credibility of the word segmentation result.

[0096] In the embodiment of the present application, when there is at least one word segmentation result, it is necessary to analyze the semantic gradient of adjacent words and the semantic adaptability of all words in the entire material data. It can be understood that two adjacent words will not be synonyms, that is, there is a semantic gap between the two adjacent words. For example, the procurement material data is "32 square meters of simple European style OPPE wardrobe for the whole house customization", and its word segmentation result is "32 square meters / simple European style / OPPE / wardrobe / whole house / customization". It can be understood that the adjacent words in the word segmentation result represent different types of attribute descriptions of the "material body" or "wardrobe", and the material attributes include "manufacturer: OPPEIN", "model: simple European style", "quantity: 32 square meters", and "service type: whole house". However, for the entire material data, each word in the word segmentation result is a different attribute description of the "material body" or "wardrobe", and the semantics represented by each word in the word segmentation result are adaptively correlated with each other. Among them, the gap in the semantic features represented by the gradient.

[0097] It can be understood that the greater the sum of the gradient values ​​corresponding to all words in the word segmentation result and the greater the semantic adaptability of all words in the word segmentation result, the greater the credibility of the word segmentation result. It should be noted that in this application, the two constraints for the gradient constraint and the semantic adaptability constraint exist simultaneously. The word segmentation result can be determined to be valid only when the gradient constraint is satisfied and the semantic adaptability constraint is satisfied at the same time. The credibility analysis of the word segmentation result can be performed based on whether the gradient of the semantic feature of the current word in the word segmentation result and the semantic feature of the second word in the adjacent position is greater than a preset gradient threshold and whether the semantic adaptability of all words in the word segmentation result is greater than a preset adaptability threshold.

[0098] In addition, it should be noted that the credibility analysis of the word segmentation results is to analyze both the first word segmentation result and the second word segmentation result mentioned above, that is, the credibility analysis is also performed on the candidate word segmentation results. When there are multiple word segmentation results, the final word segmentation result can be determined by selecting the gradient situation and the semantic adaptability situation corresponding to the word segmentation results.

[0099] In one implementation, in the above step 28, the method for obtaining the semantic adaptability of all words in the word segmentation result includes:

[0100] Step 281, based on the word embedding of all words in the word segmentation result, a first vector is formed in order.

[0101] Step 282, delete the word embedding of the current word in the word segmentation result, and sequentially compose the word embeddings of the remaining words into a second vector;

[0102] Step 283: input the first vector and the second vector into a semantic comparison analysis model, and obtain the semantic similarity between the first vector and the second vector through semantic coding as the semantic adaptability.

[0103] The word embedding model and the semantic contrast analysis model can be obtained by pre-training based on the training data. And the word embedding model and the semantic contrast analysis model can be trained synchronously. For example, a training sample is used to obtain the first vector and the second vector through the word embedding model to be trained in turn, and then the semantic contrast analysis model to be trained is input to output the semantic adaptability result, and the model parameters of the word embedding model and the semantic contrast analysis model are reversely iterated based on the error loss between the actual output result of the semantic adaptability and the predicted annotation result. It can be understood that the semantic contrast analysis model has a very large semantic adaptability for the second vector corresponding to any word segmentation in the correct word segmentation result of a material data and the first vector corresponding to the original complete material data. In this application, a complete material data is considered, and all its words have semantic coherence, such as "Nanjing Yangtze River Bridge Project". According to "Project", the word segmentation result of "Nanjing Yangtze River Bridge" can be determined as "Nanjing City / Yangtze River / Bridge". If the "Nanjing Yangtze River Bridge Project" is segmented into "Nanjing / Mayor / River Bridge / Project", the semantics after deleting the word "Project" is incompatible with the original complete semantics.

[0104] In one embodiment, the above step 5 further includes:

[0105] Step 6: After forming the standard material attribute feature library, conduct correlation matching analysis on the material data of upstream and downstream enterprises in the supply chain;

[0106] See also Figure 3 The matching analysis in step 6 includes the following steps:

[0107] Step 61, for the first material data and the second material data, respectively, a matching search is performed in a standard material attribute feature library to extract corresponding key attributes;

[0108] Step 62, for each attribute of the first material data and the second material data, determine a corresponding preset similarity calculation method according to the data type of the attribute, and obtain the similarity of the attribute of the first material data and the second material data;

[0109] Step 63: Determine the similarity between the first material data and the second material data based on the fusion of the similarity calculation results of the multiple attributes.

[0110] In the embodiment of this application, in order to solve the problem of inconsistent naming of materials by suppliers, the accurate association of upstream and downstream material data is found by using fuzzy matching of key material attributes. Natural language processing is used to realize intelligent analysis of material demand information, accurately match the material data of suppliers with the demand information of demand units, and open up the connection between supply and demand.

[0111] In the embodiment of the present application, the material data of the supply and demand sides are recorded as the first material data and the second material data respectively, and the first material data and the second material data are converted into standard material data after matching through the standard material attribute feature library. The marked material data characterizes the entity name specification standard, entity relationship specification standard, attribute item specification standard, etc. in the material data. Based on the converted standard material data, the key attributes are extracted. For different attributes, different similarity calculation methods are designed in the embodiment of the present application. It can be understood that different similarity calculation methods are adapted to the corresponding attribute characteristics. By designing different similarity calculation methods for different attributes, the fuzzy matching efficiency and matching accuracy of the material data of both parties are improved.

[0112] Specifically, the above step 61, for the first material data and the second material data, respectively performs matching searches in the standard material attribute feature library to extract corresponding key attributes, including the following steps:

[0113] Step 611, performing named entity recognition on the first material data and the second material data;

[0114] Step 612, based on the entity recognition result, matching is performed in the standard material attribute feature library to obtain the standard material data with the highest matching degree, and based on the entities and attributes in the standard material data, the key attributes in the first material data and the second material data are matched to obtain the key attributes in the first material data and the second material data.

[0115] In an embodiment of the present application, for the material data to be analyzed, after keyword extraction and entity recognition, the material data specification and standardization processing of the corresponding materials are performed in the standard material attribute feature library, and the subsequent similarity calculation is facilitated based on the first material data and the second material data after the standardized processing.

[0116] Specifically, the above step 62, for each attribute of the first material data and the second material data, determines the corresponding preset similarity calculation method according to the attribute data type, including the following steps:

[0117] Step 621, for the similarity of the text class attributes, based on the edit distance similarity as the first similarity parameter, the Jaccard similarity as the second similarity parameter, and after assigning different weights to the first similarity parameter and the second similarity parameter respectively, a comprehensive similarity result of the text class attributes is obtained based on the weighted fusion of the first similarity parameter and the second similarity parameter;

[0118] Step 622, for numerical attributes (such as specification parameters), similarity is calculated using Euclidean distance;

[0119] Step 623: For the category attributes (such as material and purpose), a semantic similarity calculation method is used, such as the Wu-Palmer algorithm based on WordNet.

[0120] In one embodiment, in the above step 621, based on the edit distance similarity as the first similarity parameter and the Jaccard similarity as the second similarity parameter, and by assigning different weights to the first similarity parameter and the second similarity parameter, the weight assignment method includes the following steps:

[0121] w1 and w2 represent the weights of edit distance similarity and Jaccard similarity respectively, and w1 + w2 = 1;

[0122] The known matching pairs and non-matching pairs are used as training data, and the supervised learning algorithm is used to learn the optimal values ​​of w1 and w2; the difference between the comprehensive similarity and the actual matching situation is used as the loss function, and w1 and w2 are continuously adjusted through the optimization algorithm until the loss function is minimized and the optimal value is determined.

[0123] In the embodiment of the present application, the similarity of text class attributes is combined with the edit distance similarity and the Jaccard similarity, and the weights of the edit distance similarity and the Jaccard similarity are trained with a large number of training samples to obtain the optimal weight parameters.

[0124] See also Figure 4 , a supply chain material data management system provided in an embodiment of the present application includes the following modules:

[0125] A data acquisition unit, used to receive material data from upstream and downstream enterprises in the supply chain;

[0126] An entity recognition unit is used to perform named entity recognition on each material data to identify the entity type in the material data, wherein the entity type includes the name of the material subject and the attributes of the material subject, wherein the attributes include attribute categories and attribute values, and wherein the attribute categories include specification model, material, and purpose;

[0127] An entity relationship recognition unit, used to analyze the semantic dependency of attributes using dependency analysis technology, wherein the categories of the semantic dependency include: synonymy, hyponymy, part-whole, and belonging;

[0128] The standard material attribute feature library unit is used to map different entities to corresponding concept nodes in the ontology and establish associations between entities and concepts. The concept nodes include material names, attributes, and attribute values. The constructed material ontology is used to perform semantic standardization on entities, map entities to corresponding concept nodes in the ontology, and replace them with standardized concept names or identifiers to form a standard material attribute feature library.

[0129] For the specific definition of the supply chain material data management system, please refer to the definition of the supply chain material data management method above, which will not be repeated here. Each unit in the above supply chain material data management system can be implemented in whole or in part by software, hardware and a combination thereof. The above units can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above units.

[0130] An electronic device provided in an embodiment of the present application includes:

[0131] processor;

[0132] a memory for storing processor-executable instructions;

[0133] The processor implements the supply chain material data management method in the above method embodiment by running the executable instructions.

[0134] The electronic device provided in the embodiment of the present application includes: at least one processor, a memory, a user interface and at least one network interface. The various components in the electronic device are coupled together through a bus system. It can be understood that the bus system is used to realize the connection and communication between these components. In addition to the data bus, the bus system also includes a power bus, a control bus and a status signal bus.

[0135] The embodiment of the present application also provides a computer-readable storage medium storing executable instructions, which implement the above-mentioned supply chain material data management method when executed by a processor. The computer-readable storage medium can be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), a magnetic tape, a floppy disk, an optical data storage node, etc.

[0136] The present invention is not limited to the above-mentioned specific implementation modes. Various changes made by ordinary technicians in this field based on the above-mentioned concepts without creative work are all within the protection scope of the present invention.

Claims

1. A supply chain material data management method, characterized in that: The steps include: Receive material data from upstream and downstream companies in the supply chain; Perform named entity recognition on each material data to identify the entity type in the material data, where the entity type includes the name of the material subject and the attributes of the material subject, where the attributes include attribute categories and attribute values, and where the attribute categories include specification model, material, and purpose; Analyzing the semantic dependency of attributes using dependency analysis technology, wherein the categories of the semantic dependency include: synonymy, hyponymy, part-whole, and belonging; Map different entities to corresponding concept nodes in the ontology and establish associations between entities and concepts. The concept nodes include material names, attributes, and attribute values. Using the constructed material ontology, the entities are semantically standardized, and the entities are mapped to the corresponding concept nodes in the ontology, replaced with standardized concept names or identifiers, to form a standard material attribute feature library; The named entity recognition is performed for each material data to identify the entity type in the material data, including: performing word segmentation and part-of-speech tagging for each material data; using dependency syntactic analysis technology to obtain syntactic dependency and semantic dependency in the material data, wherein the syntactic dependency is used to characterize the relationship between each word and the predicate, and the semantic dependency is used to characterize the semantic association relationship; based on the part-of-speech tagging, syntactic dependency, and semantic dependency, the entity type in the material data is obtained; The word segmentation and part-of-speech tagging for each material data also includes: performing credibility analysis on the word segmentation results, and the credibility analysis includes the following steps: obtaining the semantic features of each word in the word segmentation results; performing credibility analysis on the word segmentation results based on the gradient of the semantic features of the current word in the word segmentation results and the semantic features of the second word in the adjacent position and the semantic adaptability of all words in the word segmentation results, wherein the larger the gradient data corresponding to all words in the word segmentation results and the greater the semantic adaptability of all words, the greater the credibility of the word segmentation results; the semantic adaptability of all words in the word segmentation results, the acquisition method includes: based on the word embeddings of all words in the word segmentation results, forming a first vector in sequence, deleting the word embedding of the current word in the word segmentation results, and forming a second vector in sequence with the word embeddings of the remaining words; inputting the first vector and the second vector into a semantic comparison analysis model, and obtaining the semantic similarity between the first vector and the second vector through semantic encoding as the semantic adaptability.

2. A supply chain material data management method according to claim 1, characterized in that: The word segmentation and part-of-speech tagging for each material data further includes: analyzing the word segmentation results as candidate word segmentation results, wherein the candidate word segmentation result analysis process is as follows: Get the current word segmentation result and record it as the first word segmentation result; Get the current word in the word segmentation result and record it as the first word; Get the words in the adjacent positions before and after the current word position, and record them as the second word; Based on at least one character in the second word that is close to the first word and at least one character in the first word that is close to the second word, a new word is formed, which is recorded as a third word; When the frequency of the third word in the vocabulary is greater than 0, the third word is used as a candidate word; Based on the situation that the third word is a candidate word, the word segmentation results before and after the third word are re-determined and recorded as the second word segmentation results.

3. A supply chain material data management method according to claim 2, characterized in that: The method of forming a new word based on at least one character in the second word close to the first word and at least one character in the first word close to the second word, which is recorded as the third word, includes: For the first word, the word library is searched for matching words with the first character of the first word as a non-first position and matching words with the last character of the first word as a non-last position; when a matching word is found, the matching search word is used as the third word.

4. A supply chain material data management method according to claim 1, characterized in that: The method further comprises: after forming the standard material attribute feature library, performing correlation matching analysis on the material data of upstream and downstream enterprises in the supply chain; The matching analysis includes: For the first material data and the second material data, a matching search is performed in the standard material attribute feature library to extract the corresponding key attributes; For each attribute of the first material data and the second material data, determine a corresponding preset similarity calculation method according to a data type of the attribute, and obtain the similarity of the attribute of the first material data and the second material data; The similarity calculation results based on the multiple attributes are integrated to determine the similarity between the first material data and the second material data.

5. A supply chain material data management method according to claim 4, characterized in that: For each attribute of the first material data and the second material data, determining a corresponding preset similarity calculation method according to the attribute data type includes: For the similarity of text attributes, based on the edit distance similarity as the first similarity parameter and the Jaccard similarity as the second similarity parameter, different weights are assigned to the first similarity parameter and the second similarity parameter, and the comprehensive similarity result of the text attributes is obtained based on the weighted fusion of the first similarity parameter and the second similarity parameter; For numerical attributes, the Euclidean distance is used to calculate the similarity; For category attributes, the semantic similarity calculation method is used.

6. A supply chain material data management method according to claim 5, characterized in that: The method of assigning weights based on the edit distance similarity as the first similarity parameter and the Jaccard similarity as the second similarity parameter, and assigning different weights to the first similarity parameter and the second similarity parameter, comprises the following steps: w1 and w2 represent the weights of edit distance similarity and Jaccard similarity respectively, and w1 + w2 = 1; The known matching pairs and non-matching pairs are used as training data, and the supervised learning algorithm is used to learn the optimal values ​​of w1 and w2; the difference between the comprehensive similarity and the actual matching situation is used as the loss function, and w1 and w2 are continuously adjusted through the optimization algorithm until the loss function is minimized and the optimal value is determined.

7. A supply chain material data management system, characterized in that: include: A data acquisition unit, used to receive material data from upstream and downstream enterprises in the supply chain; An entity recognition unit is used to perform named entity recognition on each material data to identify the entity type in the material data, wherein the entity type includes the name of the material subject and the attributes of the material subject, wherein the attributes include attribute categories and attribute values, and wherein the attribute categories include specification model, material, and purpose; An entity relationship recognition unit, used to analyze the semantic dependency of attributes using dependency analysis technology, wherein the categories of the semantic dependency include: synonymy, hyponymy, part-whole, and belonging; The standard material attribute feature library unit is used to map different entities to corresponding concept nodes in the ontology and establish associations between entities and concepts. The concept nodes include material names, attributes, and attribute values. The constructed material ontology is used to perform semantic standardization on entities, map entities to corresponding concept nodes in the ontology, and replace them with standardized concept names or identifiers to form a standard material attribute feature library. The entity recognition unit performs named entity recognition for each material data to identify the entity type in the material data, including: performing word segmentation and part-of-speech tagging for each material data; using dependency syntactic analysis technology to obtain syntactic dependency and semantic dependency in the material data, wherein the syntactic dependency is used to characterize the relationship between each word and the predicate, and the semantic dependency is used to characterize the semantic association relationship; based on the part-of-speech tagging, syntactic dependency, and semantic dependency, the entity type in the material data is obtained; The word segmentation and part-of-speech tagging for each material data also includes: performing credibility analysis on the word segmentation results, and the credibility analysis includes the following steps: obtaining the semantic features of each word in the word segmentation results; performing credibility analysis on the word segmentation results based on the gradient of the semantic features of the current word in the word segmentation results and the semantic features of the second word in the adjacent position and the semantic adaptability of all words in the word segmentation results, wherein the larger the gradient data corresponding to all words in the word segmentation results and the greater the semantic adaptability of all words, the greater the credibility of the word segmentation results. The semantic adaptability of all words in the word segmentation results is obtained by a method including: sequentially forming a first vector based on the word embeddings of all words in the word segmentation results, deleting the word embedding of the current word in the word segmentation results, and sequentially forming a second vector with the word embeddings of the remaining words; inputting the first vector and the second vector into a semantic comparison analysis model, and obtaining the semantic similarity between the first vector and the second vector as the semantic adaptability through semantic encoding.

Images