Interaction relation extraction and demand and manufacturing matching method based on manufacturing service interaction context information

Through semi-supervised learning and probability graph model, the interaction relationships in the interactive context of manufacturing service are extracted and matched, and the problems of low efficiency of unstructured data processing and insufficient capture of complex relationships in the existing technology are solved, and efficient cross-enterprise manufacturing relationships are achieved Match the extraction and group-level requirements.

CN118916704APending Publication Date: 2024-11-08GUANGDONG UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410950697.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-16
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The prior art is inefficient when processing unstructured data in the context of manufacturing service interaction, making it difficult to accurately extract complex cross-enterprise manufacturing relationships, and lacks flexibility to effectively adapt to rapidly changing manufacturing needs and environments, especially in groups. There are shortcomings in matching requirements and capabilities at the level.

Method used

Using a two-stage automation solution: the first stage extracts entity relationships from unstructured text through semi-supervised learning method to build a heterogeneous manufacturing network; the second stage uses probability graph models to match the network to identify the requirements and capabilities Potential relationship, dynamic adaptation to the manufacturing environment.

Benefits of technology

It improves the processing efficiency of unstructured data, supports the execution of complex and dynamic manufacturing tasks, and realizes accurate capture of multi-faceted relationships across enterprises and matches group-level requirements and capabilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118916704A_ABST
    Figure CN118916704A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of enterprise information processing, and discloses an interaction relation extraction and demand and manufacturing matching method based on manufacturing service interaction context information, which comprises the following specific steps: acquiring the manufacturing service interaction context information; creating a seed list, and performing relation extraction on the manufacturing service interaction context information based on the seed list to obtain a plurality of manufacturing relation modes; constructing a heterogeneous network model based on the obtained manufacturing relation mode; and extracting interaction relation information from the manufacturing relation mode through the heterogeneous network model. The method solves the problem that the existing enterprise information processing technology cannot capture cross-enterprise and multi-aspect complex relationships, and has the characteristics of high processing efficiency and capability of supporting complex and dynamic manufacturing tasks.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of enterprise information processing, and more specifically, to a method for extracting interaction relationships and matching demand and manufacturing based on manufacturing service interaction context information. Background Art

[0002] In the manufacturing industry, with the diversification and personalization of market demands, social contexts are increasingly used in cross-enterprise manufacturing problem solving. In social manufacturing environments, a large number of manufacturing service interaction contexts (MSICs) are generated, containing rich information such as demand, capabilities, and partnerships. This information often exists in the form of unstructured text, making information extraction, processing, and utilization extremely difficult. Existing technologies have obvious deficiencies in processing these unstructured data, accurately extracting manufacturing relationships, and supporting group-level demand and capability matching.

[0003] In order to transform these massive unstructured data into decision support information for cross-enterprise manufacturing demand-capability matching, this paper proposes a two-stage automated solution: the first stage is to extract entity relationships based on semi-supervised learning methods to obtain a formalized heterogeneous manufacturing network from an unstructured text background containing high noise and irrelevant information; the second stage is to match the group-level relationships between entities in the established manufacturing network. These two systems can efficiently process and structure MSICs, accurately extract and analyze manufacturing relationships, support group-level demand-capability matching, and dynamically adapt to changes in manufacturing environment and demand.

[0004] Current solutions mainly face several limitations. First, they are not efficient enough when processing a large number of MSICs, and it is difficult to quickly and accurately extract useful information. Second, existing technologies are insufficient in deeply mining and utilizing manufacturing relationships, especially when capturing complex relationships across enterprises and multiple aspects. In addition, most solutions lack flexibility and cannot effectively adapt to rapidly changing manufacturing needs and environments, and it is difficult to support complex group-level demand and capability matching.

[0005] Therefore, how to invent an efficient interactive relationship extraction method that can capture complex relationships across enterprises and multiple aspects is a technical problem that urgently needs to be solved in this technical field. Summary of the invention

[0006] In order to solve the problem that existing enterprise information processing technologies are unable to capture complex cross-enterprise and multi-faceted relationships, the present invention provides an interactive relationship extraction and demand and manufacturing matching method based on manufacturing service interactive context information, which has the characteristics of high processing efficiency and can support complex and dynamic manufacturing tasks.

[0007] In order to achieve the above-mentioned purpose of the present invention, the technical scheme adopted is as follows:

[0008] A method for extracting interactive relationships based on manufacturing service interactive context information includes the following specific steps:

[0009] Obtaining manufacturing service interaction context information;

[0010] Create a seed list, extract the relationship of manufacturing service interaction context information based on the seed list, and obtain several manufacturing relationship patterns;

[0011] Construct a heterogeneous network model based on the obtained manufacturing relationship pattern;

[0012] Extracting interactive relationship information from manufacturing relationship patterns through heterogeneous network models.

[0013] Preferably, a seed list is created, and relationship extraction is performed on the manufacturing service interaction context information based on the seed list to obtain a plurality of manufacturing relationship patterns. The specific steps are:

[0014] Define several initial tuples representing specific manufacturing relationships and interaction processes, create an initial seed list based on the initial tuples, use unsupervised learning methods based on the seed list to induce several manufacturing relationship patterns, and build a pattern database;

[0015] Search for instances similar to existing manufacturing relationship patterns in unlabeled manufacturing service interaction context information, add these instances to the pattern database update, perform pattern enhancement iterations on the pattern database, further search for more matching instances, and continuously fill the pattern database.

[0016] Furthermore, the obtained manufacturing relationship patterns are subjected to pattern reinforcement iteration, and the specific steps are as follows:

[0017] The obtained manufacturing relations are represented as tuples:

[0018] T i =<pref;e1;inf1;e2;inf2;rel tag ;suf>

[0019] Among them, pref refers to the prefix of the sentence, inf1 and inf2 refer to the middle part of the sentence, suf refers to the suffix of the sentence, e1 and e2 refer to enterprise 1 and enterprise 2 involved in the interaction respectively, and rel tag It is the corporate logo;

[0020] During the search, the matching degree between tuple i of the unlabeled manufacturing service interaction context information and tuple j in the existing manufacturing relational schema is defined as:

[0021]

[0022] Among them, sim s (i, j) indicates similarity;

[0023] If the similarity between tuple i and tuple j is greater than the set threshold, they are considered similar and the manufacturing relationship pattern i is captured;

[0024] The pattern database is continuously updated and optimized through iterative pattern extraction of seed tuples, construction of relational patterns, pattern enhancement and similarity calculation.

[0025] Furthermore, the similarity sim s (i, j), the specific calculation method is:

[0026] For sentence X in tuple i = (x1, ..., x i , ..., x m ) and the sentence Y in tuple j = (y1, ..., y i , ..., y n ), the function sim(i, j) used to define the similarity of two sentence components is specifically:

[0027]

[0028] Among them, p(x i ) refers to the word x i The possibility of appearing in a sentence, "-" represents a space, p(x i ,y j ) represents x appearing in different sentence components i and j The probability of appearing in the same position.

[0029] A method for matching demand and manufacturing based on manufacturing service interaction context information, based on the pattern extraction method, has the following specific steps:

[0030] Aggregate the interaction relationship information into the database of the heterogeneous network model for training; obtain a heterogeneous community manufacturing network model that supports further demand-capability matching;

[0031] Through the network matching method based on probabilistic graph model, potential relationships are searched from the heterogeneous community manufacturing network model, and the demand patterns are further matched with manufacturing capabilities.

[0032] Preferably, when training a heterogeneous network model, the specific steps of the method for evaluating the interactive relationship information extracted from the heterogeneous network model are:

[0033] Calculate the confidence score of the pattern of extracted interaction relationship information:

[0034]

[0035] Among them, U i : Mode P i The number of unique tuples extracted. Unique tuples refer to the number of non-repeated tuples extracted through the pattern. The more unique tuples there are, the higher the reliability of the pattern. i : Mode P i The total number of tuples extracted. The total number of tuples refers to the number of all tuples extracted by this pattern, including duplicates and non-duplicates; log(U i ): Use a logarithmic function to adjust the number of unique tuples to avoid excessively high scores when a small number of unique tuples are extracted;

[0036] Calculate the confidence score of the tuple of extracted interaction relationship information:

[0037]

[0038] Among them, match(P i , T j ): Mode P i With tuple T j The matching degree reflects the degree to which the pattern matches the tuple T j The higher the matching degree, the closer the relationship between the pattern and the tuple is.

[0039] Furthermore, a network matching method based on a probabilistic graph model is used to search for potential relationships in the heterogeneous community manufacturing network model, and further match the demand pattern with the manufacturing capability. The specific steps are as follows:

[0040] Model the heterogeneous community manufacturing network model as a semantically heterogeneous graph;

[0041] By using graph factorization and loop belief propagation (LBP) algorithm, semantically heterogeneous graphs are mapped to structured manufacturing networks, thus finding potential manufacturing relationships in heterogeneous networks.

[0042] Based on the potential manufacturing relationships, the correspondence between demand patterns and manufacturing capabilities is identified.

[0043] Furthermore, the heterogeneous community manufacturing network model is modeled as a semantic heterogeneous graph, specifically:

[0044] The heterogeneous community manufacturing network model is defined as G = (V, E, vt, et), where V = {v1, v2, ..., v n} represents a node set, E = {e ij |e ij =v i *v j, 1≤i, j≤n} represents the edge set, vt represents the attributes of the nodes, and et represents the attributes of the edges, encoding different types of semantic relations, temporal and spatial dependencies, interactions, and influences.

[0045] Furthermore, by using graph factorization and the loop belief propagation (LBP) algorithm, the semantically heterogeneous graph is mapped to the structured manufacturing network, so as to find the mapping from query nodes to data nodes and transfer information between variable nodes and factor nodes. The specific steps are as follows:

[0046] For a given variable node x i , will be from the factor node f j Pass to x i The message mapping is

[0047]

[0048] in, Represents factor node f j The potential function, x j is with f j All variable nodes connected, x j \x i Indicates division by x i The set of all variable nodes except Represents the variable node x k Passed to factor node f j news;

[0049] will be added from the variable node x i Passed to factor node f j The message mapping is

[0050]

[0051] Among them, f i is with x i All factor nodes connected, f i \f j Indicates division by f j The set of all factor nodes except ;

[0052] The variable node x is calculated by aggregating all incoming messages i The marginal distribution b(x i ):

[0053]

[0054] Based on the above mapping of nodes to data nodes and marginal distribution of nodes, information transfer between variable nodes and factor nodes is realized, network matching and marginal distribution update are achieved, so as to find potential manufacturing relationships.

[0055] Furthermore, based on the above node-to-data node mapping and the marginal distribution of nodes, the information transfer between variable nodes and factor nodes is realized. The specific steps are:

[0056] Set the given variable node x i , from factor node f j Pass to x i The message is represented as

[0057]

[0058] in, Represents factor node f j The potential function, x j is with f j All variable nodes connected, x j \x i Indicates division by x i ;

[0059] The set of all variable nodes except Represents the variable node x k Passed to factor node f j news;

[0060] will be added from the variable node x i Passed to factor node f j The message is represented as

[0061]

[0062] Among them, f i is with x i All factor nodes connected, f i \f j Indicates division by f j The set of all factor nodes except ;

[0063] The variable node x is calculated by aggregating all incoming messages i The marginal distribution of is:

[0064]

[0065] The beneficial effects of the present invention are as follows:

[0066] The present invention discloses an interaction relationship extraction method based on manufacturing service interaction context information. By creating a seed list, the relationship extraction of the manufacturing service interaction context information is performed based on the seed list, thereby realizing the extraction of unstructured interaction data from the interaction, and building a heterogeneous network model on this basis, thereby improving the processing efficiency of unstructured data and realizing support for complex and dynamic manufacturing tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a flow chart of an interactive relationship extraction method based on manufacturing service interactive context information of the present invention.

[0068] Figure 2 It is a schematic diagram of the basic principle of an interactive relationship extraction method based on manufacturing service interactive context information of the present invention in Example 1.

[0069] Figure 3 It is a flow chart of a method for matching demand and manufacturing based on manufacturing service interaction context information of the present invention.

[0070] Figure 4 It is a factor diagram of a demand and manufacturing matching method based on manufacturing service interaction context information of the present invention.

[0071] Figure 5 It is a schematic diagram of a specific implementation flow of a method for matching demand and manufacturing based on manufacturing service interaction context information in Example 3 of the present invention. DETAILED DESCRIPTION

[0072] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0073] Example 1

[0074] like Figure 1 As shown, a method for extracting interaction relations based on manufacturing service interaction context information includes the following specific steps:

[0075] Obtaining manufacturing service interaction context information;

[0076] Create a seed list, extract the relationship of manufacturing service interaction context information based on the seed list, and obtain several manufacturing relationship patterns;

[0077] Construct a heterogeneous network model based on the obtained manufacturing relationship pattern;

[0078] Extracting interactive relationship information from manufacturing relationship patterns through heterogeneous network models.

[0079] In this embodiment, the principle of the interactive relationship extraction method is as follows Figure 2As shown, the present invention introduces a semi-supervised learning algorithm to automatically extract relationships from unstructured multi-scale integrated structures based on iterative tuple updating and approximate pattern matching, wherein the seed database stores a seed list, the context database stores manufacturing service interaction context information, and the network database stores interaction relationship information.

[0080] In a specific embodiment, a seed list is created, and relationship extraction is performed on manufacturing service interaction context information based on the seed list to obtain a plurality of manufacturing relationship patterns. The specific steps are as follows:

[0081] Define several initial tuples representing specific manufacturing relationships and interaction processes, create an initial seed list based on the initial tuples, use unsupervised learning methods based on the seed list to induce several manufacturing relationship patterns, and build a pattern database;

[0082] In this embodiment, when inducing several manufacturing relationship patterns, different unsupervised learning induction methods can be adopted according to the form of manufacturing service interaction context information. For example, for identifying the cooperation model between different manufacturing enterprises, an association rule learning algorithm can be adopted; and if grouping-related tasks are more needed, clustering algorithms such as K-Means and hierarchical clustering can be adopted.

[0083] Search for instances similar to existing manufacturing relationship patterns in unlabeled manufacturing service interaction context information, add these instances to the pattern database update, perform pattern enhancement iterations on the pattern database, further search for more matching instances, and continuously fill the pattern database.

[0084] In this embodiment, after obtaining the instance, the seed list is updated by adding the instance to the seed list, so as to extract the pattern of the next set of manufacturing service interaction context information.

[0085] In a specific embodiment, the obtained multiple manufacturing relationship patterns are subjected to pattern enhancement iteration, and the specific steps are as follows:

[0086] The obtained manufacturing relations are represented as tuples:

[0087] T i = <pref;e1;inf1;e2;inf2;rel tag ;suf>

[0088] Among them, pref refers to the prefix of the sentence, inf1 and inf2 refer to the middle part of the sentence, suf refers to the suffix of the sentence, e1 and e2 refer to enterprise 1 and enterprise 2 involved in the interaction respectively, and rel tag It is the corporate logo;

[0089] In this embodiment, taking the sentence “JUNYE provided BEIREN with deep-holedrillingservice.” as an example, it can be represented as the following tuple: {“”; <enterprise>;"provided”; <enterprise>; "with"; "service"; ""}. This pattern can summarize a considerable portion of the interaction data in MSICs. Since the number of patterns of the initial seed tuples is limited, the system retrieves context sentences matching these patterns to generate more new seed tuples, and then uses these new tuples to start the next round of iterations to capture new and more credible patterns. Here, sentence alignment methods are used to match similar patterns, and then these patterns are classified for learning.

[0090] During the search, the matching degree between tuple i of the unlabeled manufacturing service interaction context information and tuple j in the existing manufacturing relational schema is defined as:

[0091]

[0092] Among them, sim s (i,j) represents similarity;

[0093] If the similarity between tuple i and tuple j is greater than the set threshold, they are considered similar and the manufacturing relationship pattern i is captured;

[0094] The pattern database is continuously updated and optimized through iterative pattern extraction of seed tuples, construction of relational patterns, pattern enhancement and similarity calculation.

[0095] In a specific embodiment, the similarity sim s (i, j), the specific calculation method is:

[0096] For sentence X in tuple i = (x1, ..., x i , ..., x m ) and the sentence Y in tuple j = (y1, ..., y j , ..., y n ), the function sim(i, j) used to define the similarity of two sentence components is specifically:

[0097]

[0098] Among them, p(x i ) refers to the word x i The possibility of appearing in a sentence, "-" represents a space, p(x i ,y j ) represents x appearing in different sentence components i and j The probability of appearing in the same position.

[0099] In this embodiment, the extracted interaction relationships are aggregated into a heterogeneous network database including organizational, temporal, informational, and social aspects.

[0100] Example 2

[0101] like Figure 3 As shown, a method for matching demand and manufacturing based on manufacturing service interaction context information is based on the pattern extraction method, and its specific steps are:

[0102] Aggregate the interaction relationship information into the database of the heterogeneous network model for training; obtain a heterogeneous community manufacturing network model that supports further demand-capability matching;

[0103] Through the network matching method based on probabilistic graph model, potential relationships are searched from the heterogeneous community manufacturing network model, and the demand patterns are further matched with manufacturing capabilities.

[0104] In a specific embodiment, when training a heterogeneous network model, it is crucial to perform quality assessment on the extracted tuples. The quality assessment mainly includes two aspects: the confidence score of the pattern and the confidence score of the tuple. The confidence score of the pattern is used to measure the confidence of a certain pattern (pattern P i ) reliability. This score reflects the effectiveness and accuracy of the pattern in extracting unique tuples. The specific steps of the evaluation method for the interaction relationship information extracted by the heterogeneous network model are:

[0105] Calculate the confidence score of the pattern of extracted interaction relationship information:

[0106]

[0107] Among them, U i :Mode P i The number of unique tuples extracted. Unique tuples refer to the number of non-repeated tuples extracted through the pattern. The more unique tuples there are, the higher the reliability of the pattern. i :Mode P i The total number of tuples extracted. The total number of tuples refers to the number of all tuples extracted by this pattern, including duplicates and non-duplicates; log(U i ): Use a logarithmic function to adjust the number of unique tuples to avoid excessively high scores when a small number of unique tuples are extracted;

[0108] Calculate the confidence score of the tuple of extracted interaction relationship information:

[0109]

[0110] Among them, match(P i ,T j ): Mode P i With tuple T j The matching degree reflects the degree to which the pattern matches the tuple T j The higher the matching degree, the closer the relationship between the pattern and the tuple is.

[0111] In this way, the algorithm continuously expands its ability to identify specific manufacturing relationships, making the information extracted from text data increasingly rich and accurate. On this basis, the obtained pattern can obtain quite a lot of useful interaction information, through which a heterogeneous community manufacturing network that supports further demand-capability matching can be constructed.

[0112] In a specific embodiment, a network matching method based on a probabilistic graph model is used to search for potential relationships from a heterogeneous community manufacturing network model, and further match demand patterns with manufacturing capabilities. The specific steps are as follows:

[0113] Model the heterogeneous community manufacturing network model as a semantically heterogeneous graph;

[0114] By using graph factorization and loop belief propagation (LBP) algorithm, semantically heterogeneous graphs are mapped to structured manufacturing networks, thus finding potential manufacturing relationships in heterogeneous networks.

[0115] Based on the potential manufacturing relationships, the correspondence between demand patterns and manufacturing capabilities is identified.

[0116] In a specific embodiment, the heterogeneous community manufacturing network model is modeled as a semantic heterogeneous graph, specifically:

[0117] The heterogeneous community manufacturing network model is defined as G = (V, E, vt, et), where V = {v1, v2, ..., v n } represents a node set, E = {e ij |e ij =v i *v j , 1≤i, j≤n} represents the edge set, vt represents the attributes of the nodes, and et represents the attributes of the edges, encoding different types of semantic relations, temporal and spatial dependencies, interactions, and influences.

[0118] In this embodiment, the heterogeneous social manufacturing network graph (HSMN) includes multiple types of nodes and edges, where the nodes represent manufacturing enterprises, resources, demands, etc., and the edges represent the interactive relationships and matching rules between enterprises.

[0119] In a specific embodiment, the semantically heterogeneous graph is mapped to the structured manufacturing network using graph factorization and the ring belief propagation LBP algorithm, so as to find the mapping from the query node to the data node. The LBP algorithm implements local message passing by passing real-valued messages between factor and variable nodes, and calculates the marginal distribution and matching belief of the nodes. The specific steps of passing information between variable nodes and factor nodes are as follows:

[0120] For a given variable node x i , will be from the factor node f j Pass to x i The message mapping is

[0121]

[0122] in, Represents factor node f j The potential function, x j is with f j All variable nodes connected, x j \x i Indicates division by x i The set of all variable nodes except Represents the variable node x k Passed to factor node f j news;

[0123] will be added from the variable node x i Passed to factor node f j The message mapping is

[0124]

[0125] Among them, f i is with x i All factor nodes connected, f i \f j Indicates division by f j The set of all factor nodes except ;

[0126] The variable node x is calculated by aggregating all incoming messages i The marginal distribution b(x i ):

[0127]

[0128] Based on the above mapping of nodes to data nodes and marginal distribution of nodes, information transfer between variable nodes and factor nodes is realized, network matching and marginal distribution update are achieved, so as to find potential manufacturing relationships.

[0129] In this embodiment, through the above formula, the LBP algorithm effectively transfers information between variable nodes and factor nodes in the network, realizes network matching and updates marginal distribution, and thus finds potential manufacturing relationships in heterogeneous networks.

[0130] In this embodiment, in order to handle the complexity of the network and perform network search and reasoning, the system uses a network matching algorithm based on a probabilistic graph model to find potential relationships from the established heterogeneous network, and adopts a new algorithm for pattern matching between requirements and capabilities based on graph decomposition and ring belief propagation. In order to solve the problem that accurate pattern matching is not suitable for development where noisy and missing data often occur, the present invention proposes an approximate pattern detection algorithm that can identify and infer subgraphs in the data that do not fully match the query pattern. This approximate matching method is robust to noise and incomplete data and can find the most likely matching pattern in complex manufacturing network data.

[0131] In this embodiment, LBP is a local message passing algorithm on the factor graph, which is implemented by passing real-valued messages between factor and variable nodes. These messages carry beliefs about the matching of query nodes and edges in the data. Using the LBP algorithm, a mapping from query nodes to data nodes is found, and information is passed between variable nodes and factor nodes, such as Figure 4 As shown in Figure 1. In the network model, variable nodes represent manufacturing enterprises, resources or demands, such as enterprise 1, enterprise 2, etc.; factor nodes represent the interaction relationship and matching rules between enterprises, such as cooperative relationship, supply chain interaction, etc. The information transmission between variable nodes and factor nodes is realized by the ring belief propagation algorithm, which is used to calculate the marginal distribution and matching belief of nodes. The LBP algorithm performs local message transmission on the factor graph, through the variable node x i and factor node f j The marginal distribution is updated by passing information between them.

[0132] In a specific embodiment, based on the above node-to-data node mapping and the marginal distribution of nodes, information transfer between variable nodes and factor nodes is implemented, and the specific steps are as follows:

[0133] Set the given variable node x i , from factor node f j Pass to x i The message is represented as

[0134]

[0135] in, Represents factor node f j The potential function, x j is with f j All variable nodes connected, x j \x i Indicates division by x i ;

[0136] The set of all variable nodes except Represents the variable node x k Passed to factor node f j news;

[0137] will be added from the variable node x i Passed to factor node f j The message is represented as

[0138]

[0139] Among them, f i is with x i All factor nodes connected, f i \f j Indicates division by f j The set of all factor nodes except ;

[0140] The variable node x is calculated by aggregating all incoming messages i The marginal distribution of is:

[0141]

[0142] Through the above formula, the LBP algorithm can effectively transfer information between variable nodes and factor nodes in the network, realize network matching and update marginal distribution. Using these beliefs, the system can perform network matching, that is, identify the correspondence between demand patterns and manufacturing capabilities. For example, a demand pattern may require finding a manufacturer that can provide specific processing services. The LBP algorithm will determine that the manufacturer has the corresponding capabilities by reasoning about the entities and relationships in the network. A key advantage of the LBP algorithm is its scalability. It can effectively handle large-scale networks containing thousands or even tens of thousands of nodes, which makes it particularly suitable for complex manufacturing network environments. Through local computing and distributed message passing, the LBP algorithm avoids the need for global computing of the entire network, thereby greatly saving computing resources and time.

[0143] After obtaining the probability map and the corresponding information, the most likely mapping can be extracted from the updated marginal distribution through the decoding process to provide decision support for the matching between demand and capability. This method not only has advantages in accuracy and efficiency, but also has significant robustness in dealing with uncertainty and incomplete information.

[0144] Example 3

[0145] The specific implementation process of the demand and manufacturing matching method based on manufacturing service interaction context information of the present invention is as follows: Figure 5 As shown, it can be seen that in the relationship extraction stage: the present invention uses a multi-attribute graph to formulate network data, which can encode various entities and relationships. The system first extracts unstructured interaction data from the interaction, extracts the effective content in the interaction data, forms semi-structured transaction data, and on this basis, builds a structured manufacturing network graph as a heterogeneous network model to facilitate relationship matching later;

[0146] In the relationship matching stage: The relationship matching stage is based on probabilistic multi-attribute graph matching and is implemented using a distributed message passing algorithm. At the beginning of the matching stage, the user puts forward specific requirements, and the system converts these requirements into query graph representation and inputs them. Then, the query graph is mapped to the structured manufacturing network using graph factorization, LBP, and other methods to find the best matching result.

[0147] It can be seen from the specific implementation process that, in view of these limitations of the prior art, the present invention:

[0148] Improved processing efficiency of unstructured data: This invention aims to improve the speed and accuracy of extracting structured information from unstructured MSICs by introducing advanced data preprocessing and pattern learning technologies. This includes cleaning, standardization, keyword extraction, etc. of text data, as well as identifying repeated patterns and relationships in data through machine learning methods, thereby effectively reducing noise and improving data quality and availability.

[0149] Deep extraction and utilization of manufacturing relationships: The present invention uses advanced relationship extraction technology to deeply mine the complex relationships in MSICs, such as partnerships, supply and demand relationships, etc. This includes not only identifying the existence of these relationships, but also understanding their nature and strength, and how to effectively use these relationships to support decision-making and optimization. In this way, the present invention aims to provide more accurate and insightful decision support.

[0150] Achieved group-level demand and capability matching: Traditional methods often focus on individual-level matching, while ignoring the complex needs at the group or organizational level. This invention aims to effectively support group-level demand and capability matching by introducing network matching and probabilistic graph matching technology. This can not only improve the efficiency and effectiveness of resource integration, but also support more complex and dynamic manufacturing tasks.

[0151] Obviously, the above embodiments of the present invention are only examples for clearly explaining the present invention, and are not intended to limit the implementation methods of the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention shall be included in the protection scope of the claims of the present invention.< / enterprise> < / enterprise>

Claims

1. A method for extracting interactive relationships based on manufacturing service interactive context information, characterized in that: The specific steps include: Obtaining manufacturing service interaction context information; Create a seed list, extract the relationship of manufacturing service interaction context information based on the seed list, and obtain several manufacturing relationship patterns; Construct a heterogeneous network model based on the obtained manufacturing relationship pattern; Extracting interactive relationship information from manufacturing relationship patterns through heterogeneous network models.

2. The interactive relationship extraction method based on manufacturing service interactive context information according to claim 1 is characterized by: Create a seed list, extract the relationship of manufacturing service interaction context information based on the seed list, and obtain several manufacturing relationship patterns. The specific steps are as follows: Define several initial tuples representing specific manufacturing relationships and interaction processes, create an initial seed list based on the initial tuples, use unsupervised learning methods based on the seed list to induce several manufacturing relationship patterns, and build a pattern database; Search for instances similar to existing manufacturing relationship patterns in unlabeled manufacturing service interaction context information, add these instances to the pattern database update, perform pattern enhancement iterations on the pattern database, further search for more matching instances, and continuously fill the pattern database.

3. The interactive relationship extraction method based on manufacturing service interactive context information according to claim 2 is characterized by: The obtained manufacturing relationship patterns are iterated and strengthened. The specific steps are as follows: The obtained manufacturing relations are represented as tuples: T i =<pref;e1;inf1;e2;inf2;rel tag ;suf> Among them, pref refers to the prefix of the sentence, inf1 and inf2 refer to the middle part of the sentence, suf refers to the suffix of the sentence, e1 and e2 refer to enterprise 1 and enterprise 2 involved in the interaction respectively, and rel tag It is the corporate logo; During the search, the matching degree between tuple i of the unlabeled manufacturing service interaction context information and tuple j in the existing manufacturing relational schema is defined as: Among them, sim s (i,j) represents similarity; If the similarity between tuple i and tuple j is greater than the set threshold, they are considered similar and the manufacturing relationship pattern i is captured; The pattern database is continuously updated and optimized through iterative pattern extraction of seed tuples, construction of relational patterns, pattern enhancement and similarity calculation.

4. The interactive relationship extraction method based on manufacturing service interactive context information according to claim 3 is characterized by: The similarity sim s (i,j), the specific calculation method is: For sentence X in tuple i = (x1,…,x i ,…,x m ) and the sentence Y in tuple j = (y1,…,y j ,…,y n ), the function sim(i,j) used to define the similarity of two sentence components is specifically: Among them, p(x i ) refers to the word x i The possibility of appearing in a sentence, "-" represents a space, p(x i ,y j ) represents x appearing in different sentence components i and j The probability of appearing in the same position.

5. A method for matching demand and manufacturing based on manufacturing service interaction context information, characterized in that: The pattern extraction method according to any one of claims 1 to 4, wherein the specific steps are: Aggregate the interaction relationship information into the database of the heterogeneous network model for training; obtain a heterogeneous community manufacturing network model that supports further demand-capability matching; Through the network matching method based on probabilistic graph model, potential relationships are searched from the heterogeneous community manufacturing network model, and the demand patterns are further matched with manufacturing capabilities.

6. The method for matching demand and manufacturing based on manufacturing service interaction context information according to claim 5, characterized in that: When training a heterogeneous network model, the specific steps of the evaluation method for the interactive relationship information extracted by the heterogeneous network model are as follows: Calculate the confidence score of the pattern of extracted interaction relationship information: Among them, U i :Mode P i The number of unique tuples extracted. Unique tuples refer to the number of non-repeated tuples extracted through the pattern. The more unique tuples there are, the higher the reliability of the pattern. i :Mode P i The total number of tuples extracted. The total number of tuples refers to the number of all tuples extracted by this pattern, including duplicates and non-duplicates; log(U i ): Use a logarithmic function to adjust the number of unique tuples to avoid excessively high scores when a small number of unique tuples are extracted; Calculate the confidence score of the tuple of extracted interaction relationship information: Among them, match(P i ,T j ): Mode P i With tuple T j The matching degree reflects the degree to which the pattern matches the tuple T j The higher the matching degree, the closer the relationship between the pattern and the tuple is.

7. The method for matching demand and manufacturing based on manufacturing service interaction context information according to claim 6, characterized in that: Through the network matching method based on the probabilistic graph model, potential relationships are searched from the heterogeneous community manufacturing network model, and the demand pattern and manufacturing capacity are further matched. The specific steps are as follows: Model the heterogeneous community manufacturing network model as a semantically heterogeneous graph; By using graph factorization and loop belief propagation (LBP) algorithm, semantically heterogeneous graphs are mapped to structured manufacturing networks, thus finding potential manufacturing relationships in heterogeneous networks. Based on the potential manufacturing relationships, the correspondence between demand patterns and manufacturing capabilities is identified.

8. The method for matching demand and manufacturing based on manufacturing service interaction context information according to claim 7, characterized in that: The heterogeneous community manufacturing network model is modeled as a semantic heterogeneous graph, specifically: The heterogeneous community manufacturing network model is defined as G = (V, E, vt, et), where V = {v1, v2, ..., v n } represents a node set, E = {e ij |e ij =v i *v j ,1≤i,j≤n} represents the edge set, vt represents the attributes of the nodes, and et represents the attributes of the edges, encoding different types of semantic relations, temporal and spatial dependencies, interactions, and influences.

9. The method for matching demand and manufacturing based on manufacturing service interaction context information according to claim 8, characterized in that: By using graph factorization and loop belief propagation (LBP) algorithm, the semantically heterogeneous graph is mapped to the structured manufacturing network, so as to find the mapping from query nodes to data nodes and transfer information between variable nodes and factor nodes. The specific steps are as follows: For a given variable node x i , will be from the factor node f j Pass to x i The message mapping is in, Represents factor node f j The potential function, x j is with f j All variable nodes connected, x j \x i Indicates division by x i The set of all variable nodes except Represents the variable node x k Passed to factor node f j news; will be added from the variable node x i Passed to factor node f j The message mapping is Among them, f i is with x i All factor nodes connected, f i \f j Indicates division by f j The set of all factor nodes except ; The variable node x is calculated by aggregating all incoming messages i The marginal distribution b(x i ): Based on the above mapping of nodes to data nodes and marginal distribution of nodes, information transfer between variable nodes and factor nodes is realized, network matching and marginal distribution update are achieved, so as to find potential manufacturing relationships.

10. The method for matching demand and manufacturing based on manufacturing service interaction context information according to claim 9, characterized in that: Based on the above node-to-data node mapping and node marginal distribution, information transfer between variable nodes and factor nodes is realized. The specific steps are as follows: Set the given variable node x i , from factor node f j Pass to x i The message is represented as in, Represents factor node f j The potential function, x j is with f j All variable nodes connected, x j \x i Indicates division by x i ; The set of all variable nodes except Represents the variable node x k Passed to factor node f j news; will be added from the variable node x i Passed to factor node f j The message is represented as Among them, f i is with x i All factor nodes connected, f i \f j Indicates division by f j The set of all factor nodes except ; The variable node x is calculated by aggregating all incoming messages i The marginal distribution of is: