Method, device and medium for determining high correlation degree criterion

Abstract

The present application relates to a kind of high correlation degree standard determination method, equipment and medium.The method comprises: based on reference, cited and substitution relationship in standard knowledge graph, determine the one-hop candidate standard of target standard and form candidate standard set;When the number of one-hop candidate standard is less than the preset candidate number threshold, it is used as intermediate standard, two-hop candidate standard is obtained along the reference relationship and is added to the set;According to the relationship type and graph distance between target standard and each candidate standard, determine the graph relationship quantization value, and calculate semantic similarity;Fusion graph relationship quantization value and semantic similarity determine fusion correlation degree, and accordingly output the correlation standard recommendation result.The present application can expand the candidate range when direct correlation standard is insufficient, and consider structure correlation and content correlation degree, improve the integrity and accuracy of recommendation result.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph and information recommendation technology, and more specifically, to a method, device and medium for determining high relevance standards. Background Technology

[0002] With the continuous improvement of the standards system and the increasing number of standard documents, different standards have gradually formed relationships such as citation, being cited, and version replacement. When users consult or apply a certain standard, they often need to obtain other standards related to that standard in order to understand the standards system, supporting requirements, and version compatibility of the corresponding technical field. Therefore, how to accurately identify related standards from a large number of standards has become an important issue in standards information services.

[0003] In existing technologies, text matching is usually performed based on the standard name, keywords, abstract, or text content, and related standards are filtered and sorted according to the degree of text similarity. Some methods also utilize the citation information or version information recorded in the standard documents to retrieve standards directly related to the target standard and provide the search results to the user.

[0004] However, there may be limited direct correlation information between standards. When searching based solely on direct correlation, the number of candidate standards obtained is limited, and standards that are indirectly related to the target standard may be overlooked. At the same time, recommending standards based solely on text similarity or direct correlation information makes it difficult to take into account both the degree of correlation between standards and the degree of content relevance, which may result in insufficient completeness and accuracy of the recommended standards. Summary of the Invention

[0005] Therefore, it is necessary to address the above-mentioned problems by providing a method, device, and medium for determining highly relevant standards. This method expands the range of candidate standards when the number of candidate standards directly related to the target standard is insufficient, and comprehensively considers the degree of correlation and content relevance between standards to improve the completeness and accuracy of the recommended results of relevant standards.

[0006] In a first aspect, the present invention provides a method for determining a high-relevance standard, the method comprising: determining a one-hop candidate standard for a target standard based on reference relationships, referenced relationships, and substitution relationships in a standard knowledge graph, and determining a set of candidate standards based on the one-hop candidate standards; if the number of one-hop candidate standards is less than a preset candidate number threshold, using the one-hop candidate standards as intermediate standards, determining two-hop candidate standards with two-hop relationships to the target standard along the reference relationships, and adding the two-hop candidate standards to the set of candidate standards; determining a graph relationship quantification value for each candidate standard based on the relationship type and graph distance between the target standard and each candidate standard in the set of candidate standards, and calculating the semantic similarity between the target standard and each candidate standard; determining a fusion relevance degree for each candidate standard based on the graph relationship quantification value and the semantic similarity; and determining a high-relevance standard based on the fusion relevance degree.

[0007] Optionally, determining the graph relationship quantification value of each candidate standard based on the relationship type and graph distance between the target standard and each candidate standard in the candidate standard set includes: obtaining the effective relationship paths between the target standard and the candidate standards; and calculating the path score of each effective relationship path according to the following formula; ; in, This represents a valid relationship path between the target standard and the candidate standard; Indicates the effective relationship path Path scoring, Indicates the effective relationship path The relationship edge in the middle, Represents the relation edge Relationship types, Indicates the relation type The corresponding preset relationship weights, Indicates the effective relationship path Graph distance; The highest path score among all the path scores is determined as the graph relation quantification value of the candidate standard.

[0008] In the above implementation process, by multiplying the weights of different relation types in the associated path and combining them with the path length to determine the path score, the path score can be made to change with the relation type and the association level. Furthermore, by using the highest path score as the graph relation quantification value of the candidate standard, the path with a high degree of association between the target standard and the candidate standard can be characterized, reducing the undue influence of longer or weakly associated paths on the recommendation score.

[0009] Optionally, the method further includes: when adding, revising, or replacing standards, determining the changed standard nodes and relationships in the standard knowledge graph; updating the changed standard nodes and relationships; recalculating the standard semantic vectors corresponding to the changed standard nodes; identifying standards with a graph distance of less than or equal to 2 from the changed standard nodes as affected standards, and updating the associated standard recommendation results corresponding to the affected standards.

[0010] In the above implementation process, when a standard is added, revised, or replaced, by updating the changed graph nodes, graph relationships, and standard semantic vectors, and only updating the recommendation results of the affected standard whose graph distance to the changed nodes is no greater than 2, the recommendation results can be kept consistent with the changes in the standard knowledge graph. At the same time, the recalculation is limited to the affected range, reducing the amount of computation required to update the knowledge graph.

[0011] In a second aspect, the present invention also provides an electronic device, including a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the method for determining the high correlation standard as described in any of the preceding claims.

[0012] Thirdly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method for determining high correlation criteria as described in any of the preceding claims.

[0013] This invention determines one-hop candidate standards for a target standard based on reference relationships, cited relationships, and substitution relationships in a standard knowledge graph, covering different association directions and version evolution relationships. When the number of one-hop candidate standards is insufficient, two-hop candidate standards are further determined based on the one-hop candidate standards, which can supplement standards that are indirectly related to the target standard and reduce candidate omissions caused by only searching directly related standards. Furthermore, the graph relationship quantification value is determined based on the relationship type and graph distance, and the fusion association degree is determined by combining the semantic similarity between the target standard and the candidate standards. This allows the recommendation results to reflect both the structural association degree and the content relevance degree between standards, thereby expanding the range of candidate standards while reducing the possibility of standards with weak association degree or low content relevance being given priority recommendation, and improving the completeness and accuracy of the association standard recommendation results. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the conventional technology, the drawings used in the description of the embodiments or the conventional technology will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0015] Figure 1 This is a flowchart of a method for determining high correlation standards provided in one embodiment of the present invention; Figure 2 This is a flowchart of a standard knowledge graph construction process provided in one embodiment of the present invention; Figure 3 This is a flowchart of the association standard recommendation process provided in one embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided in one embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will now be described with reference to the accompanying drawings. For example, the flowcharts and block diagrams in the drawings illustrate the architecture, functions, and operations of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagram may represent a module, program segment, or part of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or can be implemented using a combination of dedicated hardware and computer instructions. In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0017] The method for determining highly relevant standards provided in this embodiment can be applied to standard information service platforms, standard management platforms, or standard retrieval systems. The method for determining highly relevant standards can be executed by a server. The server can receive the target standard selected by the user, determine candidate standards based on a standard knowledge graph, and generate recommended results for relevant standards based on the graph relationship and semantic relevance between the candidate standards and the target standard.

[0018] A standard knowledge graph is graph data composed of standards as nodes and the relationships between standards as edges. The target standard is the standard targeted in this association standard recommendation. A one-hop candidate standard is a standard connected to the target standard through one edge; a two-hop candidate standard is a standard connected to the target standard through two edges. Graph distance represents the number of edges contained in the relationship path between the target standard and candidate standards.

[0019] Please see Figure 1 , Figure 1 A flowchart illustrating a method for determining a high correlation criterion provided for an embodiment of the present invention. The method for determining the high correlation criterion can be executed by a server and includes the following steps: Step S101: Based on the reference relationships, referenced relationships, and substitution relationships in the standard knowledge graph, determine the one-hop candidate standards for the target standard, and determine the candidate standard set according to the one-hop candidate standards; if the number of one-hop candidate standards is less than a preset candidate number threshold, use the one-hop candidate standards as intermediate standards, determine the two-hop candidate standards that have a two-hop relationship with the target standard along the reference relationships, and add the two-hop candidate standards to the candidate standard set. Step S102: Based on the relationship type and graph distance between the target standard and each candidate standard in the candidate standard set, determine the graph relationship quantization value of each candidate standard, and calculate the semantic similarity between the target standard and each candidate standard; Step S103: Determine the fusion correlation degree of each candidate standard based on the graph relationship quantification value and the semantic similarity of each candidate standard; Step S104: Determine the high correlation standard based on the fusion correlation degree.

[0020] The following sections will explain each step.

[0021] Specifically, in step S101, the server can determine the target node corresponding to the target standard based on the standard number, standard name, or node identifier in the standard knowledge graph input by the user. The server can query the relationship edges connected to the target node and obtain the standard nodes connected through referencing, being referenced, and substitution relationships, respectively.

[0022] Among them, the referencing relationship represents a directed association where one standard references another standard; the referenced relationship represents a reverse association between the referenced standard and the referencing standard; and the substitution relationship represents a version evolution association between the substituted standard and the replaced standard. The server can identify standards that are directly connected to the target standard through any of the aforementioned relationships as one-hop candidate standards.

[0023] When the same standard is connected to the target standard through multiple relational edges, the server can deduplicate one-hop candidate standards based on the standard number or the node identifier in the standard knowledge graph. The server can then add the deduplicated one-hop candidate standards to the candidate standard set and count the number of one-hop candidate standards.

[0024] The preset candidate number threshold represents the lower limit of the number of candidate criteria required to perform recommendation ranking. The preset candidate number threshold can be preset according to the number of recommendation results, the size of the standard knowledge graph, or business needs. For example, it can be set to 10, 20, or other positive integers.

[0025] When the number of one-hop candidate standards reaches a preset threshold, the server can directly use the set of candidate standards composed of one-hop candidate standards as the subsequent scoring object. When the number of one-hop candidate standards is less than the preset threshold, the server can use each one-hop candidate standard as an intermediate standard to continue querying the reference relationship originating from the intermediate standard, and determine the next-level standard connected through the reference relationship as the second-hop candidate standard. The server can deduplicate the second-hop candidate standards, exclude the target standard and standards already existing in the candidate standard set, and then add the second-hop candidate standards to the candidate standard set.

[0026] Specifically, in step S102, the server can query the relationship path between the target standard and the candidate standards, and determine the relationship types and the number of relationship edges contained in the relationship path. The server can set corresponding relationship contribution values ​​for referencing relationships, referenced relationships, and substitution relationships, and determine the degree of relationship contribution between the candidate standards and the target standard based on the relationship types in the relationship path.

[0027] The server can also determine the graph distance based on the number of relation edges contained in the relation path. The smaller the graph distance between the candidate criterion and the target criterion, the more direct the structural association between the candidate criterion and the target criterion is usually. The server can combine the relation contribution value corresponding to the relation type and the distance decay value corresponding to the graph distance to obtain the graph relation quantification value of the candidate criterion.

[0028] Semantic similarity is used to characterize the degree of relevance between target and candidate standards in terms of text content. The server can obtain the standard text of both the target and candidate standards, use a text semantic encoding model to convert the standard text into semantic vectors of the same dimension, and then calculate the semantic similarity between the target and candidate standards using vector similarity. The greater the semantic similarity, the closer the text content of the target and candidate standards is.

[0029] Specifically, in step S103, the server can convert the graph relation quantification value and semantic similarity to the same numerical range. The server can set fusion weights for the graph relation quantification value and semantic similarity respectively, and weight the graph relation quantification value and semantic similarity according to the fusion weights to obtain the fusion relevance of the candidate criteria.

[0030] The fusion weights can be pre-set according to the standard recommendation scenario. In application scenarios that emphasize the association of the standard system, the fusion weight corresponding to the graph relationship quantification value can be increased; in application scenarios that emphasize standard content retrieval, the fusion weight corresponding to semantic similarity can be increased. Regardless of the fusion weight used, both the graph relationship quantification value and semantic similarity participate in determining the fusion relevance.

[0031] Specifically, in step S104, the server can sort the candidate standards from high to low according to their fusion relevance, and select a preset number of candidate standards from the sorted candidates. The server can combine the standard number, standard name, and fusion relevance of the selected candidate standards to form a recommended association standard result, and output the recommended association standard result through the display interface or data interface of the standard information service platform.

[0032] In this embodiment, first-hop candidate criteria are determined through referencing, being referenced, and substitution relationships, covering different association directions and standard version evolution relationships. When the number of first-hop candidate criteria is insufficient, second-hop candidate criteria are added to reduce candidate omissions caused by only searching directly related criteria. The graph relationship quantification value is determined based on relationship type and graph distance, and the fusion relevance is determined by combining semantic similarity, allowing the recommended criteria to simultaneously reflect the structural association degree and content relevance between criteria. Therefore, the method for determining high-relevance criteria can expand the range of candidate criteria while reducing the likelihood that criteria with weak structural association or low content relevance will be prioritized.

[0033] Alternatively, in one embodiment, please refer to Figure 2 , Figure 2 This is a flowchart of a standard knowledge graph construction process provided in one embodiment of the present invention. Figure 2The standard knowledge graph construction process shown can be executed by the server in an offline phase. In this example, the server can obtain the full text of the standard file to be added and the corresponding metadata information in the offline phase; extract standard numbers, chapters, terms, etc. from the standard text as graph node entities (corresponding to extracting standard numbers from the standard text in this application and using the standard corresponding to the standard file to be added as the standard node, where the standard number serves as the node identifier of the standard node, and the chapters, terms, etc. serve as the attribute information of the standard node); establish three types of directed relationships between the extracted graph nodes: reference, referenced, and substitution (corresponding to establishing reference, referenced, and substitution relationships between standard nodes in this application); and precipitate the extracted entities and relationships into a standard knowledge graph that supports incremental edge writing when a new standard is added to the database (wherein incremental edge writing corresponds to an implementation method of incrementally updating relationship edges when a new standard is added to the database in this application). The specific construction process of the standard knowledge graph may include the following steps: Step S201: Obtain the normative reference information and version substitution information of the standard documents to be stored.

[0034] Specifically, the standard documents included in the database can be structured standard documents, semi-structured standard documents, or standard documents that have undergone text recognition processing. The server can extract normative reference information from the normative reference section of the included standard documents. The normative reference information may include the standard number of the citing standard and the standard number of the cited standard.

[0035] The server can also extract version replacement information from the version descriptions, replacement descriptions, or standard metadata of the standard files in the database. Version replacement information may include the standard number of the replaced standard and the standard number of the replaced standard.

[0036] The server can standardize the extracted standard numbers according to a unified standard numbering format. For example, the server can standardize the character format, separators, and year identifiers in the standard numbers so that the same standard can be identified through the standardized standard numbers.

[0037] Step S202: Using the standard corresponding to the standard file in the database as a node, determine the referencing standard and the referenced standard according to the normative reference information, establish a reference relationship from the referencing standard to the referenced standard, and establish a reference relationship from the referenced standard to the referencing standard.

[0038] Specifically, the server can query the standard knowledge graph based on the normalized standard number. If the corresponding node does not exist in the standard knowledge graph, the server can create the corresponding standard node; if the corresponding node already exists in the standard knowledge graph, the server can reuse the existing standard node.

[0039] For a normative reference, the server can identify the standard containing the section of the normative reference document as the referencing standard and the standards listed in the section of the normative reference document as the cited standard. The server can establish a reference relationship from the referencing standard node to the cited standard node, and simultaneously establish a reference relationship from the cited standard node to the referencing standard node.

[0040] Using the opposite direction for the referencing and being referenced relationships allows the server to query the association criteria from the perspectives of the referrer and the referenced party respectively, avoiding the need to traverse all referencing relationships each time a reverse query is performed.

[0041] Step S203: Based on the version replacement information, establish a replacement relationship from the replaced standard to the replaced standard.

[0042] Specifically, the server can determine the replaced standard node and the replaced standard node based on the standard number in the version replacement information, and establish a replacement relationship from the replaced standard node to the replaced standard node. When there are consecutive version replacements, the server can save the replacement relationship between adjacent versions, thereby forming a standard version replacement chain.

[0043] In this embodiment, by using standards as nodes and establishing clearly defined reference, referenced, and substitution relationships, the association information scattered in the standard files can be converted into structured association data that can be used for graph queries, enabling the subsequent recommendation process to determine the associated standards from the reference direction, the referenced direction, and the version evolution direction.

[0044] Optionally, in one embodiment, determining the two-hop candidate standard that has a two-hop relationship with the target standard along the reference relationship includes: Step S301: Using each of the first-hop candidate standards as intermediate standards, determine the second-hop relationship path corresponding to the target standard along the reference relationship.

[0045] Specifically, the server can sequentially read the standard nodes corresponding to each one-hop candidate standard and query the reference relationships originating from the standard nodes. The server can then sequentially connect the target standard, the one-hop candidate standard, and the terminating standard pointed to by the reference relationships to form a two-hop relationship path.

[0046] A two-hop relationship path includes a target standard, a candidate standard serving as an intermediate standard, and a termination standard, and contains two relationship edges connecting adjacent standards. The first relationship edge can be a referencing relationship, a referenced relationship, or a substitution relationship, and the second relationship edge is a referencing relationship originating from the intermediate standard.

[0047] Step S302: Delete two-hop relationship paths that contain duplicate criteria or invalid relationships.

[0048] Specifically, a duplicate standard means that the same standard appears more than twice in the same two-hop relationship path. Two-hop relationship paths containing duplicate standards may form loops and cannot provide new candidate standards. The server can compare the standard number or node identifier of each standard in the two-hop relationship path and delete the two-hop relationship path containing duplicate standards.

[0049] An invalid relationship indicates that the referenced content of the relationship has been deleted, the standard version of the relationship has expired, or the relationship has been marked as invalid in the standard knowledge graph. The server can read the valid state of relationship edges and delete two-hop relationship paths that contain invalid relationships.

[0050] Step S303: Sequentially determine the termination criteria in the remaining two-hop relationship paths as the two-hop candidate criteria until the number of candidate criteria in the candidate criterion set reaches the preset candidate number threshold, or complete the processing of all remaining two-hop relationship paths.

[0051] Specifically, the server can process the filtered two-hop relationship paths according to a preset traversal order. For each two-hop relationship path, the server can determine the termination criterion in the two-hop relationship path as a candidate criterion. If the termination criterion is not yet included in the candidate criterion set, the server can add the termination criterion to the candidate criterion set and update the number of candidate criteria in the candidate criterion set.

[0052] When the number of candidate criteria reaches a preset threshold, the server can stop processing the remaining two-hop relationship paths. If, after processing all remaining two-hop relationship paths, the number of candidate criteria still has not reached the preset threshold, the server can stop two-hop traversal and use the already obtained set of candidate criteria to perform subsequent processing.

[0053] In this embodiment, by deleting two-hop relationship paths containing duplicate criteria or invalid relationships, loop criteria and invalid associations can be prevented from entering the candidate criterion set; by stopping the traversal when the number of candidate criteria reaches a preset candidate number threshold, the process of supplementing two-hop candidate criteria can be controlled by the number of candidates, reducing unnecessary graph traversal operations.

[0054] Optionally, in one embodiment, determining the graph relationship quantification value of each candidate standard based on the relationship type and graph distance between the target standard and each candidate standard in the candidate standard set includes: Step S401: Obtain the effective relationship path between the target standard and the candidate standard.

[0055] Specifically, the server can perform a graph query starting from the standard node corresponding to the target standard to obtain the relationship paths that connect the target standard and the candidate standards. A valid relationship path indicates that all the relationship edges contained therein are in a valid state and can connect from the target standard to the candidate standard.

[0056] When there are multiple valid relationship paths between the target standard and the same candidate standard, the server can retain each valid relationship path separately so that path scores can be calculated separately.

[0057] Step S402: Calculate the path score for each of the effective relation paths according to the following formula: ; in, This represents a valid relationship path between the target standard and the candidate standard; Indicates the effective relationship path Path scoring, Indicates the effective relationship path The relationship edge in the middle, Represents the relation edge Relationship types, Indicates the relation type The corresponding preset relationship weights, Indicates the effective relationship path The graph distance.

[0058] Specifically, the server can pre-configure relationship weights for referencing, being referenced, and substituting relationships. Relationship weights can be values ​​greater than 0 and less than or equal to 1. The server can read the relationship type of each edge in the valid relationship path, obtain the preset relationship weight corresponding to each relationship type, and calculate the product of these preset relationship weights. The server can then divide this product by the graph distance of the valid relationship path to obtain the path score of the valid relationship path.

[0059] For example, a valid relationship path contains two relationship edges, with preset relationship weights of 0.9 and 0.8 respectively. The graph distance of the valid relationship path is 2, and the path score of the valid relationship path can be 0.36. When the number of relationship edges increases or the weights corresponding to weak relationships decrease, the path score can be reduced accordingly.

[0060] Step S403: The highest path score among all the path scores is determined as the graph relation quantification value of the candidate standard.

[0061] Specifically, when there is only one valid relationship path between the target standard and the candidate standard, the server can directly determine the path score of the valid relationship path as the graph relationship quantification value of the candidate standard. When there are multiple valid relationship paths between the target standard and the candidate standard, the server can compare the scores of each path and determine the path score with the highest value as the graph relationship quantification value of the candidate standard.

[0062] In this embodiment, by multiplying the relation weights corresponding to different relation types in the effective relation path, the path score can reflect the degree of association generated by the combination of relation types; by using graph distance to constrain the relation weight product, the path score of longer relation paths can be reduced accordingly; by selecting the highest path score as the graph relation quantification value, the graph association strength between the target standard and the candidate standard can be represented by relation paths with a high degree of association, reducing the undue influence of longer or weakly associated paths on the graph relation quantification value.

[0063] Optionally, in one embodiment, calculating the semantic similarity between the target standard and each of the candidate standards may include: Step S501: Extract the standard name, scope of application, terms and definitions, and technical requirements of the target standard and each candidate standard.

[0064] Specifically, the server can locate different text sections based on the chapter identifiers, heading levels, or structured tags of the standard document. The standard name indicates the standard subject; the scope indicates the objects and boundaries to which the standard applies; terms and definitions indicate the professional concepts adopted by the standard; and the technical requirements text indicates the technical content specified by the standard.

[0065] The server can perform character format standardization, invalid character deletion, and paragraph segmentation on the extracted text to form a text sequence that can be input into the semantic coding model.

[0066] Step S502: Generate segmented semantic vectors corresponding to each extracted part of the text.

[0067] Specifically, a segmented semantic vector represents the vectorized semantic representation of a text portion in a standard document. The server can input the standard name, scope of application, terms and definitions, and technical requirements into a pre-trained text semantic encoding model to obtain the corresponding segmented semantic vectors.

[0068] When the length of the technical requirement text exceeds the single input length of the text semantic coding model, the server can divide the technical requirement text into multiple text blocks, generate semantic vectors corresponding to each text block, and then perform average pooling or weighted pooling on the semantic vectors corresponding to multiple text blocks to obtain segmented semantic vectors corresponding to the technical requirement text.

[0069] Step S503: Weight each segment semantic vector to obtain the standard semantic vector of the target standard and each candidate standard.

[0070] Specifically, the server can assign text portion weights to the standard name, scope of application, terms and definitions, and technical requirements text. The sum of the weights of each text portion can be 1. The server can multiply each segment semantic vector by its corresponding text portion weight, and then sum the weighted segment semantic vectors to obtain the standard semantic vector.

[0071] When a certain text portion is missing, the server can set the weight of the missing text portion to 0 and normalize the weights of the remaining text portions so that the sum of the weights of the remaining text portions remains 1.

[0072] Step S504: Calculate the cosine similarity between the standard semantic vector of the target standard and the standard semantic vector of each candidate standard to obtain the semantic similarity.

[0073] Specifically, the server can calculate the cosine similarity between the standard semantic vector of the target standard and the standard semantic vector of each candidate standard. The cosine similarity can be determined based on the inner product of the two standard semantic vectors and their magnitudes. The server can then use the calculated cosine similarity as the semantic similarity between the target standard and the corresponding candidate standard.

[0074] In this embodiment, by semantically encoding the standard name, scope of application, terms and definitions, and technical requirements text, the semantics of the standard can be characterized from multiple aspects such as standard theme, applicable boundaries, professional concepts, and technical content. By weighting the segmented semantic vectors of different text parts, the bias of semantic similarity caused by a single text part can be reduced, so that the semantic similarity can more completely reflect the content relevance between the target standard and the candidate standard.

[0075] Optionally, in one embodiment, determining the high correlation criterion based on the fusion correlation degree includes: Step S601: The candidate criteria for determining the graph relation quantification value, which contain alternative relations, are assigned to the alternative chain group.

[0076] Specifically, an alternative chain group represents a set of candidate standards associated with a target standard through version substitution relationships. The server can examine the relationship paths used to determine the graph relationship quantification values ​​of each candidate standard. When at least one substitution relationship is contained in the relationship path, the server can assign the corresponding candidate standard to the alternative chain group.

[0077] When multiple valid relationship paths exist between a candidate standard and a target standard, if the relationship path used to determine the graph relationship quantification value contains a substitution relationship, the server can assign the candidate standard to a substitution chain group.

[0078] Step S602: Candidate standards that are not assigned to the alternative chain group, have a graph distance of one with the target standard, and have a relationship type of referencing or being referenced are assigned to the direct reference group.

[0079] Specifically, the direct reference group represents the set of candidate standards that have a direct reference relationship with the target standard. The server can filter candidate standards from those not assigned to the alternative chain group that have a graph distance of one from the target standard and check the relationship type connecting the target standard and the candidate standards.

[0080] When the relationship type is a referencing relationship or a referenced relationship, the server can classify candidate criteria into the direct referencing group. Since the direct referencing group does not include candidate criteria already classified into the alternative chain group, there will be no duplicate candidate criteria between the alternative chain group and the direct referencing group.

[0081] Step S603: Candidate standards other than those in the substitution chain group and the direct reference group, and whose semantic similarity reaches a preset semantic threshold, are classified into the same domain semantic group.

[0082] Specifically, the domain-specific semantic group represents a set of candidate standards that have a high degree of content relevance to the target standard but are not classified into the alternative chain group or the direct reference group. The preset semantic threshold can be pre-set based on historical recommendation results or validation data.

[0083] The server can compare the semantic similarity of candidate standards with a preset semantic threshold. If a candidate standard is not assigned to the substitution chain group or the direct reference group, and its semantic similarity is greater than or equal to the preset semantic threshold, the server can assign the candidate standard to the same domain semantic group. By sequentially executing steps S601 to S603, the same candidate standard can be assigned to only one group.

[0084] Step S604: Sort the candidate criteria in each group according to the fusion correlation degree to obtain the high correlation degree criteria.

[0085] Specifically, the server can rank candidate criteria from highest to lowest fusion relevance within the substitution chain group, direct reference group, and domain-specific semantic group. The server can then combine the group names and the corresponding candidate criterion ranking results into a recommendation result for the association criteria and output the recommendation result.

[0086] In this embodiment, by prioritizing the identification of substitution relationships, then identifying direct reference relationships, and finally identifying standards within the same domain based on semantic similarity, the same candidate standard can be avoided from being repeatedly divided into multiple groups. By sorting the candidate standards in each group according to the degree of fusion relevance, the recommendation priority among candidate standards in the same group can be reflected while retaining the relationship category, enabling users to distinguish between version substitution standards, direct reference standards, and content-related standards.

[0087] Optionally, in one embodiment, the method may further include: Step S701: Determine the primary associated path with the highest path score from the effective relationship paths between the target standard and each candidate standard in the recommendation results of the associated standard.

[0088] Specifically, the primary association path represents the valid relationship path used to illustrate the main graph association between the target standard and the candidate standards. The server can obtain all valid relationship paths between the target standard and each candidate standard, and read the path score of each valid relationship path.

[0089] The server can compare the path scores of each valid relationship path and determine the valid relationship path with the highest path score as the primary relationship path for the corresponding candidate criterion. When there are two or more valid relationship paths with the same path score, the server can prioritize the valid relationship path with the smaller graph distance as the primary relationship path.

[0090] Step S702: Based on the standard order, relationship type, and graph distance in the main association path, and the semantic similarity of the candidate criteria, generate the association basis explanation field corresponding to the candidate criteria.

[0091] Specifically, the association criteria explanation field represents a structured or textual field used to describe the reasons why candidate criteria are recommended. The server can read the order of criteria in the main association path according to the direction of travel and read the relationship type between adjacent criteria. The server can write the number of relation edges contained in the main association path into the graph distance part and write the semantic similarity of candidate criteria into the semantic relevance part.

[0092] For example, when the main association path indicates that the target standard directly references the candidate standard, the graph distance is 1, and the semantic similarity is 0.82, the association basis explanation field can be expressed as "the target standard directly references the candidate standard, the graph distance is 1, and the semantic similarity is 0.82". When the main association path includes intermediate standards, the association basis explanation field can also record the relationship type between the target standard, intermediate standards, and candidate standards in the standard order.

[0093] Step S703: Output the association basis explanation field along with the corresponding candidate criteria.

[0094] Specifically, the server can establish a correspondence between the node identifiers of candidate criteria and the explanatory fields of the associated criteria. When outputting the recommendation results of the associated criteria, the server can obtain the corresponding explanatory fields of the associated criteria based on the node identifiers of the candidate criteria, and output the explanatory fields of the associated criteria along with the candidate criteria.

[0095] In this embodiment, by selecting the effective relationship path with the highest path score as the main association path, explanatory content can be generated using the main graph association criteria. By using standard order, relationship type, graph distance, and semantic similarity in the association criteria explanation field, the graph association source and content relevance of the candidate criteria can be presented simultaneously, making the association criteria recommendation results have traceable recommendation criteria.

[0096] Please see Figure 3 , Figure 3 This is a flowchart of the association standard recommendation process provided in one embodiment of the present invention. Figure 3 In the online phase, the user (corresponding to the user in this application) selects a target standard as the starting point for recommendation; recalls a set of one-hop neighbor standards (corresponding to one-hop candidate standards in this application) along three types of directed relationships (corresponding to the referencing relationship, the referenced relationship, and the substitution relationship in this application); and determines whether the number of one-hop related standards (corresponding to one-hop candidate standards in this application) reaches a preset recommendation threshold (corresponding to the preset candidate number threshold in this application). If the number of neighbors is insufficient, a two-hop traversal is performed along the referencing edge starting from the one-hop neighbor, and the candidate standard set is expanded (corresponding to the method in this application of using the one-hop candidate standard as an intermediate standard, determining the two-hop candidate standard along the referencing relationship, and adding it to the candidate standard set).

[0097] Subsequently, the cosine similarity of the semantic vectors between the target standard and each candidate standard (corresponding to the semantic similarity in this application) can be calculated, and the inverse graph distance and the cosine similarity can be weighted and fused according to configurable weights to form a fused correlation degree. Figure 3 The inverse of the graph distance is used to represent the attenuation effect of graph distance on the association strength; corresponding to the implementation of this application, the graph relationship quantification value is also determined in combination with the preset relationship weight corresponding to the relationship type, and the fusion association degree is determined based on the graph relationship quantification value and semantic similarity.

[0098] Figure 3 It also shows that the candidate standards are grouped and sorted into three groups: direct reference, semantics in the same domain, and alternative chain (corresponding to the direct reference group, semantics in the same domain, and alternative chain group in this application, respectively). For each recommended standard item, an explanatory field of the associated basis that can be understood by the user (corresponding to the user in this application) is generated, and a list of recommended associated standards (corresponding to the recommended results of associated standards in this application) and the corresponding explanatory field are output. Figure 3The listing order of the three groups does not indicate the grouping priority. The specific grouping order is subject to steps S601 to S603. The generation of candidate standard grouping and related basis interpretation fields are optional processes and do not constitute a necessary limitation of all implementation methods.

[0099] Optionally, in one embodiment, the method may further include: Step S801: When adding new standards, revising standards, or replacing standards into the database, determine the standard nodes and relationships that have changed in the standard knowledge graph.

[0100] Specifically, the server can read the standard number corresponding to a newly added, revised, or replaced standard, and query the standard knowledge graph based on the standard number. The server can compare the standard information added to the database with the node information and relationship information already stored in the standard knowledge graph.

[0101] When a new standard does not have a corresponding standard node in the standard knowledge graph, the server can identify the new standard node and its associated relationships as the changed standard nodes and relationships. When the standard content or reference information corresponding to a revised standard changes, the server can identify the standard node corresponding to the revised standard and the relationships that have been added or deleted as the changed standard nodes and relationships. When a replacement standard is added to the database, the server can identify the replaced standard node, the replaced standard node, and the newly added replacement relationships as the changed standard nodes and relationships.

[0102] Step S802: Update the standard nodes and relationships that have changed.

[0103] Specifically, the server can add or modify standard nodes that have changed, and add, delete, or modify corresponding relationship edges based on the association information in the database standards. The server can save the updated standard knowledge graph after the nodes and relationships are updated.

[0104] Step S803: Recalculate the standard semantic vector corresponding to the changed standard node.

[0105] Specifically, the server can obtain the latest standard text corresponding to the changed standard node and regenerate the standard semantic vector using a text semantic encoding model. The server can then replace the original standard semantic vector of the standard node with the regenerated standard semantic vector.

[0106] Step S804: Define the standards whose graph distance to the changed standard node is less than or equal to 2 as the affected standards, and update the associated standard recommendation results corresponding to the affected standards.

[0107] Specifically, the affected criteria refer to the criteria whose recommendation candidate range or recommendation score may change due to changes in the standard knowledge graph. The server can start from a node with a changed criterion and traverse along the relation edges in the standard knowledge graph in no more than two hops, identifying the criteria corresponding to the traversed node as the affected criterion. Since the graph distance between a node with a changed criterion and itself is 0, the criteria corresponding to a node with a changed criterion can also be considered as affected criteria.

[0108] The server can use each affected standard as a target standard, redetermine candidate standards, graph relationship quantification values, semantic similarity, and fusion correlation, and update the associated standard recommendation results corresponding to the affected standards using the recalculated results. Standards with a graph distance greater than 2 from the changed standard nodes can continue to use the already stored associated standard recommendation results.

[0109] In this embodiment, by identifying the changed standard nodes and relationships and recalculating the standard semantic vectors corresponding to the changed standard nodes, the standard knowledge graph and semantic representation can be kept consistent with the input standards. By limiting the update range of the recommendation results to the affected standards whose graph distance to the changed standard nodes is less than or equal to 2, the amount of computation generated by full recalculation can be reduced, while the recommendation results of the associated standards that may be affected by changes in nodes or relationships can be updated.

[0110] Based on the same inventive concept, embodiments of the present invention also provide an electronic device, the electronic device including a memory and a processor, the memory storing a computer program, and when the computer program is executed by the processor, causing the processor to perform the method for determining the high correlation standard described in any of the foregoing embodiments.

[0111] Please see Figure 4 , Figure 4 This is a schematic diagram of an electronic device provided in one embodiment of the present invention. The electronic device includes one or more processors 1101, a memory 1102, and a bus 1105, and may also include an input device 1103 and an output device 1104. The processor 1101, memory 1102, input device 1103, and output device 1104 can be connected via the bus 1105 or other means; the diagram shows an example of connection via the bus 1105. The components can be mounted on a common motherboard or installed in other ways as needed.

[0112] Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method for determining high correlation standards as described in any of the foregoing embodiments.

[0113] The above description is merely an embodiment of the present invention and is not intended to limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for determining a high correlation standard, characterized in that, The method includes: Based on the reference relationships, referenced relationships, and substitution relationships in the standard knowledge graph, one-hop candidate standards for the target standard are determined, and a set of candidate standards is determined according to the one-hop candidate standards. If the number of one-hop candidate standards is less than a preset candidate number threshold, the one-hop candidate standards are used as intermediate standards, and two-hop candidate standards with two-hop relationships with the target standard are determined along the reference relationships, and the two-hop candidate standards are added to the set of candidate standards. Based on the relationship type and graph distance between the target standard and each candidate standard in the candidate standard set, determine the graph relationship quantification value of each candidate standard, and calculate the semantic similarity between the target standard and each candidate standard; The fusion correlation degree of each candidate standard is determined based on the graph relationship quantification value and the semantic similarity of each candidate standard. The criteria for high correlation are determined based on the fusion correlation degree.

2. The method according to claim 1, characterized in that, The standard knowledge graph is constructed in the following way: Obtain normative reference information and version replacement information of the standard documents to be added to the database; Using the standard corresponding to the standard file in the database as a node, the referencing standard and the referenced standard are determined according to the normative reference information, a reference relationship is established from the referencing standard to the referenced standard, and a reference relationship is established from the referenced standard to the referencing standard. Based on the version replacement information, establish a replacement relationship from the replaced standard to the replaced standard.

3. The method according to claim 1, characterized in that, The step of determining the two-hop candidate standard that has a two-hop relationship with the target standard along the reference relationship includes: Using each of the aforementioned first-hop candidate standards as intermediate standards, the second-hop relationship path corresponding to the target standard is determined along the aforementioned reference relationship; Delete two-hop relationship paths that contain duplicate criteria or invalid relationships; The termination criteria in the remaining two-hop relationship paths are sequentially determined as the two-hop candidate criteria until the number of candidate criteria in the candidate criterion set reaches the preset candidate number threshold, or the processing of all remaining two-hop relationship paths is completed.

4. The method according to claim 1, characterized in that, The step of determining the graph relationship quantification value of each candidate standard based on the relationship type and graph distance between the target standard and each candidate standard in the candidate standard set includes: Obtain the effective relationship path between the target standard and the candidate standard; The path score for each of the aforementioned valid relation paths is calculated using the following formula: ； in, This represents a valid relationship path between the target standard and the candidate standard; Indicates the effective relationship path Path scoring, Indicates the effective relationship path The relationship edge in the middle, Represents the relation edge Relationship types, Indicates the relation type The corresponding preset relationship weights, Indicates the effective relationship path Graph distance; The highest path score among all the path scores is determined as the graph relation quantification value of the candidate standard.

5. The method according to claim 1, characterized in that, The calculation of the semantic similarity between the target standard and each of the candidate standards includes: Extract the standard name, scope of application, terms and definitions, and technical requirements of the target standard and each candidate standard respectively; Generate segmented semantic vectors corresponding to each extracted part of the text; The segmented semantic vectors are weighted to obtain the standard semantic vectors of the target standard and each of the candidate standards. The semantic similarity is obtained by calculating the cosine similarity between the standard semantic vector of the target standard and the standard semantic vector of each candidate standard.

6. The method according to claim 1, characterized in that, The step of determining the high correlation criterion based on the fusion correlation degree includes: Candidate criteria for determining the graph relation quantification value that contain alternative relations are assigned to alternative chain groups; Candidate standards that are not assigned to the alternative chain group, have a graph distance of one from the target standard, and have a relationship type of referencing or being referenced are assigned to the direct reference group. Candidate criteria other than those in the alternative chain group and the direct reference group, and whose semantic similarity reaches a preset semantic threshold, are classified into the same domain semantic group; The candidate criteria in each group are sorted according to the fusion correlation degree to obtain the high correlation degree criteria.

7. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, which, when executed by the processor, implements the method for determining the high correlation standard as described in any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method for determining the high correlation standard as described in any one of claims 1 to 6.

Images