A method, apparatus, terminal equipment, and storage medium for entity link verification based on electricity marketing business graph.
By using evidence continuity gating mechanisms and rule constraint verification, combined with feedback reuse verification, the problem of inaccurate entity linking results in existing technologies has been solved, thereby improving the reliability and accuracy of entity links in the power marketing business graph.
Patent Information
- Application Number
- CN202611437541.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-16
- Publication Date
- 2026-11-10
AI Technical Summary
Existing entity linking methods lack verification of the evidence linking set, making it difficult to guarantee that the linking results conform to the actual business logic, resulting in inaccurate results.
Based on the power marketing business map, the evidence completeness of candidate link objects is verified through an evidence continuity gating mechanism. Combined with rule constraints and feedback reuse verification, link credibility is generated to ensure that the link results conform to the actual business logic.
It improves the reliability and accuracy of entity linking results, avoids inflated confidence levels through a multi-dimensional verification mechanism, and enhances the compliance and reliability of the results.
Smart Images

Figure CN123070106A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power marketing mapping technology, and in particular to a method, apparatus, terminal equipment, and storage medium for verifying entity links based on power marketing business mapping. Background Technology
[0002] With the deep integration of smart grids and digital marketing, the electricity marketing business has gradually built a vast knowledge system encompassing four types of graphs: knowledge, rules, risks, and issues. Electricity marketing data comes from a wide range of sources and takes many forms, including operational data such as work orders, customer files, metering data, and complaint reports, as well as comprehensive data such as work order associations and marketing graphs. In these business scenarios, the goal of entity linking is to accurately map entities in files or data to corresponding standard nodes in the graph, and to establish consistent entity relationships between multi-source graphs.
[0003] Existing entity linking methods mostly employ similarity matching and manual review, which only judge similarity at the name level and lack the verification of evidence link sets, making it difficult to guarantee that the linking results conform to the actual business logic. Summary of the Invention
[0004] This invention provides a method, apparatus, terminal device, and storage medium for verifying entity links based on a power marketing business graph. It can effectively solve the problem that existing technologies lack evidence link sets for verification, making it difficult to ensure that the link results conform to actual business logic.
[0005] An embodiment of the present invention provides an entity link verification method based on an electricity marketing business graph, comprising: Obtain electricity marketing business data and historical link records, and match the electricity marketing business data with the target link map to generate a candidate link set; Based on the evidence continuity gating mechanism, the evidence completeness of the candidate link objects in the candidate link set is verified to obtain the semantic topology verification score of the candidate link objects. Based on the candidate link objects and the corresponding entity type, business scenario, and business occurrence time in the power marketing business data, rule constraint verification is performed to obtain rule constraint measurement parameters and rule consistency scores. Feedback reuse verification is performed based on the historical link records corresponding to the candidate link objects to obtain the feedback reuse score; The link credibility is calculated based on the semantic topology verification score, rule constraint metric parameter, rule consistency score, and feedback reuse score. The entity link verification result is determined based on the link credibility, rule constraint metric parameter, and preset credibility threshold.
[0006] Furthermore, based on the evidence continuity gating mechanism, the completeness of evidence is verified on the candidate link objects in the candidate link set to obtain the semantic topology verification score of the candidate link objects, including: For each candidate link object in the candidate link set, extract name features, contextual semantic features, entity type features, relational role features, topological neighborhood features, and time window features respectively; Based on the evidence continuity gating mechanism, the evidence continuity coefficient is calculated according to the name feature, contextual semantic feature, topological neighborhood feature, time window feature and preset feature integrity weight. Under the constraint of the evidence continuity coefficient, normalization mapping is performed based on the name feature, contextual semantic feature, entity type feature, relational role feature, topological neighborhood feature, time window feature, and preset evidence feature weights to obtain the semantic topological verification score of the candidate link object.
[0007] Furthermore, rule constraint verification is performed based on the entity type, business scenario, and business occurrence time corresponding to the candidate link objects in the electricity marketing business data, resulting in rule constraint metric parameters and rule consistency scores, including: Based on the entity type, business scenario, and business occurrence time corresponding to the candidate link object in the power marketing business data, a matching constraint is performed with the preset business rule constraint corresponding to the candidate link object to obtain the rule constraint measurement parameter; wherein, the preset business rule constraint includes hard constraints that must be followed and soft constraints that prioritize adaptation; the rule constraint measurement parameter includes: hard constraint conflict identifier and soft constraint deviation amount; The rule consistency score is calculated based on the hard constraint conflict identifier, the soft constraint deviation, and the preset penalty weight.
[0008] Furthermore, the historical link record includes: the number of times the link object is linked, the number of times the link object is verified, historical name matching conditions, historical rule adaptation conditions, and the number of times historical conflicts occur; Based on the historical link records corresponding to the candidate link objects, feedback reuse verification is performed to obtain a feedback reuse score, including: Based on the historical link records corresponding to the candidate link objects, determine the number of links and the number of verifications of link objects with the same entity type as the candidate link objects, and determine the historical link support based on the ratio of the number of links to the number of verifications of link objects with the same entity type. Determine alias matching support based on historical name matching conditions; Determine the rule exception adaptation degree based on historical rule adaptation working conditions; Conflict hit rate is determined based on the number of historical conflicts. The feedback reuse score is obtained by weighting the historical link support, alias matching support, rule exception adaptation, conflict hit rate, and preset feedback weight.
[0009] Further, the entity link verification result is determined based on the link credibility, rule constraint metric parameters, and a preset credibility threshold, including: The link credibility is compared with a preset credibility threshold. If the link credibility is not less than the credibility threshold, determine whether the hard constraint conflict flag in the rule constraint metric parameter indicates no hard constraint conflict. If so, obtain the entity link result of the entity to be verified based on the candidate link objects in the candidate link set. Otherwise, exclude the corresponding candidate link set to obtain the excluded candidate link set, and determine the candidate link objects in the excluded candidate link set as the entity link result of the entity to be verified. If the excluded candidate link set is empty, generate the first manual review prompt. If the credibility of the link is less than the credibility threshold, a second manual review prompt will be generated.
[0010] Furthermore, based on the candidate link objects in the candidate link set, the entity link results of the entity to be verified are obtained, including: Determine whether there is a one-to-one relationship between the candidate link objects in the candidate link set and the entity to be verified; If so, the corresponding candidate link object will be determined as the entity link result of the entity to be verified; If not, calculate the credibility difference between the highest and second-highest credibility of the corresponding candidate link objects, and determine whether the credibility difference is greater than a preset difference threshold; if the credibility difference is greater than the difference threshold, determine the corresponding candidate link object as the entity link result of the entity to be verified; if the credibility difference is not greater than the difference threshold, generate a third manual review prompt.
[0011] Furthermore, based on the electricity marketing business data and the target link map, a candidate link set is generated, including: Entity identification is performed based on the electricity marketing business data to obtain the entity to be verified. Using the entity to be verified as the query condition, candidate graph nodes are determined from the target link graph based on any one of the recall dimensions; where the recall dimensions include graph neighborhood relationship, entity type consistency relationship and business related object relationship; The entity to be verified, the candidate graph node, the graph neighborhood corresponding to the candidate graph node, and the business-related object are encapsulated into a candidate link object; Generate a candidate link set based on all candidate link objects.
[0012] As an improvement to the above solution, another embodiment of the present invention provides an entity link verification device based on an electricity marketing business graph, comprising: The candidate link set generation module is used to acquire electricity marketing business data and historical link records, and to generate a candidate link set by matching the electricity marketing business data with the target link map. The evidence completeness verification module is used to perform evidence completeness verification on the candidate link objects in the candidate link set based on the evidence continuity gating mechanism, and obtain the semantic topology verification score of the candidate link objects. The rule constraint verification module is used to verify the rule constraints based on the candidate link object and the corresponding entity type, business scenario, and business occurrence time in the power marketing business data, and to obtain the rule constraint measurement parameters and rule consistency score. The feedback reuse verification module is used to perform feedback reuse verification based on the historical link records corresponding to the candidate link object, and obtain the feedback reuse score. The entity link verification result generation module is used to calculate the link credibility based on the semantic topology verification score, rule constraint metric parameter, rule consistency score, and feedback reuse score, and to determine the entity link verification result based on the link credibility, rule constraint metric parameter, and preset credibility threshold.
[0013] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements an entity link verification method based on an electricity marketing business graph as described in the above embodiments.
[0014] Another embodiment of the present invention provides a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the entity link verification method based on the power marketing business graph described in the above embodiment.
[0015] By implementing this invention, at least the following beneficial effects are achieved: This invention provides a method, apparatus, terminal device, and storage medium for entity link verification based on an electricity marketing business graph. The method matches electricity marketing business data with a target link graph to generate a candidate link set. Then, based on an evidence continuity gating mechanism, it performs evidence completeness verification on the candidate link set to obtain a semantic topology verification score. This mechanism avoids artificially high confidence levels in scenarios with missing evidence, improving the reliability of the results. Next, it performs rule constraint verification based on the entity type, business scenario, and business occurrence time of the candidate link objects in the electricity marketing business data, obtaining rule constraint metric parameters and rule consistency scores. The rule constraint verification is then used to verify the candidate links. The objects undergo further verification to exclude candidate link objects that violate the rules, so that the final link results are more consistent with the actual business logic and further improve the reliability of the link results. Then, feedback reuse verification is performed based on the historical link records corresponding to the candidate link objects to obtain feedback reuse scores. The link credibility is calculated based on the semantic topology verification score, rule constraint metric parameters, rule consistency score, and feedback reuse score. The entity link verification result is determined based on the link credibility, rule constraint metric parameters, and preset credibility threshold. By comprehensively performing evidence completeness verification, rule constraint verification, and feedback reuse verification, the link verification results are improved in accuracy while ensuring compliance. Attached Figure Description
[0016] Figure 1 This is a flowchart illustrating an entity link verification method based on a power marketing business graph according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of an entity link verification device based on a power marketing business graph according to an embodiment of the present invention. Detailed Implementation
[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0018] See Figure 1 To address the problem in existing technologies where the lack of evidence link sets for verification makes it difficult to guarantee that link results conform to actual business logic, an embodiment of the present invention provides a flowchart illustrating an entity link verification method based on an electricity marketing business graph, comprising: S1. Obtain electricity marketing business data and historical link records, and match the electricity marketing business data with the target link map to generate a candidate link set; Specifically, electricity marketing business data refers to all types of structured and unstructured data generated throughout the entire electricity marketing process, including 95598 work orders, customer files, metering data, complaint reports, abnormal electricity consumption analysis reports, structured ledgers, and business relationship data. This data forms the foundation for entity identification and link verification. Historical link records refer to the complete set of results data accumulated during past entity link verification processes, including the number of times linked objects were linked, the number of times linked objects were verified, historical name matching conditions, historical rule adaptation conditions, and the number of historical conflicts. This data serves as the source for feedback and reuse. The target link graph refers to four types of business graphs used to carry electricity marketing entities and relationships: knowledge graph, rule graph, risk graph, and problem graph. The knowledge graph stores basic entity attributes and relationships; the rule graph stores business rules and engineering constraints; the risk graph stores risk events and associated entities; and the problem graph stores historical problem work orders and abnormal events. The candidate link set refers to the set of objects to be verified formed by pairing and encapsulating the entity to be verified with candidate graph nodes recalled from the target graph.
[0019] To illustrate, the process begins by extracting relevant power marketing business data and historical link records from the power marketing data platform, business systems, and historical knowledge base. After preliminary preprocessing such as cleaning and deduplication, the identified business entities are used as query units for matching and retrieval across four types of target graphs. Each entity to be verified and its corresponding candidate graph node are encapsulated as an independent candidate link object, and all candidate link objects are aggregated to form a candidate link set. This joint retrieval across four types of target graphs can cover candidate nodes across more dimensions, avoiding information blind spots from a single graph; for example, the knowledge graph handles basic entity matching, the risk graph supplements entities associated with risk events, and the problem graph supplements entities associated with historical problems.
[0020] Preferably, a candidate link set is generated by matching the electricity marketing business data with the target link map, including: Entity identification is performed based on the electricity marketing business data to obtain the entity to be verified. Using the entity to be verified as the query condition, candidate graph nodes are determined from the target link graph based on any one of the recall dimensions; where the recall dimensions include graph neighborhood relationship, entity type consistency relationship and business related object relationship; The entity to be verified, the candidate graph node, the graph neighborhood corresponding to the candidate graph node, and the business-related object are encapsulated into a candidate link object; Generate a candidate link set based on all candidate link objects.
[0021] Specifically, the entity to be verified refers to the entity object identified from the power marketing business data that needs to be matched and linked with the target graph. Recall dimensions represent the matching conditions used when retrieving candidate nodes from the target link graph; more dimensions result in a more comprehensive recall. Graph neighborhood relationships represent the neighborhood structure relationships formed by associated entities around a candidate graph node. Entity type consistency relationships are the matching relationships between the entity to be verified and the candidate graph node to see if their entity types match, used to filter nodes with mismatched types. Business association object relationships are the relationships between other objects associated with the entity in the same business document or business event, which can be used to constrain the recall scope. Candidate graph nodes represent standard nodes recalled from the target link graph that may match the entity to be verified. Graph neighborhood represents a local subgraph formed by associated entities and relationships within one or two hops of a candidate graph node, used for subsequent topology verification.
[0022] This example illustrates the process of entity extraction from electricity marketing business data. Using named entity recognition technology combined with an electricity-related dictionary, various business entities are identified, and information such as entity type, context, and associated business objects is labeled to create a list of entities to be verified. (Entities to be verified) The types cover all entities in the electricity marketing scenario, including customer type, account number type, metering point type, transformer area and line type, power supply unit type, business event type, and rules and terms type.
[0023] Using the entity to be verified as the query condition, candidate graph nodes are determined from the target link graph based on any one of the recall dimensions. These recall dimensions include graph neighborhood relationships, entity type consistency relationships, and business-related object relationships. Specifically, using the identifier, attributes, and associated objects of the entity to be verified as query conditions, candidate nodes are retrieved from four types of target graphs based on at least one recall dimension. For example, first, nodes with mismatched types are filtered out using entity type consistency relationships to narrow the recall scope; then, associated entities in the same work order or event are matched using business-related object relationships; finally, relevant nodes are expanded by combining graph neighborhood relationships. Recall dimensions can also include name or alias matching, coded field matching, rule applicability relationships, risk event association relationships, historical link records, etc., which can be flexibly combined according to business scenarios. Multi-dimensional joint recall can improve the comprehensiveness of the recall and reduce missed recalls.
[0024] The entity to be verified, candidate graph nodes, the graph neighborhood corresponding to the candidate graph node, and business-related objects are encapsulated into candidate link objects. Specifically, each entity to be verified is paired with each corresponding candidate graph node. During encapsulation, data such as the local graph neighborhood of the candidate node, basic matching information, and related business objects are included to form a complete candidate link object, which can be directly used in subsequent verification processes without repeatedly querying the graph.
[0025] A candidate link set is generated based on all candidate link objects. Specifically, all candidate link objects corresponding to all entities to be verified are summarized and organized into a unified candidate link set, which serves as the input for subsequent semantic topology verification, rule constraint verification, and feedback reuse verification.
[0026] In a preferred embodiment of the present invention, the candidate link object is represented as follows: ,in, This represents the candidate link object formed by the i-th entity to be verified and the j-th candidate graph node; This represents the i-th entity to be verified; Indicating targeting The j-th candidate graph node recalled from the knowledge graph, rule graph, risk graph, or problem graph; Indicates by , The local graph neighborhood formed by candidate graph nodes and their associated objects; This indicates the basic matching information of the candidate link object. The basic matching information includes one or more of the following: source data, entity type, business time, alias matching status, and historical link information. This indicates the accumulated evidence status of the candidate link object during subsequent verification. The system revolves around... A verification record is generated simultaneously to preserve the results of subsequent evidence extraction, rule verification, and expert review. Alias hit information is used to indicate... Did it hit? The standard name, abbreviation, historical name, manually added alias or coded alias of the corresponding node, including the status of exact hit, normalized hit, historical name hit, partial hit and no hit.
[0027] Traditional candidate retrieval relies solely on name string matching, resulting in a single retrieval dimension and a tendency to miss nodes with significantly different names but belonging to the same entity. Furthermore, retrieval only returns the node itself, without accompanying neighborhood and association information, requiring repeated graph queries in subsequent verification stages, leading to low processing efficiency. This embodiment constructs a multi-dimensional retrieval system covering multiple dimensions such as entity type, associated objects, neighborhood relationships, name, and encoding, resulting in more comprehensive retrieval. Simultaneously, it includes graph neighborhood and associated business objects during encapsulation, which can be directly used in subsequent verification stages without repeated graph queries, improving overall processing efficiency.
[0028] S2. Based on the evidence continuity gating mechanism, the evidence completeness of the candidate link objects in the candidate link set is verified to obtain the semantic topology verification score of the candidate link objects. Specifically, the evidence continuity gating mechanism controls the validity of subsequent verification scores by the completeness of the evidence chain. The more evidence is missing, the stronger the constraint on the final score, preventing overly confident conclusions from being drawn when evidence is incomplete. Evidence completeness verification is the process of verifying the completeness and validity of various semantic and topological relationship evidence in the candidate link set. The semantic and topological verification score refers to the matching degree score calculated by comprehensively considering multi-dimensional evidence such as name, context, entity type, relational role, topological neighborhood, and time window. A higher score indicates a higher degree of semantic and topological matching.
[0029] To illustrate, the candidate link objects in the candidate link set are traversed one by one. First, multi-dimensional semantic and topological feature evidence is extracted. Then, the completeness of the evidence chain is calculated through an evidence continuity gating mechanism. This is used as a constraint to perform weighted calculations on valid evidence, and finally, a semantic and topological verification score is output. The more complete the evidence chain, the higher the reference value of the score; when there is a lot of missing evidence, the score will be automatically discounted to avoid artificially inflated confidence levels.
[0030] Preferably, based on the evidence continuity gating mechanism, the candidate link objects in the candidate link set are subjected to evidence completeness verification to obtain the semantic topology verification score of the candidate link objects, including: S21. For each candidate link object in the candidate link set, extract the name feature, contextual semantic feature, entity type feature, relational role feature, topological neighborhood feature, and time window feature respectively. S22. Based on the evidence continuity gating mechanism, the evidence continuity coefficient is calculated according to the name feature, contextual semantic feature, topological neighborhood feature, time window feature and preset feature integrity weight. S23. Under the constraint of the evidence continuity coefficient, normalization mapping is performed based on the name feature, context semantic feature, entity type feature, relation role feature, topological neighborhood feature, time window feature and preset evidence feature weight to obtain the semantic topological verification score of the candidate link object.
[0031] Specifically, the name feature represents the identifier characteristics of the entity to be verified and the candidate graph nodes, including standard name, abbreviation, alias, historical name, and coded field. Contextual semantic features represent the semantic content characteristics of the business text, work order description, and business fields in which the entity to be verified resides, reflecting the business context in which the entity is located. Entity type features represent the category attributes to which the entity to be verified belongs, such as customer class, account number class, metering point class, transformer area / line class, power supply unit class, and business event class. Relationship role features represent the role that the entity to be verified plays in the business scenario or graph relationship, such as work order applicant, fault location, and responsible entity. Topological neighborhood features represent the local topological structure characteristics of the associated entities within a one-hop range around the candidate graph node, reflecting the relationship network of the entities. Time window features represent the matching characteristics between the business occurrence time and the effective time interval of the candidate node, reflecting the consistency of the time dimension.
[0032] Specifically, the evidence continuity coefficient measures the completeness of the evidence chain, with a value ranging from [0,1]. The closer the value is to 1, the more complete the evidence. The feature completeness weight represents the importance of each type of evidence in the calculation of the evidence continuity coefficient; a higher weight indicates a greater impact of that type of evidence on the completeness of the evidence chain. The evidence feature weight represents the importance of each type of evidence in the calculation of the semantic topology verification score; a higher weight indicates a greater impact of that type of evidence on the matching degree. The normalization mapping process maps the weighted calculation results to the 0-1 standard interval, ensuring a unified score range and facilitating subsequent multi-dimensional fusion.
[0033] To illustrate, for each candidate link object, six types of features are extracted from the business data side of the entity to be verified and the graph data side of the candidate node: Name features extract the standard name, abbreviation, alias, historical name, and various coded fields of both parties; Context features extract the work order description, business remarks, and historical business text of node association; Entity type features match the entity classification labels of both parties; Relationship role features extract the role of the entity in the business document and the relationship role definition in the graph; Topological neighborhood features extract the set of all associated entities within one hop of the candidate node; Time window features extract the business occurrence time point and the effective time interval of the candidate node.
[0034] Then, four core evidence categories—name, context, topological neighborhood, and time window—are selected. Each category is assessed for valid data; if valid data is available, it is marked as usable (value 1); otherwise, it is marked as unusable (value 0). The usability status value of each evidence category is multiplied by its corresponding feature integrity weight, and the sum is used to obtain the evidence continuity coefficient. Name, context, topological neighborhood, and time window are selected for continuity verification because these four categories form the foundation for semantic topological matching; the absence of any one of them significantly impacts the reliability of the results. Entity type and relational roles typically exist synchronously with the basic data and are not included in the core dimensions of continuity verification.
[0035] In a preferred embodiment of the present invention, the continuity coefficient of evidence : in, , , , These represent whether the name evidence, contextual evidence, graph relationship evidence, and time window evidence meet the usability criteria, with values of 0 or 1. to This indicates the weight of the completeness of various types of evidence. Used to measure whether the evidence chain of the same candidate link object is continuous. When evidence is missing, the effectiveness of subsequent semantic topology verification score is reduced. according to Does it contain identifiable names, codes, abbreviations, or aliases, and Determine whether there is a comparable standard name, historical name, alias, or code. according to Determine whether the text fragment, work order description, or business field contains sufficient contextual information to characterize the business scenario; according to Whether there are verifiable relationships such as customers, account numbers, metering points, transformer areas, lines, work orders, or rules and regulations can be determined; The time of the business transaction, the effective time of the node, the effective time of the rule, or the update time of the ledger are all complete and comparable.
[0036] Next, the matching scores for the six feature categories are calculated as follows: name matching score is calculated based on a combination of string edit distance and word segmentation overlap; context matching score is calculated based on the cosine similarity of text vectors; entity type consistency is determined by a binary value, with 1 for matching types and 0 for inconsistent types; relationship role consistency score is calculated based on the matching degree of role tags; topological neighborhood overlap score is calculated based on the Jaccard coefficient of the associated entity set; and time window matching score is calculated based on the overlap ratio of time intervals. The matching scores of the six feature categories are multiplied by the corresponding evidence feature weights, summed, and then multiplied by the evidence continuity coefficient for overall constraint. After adding a bias term, a sigmoid function is used for normalization mapping, finally obtaining the semantic topological verification score in the 0-1 interval.
[0037] In a preferred embodiment of the present invention, the weights for the name feature in the feature integrity weights are 0.35, 0.25 for the contextual semantic feature, 0.25 for the topological neighborhood feature, and 0.15 for the time window feature, with a total weight of 1. In the evidence feature weights, the weights for the name feature are 0.3, 0.2 for the contextual semantic feature, 0.15 for the entity type feature, 0.1 for the relational role feature, 0.15 for the topological neighborhood feature, and 0.1 for the time window feature, with a total weight of 1.
[0038] In a preferred embodiment of the present invention, under the constraint of the evidence continuity coefficient, the semantic topology verification score is calculated according to the following formula: in, Represents candidate link objects The semantic topology verification score; This represents the normalized mapping function; to 'b' represents the weight of each piece of evidence; 'b' represents the bias term. s name Name normalization similarity is used to measure and The degree of matching in standard names, abbreviations, aliases, historical names, and encoding fields can be calculated based on one or more of string edit distance, word segmentation overlap, pinyin, or encoding consistency. First, Chinese numerals, Arabic numerals, full-width and half-width characters, punctuation marks, and business suffixes such as "customer," "household number," and "metering point" are normalized. Then, edit distance similarity, word segmentation overlap, alias hit rate, and encoding consistency are calculated separately, and combined using name evidence parameters in the verification parameter set Theta. For example, "Dongcheng Garden Building 3" and the candidate node alias "Dongcheng Garden Building 3" have the same name subject after numeral normalization. If they also match the last digit of the household number or the customer abbreviation, then s name The corresponding increase.
[0039] s ctx Contextual semantic similarity is used to measure The text snippet, work order description, or business field contains the same information as the work order description. The semantic relevance between node descriptions, historical related work orders, or linked contexts can be calculated based on one or more of text vector representation, domain dictionary matching, or business keyword overlap; [Excerpt] Extract business keywords and text vectors from the context window of the sentence, adjacent fields, or the same work order. Combine the cosine similarity of the text vectors, the domain dictionary hit rate, and the overlap of business keywords to obtain s. ctx For example, if the context simultaneously includes "out of service", "data still being collected", and "metering anomaly", then if the candidate metering points have the same or similar anomaly records, then a high level of contextual support is formed.
[0040] s type Entity type consistency is used to determine... Entity type and The node types are consistent. If the two types are completely consistent, for example, "metering point number" corresponds to "metering point node", then the higher consistent value is taken. If the two are compatible upper and lower type, for example, "customer name" corresponds to "electricity customer node", then the middle consistent value is taken. If the two are incompatible, for example, "metering point" is linked as "line node", then the lower consistent value or 0 is taken.
[0041] s rel Relationship role consistency is used to determine... The role in the current business text or work order, and Whether the roles are consistent in the four types of graphs or work order relationships; identify role relationships from the current data, such as "customer has account number", "account number is linked to metering point", "metering point belongs to transformer area", "work order points to abnormal event", and compare them with the role relationships of candidate graph nodes. Calculate s based on the number of matched key role relationships, relationship importance, and the total number of relationships to be verified.rel .
[0042] s topo Topological neighborhood overlap is used to measure... Whether the related objects such as customers, account numbers, metering points, transformer areas, lines, work orders or business events are consistent or partially consistent; s time The time window matching score is used to determine whether the time of the business event mentioned by the entity to be linked matches the valid time range of the candidate graph node, relationship, or rule clause. If the business event time is within the valid time range of the node or relationship, s time Take the highest matching value; if it is within the allowable tolerance window, reduce the value according to the degree of deviation beyond the time window; if it is significantly earlier than the effective time or later than the expiration time, or if the candidate node has been deregistered, migrated or expired when the business occurs, take the lowest matching value or 0.
[0043] Traditional semantic matching methods directly sum various features by weight, without distinguishing between complete and incomplete evidence. When key evidence is missing, they still output high matching scores, leading to inflated credibility. This embodiment achieves a gating effect through an evidence continuity coefficient; the more complete the evidence, the more effective the score. When evidence is missing, the score is automatically discounted, thus avoiding misjudgments in data-deficient scenarios. It covers six dimensions: name, context, type, role, topology, and time, taking into account literal matching, semantic matching, and structural matching. Compared to methods that only match names, it significantly improves both recall and precision.
[0044] S3. Based on the candidate link object and the corresponding entity type, business scenario and business occurrence time in the power marketing business data, perform rule constraint verification to obtain the rule constraint measurement parameters and rule consistency score. Specifically, the rule constraint metric parameter refers to the set of quantitative results output after rule constraint verification, including hard constraint conflict identifiers and soft constraint deviations, used to characterize the compliance of candidate links with business rules. The rule consistency score refers to the rule compliance score calculated based on the hard constraint conflict situation and the degree of soft constraint deviation; the higher the score, the higher the degree of compliance with business rules.
[0045] To illustrate, based on the entity type, business scenario, and occurrence time of the candidate link, the corresponding business rules and engineering constraint sets are matched from the rule graph. The compliance of hard mandatory constraints and soft reference constraints is verified respectively. Two types of measurement parameters are output: hard constraint conflict identifier and soft constraint deviation. Then, the rule consistency score is calculated based on the penalty weight.
[0046] Preferably, rule constraint verification is performed based on the entity type, business scenario, and business occurrence time corresponding to the candidate link object in the electricity marketing business data to obtain rule constraint measurement parameters and rule consistency scores, including: S31. Based on the entity type, business scenario, and business occurrence time corresponding to the candidate link object in the power marketing business data, match the preset business rule constraints corresponding to the candidate link object to obtain the rule constraint measurement parameters; wherein, the preset business rule constraints include hard constraints that must be followed and soft constraints that are preferred to be adapted; the rule constraint measurement parameters include: hard constraint conflict identifier and soft constraint deviation amount. S32. Calculate the rule consistency score based on the hard constraint conflict identifier, the soft constraint deviation amount, and the preset penalty weight.
[0047] Specifically, the preset business rule constraints are the power marketing business rules and engineering constraints pre-configured in the rule graph, categorized into hard constraints and soft constraints according to their enforceability. Hard constraints must be strictly followed and cannot be violated; any violation will invalidate the link relationship, representing a compliance baseline requirement. Soft constraints are reference constraints that prioritize adaptation and allow for reasonable business exceptions; they only measure the degree of deviation and do not directly reject the link relationship, representing an optimization requirement. The hard constraint conflict identifier is a binary identifier used to characterize whether a candidate link violates a hard constraint; a value of 1 indicates a conflict, and a value of 0 indicates no conflict. The soft constraint deviation measures the overall degree of violation of the soft constraint set by the candidate link, ranging from 0 to 1, with larger values indicating more severe deviations from ideal conditions. The penalty weight represents the penalty coefficient for hard constraint conflicts and soft constraint deviations in the rule consistency score calculation; higher weights indicate a greater impact of the corresponding constraint on the score.
[0048] To illustrate, firstly, based on the entity type, business scenario, and business occurrence time of the candidate link, the corresponding set of effective rules is matched from the rule graph to ensure that the rules are compatible with the scenario and within their validity period. Then, each hard constraint is verified one by one. Common hard constraints include hierarchical affiliation constraints such as customer-account number-metering point-transformer area, transformer area line affiliation constraints, metering point status constraints, rule validity time constraints, and power supply unit jurisdiction constraints. If any hard constraint is not met, the hard constraint conflict flag is set to 1. Next, each soft constraint is verified one by one. Common soft constraints include voltage level matching, capacity range matching, business status matching, and power supply radius reference constraints. The deviation degree of each soft constraint is calculated, and then weighted and summed according to their corresponding weights to obtain the comprehensive soft constraint deviation.
[0049] The calculation logic for the rule consistency score is as follows: a base score of 1 point is subtracted from the penalty points for hard constraint conflicts and soft constraint deviations, resulting in a final score range of 0-1. When hard constraint conflicts exist, the score is significantly reduced, reflecting the enforceability of hard constraints; the greater the deviation from soft constraints, the lower the score, reflecting the optimization orientation. A typical value for the hard constraint penalty weight is 0.7, and a typical value for the soft constraint penalty weight is 0.3.
[0050] In a preferred embodiment of the present invention, it is set that For candidate link objects The set of hard constraints, For applicable candidate link objects The set of soft constraints. Hard constraint conflict identifiers are shown as... The deviation of soft constraints is expressed as : in, This indicates the degree of violation of the k-th soft constraint, and its value ranges from [0,1]. This represents the weight of the k-th soft constraint; Represents candidate link objects In the set of soft constraints The overall deviation. For hard constraints, if any hard constraint is violated, then... =1; if there is no violation of hard constraints, then =0.
[0051] The rule consistency score is calculated using the following formula: ,in, Represents candidate link objects The score for rule consistency; This indicates the penalty weight for hard constraint conflicts in the rule consistency score. This indicates the penalty weight of the deviation of soft constraints in the rule consistency score. and Belongs to the verification parameter set The set of rule constraint weights The determination can be made based on the priority of business rules, the strength of engineering constraints, and historical audit feedback. This indicates the penalty weight for hard constraint conflicts in the rule consistency score; This indicates the penalty weight of the deviation of soft constraints in the rule consistency score.
[0052] Traditional rule validation typically employs a uniform binary judgment model, either enforcing all rules rigidly, leading to the wrong exclusion of many reasonable business exceptions, or enforcing all rules leniently, resulting in the violation of core business rules and posing compliance risks. This embodiment constructs a binary constraint system, with hard constraints upholding the bottom line of business compliance and soft constraints accommodating business flexibility. This avoids compliance risks and significantly reduces the false exclusion rate. Simultaneously, it quantifies the degree of deviation of soft constraints, distinguishing the priority of different rules through weights, resulting in more refined rule adaptation and supporting differentiated business scenario requirements.
[0053] S4. Perform feedback reuse verification based on the historical link records corresponding to the candidate link objects to obtain the feedback reuse score; Specifically, the feedback reuse verification calls upon historical link data to validate the support level of the current candidate link. The feedback reuse score refers to the experience support score calculated based on four types of historical data: historical link records, alias matching, rule exceptions, and difficult case conflicts. The higher the score, the higher the degree of support of historical experience for the current link.
[0054] To illustrate, the system matches similar entities and scenario samples from historical link records, calculates the degree of experience support from four dimensions: historical confirmation, alias matching, rule exceptions, and difficult case conflicts, and obtains a feedback reuse score after weighted fusion, thereby realizing the reuse of expert experience.
[0055] Preferably, the historical link record includes: the number of times the link object is linked, the number of times the link object is verified, historical name matching conditions, historical rule adaptation conditions, and the number of times historical conflicts occur; Based on the historical link records corresponding to the candidate link objects, feedback reuse verification is performed to obtain a feedback reuse score, including: S41. Based on the historical link records corresponding to the candidate link objects, determine the number of links and the number of verifications of the link objects with the same entity type as the candidate link objects, and determine the historical link support based on the ratio of the number of links to the number of verifications of the link objects with the same entity type. S42. Determine the alias matching support based on historical name matching conditions; S43. Determine the rule exception adaptation degree based on historical rule adaptation conditions; S44. Determine the conflict hit rate based on the number of historical conflicts; S45. The feedback reuse score is obtained by weighting the historical link support, alias matching support, rule exception adaptation, conflict hit rate and preset feedback weight.
[0056] Specifically, the "Link Object Link Count" represents the number of times similar entity link relationships have been linked after historical review. The "Link Object Verification Count" represents the total number of times similar entity link relationships have been reviewed historically, including all review results such as confirmation, exclusion, and link modification. "Historical Name Matching Status" records historical name matching success rates, used to support matching determination in name variation scenarios. "Historical Rule Adaptation Status" records historically confirmed rule exceptions and applicable scenarios, used to support rule exemption determination in special scenarios. "Historical Conflict Occurrence Count" represents the number of times similar links have clashed and been excluded historically.
[0057] Specifically, historical link support represents the confirmation ratio of similar historical links, measuring the positive support of historical experience for the current link. Alias matching support represents the degree of support of the historical alias dictionary for matching the current entity name; the higher the matching degree, the higher the support. Rule exception fit represents the applicability of historical rule exception conditions to the current business scenario; the higher the fit, the more the current scenario matches the historical exceptions. Conflict hit represents the degree of conflict matching between historical difficult examples and excluded records and the current link; the higher the hit, the more likely that such links are erroneous links. Feedback weight represents the importance weight of the four types of feedback indicators in the calculation of feedback reuse score.
[0058] To illustrate, historical samples of the same entity type and business scenario are matched from historical link records. The number of times each sample is confirmed and the total number of reviews are counted. Historical link support is calculated as: number of confirmations / total number of reviews, with a value ranging from 0 to 1. The higher the confirmation rate of similar samples, the higher the historical acceptance of that type of link.
[0059] The matching function matches aliases and historical name records in the matching context. If an exact alias is matched, the support score is 1; if a normalized alias is matched, the support score is 0.8; if a historical name is matched, the support score is 0.7; and if no match is found, the support score is 0.
[0060] The exception rule entries in the historical rule adaptation conditions are matched. If all four dimensions of entity type, business scenario, rule number, and effective time match, the exception adaptation degree is 1; if there is a partial match, it is calculated according to the proportion of the matching dimensions; if there is no match at all, it is 0.
[0061] This counts the number of times a link of the same type has been conflicted or excluded in the past. The conflict hit rate is calculated as: (Number of conflicts in the same type in the past) / (Total number of samples in the same type in the past), with a value ranging from 0 to 1. A higher hit rate indicates a higher probability of historical conflict for that type of link, and it is more likely to be a faulty link.
[0062] The first three categories of historical link support, alias matching support, and rule exception adaptation are positive indicators. They are multiplied by their corresponding weights and summed. The negative conflict hit rate multiplied by its corresponding weight is then subtracted. Finally, the result is constrained to the range of 0-1 by the value range truncation function to obtain the feedback reuse score.
[0063] In a preferred embodiment of the present invention, the feedback reuse score is: in, Represents candidate link objects Feedback reuse score; This indicates the degree of support confirmed for similar links in historical link records, and can be determined by the ratio of the number of times the link object of the same type of link object is linked to the number of times the link object is verified. Indicates alias dictionary pairs and The level of support can be determined based on exact alias hits, normalized alias hits, and historical name hits or misses. This indicates the applicability of the rule exceptions confirmed by experts to the current candidate link object, which can be determined based on whether the entity type, business scenario, rule number, and validity period match. This indicates the degree of hit rate for similar conflicts in the difficult sample pool or historical exclusion records; to This indicates the weight of each feedback item; clip indicates that the calculation result is limited to the range of 0 to 1.
[0064] Traditional entity linking methods do not utilize historical experience, calculating each link from scratch, requiring repeated reviews for the same type of link, resulting in low processing efficiency. Furthermore, expert review experience cannot be distilled into reusable knowledge, model parameters cannot be continuously iterated and optimized, and accuracy is difficult to improve with data accumulation. This embodiment transforms historical review experience into quantifiable feedback reuse scores, reducing repetitive manual reviews and improving overall processing efficiency. It covers four dimensions: historical confirmation, alias matching, rule exceptions, and difficult case conflicts, balancing positive support and negative exclusion, and fully utilizing the value of historical data.
[0065] S5. Calculate the link credibility based on the semantic topology verification score, rule constraint metric parameter, rule consistency score, and feedback reuse score. Determine the entity link verification result based on the link credibility, rule constraint metric parameter, and preset credibility threshold.
[0066] Specifically, link credibility refers to the final matching credibility obtained by integrating semantic topology verification score, rule consistency score, and feedback reuse score. Credibility threshold refers to the preset score boundary used to determine link confirmation and exclusion.
[0067] Intuitively, the semantic topology verification score, rule consistency score, and feedback reuse score are fused together to obtain the final link credibility. In the result judgment stage, the hard constraint conflict identifier in the rule constraint metric is used to make a graded judgment according to a preset credibility threshold, outputting three types of results: confirmed link, excluded link, or manual review. When calculating link credibility, the core inputs are the three types of scores. The rule constraint metric is mainly used for the final result rejection judgment; that is, when a hard constraint conflict exists, the exclusion logic is directly triggered, and it does not participate in the credibility weighting calculation, avoiding the dilution of the binding force of hard constraints by the score and ensuring the bottom line of business compliance.
[0068] In a preferred embodiment of the present invention, the link credibility is calculated according to the following formula: in, Represents candidate link objects The ultimate credibility; Indicates the verification parameter set The credibility fusion function of the control; , and These represent the semantic topology verification score, rule consistency score, and feedback reuse score, respectively. Indicates model uncertainty; Indicates a hard constraint conflict. This indicates the deviation from the soft constraint. Piecewise rule functions, weighted logic functions, tree model functions, or classification functions calibrated using expert-reviewed feedback samples can be used. Model uncertainty can be determined based on the difference between the highest and second-highest candidate confidence scores for the same entity to be verified, or based on the candidate distribution entropy.
[0069] In a preferred embodiment of the present invention, the verification parameter set Represented as: in, This represents the set of weights corresponding to semantic evidence and graph relation evidence. This represents the set of weights corresponding to business rules and engineering constraints. This represents the set of feedback reuse weights corresponding to historical link records, alias dictionary, rule exception conditions, and difficult example sample pool. Represents the final credibility fusion function The corresponding set of decision weights, where T represents the set of thresholds.
[0070] In the parameter determination process, initial verification parameters are first formed based on the priority of business rules, the strength of engineering constraints, and the reliability of evidence sources. Subsequently, based on historical link confirmation rate, historical exclusion rate, alias hit rate, rule conflict frequency, and expert review feedback samples, the results were analyzed. Perform periodic calibration. The periodic calibration can be expressed as: in, B represents the calibrated set of verification parameters, and B represents the set of feedback samples formed within a preset time window or preset sample size. Represents candidate link objects The corresponding feedback samples, Let L represent the expert review label, and let L represent the loss function used to measure the difference between the expert review label and the system credibility result. This represents the constraint coefficient that maintains the stability of the initial business parameters. A single expert review feedback does not directly change the underlying model parameters; instead, it is first structured and written back as feedback samples. After reaching a preset sample size or a preset time window, the verification parameter set is then... Perform version calibration.
[0071] The feedback sample is represented as: in, Represents candidate link objects Corresponding feedback samples; The expert review label includes confirmation, exclusion, link modification, addition of aliases, addition of rule exceptions, or marking of difficult cases. Indicate the reason for the review; Indicates the feedback time; Represents a set of semantic evidence; This represents the set of evidence constrained by the rules. The system, based on... Update historical link records, alias dictionary, rule exception conditions, verification thresholds, and hard case sample pool, and then verify them in subsequent candidate link objects of the same type. Call the feedback result.
[0072] Preferably, the entity link verification result is determined based on the link credibility, rule constraint metric parameters, and a preset credibility threshold, including: The link credibility is compared with a preset credibility threshold. If the link credibility is not less than the credibility threshold, determine whether the hard constraint conflict flag in the rule constraint metric parameter indicates no hard constraint conflict. If so, obtain the entity link result of the entity to be verified based on the candidate link objects in the candidate link set. Otherwise, exclude the corresponding candidate link set to obtain the excluded candidate link set, and determine the candidate link objects in the excluded candidate link set as the entity link result of the entity to be verified. If the excluded candidate link set is empty, generate the first manual review prompt. If the credibility of the link is less than the credibility threshold, a second manual review prompt will be generated.
[0073] Specifically, the first manual review prompt is a manual review alert triggered when the candidate set is empty after hard constraint exclusion, used to handle boundary scenarios where there are no valid candidates. The second manual review prompt is a manual review alert triggered when the link's credibility does not meet the confirmation criteria, used to handle low-confidence scenarios.
[0074] In a schematic manner, the calculated link credibility is compared with a preset credibility threshold as the basis for hierarchical judgment. If the link credibility is not less than the credibility threshold, the hard constraint conflict flag in the rule constraint metric parameter indicates the absence of hard constraint conflict. If so, the entity link result of the entity to be verified is obtained based on the candidate link objects in the candidate link set; otherwise, the corresponding candidate link set is excluded, resulting in an excluded candidate link set, and the candidate link objects in the excluded candidate link set are determined as the entity link result of the entity to be verified. If the excluded candidate link set is empty, a first manual review prompt is generated. In specific implementation, when the link credibility reaches the credibility threshold, the hard constraint conflict flag is further verified: if there is no hard constraint conflict, the candidate link is directly determined as the final entity link result; if a hard constraint conflict exists, the candidate link is excluded from the set, and the above judgment logic is repeated for the remaining candidate links. If all candidates are excluded due to hard constraints, resulting in an empty candidate set, a first manual review prompt is generated, and this boundary scenario is handled manually.
[0075] When a link's credibility does not reach the confirmation threshold, it indicates that the automatic judgment lacks sufficient confidence, generating a second manual review prompt and submitting it for manual verification. In practical applications, an exclusion threshold can be added; candidate links with credibility below the exclusion threshold are automatically excluded without requiring manual review, further improving automation efficiency.
[0076] Traditional entity linking relies solely on a single score threshold for judgment, without considering business rules for rejection. This can lead to erroneous links with high scores that violate core business rules. Furthermore, it fails to account for scenarios where all candidates are excluded (empty set boundary scenarios), easily causing process interruptions or incorrect output. This embodiment employs a dual judgment mechanism of credibility threshold and hard constraint rejection, ensuring both matching accuracy requirements and upholding business compliance standards, thus preventing high-scoring but erroneous links. It also fully covers scenarios where all candidates are excluded (empty set boundary scenarios), ensuring a flawless closed-loop process logic and adaptability to complex real-world business scenarios.
[0077] Preferably, obtaining the entity link result of the entity to be verified based on the candidate link objects in the candidate link set includes: Determine whether there is a one-to-one relationship between the candidate link objects in the candidate link set and the entity to be verified; If so, the corresponding candidate link object will be determined as the entity link result of the entity to be verified; If not, calculate the credibility difference between the highest and second-highest credibility of the corresponding candidate link objects, and determine whether the credibility difference is greater than a preset difference threshold; if the credibility difference is greater than the difference threshold, determine the corresponding candidate link object as the entity link result of the entity to be verified; if the credibility difference is not greater than the difference threshold, generate a third manual review prompt.
[0078] Specifically, a one-to-one relationship means that one entity to be verified corresponds to only one candidate link object. The credibility difference represents the difference between the highest and second-highest credibility scores among multiple candidate link objects, reflecting the discriminatory power between candidates. The difference threshold is the minimum difference standard for determining whether the candidate with the highest credibility can be directly selected; a credibility difference reaching this threshold indicates sufficient discriminatory power among candidates.
[0079] To illustrate, count the number of candidate link objects corresponding to a single entity to be verified. If the number is 1, it belongs to a one-to-one relationship; if the number is greater than 1, it belongs to a multi-candidate relationship.
[0080] If it's a one-to-one relationship, the corresponding candidate link object is determined as the entity link result of the entity to be verified. If it's not a one-to-one relationship, the credibility difference between the highest and second-highest credibility of the corresponding candidate link objects is calculated, and it's determined whether the credibility difference is greater than a preset difference threshold. If the credibility difference is greater than the difference threshold, the corresponding candidate link object is determined as the entity link result of the entity to be verified. If the credibility difference is not greater than the difference threshold, a second manual review prompt is generated. Specifically, in a multi-candidate scenario, all candidates are first sorted from highest to lowest credibility, and the difference between the highest and second-highest credibility is calculated. If the credibility difference is greater than or equal to the difference threshold, it indicates that the highest-credible candidate has a significant advantage and sufficient differentiation, and the candidate with the highest credibility can be directly confirmed as the link result. If the credibility difference is less than the difference threshold, it indicates that the credibility of multiple candidates is close, the risk of automatic judgment is high, a third manual review prompt is generated, and manual judgment is submitted.
[0081] In a preferred embodiment of the present invention, a work order (95598) records: A customer in Building 3 of Dongcheng Garden reported an abnormal power supply; the last four digits of the household number are 7788. The metering point is listed as out of service but is still collecting data. The system identifies the entity to be verified from this work order. : Dongcheng Garden Building 3 / Unit Number ending in 7788 / Metering Point. Two candidate graph nodes were recalled from the four types of graphs: For customer A, associated account number 3207788, metering point MP-01, and transformer area T-01; For customer B, the associated account number is 4107788, metering point is MP-09, and transformer area is T-03. The system encapsulates candidate link objects accordingly. and .
[0082] for Name evidence, contextual evidence, graph relation evidence, and time window evidence are all usable, therefore =1、 =1、 =1、 =1. Among them, Building 3 of Dongcheng Garden and Building 3 of Dongcheng Garden belong to the normalized alias hits, which can be recorded as 0.8 in this embodiment. In this example, we take... =0.25、 =0.25、 =0.30、 =0.20, then =1. For Name and contextual evidence are available, but the relationship between the substation area and the household number cannot provide valid support; therefore, it is acceptable. =1、 =1、 =0、 =1, then =(0.25+0.25+0+0.20) / (0.25+0.25+0.30+0.20)=0.70.
[0083] In this example, take to The values are 0.18, 0.16, 0.14, 0.18, 0.22, and 0.12, respectively. For The system first normalizes the numbers for Building 3 and Building No. 3, and then matches the historical alias of Customer A. Simultaneously, the last four digits of the account number, 7788, match the account number associated with Customer A, 3207788. name Recorded as 0.90; the "abnormal power consumption," "outage," and "data still being collected" in the work order context are similar to the historical abnormal records of customer A's associated metering point MP-01, s ctx Recorded as 0.85; The object type in s is compatible with the candidate node type. type Recorded as 1.00; the work order text simultaneously includes the customer, the last four digits of the account number, the metering point status, and the abnormal event. The data contains relationships such as Customer A - Account Number 3207788, Account Number 3207788 - Metering Point MP-01, and Metering Point MP-01 - Substation T-01. Key role relationships are basically matched. rel Recorded as 0.90; the last digit of the household number, the status of the metering point, and the neighborhood of the transformer area highly overlap with the local neighborhood of the candidate node associated with the entity to be linked, s topo Recorded as 0.92; the transaction occurred within the valid time range of customer A, account number 3207788, and metering point MP-01, s time Recorded as 1.00.
[0084] for Although the name or alias of candidate customer B is somewhat similar to that of Building 3, Dongcheng Garden, name Recorded as 0.82, and the context can also form a general semantic relationship, s ctx The value is recorded as 0.70; however, customer B's associated account number 4107788, metering point MP-09, and transformer area T-03 do not adequately match the account number endings, metering point status, and transformer area neighborhood in the work order. Therefore, s rel Recorded as 0.40, s topo Recorded as 0.25; since both candidate nodes are in a valid state in terms of time, s time All can be recorded as 1.00. Based on this, we obtain... of Higher than The reasons for the above sub-items will be written into the verification record of the candidate link object.
[0085] right The system from The system reads the valid edges of customer A - account number 3207788 - metering point MP-01 - transformer area T-01. It also reads the metering point status, capacity, voltage level, and other fields from the ledger, and loads the applicable state constraints and hierarchical constraints for the current business scenario from the rule base. After comparison, the hierarchical relationship is established at the time of the business transaction, the candidate nodes are valid, and the entity types are compatible. Therefore... =0. For operations that are suspended but still collecting data, if the rule identifies it as an interpretable state deviation in an abnormal work order, rather than an absolute exclusion condition, it can be used as a soft constraint in the calculation; for example, if the degree of violation of this state deviation is recorded as 0.20, and other soft constraints are not violated, then... =0.20.
[0086] right The system reads the candidate customer B's corresponding account number 4107788, metering point MP-09, and transformer area T-03. However, these do not form the same customer-account-metering point-transformer area link as the account number's last digit, metering point status, and local neighborhood in the work order. Furthermore, key hierarchical relationships are missing or conflicting. Therefore... =1; Meanwhile, if its status fields, business events, or context deviate partially, this can be recorded as the soft constraint deviation. =0.60. In this embodiment, we take... =0.80、 =0.20, then =1-(0.80*0+0.20*0.20) / (0.80+0.20)=0.96, =1-(0.80*1+0.20*0.60) / (0.80+0.20)=0.08.
[0087] but The rule consistency score was significantly higher than .
[0088] If historical link records show that Client A and Building 3 of Dongcheng Garden have been confirmed by experts, and the alias dictionary contains an alias relationship between Building 3 of Dongcheng Garden and Client A, then... It can obtain high feedback and reuse support; If historical exclusion records exist, then Increase. In this example, take... =0.35、 =0.25、 =0.25、 =0.15, and set the confirmation threshold. =0.82, Exclusion Threshold =0.40, Difference Threshold =0.12. Calculated... =0.88、 =0.31, and and The difference is greater than The system will then confirm. Corresponding candidate graph nodes for The link target, and exclude .
[0089] If the expert review confirms Correct. Adding "Building 3, Dongcheng Garden" as a common alias for Customer A, the system generates a feedback sample. The system will add the alias to the alias dictionary and the confirmation result to the historical link record. If experts believe that the suspension of operations but the continued collection of data constitutes an acceptable exception in this type of work order, the exception condition will be added to the rule exception conditions or the rule version pending review. When similar entities to be linked appear subsequently, the system can... Call the above feedback results, and through Version calibration adjusts the weights and thresholds of relevant evidence.
[0090] By implementing this embodiment, a candidate link set is generated by matching electricity marketing business data with the target link graph. Then, based on the evidence continuity gating mechanism, the candidate link set is checked for evidence completeness, and a semantic topology check score is obtained. This mechanism avoids inflated confidence levels in scenarios with missing evidence, thus improving the reliability of the results. Next, rule constraint checks are performed on the candidate link objects according to the corresponding entity type, business scenario, and business occurrence time in the electricity marketing business data, obtaining rule constraint metric parameters and rule consistency scores. The candidate link objects are further checked through rule constraint checks to eliminate those that violate the rules. Selecting link objects ensures that the final link results better align with actual business logic, further improving the reliability of the link results. Then, feedback reuse verification is performed based on the historical link records corresponding to the candidate link objects to obtain a feedback reuse score. The link credibility is calculated based on the semantic topology verification score, rule constraint metric parameters, rule consistency score, and feedback reuse score. The entity link verification result is determined based on the link credibility, rule constraint metric parameters, and a preset credibility threshold. By comprehensively performing evidence completeness verification, rule constraint verification, and feedback reuse verification, the link verification results not only ensure compliance but also improve accuracy.
[0091] See Figure 2 This is a schematic diagram of the structure of an entity link verification device based on a power marketing business graph according to an embodiment of the present invention, comprising: The candidate link set generation module 101 is used to acquire electricity marketing business data and historical link records, and generate a candidate link set by matching the electricity marketing business data with the target link map. The evidence completeness verification module 102 is used to perform evidence completeness verification on the candidate link objects in the candidate link set based on the evidence continuity gating mechanism, and obtain the semantic topology verification score of the candidate link objects. The rule constraint verification module 103 is used to verify the rule constraints based on the candidate link object and the corresponding entity type, business scenario and business occurrence time in the power marketing business data, and to obtain the rule constraint measurement parameters and rule consistency score. The feedback reuse verification module 104 is used to perform feedback reuse verification based on the historical link records corresponding to the candidate link object and obtain the feedback reuse score. The entity link verification result generation module 105 is used to calculate the link credibility based on the semantic topology verification score, rule constraint metric parameter, rule consistency score and feedback reuse score, and to determine the entity link verification result based on the link credibility, rule constraint metric parameter and preset credibility threshold.
[0092] This invention provides an entity link verification device based on an electricity marketing business graph. The device comprises a candidate link set generation module, which acquires electricity marketing business data and historical link records, and matches the electricity marketing business data with a target link graph to generate a candidate link set. In the evidence completeness verification module, based on an evidence continuity gating mechanism, the candidate link objects in the candidate link set are verified for evidence completeness, yielding a semantic topology verification score. In the rule constraint verification module, rule constraint verification is performed based on the entity type, business scenario, and business occurrence time corresponding to the candidate link objects in the electricity marketing business data, yielding a rule constraint metric parameter and a rule consistency score. In the feedback reuse verification module, feedback reuse verification is performed based on the historical link records corresponding to the candidate link objects, yielding a feedback reuse score. Finally, in the entity link verification result generation module, the link credibility is calculated based on the semantic topology verification score, rule constraint metric parameter, rule consistency score, and feedback reuse score. The entity link verification result is determined based on the link credibility, rule constraint metric parameter, and a preset credibility threshold. By matching power marketing business data with the target link graph, a candidate link set is generated. Then, based on an evidence continuity gating mechanism, the candidate link set is checked for evidence completeness, resulting in a semantic topology check score. This mechanism avoids inflated confidence levels in scenarios with missing evidence, improving the reliability of the results. Next, rule constraint checks are performed on the candidate link objects and their corresponding entity types, business scenarios, and business occurrence times in the power marketing business data, resulting in rule constraint measurement parameters and rule consistency scores. The rule constraint check further verifies the candidate link objects, excluding those that violate the rules, so that the final link results are more consistent with the actual business logic, further improving the reliability of the link results. Then, feedback reuse checks are performed on the historical link records corresponding to the candidate link objects, resulting in a feedback reuse score. The link credibility is calculated based on the semantic topology check score, rule constraint measurement parameters, rule consistency score, and feedback reuse score. The entity link verification result is determined based on the link credibility, rule constraint measurement parameters, and a preset credibility threshold. By comprehensively performing evidence completeness checks, rule constraint checks, and feedback reuse checks, the link verification results are improved in accuracy while ensuring compliance.
[0093] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.
[0094] Those skilled in the art will understand that, for convenience and brevity, the specific working process of the device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.
[0095] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements an entity link verification method based on an electricity marketing business graph as described in the above embodiments. The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.
[0096] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.
[0097] The memory can be used to store the computer program. The processor implements various functions of the terminal device by running or executing the computer program stored in the memory and calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device or other volatile solid-state storage device.
[0098] Another embodiment of the present invention provides a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the entity link verification method based on the power marketing business graph described in the above embodiment.
[0099] The storage medium is a computer-readable storage medium, and the computer program is stored in the computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.
[0100] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.
Claims
1. A method for verifying entity links based on an electricity marketing business graph, characterized in that, include: Obtain electricity marketing business data and historical link records, and match the electricity marketing business data with the target link map to generate a candidate link set; Based on the evidence continuity gating mechanism, the completeness of evidence is verified for the candidate link objects in the candidate link set to obtain the semantic topology verification score of the candidate link objects. Based on the candidate link objects and the corresponding entity type, business scenario, and business occurrence time in the power marketing business data, rule constraint verification is performed to obtain rule constraint measurement parameters and rule consistency scores. Feedback reuse verification is performed based on the historical link records corresponding to the candidate link objects to obtain the feedback reuse score; The link credibility is calculated based on the semantic topology verification score, rule constraint metric parameter, rule consistency score, and feedback reuse score. The entity link verification result is determined based on the link credibility, rule constraint metric parameter, and preset credibility threshold.
2. The entity link verification method based on the electricity marketing business graph as described in claim 1, characterized in that, Based on the evidence continuity gating mechanism, the completeness of evidence is verified for candidate link objects in the candidate link set to obtain the semantic topology verification score of the candidate link objects, including: For each candidate link object in the candidate link set, extract name features, contextual semantic features, entity type features, relational role features, topological neighborhood features, and time window features respectively; Based on the evidence continuity gating mechanism, the evidence continuity coefficient is calculated according to the name feature, contextual semantic feature, topological neighborhood feature, time window feature and preset feature integrity weight. Under the constraint of the evidence continuity coefficient, normalization mapping is performed based on the name feature, contextual semantic feature, entity type feature, relational role feature, topological neighborhood feature, time window feature, and preset evidence feature weights to obtain the semantic topological verification score of the candidate link object.
3. The entity link verification method based on the electricity marketing business graph as described in claim 1, characterized in that, Based on the candidate link objects and their corresponding entity types, business scenarios, and business occurrence times in the electricity marketing business data, rule constraint verification is performed to obtain rule constraint metric parameters and rule consistency scores, including: Based on the entity type, business scenario, and business occurrence time corresponding to the candidate link object in the power marketing business data, a matching constraint is performed with the preset business rule constraint corresponding to the candidate link object to obtain the rule constraint measurement parameter; wherein, the preset business rule constraint includes hard constraints that must be followed and soft constraints that prioritize adaptation; the rule constraint measurement parameter includes: hard constraint conflict identifier and soft constraint deviation amount; The rule consistency score is calculated based on the hard constraint conflict identifier, the soft constraint deviation, and the preset penalty weight.
4. The entity link verification method based on the electricity marketing business graph as described in claim 1, characterized in that, The historical link records include: the number of times the link object was linked, the number of times the link object was verified, historical name matching conditions, historical rule adaptation conditions, and the number of times historical conflicts occurred. Based on the historical link records corresponding to the candidate link objects, feedback reuse verification is performed to obtain a feedback reuse score, including: Based on the historical link records corresponding to the candidate link objects, determine the number of links and the number of verifications of link objects with the same entity type as the candidate link objects, and determine the historical link support based on the ratio of the number of links to the number of verifications of link objects with the same entity type. Determine alias matching support based on historical name matching conditions; Determine the rule exception adaptation degree based on historical rule adaptation conditions; Conflict hit rate is determined based on the number of historical conflicts. The feedback reuse score is obtained by weighting the historical link support, alias matching support, rule exception adaptation, conflict hit rate, and preset feedback weight.
5. The entity link verification method based on the electricity marketing business graph as described in claim 3, characterized in that, The entity link verification result is determined based on the link credibility, rule constraint metric parameters, and a preset credibility threshold, including: The link credibility is compared with a preset credibility threshold. If the link credibility is not less than the credibility threshold, determine whether the hard constraint conflict flag in the rule constraint metric parameter indicates no hard constraint conflict. If so, obtain the entity link result of the entity to be verified based on the candidate link objects in the candidate link set. Otherwise, exclude the corresponding candidate link set to obtain the excluded candidate link set, and determine the candidate link objects in the excluded candidate link set as the entity link result of the entity to be verified. If the excluded candidate link set is empty, generate the first manual review prompt. If the credibility of the link is less than the credibility threshold, a second manual review prompt will be generated.
6. The entity link verification method based on the electricity marketing business graph as described in claim 5, characterized in that, The entity link results of the entity to be verified are obtained based on the candidate link objects in the candidate link set, including: Determine whether there is a one-to-one relationship between the candidate link objects in the candidate link set and the entity to be verified. If so, the corresponding candidate link object will be determined as the entity link result of the entity to be verified; If not, calculate the credibility difference between the highest and second-highest credibility of the corresponding candidate link objects, and determine whether the credibility difference is greater than a preset difference threshold; if the credibility difference is greater than the difference threshold, determine the corresponding candidate link object as the entity link result of the entity to be verified; if the credibility difference is not greater than the difference threshold, generate a third manual review prompt.
7. The entity link verification method based on the electricity marketing business graph as described in claim 1, characterized in that, Based on the matching of the aforementioned electricity marketing business data with the target link map, a candidate link set is generated, including: Entity identification is performed based on the electricity marketing business data to obtain the entity to be verified. Using the entity to be verified as the query condition, candidate graph nodes are determined from the target link graph based on any one of the recall dimensions; where the recall dimensions include graph neighborhood relationship, entity type consistency relationship and business related object relationship; The entity to be verified, the candidate graph node, the graph neighborhood corresponding to the candidate graph node, and the business-related object are encapsulated into a candidate link object; Generate a candidate link set based on all candidate link objects.
8. A device for verifying entity links based on an electricity marketing business graph, characterized in that, include: The candidate link set generation module is used to acquire electricity marketing business data and historical link records, and to generate a candidate link set by matching the electricity marketing business data with the target link map. The evidence completeness verification module is used to perform evidence completeness verification on the candidate link objects in the candidate link set based on the evidence continuity gating mechanism, and obtain the semantic topology verification score of the candidate link objects. The rule constraint verification module is used to verify the rule constraints based on the candidate link object and the corresponding entity type, business scenario, and business occurrence time in the power marketing business data, and to obtain the rule constraint measurement parameters and rule consistency score. The feedback reuse verification module is used to perform feedback reuse verification based on the historical link records corresponding to the candidate link object, and obtain the feedback reuse score. The entity link verification result generation module is used to calculate the link credibility based on the semantic topology verification score, rule constraint metric parameter, rule consistency score, and feedback reuse score, and to determine the entity link verification result based on the link credibility, rule constraint metric parameter, and preset credibility threshold.
9. A terminal device, characterized in that, The device includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements an entity link verification method based on an electricity marketing business graph as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform an entity link verification method based on an electricity marketing business graph as described in any one of claims 1 to 7.
