Method and apparatus for entity relationship prediction based on grid public data model

Abstract

The application provides a kind of entity relationship prediction method based on power grid public data model, comprising: obtaining the entity file and relationship file of power grid public data model, extracting first type triple data set;First type triple data set is matched with the entity file of power grid public data model, and second type triple data set is extracted;First type triple data set and second type triple data set are input into different semantic recognition models, and the relationship prediction score of first type triple data set and second type triple data set is output;The relationship prediction score of first type triple data set is fused with the relationship prediction score of second type triple data set, and the relationship prediction result is output.The application reduces the error that may occur in artificial relationship reasoning, and the matching error caused by the insufficient understanding of personnel to power grid public data model, to provide reliable basis for the expansion of power grid public data model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of big data, and in particular to a method and device for predicting entity relationship based on a power grid public data model. BACKGROUND

[0002] The enterprise public data model (SG-CIM) is a unified information view formed by State Grid Corporation of China in accordance with actual business needs and in compliance with international unified standards, and is rated as the best application practice in the industry. It is an object-oriented enterprise-level data model. At present, it contains more than one hundred information classes abstracted from problem domains as subject and sub-subject domains, such as safety, finance, power grid, customers, personnel, market, materials, projects, assets, and comprehensive domains. Each domain covers nearly a thousand related entities and nearly ten thousand attributes, and is used as a unified coding specification for State Grid, providing a unified data model for business application systems.

[0003] The enterprise public data model (SG-CIM) mainly has two different forms: a logical model and a physical model. Their names mainly come from the definitions in data warehouses. They are introduced as follows:

[0004] (1) The logical model is the logical structure of the data model. Here, the related entities and their information under the domain in the SG-CIM model are expressed.

[0005] (2) The physical model, which can also be directly referred to as the model, is used to simulate physical objects. Here, the actual existing data tables and their related information are expressed.

[0006] Specifically, the main components of the entity data of the enterprise public data model (SG-CIM) include entity Chinese name, entity English name, description information, and characteristic information. The relationship data mainly includes head entity Chinese and English names, tail entity Chinese and English names, and relationship types. The graph structure relationship types mainly include five types of relationships: association, dependency, generalization, combination, and aggregation.

[0007] The Transformer proposed by Google in 2017 is a deep network based on a self-attention mechanism. The Transformer provides enhanced parallelization and better long-range dependency modeling, mainly for natural language processing. In 2018, based on the Transformer, Google proposed the Bidirectional Encoder Representation from Transforme (BERT). The BERT model is a language model trained by Google in an unsupervised manner using a large amount of unlabeled text. Its architecture is mainly based on the Encoder part in the Transformer. This model has a wide range of applications in semantic information recognition, and a large number of variant models have also appeared. Baidu proposed an Enhanced Representation through Knowlekge Integration (ERNIE) voice representation model based on knowledge enhancement. The ERNIE model learns the semantic relationships in the real world by modeling the prior semantic knowledge of entities in a large amount of data. Specifically, the ERNIE model masks the semantic units such as words and entities, so that the model learns the semantic representation of complete concepts. Compared with BERT learning the original language signal, ERNIE directly models the prior semantic knowledge units, enhancing the model's semantic representation ability.

[0008] The construction and improvement of the SG-CIM public data model is the basis for the construction of the data center of State Grid, and the data center provides unified data services for various data analysis applications of enterprises. In the prior art, the SG-CIM model is a comprehensive abstraction of the company's enterprise-level data. In the process of updating the model version, new entities are introduced, and the newly introduced entities are linked. However, the existing relationship linking is usually manually annotated and judged, which consumes a lot of manual cost and is prone to errors, limiting the number of newly introduced entities in the new version of the SG-CIM data model. To realize the update and iteration of the SG-CIM data model, it is necessary to efficiently complete the task of complex relationship prediction in the SG-CIM data model. SUMMARY

[0009] The present application provides a method for predicting entity relationships based on a power grid public data model to solve the problem of lack of efficiency and accuracy caused by relying on manual work in the process of associating power grid entities in the power grid public model data in the prior art.

[0010] One object of the present application is to provide a method for predicting entity relationships based on a power grid public data model, which comprises:

[0011] An entity file and a relationship file of a power grid public data model are acquired, and a first type of triple data set is generated by taking head entity semantic information and tail entity semantic information as input and a relationship category as output;

[0012] The first type of triple data set is matched with the entity file of the power grid public data model, and a second type of triple data set is extracted by matching the first type of triple data set;

[0013] The first type of triple data set and the second type of triple data set are input into different semantic recognition models, and relationship prediction scores of the first type of triple data set and the second type of triple data set are output;

[0014] The relationship prediction scores of the first type of triple data set and the relationship prediction scores of the second type of triple data set are fused, and a relationship prediction result is output.

[0015] In one embodiment, the first type of triple data set includes English name semantic data and Chinese name semantic data;

[0016] The second type of triple data set includes description data and attribute data.

[0017] In one embodiment, the Chinese name semantic data in the first type of triple data set is matched with the entity file of the power grid public data model, and the second type of triple data set is extracted.

[0018] In one embodiment, the English name semantic data, the Chinese name semantic data, the description data and the attribute data are respectively input into different semantic recognition models, and the following are respectively output:

[0019] Prediction scores of multiple relationship types corresponding to the English name semantic data, prediction scores of multiple relationship types corresponding to the Chinese name semantic data, prediction scores of multiple relationship types corresponding to the description data and prediction scores of multiple relationship types corresponding to the attribute data.

[0020] In one embodiment, the prediction scores of multiple relationship types of the English name semantic data, the Chinese name semantic data, the description data and the attribute data are weighted and fused, and prediction results of multiple relationship types are output

[0021] In one embodiment, the relationship type with the largest probability value is selected as the final relationship prediction result from the output prediction results of multiple relationship types.

[0022] Another aspect of the present application is to provide a method suitable for entity relationship prediction of parallel type relationship, the method comprising:

[0023] obtaining an entity file of a power grid public data model, taking a plurality of head entities and a plurality of tail entities existing in a one-to-one relationship as an item set, wherein each entity is an item in the item set,

[0024] calculating a support probability of different entities in each item set, and a confidence probability of different entities, and

[0025] calculating a support probability of connected entities between connected item sets, and a confidence probability of connected entities;

[0026] judging a possibility of a relationship between two entities by the support probability and the confidence probability,

[0027] performing relationship prediction on entities that may have a relationship.

[0028] In one embodiment, the support probability and the confidence probability of the head entity and the tail entity in the item set are calculated,

[0029] When the support probability and the confidence probability of a certain head entity and a certain tail entity are both greater than a preset threshold, the head entity and the tail entity may have a relationship.

[0030] In one embodiment, the support probability and the confidence probability of different entities connected between connected item sets are calculated,

[0031] When the support probability and the confidence probability of the entities connected in different item sets are greater than a preset threshold, the connected entities may have a relationship.

[0032] Still another aspect of the present application provides a method suitable for entity relationship prediction of inheritance type relationship, which comprises:

[0033] obtaining an entity file of a power grid public data model, screening entities existing in a one-to-many relationship as a first item set, wherein a tail entity is a first parent entity, and a head entity is a first child entity;

[0034] obtaining an entity file of a power grid public data model, removing each entity in the first item set, and taking the remaining entities as a second item set;

[0035] selecting an entity similar to the first child entity in the first item set in the second item set as a second child entity, and

[0036] taking a first parent entity corresponding to the first child entity of the similar entity in the second item set as a second parent entity,

[0037] performing relationship prediction on the second child entity and the second parent entity.

[0038] In one embodiment, the similarity between each entity in the second set and the first sub-entity in the first set is determined by an edit distance or an edit semantic distance.

[0039] In one embodiment, the second sub-entity is taken as the head entity and the second parent entity is taken as the tail entity for relationship prediction.

[0040] Still another object of the present application is to provide an apparatus for entity relationship prediction based on a power grid common data model, which comprises:

[0041] a file acquisition module configured to acquire an entity file and a relationship file of the power grid common data model;

[0042] a data extraction module configured to take the head entity semantic information and the tail entity semantic information as input, the relationship category as output, and the generated triple data set as the first type of triple data set;

[0043] match the first type of triple data set with the entity file of the power grid common data model, extract the matched triple data set as the second type of triple data set;

[0044] a semantic recognition module configured to input the first type of triple data set and the second type of triple data set into different semantic recognition models, and output the relationship prediction scores of the first type of triple data set and the second type of triple data set;

[0045] a relationship prediction module configured to fuse the relationship prediction scores of the first type of triple data set and the relationship prediction scores of the second type of triple data set, and output the relationship prediction result.

[0046] In one embodiment, the first type of triple data set includes English name semantic data and Chinese name semantic data;

[0047] The second type of triple data set includes description data and attribute data.

[0048] In one embodiment, the Chinese name semantic data in the first type of triple data set is matched with the entity file of the power grid common data model to extract the second type of triple data set.

[0049] In one embodiment, the English name semantic data, the Chinese name semantic data, the description data and the attribute data are respectively input into different semantic recognition models, and the following are respectively output:

[0050] the prediction scores of the multiple relationship types corresponding to the English name semantic data, the prediction scores of the multiple relationship types corresponding to the Chinese name semantic data, the prediction scores of the multiple relationship types corresponding to the description data, and the prediction scores of the multiple relationship types corresponding to the attribute data.

[0051] In one embodiment, the prediction scores of the various relationship types of the obtained English name semantic data, Chinese name semantic data, description data and attribute data are weighted and fused to output prediction results of the various relationship types.

[0052] In one embodiment, the relationship type with the largest probability value is selected as the final relationship prediction result from the output prediction results of the various relationship types.

[0053] Still another aspect of the present application is to provide a device suitable for entity relationship prediction of parallel type relationship, which comprises:

[0054] The first file acquisition module is configured to acquire an entity file of a power grid common data model, and to take a plurality of head entities and a plurality of tail entities existing in a one-to-one relationship as item sets, wherein each entity is an item in the item set,

[0055] The calculation module is configured to calculate the support probability of different entities in each item set, and the confidence probability of different entities, and

[0056] The calculation module is configured to calculate the support probability of different entities in each item set, and the confidence probability of different entities, and

[0057] The entity relationship judgment module is configured to judge the possibility of the existence of a relationship between two item entities by using the support probability and the confidence probability,

[0058] The entity relationship judgment module is configured to judge the possibility of the existence of a relationship between two item entities by using the support probability and the confidence probability,

[0059] In one embodiment, the support probability and the confidence probability of the head entity and the tail entity in the item set are calculated,

[0060] When the support probability and the confidence probability of a certain head entity and a certain tail entity are both greater than a preset threshold, the head entity and the tail entity are likely to have a relationship;

[0061] The calculation module is configured to calculate the support probability of different entities in each item set, and the confidence probability of different entities, and

[0062] When the support probability and the confidence probability of the different entities connected in the different item sets are greater than a preset threshold, the connected entities are likely to have a relationship.

[0063] Still another aspect of the present application is to provide a device suitable for entity relationship prediction of inheritance type relationship, which comprises:

[0064] The second file acquisition module is configured to acquire an entity file of a power grid common data model,

[0065] An entity division module is configured to screen entities with a one-to-many relationship as a first item set, wherein a tail entity is a first parent entity and a head entity is a first child entity;

[0066] Each item entity in the first item set is removed, and the remaining entities are taken as a second item set;

[0067] An entity similar to the first child entity in the first item set is selected from the second item set as a second child entity, and

[0068] A first parent entity corresponding to the first child entity of the similar entity in the second item set is taken as a second parent entity,

[0069] The second child entity and the second parent entity are subjected to relationship prediction.

[0070] In an embodiment, the similarity between each entity in the second item set and the first child entity in the first item set is determined by an edit distance or a semantic distance.

[0071] The method and device for entity relationship prediction based on a power grid public data model provided by the application can ensure that a business application meets the requirements of the power grid public data model, reduce errors that may occur in manual relationship reasoning, and errors in matching caused by insufficient understanding of the power grid public data model by personnel, and provide a reliable basis for the expansion of the power grid public data model.

[0072] The method and device for entity relationship prediction based on a power grid public data model provided by the application can realize continuous improvement and rapid updating of the power grid public data model.

[0073] The method and device for entity relationship prediction based on a power grid public data model provided by the application can ensure that a business application meets the requirements of the power grid public data model, reduce errors that may occur in manual relationship reasoning, and errors in matching caused by insufficient understanding of the power grid public data model by personnel, and provide a reliable basis for the expansion of the power grid public data model, and promote continuous improvement and upgrading of the power grid public data model.

[0074] The method and device for entity relationship prediction based on a power grid public data model provided by the application can promote the actual business needs and intelligent landing projects of the power industry production, effectively promote the power grid big data driven business application, solve the problems of difficult model application conversion, landing, low efficiency, and low intelligent level, promote the landing of the power grid public data model, improve the compliance of business application to the power grid public data model, realize standardized business information sharing, reduce the cost of information exchange between business applications, strengthen the unified control of business applications, and ultimately promote the standardized construction of business applications.

[0075] The application discloses an entity relationship prediction and optimization method based on a power grid public data model.

[0076] The application provides an entity relationship prediction method suitable for a parallel relationship and an entity relationship prediction method suitable for an inheritance relationship, clusters different relationships between entities of the power grid public data model into multiple types, adopts different prediction algorithms for different types of relationships, and thus improves the accuracy of category prediction. BRIEF DESCRIPTION OF DRAWINGS

[0077] In order to more clearly illustrate the specific embodiments of the application or the technical solutions in the prior art, the following will briefly introduce the drawings needed to be used in the specific embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0078] Figure 1 A schematic diagram for extracting a first triple data set from the power grid public data model according to the application;

[0079] Figure 2 A schematic diagram for extracting a second triple data set from the first triple data set according to the application;

[0080] Figure 3 A flowchart of an embodiment of the entity relationship prediction based on the power grid public data model according to the application;

[0081] Figure 4 A schematic diagram of using a BERT model for English semantic recognition in an embodiment of the application;

[0082] Figure 5 A schematic diagram of parallel and inheritance relationships based on the power grid public data model according to the application.

[0083] In the above drawings, the components represented by the numbers are as follows:

[0084] 100, entity file; 200, relationship file; 300, English name semantic data; 400, Chinese name semantic data; 500, description data; 600, attribute data. DETAILED DESCRIPTION

[0085] In order to make the above and other features and advantages of the present application more apparent, the present application will be further described below with reference to the accompanying drawings. It should be understood that the specific embodiments given herein are for illustrative purposes only and are not to be construed as limiting.

[0086] In order to solve the technical problems of lacking efficiency and accuracy caused by relying on manual work in the face of power grid public model data in power grid big data and power grid entity relationship association in the prior art, according to the embodiment of the present application, a method and device for entity relationship prediction based on a power grid public data model are provided for intelligent matching and relationship prediction of the power grid public data model.

[0087] First, data preprocessing and format conversion are performed, entity-relationship identification and extraction are performed on original SG-CIM model data, and a standardized triple data set is generated. Secondly, graph structure feature relationships and semantic information are extracted, four kinds of data sets are generated for the four kinds of semantic information of Chinese name, English name, description information (Description) and attribute information (Attributes), and the relationship categories between the head entity and the tail entity are predicted by combining the BERT model and the ERNIE model for fusion.

[0088] The present application mainly focuses on relationship prediction of the logical model of SG-CIM model data, and takes SG-CIM model data as the data basis for relationship mining and prediction.

[0089] Entity relationship prediction based on grid common data model

[0090] According to the embodiment of the present application, a method for entity relationship prediction based on a power grid public data model comprises the following method steps:

[0091] Step 1, obtaining the entity file and the relationship file of the power grid public data model, and extracting the first type of triple data set. The first type of triple data set refers to a triple data set including English name semantic data and Chinese name semantic data.

[0092] The SG-CIM model mainly consists of two parts: the SG-CIM model entity file and the SG-CIM model relation file. The SG-CIM model entity file records the Chinese and English names, descriptions, and attributes of each entity. The SG-CIM model relation file records the Chinese and English names of the head entity and the tail entity, as well as the relation names. The graph structure relations mainly include five types: Association, Aggregation, Generalization, Dependency, and Composition.

[0093] like Figure 1 The diagram illustrates the extraction of a first-class triplet dataset from a power grid public data model according to the present invention. It obtains entity file 100 and relation file 200 from the power grid public data model and extracts the first-class triplet dataset. According to an embodiment of the present invention, using head entity semantic information and tail entity semantic information as input and the relation category of the graph structure as output, a first-class triplet dataset is generated. The first-class triplet dataset includes English name semantic data 300 and Chinese name semantic data 400.

[0094] Step 2: Match the first type of triplet dataset with the entity files of the power grid public data model to extract the second type of triplet dataset. The second type of triplet dataset refers to a triplet dataset that includes description datasets and attribute datasets.

[0095] The first type of triplet dataset extracts the Chinese and English names and relation categories from relation files and entity files. The second type of triplet dataset extracts the descriptive data and attribute data from the results of the first type of Chinese and English names and relation categories and entity files.

[0096] Since the entity's description and attribute information are stored in the SG-CIM model entity file, according to an embodiment of the present invention, such as Figure 2 The diagram illustrates how this invention extracts a second type of triplet dataset from a first type of triplet dataset. The semantic data of Chinese names in the first type of triplet dataset is matched with entity files from the power grid public data model to extract the second type of triplet dataset. The semantic information of the head and tail entities in the first type of triplet dataset serves as input, and the relation category is used as output to generate the second type of triplet dataset.

[0097] That is, the first type of triple data is extracted from the SG-CIM model, the corresponding description information and attribute information are found in the SG-CIM model entity file, and the description information and attribute information are summarized into the second type of triple data.

[0098] According to an embodiment of the present invention, the second type of triplet data includes description datasets 500 and attribute datasets 600.

[0099] In some specific implementations, when performing information search and matching, domain-based block partitioning and binary search methods can be used to improve the efficiency of extracting description and attribute information. In the SG-CIM model, entities and relationships are divided into blocks based on ten domain names, such as Security, Finance, Power Grid, Customer, Personnel, Market, Materials, Project, Asset, and Comprehensive Domain, depending on the domain and scope. Furthermore, there are no relational connections between entities within these domains. Therefore, during a search, it is only necessary to search for matches within the domain. Within each domain, the Chinese names of entities are sorted in lexicographical order. When searching by Chinese name, a binary search method can be used instead of a sequential search, improving search efficiency.

[0100] Step 3: Input the first type of triplet dataset and the second type of triplet dataset into different semantic recognition models, and output the relationship prediction scores of the first type of triplet dataset and the second type of triplet dataset.

[0101] like Figure 3 The diagram shown illustrates the entity relationship prediction process based on a power grid public data model in one embodiment of the present invention. The graph structure and semantic information in the SG-CIM model are extracted into four datasets: English name semantic data (300 sets), Chinese name semantic data (400 sets), description data (500 sets), and attribute data (600 sets). All four datasets use head entity semantic information and tail entity semantic information as input, and the relationship prediction type as the output classification label.

[0102] According to an embodiment of the present application, the English name semantic data 300, the Chinese name semantic data 400, the description data 500 and the attribute data 600 are respectively input into different semantic recognition models, and the following are respectively output: the prediction scores of the multiple relationship types corresponding to the English name semantic data, the prediction scores of the multiple relationship types corresponding to the Chinese name semantic data, the prediction scores of the multiple relationship types corresponding to the description data and the prediction scores of the multiple relationship types corresponding to the attribute data.

[0103] For example, the English name semantic data 300 recognition model selects a BERT model, and the Chinese name semantic data 400, the description data 500 and the attribute data 600 are in Chinese, so the ERNIE model which has better processing effect on Chinese is used for semantic information recognition. The ERNIE is obtained on the basis of the BERT model, and is respectively focused on the English and Chinese semantic information processing tasks.

[0104] According to an embodiment of the present application, the extracted data is input into different semantic recognition models, and each semantic recognition model outputs the connection prediction result of the corresponding data set, and outputs the probability of the five relationship categories after normalization.

[0105] The semantic recognition model is an automatic recognition model for recognizing the input data and outputting the connection prediction result, such as a deep network based on self-attention mechanism, a bidirectional encoder model (Bidirectional Encoder Representation from Transforme BERT) and a knowledge enhanced representation model (Enhanced Representation through Knowledge Integration ERNIE) based on knowledge enhancement.

[0106] Taking the BERT model for recognizing the English name semantic data 300 as an example, as shown in FIG. 2, a schematic diagram of the English semantic recognition by the BERT model in an embodiment of the present application is shown, and the output is a probability score in the following format: [0.83 0.10 0.01 0.03 0.03], which respectively represents the probabilities of the five relationships of association, dependency, generalization, combination and aggregation according to the positions. Figure 4 Similarly, the Chinese name semantic data 400, the description data 500 and the attribute data 600 are respectively input into the semantic recognition model ERNIE model for semantic recognition, and the following are respectively output: the probabilities of the five relationship types corresponding to the Chinese name semantic data 400, the probabilities of the five relationship types corresponding to the description data 500 and the probabilities of the five relationship types corresponding to the attribute data 600.

[0107]

[0108] ​Step 4, the relationship prediction score of the first type of triple data set is fused with the relationship prediction score of the second type of triple data set, and a relationship prediction result is output.

[0109] According to the embodiment of the application, the prediction scores of various relationship types of the obtained English name semantic data, Chinese name semantic data, description data and attribute data are weighted and fused, and a prediction result of the relationship type is output. From the output prediction result of the various relationship types, the relationship type with the maximum probability value is selected as the final relationship prediction result.

[0110] Specifically, in the embodiment, the Chinese name relationship prediction score, the English name relationship prediction score, the description (Description) relationship prediction score and the attribute (Attribute) relationship prediction score are weighted and fused by the following algorithm:

[0111] ω1×E+ω2×C+ω3×D+ω4×A=S,

[0112] wherein E, C, D and A represent the Chinese name relationship prediction score, the English name relationship prediction score, the description (Description) relationship prediction score and the attribute (Attribute) relationship prediction score respectively, S represents the final relationship prediction score, ω1, ω2, ω3 and ω4 represent the weighted average parameters respectively, and ω1+ω2+ω3+ω4=1. In the final link prediction score vector, the maximum probability is selected as the final relationship prediction result.

[0113] In order to improve the relationship prediction efficiency, the application further provides an entity relationship prediction method suitable for a parallel type relationship and an entity relationship prediction method suitable for an inheritance type relationship. According to the structural relationship between entities (association, dependency, generalization, combination and aggregation), the five types of relationships are divided into parallel type relationships (association and dependency) and inheritance type relationships (generalization, combination and aggregation), as shown in the schematic diagram of the parallel type relationship and the inheritance type relationship based on the power grid public data model. Figure 5

[0114] The K-neighbor method is used to mine the entity pairs that may have the inheritance type relationship, and then a model is used for specific relationship category prediction.

[0115] Entity relationship prediction for parallel relationship

[0116] ​It should be noted that the parallel type relationship prediction in the application is different from the relationship prediction in the above, and the parallel type relationship prediction refers to mining of entity pairs that may exist a relationship, and is an optimization method for reducing the target range, and is different from the prediction of specific relationship types in the relationship prediction in the above.

[0117] According to an embodiment of the application, a method suitable for entity relationship prediction of parallel type relationship comprises:

[0118] Step 1, obtaining an entity file of a power grid public data model, and taking a plurality of head entities and a plurality of tail entities that exist a one-to-one relationship as item sets, wherein each entity is an item in the item set.

[0119] In the parallel type relationship, most of the entities are in a one-to-one relationship, and the frequent item mining has a good performance effect on mining the relatively independent entity pairs.

[0120] Step 2, calculating the support probability of different entities in each item set, the confidence probability of different entities, the support probability of connected entities between connected item sets, and the confidence probability of connected entities.

[0121] The application uses two values of support and confidence to measure the possibility of the existence of the association relationship, and the application is divided into two cases of head entities and tail entities in the item set and the connected items between the connected item sets.

[0122] For the case of head entities and tail entities in the item set, the support probability and the confidence probability of the head entities and the tail entities in the item set are calculated, and specifically, the support probability is represented by counting and calculating the intersection ratio:

[0123] Support(A, B) = support_count(A∩B) / support_count(A∪B),

[0124] Wherein, A is the head entity, and B is the tail entity.

[0125] The confidence probability is calculated by taking the proportion of the co-occurrence number of the head entity A and the tail entity B in the total number of the head entity A and the tail entity B:

[0126]

[0127]

[0128] For the case of the connected items between the connected item sets, the support probability and the confidence probability of the connected different entities between the connected item sets are calculated, and the specific calculation method is the same as the support probability and the confidence probability calculation method of the case of the head entities and the tail entities in the item set, which is not repeated here.

[0129] Step 3, judging the possibility of the relationship between two entities by the support probability and the confidence probability.

[0130] For the case of the head entity and the tail entity in the item set, when the support probability and the confidence probability of a certain head entity and a certain tail entity are greater than the preset threshold, the head entity and the tail entity may have a relationship.

[0131] If the support and the confidence of the head entity A and the tail entity B are calculated to be greater than the preset threshold, then the head entity A and the tail entity B are considered to be frequent items, and the two may have a relationship.

[0132] For the case of the connected items in the connected item set, when the support probability and the confidence probability of the connected entities in different item sets are greater than the preset threshold, the connected entities may have a relationship.

[0133] In the above two cases, the connected entity items are merged into a new item set, and the final frequent item set is obtained, and it is assumed that there is a strong association rule between each two of them.

[0134] Step 4, relationship prediction for the entities that may have a relationship.

[0135] The frequent item set mined is subjected to relationship prediction by the method for predicting entity relationship based on the power grid public data model provided by the application, that is, the frequent item set mined is input into the relationship prediction model for relationship prediction. Only the relationship prediction probability of association and dependence is greater than the relationship prediction probability of generalization, combination and aggregation, and if the prediction probability result of the inheritance type relationship is greater than the prediction probability of the parallel type relationship, it is considered that there is no parallel type relationship.

[0136] Entity relationship prediction for inheritance relationship

[0137] It should be noted that the inheritance type relationship prediction in the application refers to the mining of entities that may have a relationship, which is an optimization method for reducing the target range.

[0138] In the inheritance type relationship, most of the entities are in a one-to-many relationship, and in this type of relationship, there is a clear upper and lower inheritance type between entities, and multiple child entities have obvious semantic similarity. According to this property, all known inheritance type relationship models in the SG-CIM model, that is, all entities and relationships including generalization, combination and aggregation, are selected as the basic graph.

[0139] According to an embodiment of the application, a method for predicting entity relationship suitable for inheritance type relationship, comprising:

[0140] Step 1, obtaining an entity file of a power grid public data model, screening an entity with a one-to-many relationship as a first item set, wherein a tail entity is a first parent entity, and a head entity is a first child entity.

[0141] Step 2, obtaining an entity file of a power grid public data model, removing each entity in the first item set, and taking the remaining entities as a second item set.

[0142] Step 3, selecting an entity similar to the first child entity in the first item set in the second item set as a second child entity, and

[0143] taking the first parent entity corresponding to the first child entity of the similar entity in the second item set as a second parent entity.

[0144] In a specific embodiment, the similarity between each entity in the second item set and the first child entity in the first item set is determined by an edit distance or an edit semantic distance (K-neighbor algorithm), and K entities most similar to the first child entity in the first item set are found in the second item set as second child entities.

[0145] In the first item set, the K first child entities with similarity in the second item set are found, and the first parent entity corresponding to the K first child entities is taken as a second parent entity.

[0146] Step 4, relationship prediction is performed on the second child entity and the second parent entity.

[0147] According to the embodiment of the present application, the second child entity is taken as a head entity, and the second parent entity is taken as a tail entity, and the relationship prediction is performed using the method for predicting the relationship of the entity based on the power grid public data model provided by the present application.

[0148] According to the embodiment of the present application, the present application provides a device for predicting the relationship of an entity based on a power grid public data model, comprising:

[0149] A file acquisition module is configured to acquire an entity file and a relationship file of a power grid public data model.

[0150] A data extraction module is configured to take head entity semantic information and tail entity semantic information as input, a relationship category as output, generate a triple data set as a first type of triple data set, and

[0151] The first type of triple data set is matched with the entity file of the power grid public data model, and a matched triple data set is extracted as a second type of triple data set.

[0152] The first type of triple data set includes English name semantic data and Chinese name semantic data; and the second type of triple data set includes description data and attribute data.

[0153] The Chinese name semantic data in the first type of triple data set is matched with entity files of a power grid public data model to extract a second type of triple data set.

[0154] The semantic recognition module is configured to input the first type of triple data set and the second type of triple data set into different semantic recognition models, and output relationship prediction scores of the first type of triple data set and the second type of triple data set.

[0155] The English name semantic data, the Chinese name semantic data, the description data and the attribute data are respectively input into different semantic recognition models, and the following are respectively output:

[0156] The prediction scores of the plurality of relationship types corresponding to the English name semantic data, the prediction scores of the plurality of relationship types corresponding to the Chinese name semantic data, the prediction scores of the plurality of relationship types corresponding to the description data and the prediction scores of the plurality of relationship types corresponding to the attribute data.

[0157] The relationship prediction module is configured to fuse the relationship prediction scores of the first type of triple data set and the relationship prediction scores of the second type of triple data set, and output a relationship prediction result.

[0158] The prediction scores of the plurality of relationship types of the English name semantic data, the Chinese name semantic data, the description data and the attribute data are weighted and fused to output prediction results of the plurality of relationship types.

[0159] The relationship type with the maximum probability value is selected as a final relationship prediction result from the output prediction results of the plurality of relationship types.

[0160] According to the embodiments of the present application, the device for predicting entity relationship suitable for parallel relationship provided by the present application comprises:

[0161] The first file acquisition module is configured to acquire entity files of a power grid public data model, and to take a plurality of head entities and a plurality of tail entities existing in a one-to-one relationship as item sets, wherein each entity is an item in the item set.

[0162] The calculation module is configured to calculate support probability of different entities in each item set, confidence probability of different entities, and

[0163] The calculation module is configured to calculate support probability of connected entities between connected item sets, and confidence probability of the connected entities, wherein

[0164] The calculation module is configured to calculate support probability and confidence probability of the head entity and the tail entity in the item set,

[0165] When the support probability and the confidence probability of a certain head entity and a certain tail entity are greater than a preset threshold, the head entity and the tail entity can have a relationship.

[0166] The support probability and the confidence probability of different entities connected between the connected item sets are calculated.

[0167] When the support probability and the confidence probability of the connected entities in different item sets are greater than a preset threshold, the connected entities can have a relationship.

[0168] The entity relationship judgment module is configured to judge the possibility of the existence of a relationship between two entities by using the support probability and the confidence probability.

[0169] The device for predicting the entity relationship based on the power grid public data model is used to predict the relationship of the mined frequent item set, that is, the mined frequent item set is input into the relationship prediction model to predict the relationship. Only the relationship prediction probability of association and dependence is greater than the relationship prediction probability of generalization, combination and aggregation. If the prediction probability result of the inheritance type relationship is greater than the prediction probability of the parallel type relationship, it is considered that the parallel type relationship does not exist.

[0170] According to the embodiment of the present application, the device for predicting the entity relationship of the inheritance type provided by the present application comprises:

[0171] The second file acquisition module is configured to acquire the entity file of the power grid public data model.

[0172] The entity division module is configured to filter the entities having a one-to-many relationship as the first item set, wherein the tail entity is taken as the first parent entity and the head entity is taken as the first child entity.

[0173] Each entity in the first item set is removed, and the remaining entities are taken as the second item set.

[0174] The entity similar to the first child entity in the first item set is selected from the second item set as the second child entity, and

[0175] The first parent entity corresponding to the first child entity of the similar entity in the second item set is taken as the second parent entity.

[0176] According to the embodiment of the present application, the similarity between each entity in the second item set and the first child entity in the first item set is judged by using the edit distance or the semantic distance.

[0177] According to the embodiment of the present application, the second child entity is taken as the head entity, the second parent entity is taken as the tail entity, and the device for predicting the entity relationship based on the power grid public data model is used to predict the relationship.

[0178] The application provides an entity relationship prediction method and device based on a power grid public data model, which is used as an automatic design tool of the power grid public data model to automatically match and intelligently push the model required by a business application.

[0179] The application provides an entity relationship prediction method and device based on a power grid public data model, which is used as an automatic design tool of the power grid public data model to automatically match and intelligently push the model required by a business application.

[0180] The application provides an entity relationship prediction method and device based on a power grid public data model, which is used as an automatic design tool of the power grid public data model to automatically match and intelligently push the model required by a business application.

[0181] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present application can be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0182] The present application is described with reference to flowcharts and / or block diagrams according to the methods, devices (systems), and computer program products of the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one flow or multiple flows and / or blocks Figure 1 The functions specified in one flow or multiple flows and / or blocks

[0183] These computer program instructions can also be stored in a computer-readable memory that can guide the computer or other programmable data processing devices to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including instruction devices that implement the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one flow or multiple flows and / or blocksFigure 1 the function specified in one or more blocks.

[0184] These computer program instructions can also be loaded into computer or other programmable data processing devices, so that a series of operation steps are performed on the computer or other programmable data processing devices to generate computer-implemented processes, thus the instructions executed on the computer or other programmable data processing devices provide processes for implementing the flow Figure 1 one or more flows and / or blocks Figure 1 the function specified in one or more blocks.

[0185] Although the preferred embodiments of the application have been described, those skilled in the art will be able to make additional modifications and variations to these embodiments without departing from the spirit and scope of the application. Accordingly, it is intended that the appended claims be construed to cover all such modifications and variations as fall within the scope of the application.

[0186] Obviously, various modifications and changes can be made to the present application by those skilled in the art without departing from the spirit and scope of the present application. Thus, it is intended that the present application encompass all such modifications and changes as fall within the scope of the claims and their equivalents.

[0187] Although the embodiments of the present application have been shown and described above, it is to be understood that the above-described embodiments are merely exemplary, and are not to be interpreted as limiting the present application, and that variations, modifications, substitutions and changes can be made to the above-described embodiments without departing from the scope of the present application.

Claims

1. A method for entity relationship prediction based on grid common data model, characterized in that, The method comprises: acquiring an entity file and a relationship file of a power grid public data model, extracting a generated triple data set with head entity semantic information and tail entity semantic information as input, a relationship category as output, as a first type of triple data set; matching the first type of triple data set with the entity file of the power grid public data model, extracting a matched triple data set as a second type of triple data set; inputting the first type of triple data set and the second type of triple data set into different semantic recognition models, and outputting relationship prediction scores of the first type of triple data set and the second type of triple data set; fusing the relationship prediction scores of the first type of triple data set and the relationship prediction scores of the second type of triple data set, and outputting a relationship prediction result.

2. The method of claim 1, wherein, The first type of triple data set comprises English name semantic data and Chinese name semantic data; The second type of triple data set comprises description data and attribute data.

3. The method of claim 2, wherein, The Chinese name semantic data in the first type of triple data set is matched with the entity file of the power grid public data model to extract the second type of triple data set.

4. The method of claim 2, wherein, The English name semantic data, the Chinese name semantic data, the description data and the attribute data are respectively input into different semantic recognition models, and the following are respectively output: prediction scores of multiple relationship types corresponding to the English name semantic data, prediction scores of multiple relationship types corresponding to the Chinese name semantic data, prediction scores of multiple relationship types corresponding to the description data, and prediction scores of multiple relationship types corresponding to the attribute data.

5. The method of claim 4, wherein, The prediction scores of multiple relationship types of the English name semantic data, the Chinese name semantic data, the description data and the attribute data are weighted and fused to output prediction results of multiple relationship types.

6. The method of claim 5, wherein, From the output prediction results of multiple relationship types, a relationship type with the largest probability value is selected as the final relationship prediction result.

7. A method for entity relationship prediction of a parallel relationship, the parallel relationship being association or dependency, characterized in that, The method comprises: acquiring an entity file of a power grid public data model, and taking a plurality of head entities and a plurality of tail entities existing in a one-to-one relationship as item sets, wherein each entity is an item in the item set, calculating support probability of different entities in each item set, and confidence probability of different entities, and calculating support probability of connected entities between connected item sets, and confidence probability of connected entities; judging the possibility of the existence of a relationship between two entities by using the support probability and the confidence probability, using the method of any one of claims 1 to 6 to perform relationship prediction on entities that may exist in a relationship.

8. The method of claim 7, wherein, calculating support probability and confidence probability of head entities and tail entities in an item set, when the support probability and the confidence probability of a certain head entity and a certain tail entity are both greater than a preset threshold, the head entity and the tail entity are screened as candidate entities that exist in a relationship; calculating support probability and confidence probability of different entities connected between connected item sets; when the support probability and the confidence probability of the entities connected between different item sets are both greater than a preset threshold, the connected entities are screened as candidate entities that exist in a relationship.

9. A method for entity relationship prediction suitable for inheritance type relationships, the inheritance type relationships being generalization, composition or aggregation, characterized in that, The method comprises: Obtaining an entity file of a power grid public data model, screening entities with one-to-many relationship as a first item set, wherein a tail entity is a first parent entity and a head entity is a first child entity; Obtaining an entity file of a power grid public data model, screening entities with one-to-many relationship as a first item set, wherein a tail entity is a first parent entity and a head entity is a first child entity; Selecting an entity similar to the first child entity in the first item set in the second item set as a second child entity, and Selecting an entity similar to the first child entity in the first item set in the second item set as a second child entity, and Using the method of any one of claims 1 to 6 to perform relationship prediction on the second child entity and the second parent entity.

10. The method of claim 9, wherein, The similarity between each entity in the second item set and the first child entity in the first item set is determined by editing distance or semantic distance.

11. An apparatus for entity relationship prediction based on grid common data model, the apparatus comprising: The device comprises: A file obtaining module for obtaining an entity file and a relationship file of a power grid public data model; A data extraction module for generating a first type of triple data set by taking head entity semantic information and tail entity semantic information as input, relationship category as output; Matching the first type of triple data set with the entity file of the power grid public data model, and extracting a matched triple data set as a second type of triple data set; A semantic recognition module for inputting the first type of triple data set and the second type of triple data set into different semantic recognition models, and outputting relationship prediction scores of the first type of triple data set and the second type of triple data set; A relationship prediction module for fusing the relationship prediction scores of the first type of triple data set and the relationship prediction scores of the second type of triple data set, and outputting a relationship prediction result.

12. The apparatus of claim 11, wherein, The first type of triple data set comprises English name semantic data and Chinese name semantic data; The second type of triple data set comprises description data and attribute data.

13. The apparatus of claim 12, wherein, The Chinese name semantic data in the first type of triple data set is matched with the entity file of the power grid public data model to extract the second type of triple data set.

14. The apparatus of claim 12, wherein, The English name semantic data, Chinese name semantic data, description data and attribute data are respectively input into different semantic recognition models, and the following are respectively output: Prediction scores of multiple relationship types corresponding to the English name semantic data, prediction scores of multiple relationship types corresponding to the Chinese name semantic data, prediction scores of multiple relationship types corresponding to the description data, and prediction scores of multiple relationship types corresponding to the attribute data.

15. The apparatus of claim 14, wherein, The prediction scores of multiple relationship types of the obtained English name semantic data, Chinese name semantic data, description data and attribute data are weighted and fused to output prediction results of multiple relationship types.

16. The apparatus of claim 15, wherein, From the output prediction results of multiple relationship types, the relationship type with the largest probability value is selected as the final relationship prediction result.

17. An apparatus for entity relationship prediction of a parallel relationship, the parallel relationship being association or dependency, the apparatus comprising: The device comprises: A first file obtaining module for obtaining an entity file of a power grid public data model, and taking a plurality of head entities and a plurality of tail entities with one-to-one relationship as an item set, wherein each entity is an item in the item set, calculating the support probability and the confidence probability of different entities in each item set, and calculating the support probability and the confidence probability of different entities between connected item sets; judging the possibility of the existence of a relationship between two entities by using the support probability and the confidence probability, using the device of any one of claims 11 to 16 to predict the relationship of the entities that may have a relationship.

18. The apparatus of claim 17, wherein, calculating the support probability and the confidence probability of the head entity and the tail entity in an item set, when the support probability and the confidence probability of a certain head entity and a certain tail entity are both greater than a preset threshold, the head entity and the tail entity are screened as candidate entities that have a relationship; calculating the support probability and the confidence probability of different entities between connected item sets; when the support probability and the confidence probability of different entities between connected item sets are both greater than a preset threshold, the different entities are screened as candidate entities that have a relationship.

19. An apparatus adapted for entity relationship prediction of inheritance type relationships, the inheritance type relationships being generalization, composition or aggregation, characterized in that, The device comprises: a second file acquisition module configured to acquire an entity file of a power grid common data model, an entity division module configured to screen entities that have a one-to-many relationship as a first item set, wherein a tail entity is a first parent entity and a head entity is a first child entity; eliminating each entity in the first item set and taking the remaining entities as a second item set; selecting an entity similar to the first child entity in the first item set in the second item set as a second child entity, and taking the first parent entity corresponding to the first child entity of the similar entity in the second item set as a second parent entity, using the device of any one of claims 11 to 16 to predict the relationship of the second child entity and the second parent entity.

20. The apparatus of claim 19, wherein, The similarity between each entity in the second item set and the first child entity in the first item set is determined by using an edit distance or a semantic distance.

Images