Knowledge graph completion method and device, computer equipment, readable storage medium and program product

Abstract

The invention relates to a knowledge graph completion method and device, computer equipment, a computer readable storage medium and a computer program product. Comprising the following steps: acquiring a knowledge graph which comprises a plurality of triples; through a preset graph convolutional network, based on the triple, extracting a relationship deep feature of each relationship and an entity deep feature of each entity; the entity deep features are generated based on fusion of multiple relation vectors and entity vectors; determining at least two task triples according to a preset completion task; calculating a target function value according to the entity deep feature and the relationship deep feature corresponding to each task triple; and according to the target function value, determining at least one target triple of which the probability is higher than a preset threshold value, and complementing the knowledge graph. According to the method, graph structure information and multi-relation characteristics of the knowledge graph are considered, relation deep-layer characteristics and entity deep-layer characteristics containing rich and effective information are mined, the relation reasoning capability is enhanced, and the knowledge graph completion accuracy is improved.

Description

Technical Field

[0001] The present application relates to the field of knowledge graph technology, and in particular to a knowledge graph completion method, apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Art

[0002] With the rapid development of internet technology, network data has also experienced explosive growth. However, this data is characterized by heterogeneity, diversity, and loose organization, making it difficult to obtain effective information. Knowledge graph technology can tightly organize the effective knowledge in network data. As a result, a large number of knowledge graphs have been constructed and widely used in the field of artificial intelligence.

[0003] However, the current knowledge graph completion methods do not mine the natural graph structure information of the knowledge graph. At the same time, there is a lack of exploration and research on the multiple complex relationships existing in some large professional field knowledge graphs, resulting in poor knowledge graph completion effects. Summary of the Invention

[0004] Based on this, it is necessary to provide a knowledge graph completion method, device, computer equipment, computer-readable storage medium and computer program product to address the above technical problems, which can mine the deep features of relationships and entities containing rich and effective information, enhance the relationship reasoning ability and improve the accuracy of knowledge graph completion.

[0005] In a first aspect, the present application provides a knowledge graph completion method, comprising:

[0006] Obtain a knowledge graph; the knowledge graph includes multiple triplets, each of which includes a head entity vector, a relationship vector, and a tail entity vector;

[0007] Through a preset graph convolutional network, based on triples, the deep relational features of each relationship and the deep entity features of each entity are extracted. The deep entity features are generated by fusing multiple relation vectors and entity vectors.

[0008] Determine at least two task triples according to a preset completion task; the preset completion task is used to indicate any two of the head entity, relationship, and tail entity corresponding to the task triple; the task triple includes any two entity deep features corresponding to the preset completion task, and any one relationship deep feature corresponding to the preset completion task;

[0009] Calculate the objective function value based on the entity deep features and relationship deep features corresponding to each task triple; the objective function value is used to represent the probability information of the task triple;

[0010] Determining, based on the objective function value, at least one target triplet having a probability higher than a preset threshold;

[0011] Complete the knowledge graph based on at least one target triple.

[0012] In one embodiment, a preset graph convolutional network is used to extract deep relational features of each relationship and deep entity features of each entity based on triples, including:

[0013] Through the preset graph convolutional network, the relationship vector and tail entity vector in each triple are jointly processed to obtain the joint feature;

[0014] According to the relationship weight information, the deep relationship features of each relationship are extracted from the relationship vector of each triple;

[0015] For each entity, the joint features of the entity's corresponding triples and the entity's initial features are fused to obtain the entity's deep features; the initial features are the head entity vector or tail entity vector corresponding to the entity.

[0016] In one embodiment, for each entity, the joint features of the triples corresponding to the entity and the initial features of the entity are fused to obtain the entity's deep features, including:

[0017] For each entity, according to the importance score, the joint features of the triples corresponding to the entity and the initial features of the entity are fused to obtain the deep features of the entity.

[0018] In one embodiment, the relationship weight information includes forward relationship weight information, reverse relationship weight information, and self-loop weight information.

[0019] In one embodiment, the method further comprises:

[0020] Obtain an initial knowledge graph; the initial knowledge graph includes multiple initial triples, and the initial triples include a head entity initial vector, a relationship initial vector, and a tail entity initial vector;

[0021] Based on the initial knowledge graph, the initial graph convolutional network is trained to obtain a trained preset graph convolutional network;

[0022] By presetting the graph convolutional network, based on the initial triples, the deep features of each relationship and each entity are extracted;

[0023] Based on the deep features of each relationship and the deep features of each entity, the initial knowledge graph is updated to obtain a knowledge graph.

[0024] In one embodiment, based on the initial knowledge graph, an initial graph convolutional network is trained to obtain a trained preset graph convolutional network, including:

[0025] In the kth iteration, the initial graph convolutional network is used to extract the sample deep features of each relationship and the sample deep features of each entity based on the initial triples;

[0026] Determine multiple sample triplets based on the sample deep features of each relationship and the sample deep features of each entity;

[0027] Calculate the loss function value based on the deep features of the samples corresponding to multiple sample triplets;

[0028] When the loss function value is greater than the preset loss value, the model parameters of the initial graph convolutional network are adjusted, and the k+1th iteration process is performed until the preset iteration stop condition is met to obtain a trained preset graph convolutional network.

[0029] In a second aspect, the present application also provides a knowledge graph completion device, comprising:

[0030] An acquisition module is used to acquire a knowledge graph; the knowledge graph includes multiple triples, each of which includes a head entity vector, a relationship vector, and a tail entity vector;

[0031] The feature extraction module is used to extract the deep relational features of each relationship and the deep entity features of each entity based on triples through a preset graph convolutional network. The deep entity features are generated by fusing multiple relation vectors and entity vectors.

[0032] A task determination module is configured to determine at least two task triples based on a preset completion task; the preset completion task is used to indicate any two of the head entity, relationship, and tail entity corresponding to the task triple; and the task triple includes any two entity deep features corresponding to the preset completion task and any one relationship deep feature corresponding to the preset completion task;

[0033] A calculation module is configured to calculate an objective function value based on the entity deep features and relationship deep features corresponding to each task triple; the objective function value is used to represent the probability information of the task triple; and based on the objective function value, at least one target triple is determined whose probability is higher than a preset threshold;

[0034] The completion module is used to complete the knowledge graph based on at least one target triple.

[0035] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method described in the first aspect above when executing the computer program.

[0036] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect above.

[0037] In a fifth aspect, the present application also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the method described in the first aspect above.

[0038] The above-mentioned knowledge graph completion method, device, computer equipment, computer-readable storage medium and computer program product obtain a knowledge graph; the knowledge graph includes multiple triples, and the triples include a head entity vector, a relationship vector and a tail entity vector; through a preset graph convolutional network, based on the triples, the relationship deep features of each relationship and the entity deep features of each entity are extracted; wherein the entity deep features are generated based on the fusion of multiple relationship vectors and entity vectors; according to the preset completion task, at least two task triples are determined; the preset completion task is used to indicate any two of the head entity, relationship and tail entity corresponding to the task triple; the task triple includes any two entity deep features corresponding to the preset completion task, and any one relationship deep feature corresponding to the preset completion task; according to the entity deep features and relationship deep features corresponding to each task triple, the objective function value is calculated; the objective function value is used to characterize the probability information of the task triple; according to the objective function value, at least one target triple with a probability higher than a preset threshold is determined; according to at least one target triple, the knowledge graph is completed. Through the above method, the graph structure information and multiple relationship features of the knowledge graph are taken into consideration, and the deep relationship features and entity deep features containing rich and effective information are mined, so as to efficiently complete the knowledge graph completion task, enhance the relationship reasoning ability, improve the accuracy of knowledge graph completion, and provide data support for the subsequent application of knowledge graphs. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments of the present application or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying any creative work.

[0040] Figure 1 These are some examples of biomedical knowledge graphs;

[0041] Figure 2 Schematic diagram of a process of a knowledge graph completion method in one embodiment;

[0042] Figure 3 A schematic diagram showing the physical features of an embodiment;

[0043] Figure 4 Schematic diagram of a knowledge graph completion process based on a graph convolutional network in one embodiment;

[0044] Figure 5 Schematic diagram of the process of a knowledge graph completion method in another embodiment;

[0045] Figure 6 1 is a structural block diagram of a knowledge graph completion device in one embodiment;

[0046] Figure 7 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0048] It is understandable that the knowledge graph completion mentioned in the embodiments of this application refers to the improvement and supplementation of the existing knowledge graph. Specifically, the knowledge graph is generally composed of a large number of triples (head entity, relationship, tail entity), and the knowledge graph completion task is to find the missing parts of certain triples in the knowledge graph. For example, for the triple of (head entity, relationship,?), find the corresponding tail entity; for the triple of (?, relationship, tail entity), find the corresponding head entity; for the triple of (head entity,?, tail entity), find the corresponding relationship.

[0049] The multiple complex relationships mentioned in the embodiment of this application refer to the many-to-many relationships between entities in the knowledge graph, that is, there are many different types of relationships between entities. Figure 1 In the biomedical knowledge graph, the entity "calcium" has two relationships with the entity "caspase 3": "affects response" and "affects co-processing." The entity "calcium" has a "reduces response" relationship with the entity "AKT serine / threonine kinase 1." The entity "zinc" also has two relationships with the entity "AKT serine / threonine kinase 1": "reduces phosphorylation" and "increases response." The entity "zinc" has an "increases cleavage" relationship with the entity "caspase 3." The entity "calcium" has an "increases expression" relationship with the entity "cyclin D1." The entity "zinc" has a "reduces expression" relationship with the entity "cyclin D1."

[0050] In an exemplary embodiment, Figure 2As shown, a knowledge graph completion method is provided. This embodiment uses the method applied to a terminal as an example for illustration. It is understandable that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and implemented through the interaction between the terminal and the server. In this embodiment, the method includes:

[0051] Step 202, obtain a knowledge graph; the knowledge graph includes multiple triples, and the triples include a head entity vector, a relationship vector, and a tail entity vector.

[0052] Among them, the knowledge graph of this embodiment corresponds to the existing knowledge graph to be completed, and the knowledge graph includes multiple triples, each of which includes a head entity vector, a relationship vector and a tail entity vector.

[0053] In an optional implementation, the entities and relationships in the knowledge graph are processed through a model that generates word vectors to generate entity vectors and relationship vectors, and multiple triples are formed based on the graph structure information in the knowledge graph.

[0054] Optionally, a preset graph convolutional network is used to extract features of entities and relationships in the knowledge graph, generate entity vectors and relationship vectors, and form multiple triples based on the graph structure information in the knowledge graph.

[0055] In step 204, a preset graph convolutional network is used to extract the deep relational features of each relation and the deep entity features of each entity based on triples; wherein the deep entity features are generated based on the fusion of multiple relation vectors and entity vectors.

[0056] Among them, the preset graph convolutional network is a scalable semi-supervised learning method for graph-structured data, and is also an efficient variant of convolutional neural networks. It can directly perform convolution operations on Euclidean data and can efficiently extract features. It can be understood that the graph convolutional network consists of multiple graph convolution layers, which can perform multiple graph convolutions on the data of the knowledge graph and obtain the structural features of the graph through the propagation of node information in the hidden layer. The way information is propagated between layers in the graph convolutional neural network can be defined as the following formula (1):

[0057] (1);

[0058] in, represents the adjacency matrix of the graph, , is the identity matrix, represents the parameter matrix, express The degree matrix of represents the feature matrix, Represents a nonlinear activation function. In this way, each node in the graph can extract the characteristic information of the graph structure to complete various tasks.

[0059] In an optional implementation, the preset graph convolutional network of this embodiment can be a relation-based joint graph attention network model (RJGANM), which can simultaneously extract graph structure features and multiple complex relationship features in the knowledge graph. The RJGANM model includes a joint graph attention network (JGAN) module based on a graph convolutional network and a prediction module. The JGAN module takes into account the multiple relationship features of the knowledge graph and fuses multiple relationship vectors with corresponding entity vectors, so that the deep entity features contain valid information from the multiple relationship vectors and valid information related to the graph structure.

[0060] Step 206: Determine at least two task triples based on the preset completion task; the preset completion task is used to indicate any two of the head entity, relationship, and tail entity corresponding to the task triple; the task triple includes any two entity deep features corresponding to the preset completion task, and any one relationship deep feature corresponding to the preset completion task.

[0061] Preset completion tasks are tasks set based on actual needs, including head entity completion tasks, relationship completion tasks, and tail entity completion tasks. Preset completion tasks are used to indicate any two of the head entity, relationship, and tail entity corresponding to a task triple. For example, a preset completion task might be a tail entity completion task (head entity, relationship, ?), which indicates the head entity and relationship corresponding to a task triple.

[0062] Based on the deep features extracted in step 204 and any two of the head entity, relationship, and tail entity indicated by the preset completion task, at least two task triples are constructed, each comprising the entity deep features and the relationship deep features. The at least two task triples contain all prediction results corresponding to the preset completion task. For example, if the preset completion task is the tail entity completion task (head entity, relationship, ?), the at least two task triples contain all possible tail entities: (head entity, relationship, tail entity 1), (head entity, relationship, tail entity 2), ..., (head entity, relationship, tail entity n).

[0063] Step 208 : Calculate the objective function value based on the entity deep features and relationship deep features corresponding to each task triple; the objective function value is used to characterize the probability information of the task triple.

[0064] Among them, the entity deep features and relationship deep features corresponding to each task triple are input into the objective function to calculate the probability information and obtain the objective function value.

[0065] Alternatively, assume that the task triple is ( , , ), based on the following formula (2), the entity deep features of the task triplet are and Reshape into a two-dimensional vector and use multiple filters for feature extraction:

[0066] (2);

[0067] in yes The two-dimensional representation of yes The two-dimensional representation of W represents the weight matrix. Represents a filter.

[0068] Taking the prediction relationship as an example, the prediction module adopts As the activation function, the objective function value is calculated based on formula (3):

[0069] (3);

[0070] Among them, the objective function value m is used to represent the triple The probability information of being true.

[0071] Step 210: Determine at least one target triplet with a probability higher than a preset threshold according to the objective function value.

[0072] The preset threshold is a critical value set in advance based on experiments or experience to evaluate the probability. The task triples with a probability higher than the preset threshold are determined as target triples.

[0073] Step 212: Complete the knowledge graph based on at least one target triple.

[0074] Among them, based on the determined target triples, the original knowledge graph is completed to complete the knowledge graph completion task.

[0075] In the above-mentioned knowledge graph completion method, a knowledge graph is obtained; the knowledge graph includes multiple triples, and the triples include a head entity vector, a relationship vector and a tail entity vector; through a preset graph convolutional network, based on the triples, the relationship deep features of each relationship and the entity deep features of each entity are extracted; wherein, the entity deep features are generated based on the fusion of multiple relationship vectors and entity vectors; according to the preset completion task, at least two task triplets are determined; the preset completion task is used to indicate any two of the head entity, relationship and tail entity corresponding to the task triplet; the task triplet includes any two entity deep features corresponding to the preset completion task, and any one relationship deep feature corresponding to the preset completion task; according to the entity deep features and relationship deep features corresponding to each task triplet, the objective function value is calculated; the objective function value is used to characterize the probability information of the task triplet; according to the objective function value, at least one target triplet with a probability higher than a preset threshold is determined; according to the at least one target triplet, the knowledge graph is completed. Through the above method, the graph structure information and multiple relationship features of the knowledge graph are taken into consideration, and the deep relationship features and entity deep features containing rich and effective information are mined, so as to efficiently complete the knowledge graph completion task, enhance the relationship reasoning ability, improve the accuracy of knowledge graph completion, and provide data support for the subsequent application of knowledge graphs.

[0076] In an exemplary embodiment, step 204 includes: jointly processing the relationship vector and the tail entity vector in each triple through a preset graph convolutional network to obtain a joint feature; extracting the deep relationship feature of each relationship from the relationship vector of each triple based on the relationship weight information; for each entity, fusing the joint feature of the triple corresponding to the entity and the initial feature of the entity to obtain the deep entity feature of the entity; the initial feature is the head entity vector or the tail entity vector corresponding to the entity.

[0077] The knowledge graph of this embodiment is different from other graph-structured data. The edges in the knowledge graph contain semantic information, representing that the relationship also has a direction, allowing information to propagate in directed relationships, thereby expanding the reverse relationship and self-loop relationship between two entities. The knowledge graph of this embodiment is represented by the following formulas (4) and (5):

[0078] (4);

[0079] (5);

[0080] Among them, i and j represent nodes, that is, entities in the knowledge graph. represents a reverse edge, and L represents a self-loop edge.

[0081] In many large knowledge graphs, there may be many different complex relationships between entities. Therefore, this embodiment uses the relationship feature As the initial relationship representation, use Represents the relationship vector representation, represents the entity vector representation, where In order for entities to obtain the feature vectors of their corresponding different relations, such as Figure 3 As shown in Figure 2, the JGAN module jointly processes relations with their corresponding entities through the operations used in the knowledge graph embedding method, which can be performed through formula (6):

[0082] (6);

[0083] in, Indicates a join operation, Indicates its corresponding vector representation.

[0084] The update rule of the traditional graph convolutional network is shown in the following formula (7):

[0085] (7);

[0086] in, is a parameter related to the relationship, and N(i) represents the set of adjacent nodes of node i. That is, the feature of the current node is generated by fusing the features of the adjacent nodes. This method cannot obtain the relationship features in the knowledge graph of multiple complex relationships.

[0087] Reference Figure 4 In the preset graph convolutional network of this embodiment, the JGAN module combines the features of the relationship and its corresponding adjacent nodes through a joint operation, that is, , and also calculates the importance scores of different edges of each pair of adjacent nodes, namely express , express , express Therefore, the update rule of the graph convolutional neural network layer in the JGAN module is shown in the following formula (8):

[0088] (8);

[0089] in, represents the initialization vector of entity node j, is the initialization vector of relation r, represents the updated vector representation of entity node i, is a parameter that depends on the relation r.

[0090] It is understandable that the relationship of the knowledge graph in this embodiment has a direction, and the weights corresponding to different adjacency relationships are Different. Optionally, different weights are defined based on the importance scores of different adjacency relationships and the weight information of different adjacency relationships. .

[0091] The update rule for relation r is shown in the following formula (9):

[0092] (9);

[0093] in, represents the weight of relation r, Represents the initialization vector representation of the relationship r. Optionally, different weights are defined based on the importance scores of different adjacency relationships and the weight information of different adjacency relationships. In this way, deep features of all entities and relations are obtained.

[0094] Reference Figure 4 , for a triple in a given knowledge graph , through the combination operation, the tail entity The corresponding relationship Fusion into a vector representation containing relational features, while the tail entity The corresponding relationship Fusion is performed to obtain vector representation, combined with entity The vector representation of two different neighborhood relationships is used to obtain its embedded representation containing neighborhood information, and then the entity is calculated The importance of different neighbor nodes and , combining the neighborhood relationship features of the entity into the entity representation to obtain an entity representation that contains neighborhood relationships and multiple complex relationship features, and inputting the vector representation of the entity and relationship into the objective function to calculate the score.

[0095] In an exemplary embodiment, for each entity, the joint features of the triples corresponding to the entity and the initial features of the entity are fused to obtain the entity deep features of the entity, including: for each entity, according to the importance score, the joint features of the triples corresponding to the entity and the initial features of the entity are fused to obtain the entity deep features of the entity.

[0096] It is understandable that in the graph structure of the knowledge graph, the importance of different neighboring nodes of each entity is different. Under the action of the attention mechanism, the preset graph convolutional network of this embodiment assigns different importance scores to different neighboring nodes of the same entity. For a given entity h, the set of neighboring nodes of node h is defined as ,in, , the preset graph convolutional network can calculate the attention coefficient through formula (10):

[0097] (10);

[0098] in, represents the attention coefficient of node j to node i, represents the vector representation of node i, represents the vector representation of node j, is the weight matrix, a represents the parameterized feedforward neural network, and LeakyReLU(·) represents the activation function.

[0099] Based on the principle of additive attention mechanism, an alignment score function is defined in the knowledge graph, so that the preset graph convolutional network can pay more attention to the graph structure features that are helpful for the knowledge graph completion task. The calculation function of the importance score refers to the following formula (11):

[0100] (11);

[0101] in, represents the attention coefficient of node j to node i, represents the importance score of node i to node j.

[0102] Specifically, the calculation function of the importance score refers to the following formula (12):

[0103] (12);

[0104] Among them, || represents the connection operation and a represents the parameterized feedforward neural network.

[0105] Reference Figure 3 , for the entity node h given in the figure, the set of its adjacent nodes is , after calculating the function shown in formula (12), the adjacent nodes of the entity node h can be obtained Importance score In this way, the preset graph convolutional network can obtain entity vector representations containing different neighborhood relationships. At the same time, since the JGAN module combines the features of the entity and its corresponding relationship, different relationship vector representations also receive different attention scores.

[0106] It is understandable that the relationships in the knowledge graph have directions, and different adjacency relationships have different importance. The JGAN module defines different weights for different relationships based on the importance scores of different relationships. , the weight of the relationship r between entity node i and entity node j The definition is as shown in the following formula (13):

[0107] (13);

[0108] in, Represents the positive relationship weight information, Represents the reverse relationship weight information, Represents the self-loop relationship weight information. The calculation method is the same as Combining formula (8) and formula (13), according to the importance score, the joint features of the triples corresponding to the entity and the initial features of the entity are fused to obtain the entity's deep features.

[0109] In this embodiment, an attention mechanism is used to measure the importance of different multiple adjacency relationships between entities, obtain the local structural information of the graph in the knowledge graph, and use a graph convolutional neural network to explore the natural graph structural features in the knowledge graph, thereby obtaining entity and relationship embedding representations that contain the knowledge graph structural information and multiple relationship features.

[0110] In an exemplary embodiment, the relationship weight information includes forward relationship weight information, reverse relationship weight information, and self-loop weight information.

[0111] Wherein, referring to the above formula (13), the relationship weight information includes the positive relationship weight information , reverse relationship weight information and self-circulating weight information .

[0112] In an exemplary embodiment, referring to Figure 5 , the method further comprises:

[0113] Step 502, obtain an initial knowledge graph; the initial knowledge graph includes multiple initial triples, and the initial triples include a head entity initial vector, a relationship initial vector, and a tail entity initial vector.

[0114] Step 504: Based on the initial knowledge graph, the initial graph convolutional network is trained to obtain a trained preset graph convolutional network.

[0115] Step 506 , extracting the deep features of each relationship and each entity based on the initial triples through a preset graph convolutional network.

[0116] Here, the above formulas (8), (9) and (13) are combined to extract deep features from the original entities and relationships.

[0117] Step 508: Based on the deep features of each relationship and the deep features of each entity, the initial knowledge graph is updated to obtain a knowledge graph.

[0118] In an exemplary embodiment, step 504 includes: during the kth iteration, extracting the sample deep features of each relationship and the sample deep features of each entity based on the initial triples through the initial graph convolutional network; determining multiple sample triplets according to the sample deep features of each relationship and the sample deep features of each entity; calculating the loss function value according to the sample deep features corresponding to the multiple sample triplets; when the loss function value is greater than the preset loss value, adjusting the model parameters of the initial graph convolutional network, and performing the k+1th iteration process until the preset iteration stop condition is met to obtain a trained preset graph convolutional network.

[0119] Wherein, k is an integer greater than or equal to 1, the kth iteration process refers to the current iteration process, and the k+1th iteration process refers to the next iteration process. In this embodiment, the loss function value can be calculated using the following formula (14):

[0120] (14);

[0121] Combined with the above formula (3), the loss function can be expressed by the following formula (15):

[0122] (15);

[0123] Among them, p represents the label vector. If the relationship exists, the corresponding element in the label vector is 1, otherwise it is 0.

[0124] The preset iteration stop condition can be that the number of iterations reaches the preset maximum number of iterations or the loss function value after the iteration is not greater than the preset loss value. Through multiple iterations, the model parameters of the initial graph convolutional network are adjusted, the model parameters of the output initial graph convolutional network are adjusted,

[0125] For example, it should be noted that the performance of the RJGANM model provided in this embodiment was evaluated based on the following operating environment: a 6-core Intel Xeon E5-2620 processor, 64GB of memory, a Titan X graphics card, and an Ubuntu 16.04 operating system. Two publicly available test datasets in the knowledge graph field were selected: WN18RR and FB15K237. Alternatively, the Biomedical dataset and the CNMedical dataset could be selected. Among them:

[0126] (1) The WN18RR dataset is a benchmark dataset obtained by removing the reversible relations in the WN18 dataset. It contains 140,943 entities and 11 relationship types. The WN18 dataset is an English dictionary containing semantic information extracted from WordNet. It covers 40,943 entities and 18 relationship types. Each entity corresponds to a different semantics, and there is a certain relationship between different semantics.

[0127] (2) The FB15K237 dataset is a benchmark dataset extracted from FB15K, containing 14,541 entities and 237 relation types. The FB15K dataset is a large knowledge base consisting of metadata of large-scale commonsense knowledge and a subsample of the FreeBase dataset, which contains 14,951 entities and 1,345 relation types.

[0128] (3) The Biomedical dataset is a biomedical knowledge graph that contains biomedical datasets describing the relationships between diseases, chemicals, and genes, covering 14,858 entities and 134 relationship types.

[0129] (4) CNMedical is a drug knowledge graph dataset that describes the relationship between diseases, drugs, and symptoms. The dataset contains 27,100 entities and 41 relationship types.

[0130] This evaluation uses the following metrics: Mean Rank (MR), Mean Reciprocal Ranking (MRR), and Average Hit Rate (Hit@k). Taking the prediction of tail entities as an example, for a triple (h, r, ?), each entity in the knowledge graph is substituted into the triple (h, r, ?) as the tail entity and the objective function value f(h, r, t) is calculated to obtain the corresponding score for each entity (i.e., the objective function value). The mean rank refers to the average ranking of the correct entities in the objective function scores, while the mean reciprocal ranking (MRR) refers to the sum of the reciprocals of the mean ranks. Similarly, the average hit rate is calculated based on the mean rank. Specifically, the number of hits is calculated by checking whether the correct answer for each test triple is ranked in the top k of the sequence. If so, the hit count is incremented by 1, and the average hit rate is calculated based on the number of hits. Lower mean rank (MR), higher mean reciprocal ranking (MRR), and higher average hit rate indicate better performance.

[0131] The RJGANM model and all baseline methods provided in this example were tested using the Biomedical and CNMedical datasets. Table 1 shows the test results based on the Biomedical and CNMedical datasets. On the Biomedical dataset, the RJGANM model achieved the best results in terms of MR, Hit@10, Hit@3, and Hit@1. Specifically, compared to the SACN method with the best MR value, the RJGANM model's MR value increased by 54 times; compared to the RotatE method with the best Hit@3, the RJGANM model's Hit@3 value increased by 2.3%. Furthermore, on the CNMedical dataset, the RJGANM model also achieved good results in terms of MR and Hit@k. Specifically, compared to the best baseline method's result of 2508, the RJGANM model's MR value increased by 82 points; the RJGANM method achieved the best Hit@3 value of 0.197 and Hit@1 value of 0.132.

[0132] Table 1:

[0133]

[0134] In order to further verify the versatility of the RJGANM model, the RJGANM model provided in this embodiment and all baseline methods were tested using the WN18RR dataset and the FB15K237 dataset. Refer to Table 2, which shows the test results based on the WN18RR and FB15K237 datasets. Among them, on the FB15K237 dataset, the RJGANM model achieved the best results in MR value, MRR, Hit@3, and Hit@1. Specifically, compared with the RotatE method with the best MR value, the MR value of the RJGANM model increased by 27 points, and also obtained the best Hit@k value. In the WN18RR dataset, the RJGANM model also obtained the best MRR value of 2419, and its performance in other indicators was average.

[0135] Table 2:

[0136]

[0137] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0138] Based on the same inventive concept, the embodiment of the present application also provides a knowledge graph completion device for implementing the above-mentioned knowledge graph completion method. The implementation solution provided by the device is similar to the implementation solution described in the above-mentioned method. Therefore, the specific limitations in the one or more knowledge graph completion device embodiments provided below can be found in the above-mentioned limitations on the knowledge graph completion method, and will not be repeated here.

[0139] In an exemplary embodiment, Figure 6 As shown, a knowledge graph completion device is provided, comprising:

[0140] Acquisition module 602 is used to acquire a knowledge graph; the knowledge graph includes multiple triples, and the triples include a head entity vector, a relationship vector and a tail entity vector.

[0141] The feature extraction module 604 is used to extract the relationship deep features of each relationship and the entity deep features of each entity based on triples through a preset graph convolutional network; wherein the entity deep features are generated based on the fusion of multiple relationship vectors and entity vectors.

[0142] The task determination module 606 is used to determine at least two task triples based on the preset completion task; the preset completion task is used to indicate any two of the head entity, relationship and tail entity corresponding to the task triple; the task triple includes any two entity deep features corresponding to the preset completion task, and any one relationship deep feature corresponding to the preset completion task.

[0143] The calculation module 608 is used to calculate the objective function value based on the entity deep features and relationship deep features corresponding to each task triple; the objective function value is used to characterize the probability information of the task triple; based on the objective function value, determine at least one target triple with a probability higher than a preset threshold.

[0144] The completion module 610 is used to complete the knowledge graph according to at least one target triple.

[0145] In the above-mentioned knowledge graph completion device, the graph structure information and multiple relationship features of the knowledge graph are taken into consideration, and the deep relationship features and entity deep features containing rich and effective information are mined, so as to efficiently complete the knowledge graph completion task, enhance the relationship reasoning ability, improve the accuracy of knowledge graph completion, and provide data support for subsequent knowledge graph applications.

[0146] In an exemplary embodiment, the feature extraction module 604 is also used to jointly process the relationship vector and the tail entity vector in each triple through a preset graph convolutional network to obtain a joint feature; based on the relationship weight information, the deep relationship features of each relationship are extracted from the relationship vector of each triple; for each entity, the joint features of the triple corresponding to the entity and the initial features of the entity are fused to obtain the entity deep features of the entity; the initial features are the head entity vector or the tail entity vector corresponding to the entity.

[0147] In an exemplary embodiment, the feature extraction module 604 is further configured to fuse, for each entity, the joint features of the triplet corresponding to the entity and the initial features of the entity according to the importance score, to obtain the deep features of the entity.

[0148] In an exemplary embodiment, the relationship weight information includes forward relationship weight information, reverse relationship weight information, and self-loop weight information.

[0149] In an exemplary embodiment, the knowledge graph completion device also includes a training module, which is used to obtain an initial knowledge graph; the initial knowledge graph includes multiple initial triples, and the initial triples include a head entity initial vector, a relationship initial vector, and a tail entity initial vector; based on the initial knowledge graph, the initial graph convolutional network is trained to obtain a trained preset graph convolutional network; through the preset graph convolutional network, based on the initial triples, the deep features of each relationship and the deep features of each entity are extracted; based on the deep features of each relationship and the deep features of each entity, the initial knowledge graph is updated to obtain a knowledge graph.

[0150] In an exemplary embodiment, the training module is also used to extract the sample deep features of each relationship and the sample deep features of each entity based on the initial triples through the initial graph convolutional network during the k-th iteration; determine multiple sample triplets based on the sample deep features of each relationship and the sample deep features of each entity; calculate the loss function value based on the sample deep features corresponding to the multiple sample triplets; when the loss function value is greater than the preset loss value, adjust the model parameters of the initial graph convolutional network and perform the k+1-th iteration process until the preset iteration stop condition is met to obtain a trained preset graph convolutional network.

[0151] Each module in the above-mentioned knowledge graph completion device can be implemented in whole or in part through software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each of the above modules.

[0152] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 7 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store intermediate data processed by the graph convolutional network. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a knowledge graph completion method is implemented.

[0153] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0154] In an exemplary embodiment, a computer device is provided, comprising a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented: obtaining a knowledge graph; the knowledge graph comprises multiple triples, each triple comprising a head entity vector, a relationship vector, and a tail entity vector; extracting, based on the triples, relational deep features of each relationship and entity deep features of each entity through a preset graph convolutional network; wherein the entity deep features are generated based on the fusion of multiple relationship vectors and entity vectors; determining at least two task triples according to a preset completion task; the preset completion task is used to indicate any two of the head entity, relationship, and tail entity corresponding to the task triple; the task triple comprises any two entity deep features corresponding to the preset completion task, and any one relationship deep feature corresponding to the preset completion task; calculating an objective function value according to the entity deep features and relationship deep features corresponding to each task triple; the objective function value is used to characterize the probability information of the task triple; determining, according to the objective function value, at least one target triple with a probability higher than a preset threshold; and completing the knowledge graph according to the at least one target triple.

[0155] In one embodiment, when the processor executes the computer program, it also implements the following steps: through a preset graph convolutional network, the relationship vector and the tail entity vector in each triple are jointly processed to obtain a joint feature; based on the relationship weight information, the deep relationship features of each relationship are extracted from the relationship vector of each triple; for each entity, the joint features of the triple corresponding to the entity and the initial features of the entity are fused to obtain the entity deep features of the entity; the initial features are the head entity vector or the tail entity vector corresponding to the entity.

[0156] In one embodiment, when executing the computer program, the processor further implements the following steps: for each entity, based on the importance score, fusing the joint features of the triplet corresponding to the entity and the initial features of the entity to obtain the entity deep features of the entity.

[0157] In one embodiment, when the processor executes the computer program, it also implements the following steps: obtaining an initial knowledge graph; the initial knowledge graph includes multiple initial triples, and the initial triples include a head entity initial vector, a relationship initial vector, and a tail entity initial vector; based on the initial knowledge graph, the initial graph convolutional network is trained to obtain a trained preset graph convolutional network; through the preset graph convolutional network, based on the initial triples, the deep features of each relationship and the deep features of each entity are extracted; based on the deep features of each relationship and the deep features of each entity, the initial knowledge graph is updated to obtain a knowledge graph.

[0158] In one embodiment, when the processor executes the computer program, the following steps are also implemented: during the kth iteration, based on the initial triples, the sample deep features of each relationship and the sample deep features of each entity are extracted through the initial graph convolutional network; multiple sample triplets are determined according to the sample deep features of each relationship and the sample deep features of each entity; the loss function value is calculated according to the sample deep features corresponding to the multiple sample triplets; when the loss function value is greater than the preset loss value, the model parameters of the initial graph convolutional network are adjusted, and the k+1th iteration process is performed until the preset iteration stop condition is met to obtain a trained preset graph convolutional network.

[0159] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the following steps: obtaining a knowledge graph; the knowledge graph includes multiple triples, each triple including a head entity vector, a relationship vector, and a tail entity vector; extracting, based on the triples, the relationship deep features of each relationship and the entity deep features of each entity through a preset graph convolutional network; wherein the entity deep features are generated based on the fusion of multiple relationship vectors and entity vectors; determining at least two task triples according to a preset completion task; the preset completion task is used to indicate any two of the head entity, relationship, and tail entity corresponding to the task triple; the task triple includes any two entity deep features corresponding to the preset completion task, and any one relationship deep feature corresponding to the preset completion task; calculating an objective function value according to the entity deep features and relationship deep features corresponding to each task triple; the objective function value is used to characterize the probability information of the task triple; according to the objective function value, determining at least one target triple with a probability higher than a preset threshold; and completing the knowledge graph according to the at least one target triple.

[0160] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: the relationship vector and the tail entity vector in each triple are jointly processed through a preset graph convolutional network to obtain a joint feature; based on the relationship weight information, the deep relationship features of each relationship are extracted from the relationship vector of each triple; for each entity, the joint features of the triple corresponding to the entity and the initial features of the entity are fused to obtain the entity deep features of the entity; the initial features are the head entity vector or the tail entity vector corresponding to the entity.

[0161] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: for each entity, based on the importance score, the joint features of the triples corresponding to the entity and the initial features of the entity are fused to obtain the deep entity features of the entity.

[0162] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: obtaining an initial knowledge graph; the initial knowledge graph includes multiple initial triples, and the initial triples include a head entity initial vector, a relationship initial vector, and a tail entity initial vector; based on the initial knowledge graph, the initial graph convolutional network is trained to obtain a trained preset graph convolutional network; through the preset graph convolutional network, based on the initial triples, the deep features of each relationship and the deep features of each entity are extracted; based on the deep features of each relationship and the deep features of each entity, the initial knowledge graph is updated to obtain a knowledge graph.

[0163] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: during the kth iteration, the sample deep features of each relationship and the sample deep features of each entity are extracted based on the initial triples through the initial graph convolutional network; multiple sample triplets are determined according to the sample deep features of each relationship and the sample deep features of each entity; the loss function value is calculated according to the sample deep features corresponding to the multiple sample triplets; when the loss function value is greater than the preset loss value, the model parameters of the initial graph convolutional network are adjusted, and the k+1th iteration process is performed until the preset iteration stop condition is met to obtain a trained preset graph convolutional network.

[0164] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps: obtaining a knowledge graph; the knowledge graph comprises a plurality of triples, each triple comprising a head entity vector, a relationship vector, and a tail entity vector; extracting, based on the triples, relational deep features of each relationship and entity deep features of each entity through a preset graph convolutional network; wherein the entity deep features are generated based on the fusion of multiple relationship vectors and entity vectors; determining at least two task triples according to a preset completion task; the preset completion task is used to indicate any two of the head entity, relationship, and tail entity corresponding to the task triple; the task triple comprises any two entity deep features corresponding to the preset completion task, and any one relationship deep feature corresponding to the preset completion task; calculating an objective function value according to the entity deep features and relationship deep features corresponding to each task triple; the objective function value is used to characterize the probability information of the task triple; determining, according to the objective function value, at least one target triple with a probability higher than a preset threshold; and completing the knowledge graph according to the at least one target triple.

[0165] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: the relationship vector and the tail entity vector in each triple are jointly processed through a preset graph convolutional network to obtain a joint feature; based on the relationship weight information, the deep relationship features of each relationship are extracted from the relationship vector of each triple; for each entity, the joint features of the triple corresponding to the entity and the initial features of the entity are fused to obtain the entity deep features of the entity; the initial features are the head entity vector or the tail entity vector corresponding to the entity.

[0166] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: for each entity, based on the importance score, the joint features of the triples corresponding to the entity and the initial features of the entity are fused to obtain the deep entity features of the entity.

[0167] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: obtaining an initial knowledge graph; the initial knowledge graph includes multiple initial triples, and the initial triples include a head entity initial vector, a relationship initial vector, and a tail entity initial vector; based on the initial knowledge graph, the initial graph convolutional network is trained to obtain a trained preset graph convolutional network; through the preset graph convolutional network, based on the initial triples, the deep features of each relationship and the deep features of each entity are extracted; based on the deep features of each relationship and the deep features of each entity, the initial knowledge graph is updated to obtain a knowledge graph.

[0168] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: during the kth iteration, the sample deep features of each relationship and the sample deep features of each entity are extracted based on the initial triples through the initial graph convolutional network; multiple sample triplets are determined according to the sample deep features of each relationship and the sample deep features of each entity; the loss function value is calculated according to the sample deep features corresponding to the multiple sample triplets; when the loss function value is greater than the preset loss value, the model parameters of the initial graph convolutional network are adjusted, and the k+1th iteration process is performed until the preset iteration stop condition is met to obtain a trained preset graph convolutional network.

[0169] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0170] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the memory, database or other media mentioned in each embodiment provided by this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), quantum computing-based data processing logic devices, artificial intelligence (AI) processors, and the like.

[0171] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0172] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A knowledge graph completion method, characterized in that: The method comprises: Obtain a knowledge graph; the knowledge graph includes multiple triples, and the triples include a head entity vector, a relationship vector, and a tail entity vector; By using a preset graph convolutional network, based on the triples, the deep relational features of each relation and the deep entity features of each entity are extracted; wherein the deep entity features are generated based on the fusion of multiple relation vectors and entity vectors; Determine at least two task triples according to a preset completion task; the preset completion task is used to indicate any two of the head entity, relationship, and tail entity corresponding to the task triple; the task triple includes any two entity deep features corresponding to the preset completion task and any one relationship deep feature corresponding to the preset completion task; Calculate the objective function value based on the entity deep features and relationship deep features corresponding to each task triple; the objective function value is used to characterize the probability information of the task triple; Determining, based on the objective function value, at least one target triplet having a probability higher than a preset threshold; The knowledge graph is completed according to the at least one target triple.

2. The method according to claim 1, characterized in that The method of extracting the deep relationship features of each relationship and the deep entity features of each entity based on the triples by using a preset graph convolutional network includes: Through the preset graph convolutional network, the relationship vector and tail entity vector in each triple are jointly processed to obtain the joint feature; According to the relationship weight information, the deep relationship features of each relationship are extracted from the relationship vector of each triple; For each entity, the joint features of the triplet corresponding to the entity and the initial features of the entity are fused to obtain the entity deep features of the entity; the initial features are the head entity vector or the tail entity vector corresponding to the entity.

3. The method according to claim 2, characterized in that For each entity, the joint features of the triples corresponding to the entity and the initial features of the entity are fused to obtain the entity deep features of the entity, including: For each entity, according to the importance score, the joint features of the triples corresponding to the entity and the initial features of the entity are fused to obtain the entity deep features of the entity.

4. The method according to claim 2, characterized in that The relationship weight information includes forward relationship weight information, reverse relationship weight information and self-loop weight information.

5. The method according to claim 1, characterized in that The method further comprises: Obtain an initial knowledge graph; the initial knowledge graph includes multiple initial triples, and the initial triples include a head entity initial vector, a relationship initial vector, and a tail entity initial vector; Based on the initial knowledge graph, an initial graph convolutional network is trained to obtain a trained preset graph convolutional network; Extracting deep features of each relationship and deep features of each entity based on the initial triples through the preset graph convolutional network; Based on the deep features of each relationship and the deep features of each entity, the initial knowledge graph is updated to obtain the knowledge graph.

6. The method according to claim 5, characterized in that The initial graph convolutional network is trained based on the initial knowledge graph to obtain a trained preset graph convolutional network, including: In the kth iteration process, the initial graph convolutional network is used to extract sample deep features of each relationship and sample deep features of each entity based on the initial triples; Determining a plurality of sample triples according to the sample deep features of each relationship and the sample deep features of each entity; Calculating a loss function value according to the deep features of the samples corresponding to the multiple sample triplets; When the loss function value is greater than the preset loss value, the model parameters of the initial graph convolutional network are adjusted, and the k+1th iteration process is performed until the preset iteration stop condition is met to obtain a trained preset graph convolutional network.

7. A knowledge graph completion device, characterized in that: The device comprises: An acquisition module, configured to acquire a knowledge graph; the knowledge graph includes a plurality of triples, each of which includes a head entity vector, a relationship vector, and a tail entity vector; A feature extraction module, configured to extract, based on the triples, the relational deep features of each relationship and the entity deep features of each entity using a preset graph convolutional network; wherein the entity deep features are generated based on the fusion of multiple relation vectors and entity vectors; A task determination module is configured to determine at least two task triples according to a preset completion task; the preset completion task is used to indicate any two of the head entity, relationship, and tail entity corresponding to the task triple; the task triple includes any two entity deep features corresponding to the preset completion task and any one relationship deep feature corresponding to the preset completion task; a calculation module, configured to calculate an objective function value based on the entity deep features and relationship deep features corresponding to each task triple; the objective function value is used to characterize the probability information of the task triple; and based on the objective function value, determine at least one target triple whose probability is higher than a preset threshold; A completion module is used to complete the knowledge graph according to the at least one target triple.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images