Processing method, system, device and medium for knowledge graph completion

Abstract

The application discloses a processing method, system and device for knowledge graph completion and a medium, the processing method is applied to a tourism knowledge graph, the tourism knowledge graph comprises triples, the triples comprise a head entity, a tail entity and a relation entity, and the processing method comprises the following steps: acquiring tourism data, preprocessing the tourism data to obtain first data and second data, the first data comprises triples containing two or more entities, and the remaining data forms the second data; constructing a tourism knowledge graph according to the first data; inputting the triples with missing entities in the first data and the second data into a weighted graph convolutional neural network model, and calculating a link possibility score of the triples from candidate entities in the second data and the missing entities; and supplementing the candidate entity with the highest score into the corresponding triples of the missing entities as a target entity. The method can obtain high-quality entity and relation link possibilities, and obviously improves the knowledge graph completion effect.

Description

TECHNICAL FIELD

[0001] The present application relates to a processing method for a knowledge graph, in particular to a processing method, system, device and medium for knowledge graph completion. BACKGROUND

[0002] With the continuous development of artificial intelligence, knowledge graphs have gradually attracted attention, and knowledge graphs are becoming more and more popular in various downstream tasks, such as intelligent question answering, information retrieval and recommendation, etc. However, due to the large scale of the knowledge graph, it is impossible to manually check and complete all fact triples, so the knowledge graph is usually incomplete, and this incompleteness will lead to a decrease in the performance of various downstream tasks, thereby leading to poor application results. Therefore, the knowledge graph completion task becomes increasingly important, and people are becoming more and more interested in the task of knowledge graph completion. The task of the present application can be described as: given a KG, learning the embedding representation of entities and relationships, automatically adding new knowledge to complete the triple completion task (h, r, t) in the knowledge graph without additional knowledge, that is, given certain head entities and relationships, finding the most likely tail entity corresponding thereto.

[0003] Knowledge graph embedding has always been an active research topic in knowledge graph completion. From the continuous improvement of the initial TransE, TransH, DistMult, etc. to the most advanced technology RotatE, ConvE, etc., a lot of work has been done. In traditional knowledge embedding models, additive models such as TransE and TransH, in which the relationship is interpreted as a simple translation on the hidden entity representation. There are also multiplicative models such as DistMult and ComplEx, which can see their performance better than these simple additive models. In addition to these models, RotatE defines the relationship as a simple rotation, so that the head entity can be rotated in the complex embedding space to match the tail entity, which has been proven to satisfy many useful semantic properties, such as the compositionality of the relationship. However, all these models process each triple independently. Therefore, these methods cannot capture semantically rich neighborhood structure information, resulting in low-quality embeddings. SUMMARY

[0004] The technical problem to be solved by the present application is to overcome the defect that the model in the prior art can only process each triple independently, and provide a processing method, system, device and medium for knowledge graph completion.

[0005] The present application solves the above technical problems by the following technical solutions:

[0006] The first aspect provides a processing method for knowledge graph completion, applied to a tourism knowledge graph, the tourism knowledge graph comprising triples, the triples comprising a head entity, a tail entity and a relation entity, and the processing method comprising:

[0007] obtaining tourism data, preprocessing the tourism data to obtain first data and second data, the first data comprising triples containing two or more entities, and the remaining data forming the second data;

[0008] constructing a tourism knowledge graph according to the first data;

[0009] inputting the triples with missing entities in the first data and the second data into a weighted graph convolutional neural network model, and calculating a link possibility score of the triples of the missing entities from candidate entities in the second data;

[0010] filling the candidate entity with the highest score as a target entity into the triples of the missing entities.

[0011] Preferably, the training step of the weighted graph convolutional neural network model comprises:

[0012] obtaining training data, the training data comprising a tourism-related knowledge graph;

[0013] inputting the training data into a W-GCN model to obtain a graph structure representation model, the graph structure representation model taking a knowledge graph as input and taking node embedding as output;

[0014] inputting the node embedding into a RotatE model to obtain an embedding model, the triple representation model taking node embedding as input and outputting a link possibility score between entities.

[0015] Preferably, the RotatE model is trained using a self-adversarial negative sampling method, and the training step comprises:

[0016] defining a scoring function to obtain positive triples with positive scores and negative triples with negative scores, the function formula being:

[0017]

[0018] wherein {h, r, t} = e iθ = cosθ+isinθ, h is a head entity, t is a tail entity, r is a relation entity, is a Hadamard inner product, and the imaginary unit i is regarded as a rotation in a complex space, in which e represents continuity, i represents rotation, and e i represents continuous rotation;

[0019] sampling the negative triples, and the negative sampling probability calculation formula being:

[0020]

[0021] wherein, a is a sampling probability, (h j , r, t j ) represents a negative sample corresponding to (h i , r i , t i ), h i represents the i-th head entity;

[0022] According to the positive triple and the scoring calculation model of the triple, a loss is calculated, and a sampling probability is added to a loss function to obtain a calculation formula of a sampling loss function:

[0023]

[0024] wherein, σ is a sigmoid function, γ is a margin distance hyperparameter, an Adam algorithm is used to minimize the sampling loss function formula, and a final model is obtained through iterative training.

[0025] Preferably, the parameters of the RotatE model are selected as follows: the dimension of embedding (vector) is set to 128 dimensions, and the batch_size size is 1024.

[0026] The second aspect provides a processing system for knowledge graph completion, which is applied to a tourism knowledge graph, the tourism knowledge graph includes triples, and each triple includes a head entity, a tail entity and a relation entity. The processing system includes:

[0027] A obtaining module is configured to obtain tourism data, preprocess the tourism data to obtain first data and second data, the first data includes triples containing two or more entities, and the remaining data forms the second data.

[0028] A knowledge graph module is configured to construct a tourism knowledge graph according to the first data.

[0029] A weighted graph convolutional neural network module is configured to input the triples with missing entities in the first data and the second data into a weighted graph convolutional neural network model, and calculate a link possibility score of the triples of the missing entities from the candidate entities in the second data.

[0030] An embedding module is configured to supplement the triple of the missing entity with the candidate entity with the highest score as a target entity.

[0031] Preferably, the training step of the weighted graph convolutional neural network model in the weighted graph convolutional neural network module includes:

[0032] Obtaining training data, the training data comprising a tourism-related knowledge graph;

[0033] Inputting the training data into a W-GCN model to obtain a graph structure representation model, the graph structure representation model taking the knowledge graph as input and outputting node embedding.

[0034] Inputting the node embedding into a RotatE model to obtain an embedding model, the triple representation model taking the node embedding as input and outputting a link possibility score between entities.

[0035] Preferably, the RotatE model is trained using a self-adversarial negative sampling method, and the training steps include:

[0036] Defining a scoring function to obtain positive triplets with positive scores and negative triplets with negative scores, the function formula being:

[0037]

[0038] where {h, r, t}=e iθ cosθ+isinθ, h is the head entity, t is the tail entity, r is the relationship entity, is the Hadamard inner product, and the imaginary unit i is regarded as a rotation in the complex space, e i represents a continuous rotation.

[0039] Sampling the negative triplets, and the negative sampling probability calculation formula being:

[0040]

[0041] where α is the sampling probability, (h′ j ,r,t′ j ) represents the negative sample corresponding to (h i ,r i ,t i ), h i represents the i-th head entity.

[0042] According to the scoring calculation model of the positive triplets and the triplets, a loss is calculated, and the negative sampling probability is added to the loss function to obtain the calculation formula of the sampling loss function:

[0043]

[0044] where σ is a sigmoid function, γ is a margin distance hyperparameter, the sampling loss function formula is minimized using an Adam algorithm, and a final model is obtained through iterative training.

[0045] Preferably, the parameters of the RotatE model are selected as follows: the dimension of the embedding is set to 128 dimensions, and the batch_size size is 1024.

[0046] The third aspect provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the processing method for knowledge graph completion as described above when executing the computer program.

[0047] The fourth aspect provides a computer storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement the processing method for knowledge graph completion as described above.

[0048] The positive progress effect of the present application is that the method of constructing a tourism field knowledge graph and performing link prediction based on an existing knowledge graph to realize knowledge graph completion better utilizes the field structure information of the knowledge graph compared to traditional knowledge graph completion methods. In the tourism scenario, the method proposed in the present application jointly learns the field structure and related knowledge of the tourism knowledge graph through W-GCN and RotatE to obtain high-quality entity and relationship embeddings. The effect of knowledge graph completion is significantly improved. The method can analyze existing tourism knowledge through reasoning, discover new knowledge, and continuously improve the knowledge graph, thereby establishing a foundation for subsequent intelligent question answering, information retrieval and recommendation tasks, thereby accelerating the full application of the tourism knowledge graph in the downstream and greatly saving manpower costs. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 A flowchart of the processing method for knowledge graph completion of embodiment 1 of the present application;

[0050] Figure 2 A module schematic diagram of the processing system for knowledge graph completion of embodiment 2 of the present application;

[0051] Figure 3 A hardware structure schematic diagram of the electronic device of embodiment 3 of the present application. DETAILED DESCRIPTION

[0052] The present application will be further described below by way of examples, but the present application is not limited in the scope of the examples.

[0053] Embodiment 1

[0054] The present embodiment provides a processing method for knowledge graph completion, as shown in Figure 1 applied to a tourism knowledge graph, the tourism knowledge graph comprising triples, the triples comprising head entities, tail entities and relationship entities, and the processing method comprising:

[0055] S101, acquire tourism data, preprocess the tourism data to obtain first data and second data, the first data includes triples containing two or more entities, and the remaining data forms the second data;

[0056] S102, construct a tourism knowledge graph according to the first data;

[0057] S103, input the triples with missing entities in the first data and the second data into a weighted graph convolutional neural network model, and calculate the link possibility score of the triples of the missing entities from the candidate entities in the second data;

[0058] S104, the candidate entity with the highest score is supplemented as the target entity into the triples of the missing entities.

[0059] Specifically, in the process of acquiring data related to the tourism field, such as group tour, free tour, private tour, semi-self-help tour and other products, as well as scenic spots, destinations, tags, etc., the data is preprocessed and cleaned to remove part of the invalid data, and the remaining data is arranged in the form of triples, and a knowledge graph is formed through triples.

[0060] Among them, the knowledge graph is a collection of factual information in the form of relationship triples, represented by G=(V,E), where V represents nodes and E represents edges. Each relationship triple can be represented as (h,r,t), where h and t represent entities in the knowledge graph, and r represents the relationship between h and t. For the triples with missing entities in the knowledge graph, the link possibility of the triples containing the missing entities is calculated in the tourism data through the weighted graph convolutional neural network model.

[0061] As an implementable way, the training step of the weighted graph convolutional neural network model in step S103 includes:

[0062] Obtain training data, the training data includes a knowledge graph related to tourism;

[0063] Input the training data into the W-GCN model to obtain a graph structure representation model, the graph structure representation model takes the knowledge graph as input and takes node embedding as output;

[0064] Input the node embedding into the RotatE model to obtain an embedding model, the triple representation model takes the node embedding as input and outputs the link possibility score between entities.

[0065] Specifically, GCN (Graph Convolutional Network) is a type of graph neural network that uses convolutional operations. It belongs to the category of graph neural networks and can be applied to graph embedding. By using each node as the aggregation center node, and for each center node, aggregating the current-layer feature representations of its neighboring nodes, we obtain the next-layer feature representation for that center node. That is:

[0066]

[0067] Where, N i h represents the set of neighboring nodes of node i (including node i itself). l Let g be the vector representation of a node at layer l. g(·,·) represents the information transfer function. Through a summation operation, the linearly transformed vector representations of neighboring nodes are shared with the central node, realizing the aggregation operation in the GCN layer. Forward propagation is achieved through the stacking of GCN layers. Each GCN layer aggregates neighbor information. After the first layer aggregates first-order neighbor information, when the second layer aggregates neighbors, its neighboring nodes already have information about their neighboring nodes; therefore, the second GCN layer aggregates second-order neighbor information. The more GCN layers are stacked, the wider the range of neighbors aggregated by the central node.

[0068] In this embodiment, graph structure representation learning employs W-GCN (Weighted Graph Convolutional Networks), which represents entities by aggregating connected entities specified by relations in the knowledge graph. Different types of relations are weighted during aggregation, and the weights of different relations can be adaptively learned during network training. Through weight learning, W-GCN can control the amount of information from neighboring nodes. The l-th W-GCN layer takes the output vector of each node in the previous layer as input and generates a new representation of the node. Therefore, the forward iteration formula is:

[0069]

[0070] in, Represents the node v at level l. i The vector representation of v. j It is node v i Domain nodes, Represents the node v at level l. i The domain node vector representation, with weights defined as follows: 1≤t≤T, where T is the total number of relations. As a learnable parameter, the g function specifies how to aggregate information from neighborhood nodes, defined as follows:

[0071]

[0072] If the center node and the domain node are separated, the formula is:

[0073]

[0074] The output of the lth layer is a node feature matrix: And is the lth row of H l+1 , which represents the feature of node v j in the l+1th layer.

[0075] The above process can be organized as a matrix multiplication to simultaneously compute the embedding of all nodes through the adjacency matrix. For each type of relationship (edge), the adjacency matrix A t is a binary matrix, and its ijth entry is 1 if there is an edge connecting v i and v j , otherwise 0. Write it in matrix form as follows:

[0076]

[0077] where I is an N x N identity matrix. A l is the weighted sum of subgraph adjacency matrices plus self-connection. In the present application, all first-order neighbors in the linear transformation of each layer are considered. If there are L layers in total in the W-GCN, the output H l+1 of the Lth layer is the final embedding:

[0078] H l+1 = σ (A l H l W l )

[0079] As an implementable way, the RotatE model adopts a self-antagonistic negative sampling method for training, and the training steps include:

[0080] Define the scoring function to obtain positive triplets with positive scores and negative triplets with negative scores, and the function formula is:

[0081]

[0082] Where {h, r, t} = e iθ = cosθ+isinθ, h is the head entity, t is the tail entity, r is the relationship entity, is the Hadamard inner product, and the imaginary unit i is regarded as a rotation in the complex space, and in this identity e represents continuity, i represents rotation, and e i just represents continuous rotation;

[0083] The negative triplet is sampled, and the negative sampling probability is calculated using the following formula:

[0084]

[0085] Where α is the sampling probability, (h′ j ,r,t′ j ) indicates (h i r i , t i The corresponding negative sample, h i Represents the i-th head entity;

[0086] The model loss is calculated based on the positive triplet and the triplet's score, and the subsampling probability is added to the loss function to obtain the calculation formula for the sampling loss function:

[0087]

[0088] Where σ is the sigmoid function and γ is the margin distance hyperparameter, the Adam algorithm is used to minimize the sampling loss function formula, and the final model is obtained through iterative training.

[0089] As one possible approach, the parameters of the RotatE model are selected as follows: the embedding dimension is set to 128 dimensions, and the batch size is 1024.

[0090] The knowledge graph completion method described in this embodiment constructs a tourism domain knowledge graph and performs link prediction based on the existing knowledge graph to achieve knowledge graph completion. Compared with traditional knowledge graph completion methods, it makes better use of the domain structure information of the knowledge graph. In the tourism scenario, W-GCN and RotatE are used to jointly learn the domain structure and related knowledge of the tourism knowledge graph to obtain high-quality entity and relation embeddings. This significantly improves the effect of knowledge graph completion. This method can discover new knowledge by reasoning and analyzing existing tourism knowledge, continuously improve the knowledge graph, and lay the foundation for subsequent tasks such as intelligent question answering, information retrieval, and recommendation, thereby accelerating the full application of tourism knowledge graphs downstream and greatly saving labor costs.

[0091] Example 2

[0092] This embodiment provides a processing system 200 for knowledge graph completion, such as... Figure 2 As shown, this is applied to a tourism knowledge graph, which includes triples, each triple consisting of a head entity, a tail entity, and a relation entity. The processing system includes:

[0093] The acquisition module 201 is used to acquire tourism data, preprocess the tourism data to obtain first data and second data, wherein the first data includes triples containing two or more entities, and the remaining data forms the second data;

[0094] Knowledge graph module 202 is used to construct a tourism knowledge graph based on the first data;

[0095] The weighted graph convolutional neural network module 203 is used to input the triplet of the missing entity in the first data and the second data into the weighted graph convolutional neural network model to calculate the link probability score between the candidate entity in the second data and the triplet of the missing entity.

[0096] The embedding module 204 is used to fill the corresponding triple of the missing entity with the candidate entity with the highest score as the target entity.

[0097] As one possible approach, the training steps of the weighted graph convolutional neural network model in the weighted graph convolutional neural network module include:

[0098] Acquire training data, which includes a tourism-related knowledge graph;

[0099] The training data is input into the W-GCN model to obtain a graph structure representation model, which takes a knowledge graph as input and node embeddings as output.

[0100] The node embeddings are input into the RotatE model to train the embedding model. The triples represent the model taking the node embeddings as input and outputting a link probability score between entities.

[0101] As one possible approach, the RotatE model is trained using a self-adversarial negative sampling method, and the training steps include:

[0102] Define a rating function to obtain positive triples with positive ratings and negative triples with negative ratings. The function formula is:

[0103]

[0104] Where {h,r,t}=e iθ =cosθ + isinθ, where h is the head entity, t is the tail entity, and r is the relation entity. This is the Hadamard inner product, which treats the imaginary unit *i* as a rotation in complex space. In this identity, *e* represents continuity, *i* represents rotation, and *e* represents... i This represents continuous rotation;

[0105] The negative triplet is sampled, and the negative sampling probability is calculated using the following formula:

[0106]

[0107] Where α is the sampling probability, (h′ j ,r,t′ j ) indicates (h i ,r i ,t i The corresponding negative sample, h i Represents the i-th head entity;

[0108] The model loss is calculated based on the positive triplet and the triplet's score, and the subsampling probability is added to the loss function to obtain the calculation formula for the sampling loss function:

[0109]

[0110] Where σ is the sigmoid function and γ is the margin distance hyperparameter, the Adam algorithm is used to minimize the sampling loss function formula, and the final model is obtained through iterative training.

[0111] As one possible approach, the parameters of the RotatE model are selected as follows: the embedding dimension is set to 128 dimensions, and the batch size is 1024.

[0112] The knowledge graph completion processing system described in this embodiment constructs a tourism domain knowledge graph and performs link prediction based on the existing knowledge graph to achieve knowledge graph completion. Compared with traditional knowledge graph completion methods, this method makes better use of the domain structure information of the knowledge graph. In the tourism scenario, W-GCN and RotatE are used to jointly learn the domain structure and related knowledge of the tourism knowledge graph, obtaining high-quality entity and relation embeddings. This significantly improves the effect of knowledge graph completion. This method can discover new knowledge by reasoning and analyzing existing tourism knowledge, continuously improving the knowledge graph and laying the foundation for subsequent tasks such as intelligent question answering, information retrieval, and recommendation. This accelerates the full application of tourism knowledge graphs downstream and greatly saves labor costs.

[0113] Example 3

[0114] like Figure 3 The diagram shown is a structural schematic of an electronic device provided in Embodiment 3 of the present invention. It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the processing method for knowledge graph completion described in Embodiment 1. Figure 3 The electronic device 30 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of the present invention.

[0115] The electronic device 30 may be in the form of a general-purpose computing device, such as a server device. The components of the electronic device 30 may include, but are not limited to: at least one processor 31, at least one memory 32, and a bus 33 connecting different system components (including memory 32 and processor 31).

[0116] Bus 33 includes a data bus, an address bus, and a control bus.

[0117] The memory 32 may include volatile memory, such as random access memory (RAM) 321 and / or cache memory 322, and may further include read-only memory (ROM) 323.

[0118] The memory 32 may also include a program / utility 325 having a set (at least one) of program modules 324, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0119] The processor 31 executes various functional applications and data processing by running computer programs stored in the memory 32, such as the processing method for knowledge graph completion in Embodiment 1 of the present invention.

[0120] Electronic device 30 can also communicate with one or more external devices 34 (e.g., keyboard, pointing device, etc.). This communication can be performed via input / output (I / O) interface 35. Furthermore, the model-generated device 30 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public network, such as the Internet) via network adapter 36. As shown, network adapter 36 communicates with other modules of the model-generated device 30 via bus 33. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the model-generated device 30, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems.

[0121] It should be noted that although several units / modules or sub-units / modules of the electronic device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of the present invention, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.

[0122] Example 4

[0123] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the processing method for knowledge graph completion described in Embodiment 1.

[0124] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0125] In a possible implementation, the present invention can also be implemented as a program product comprising program code, wherein when the program product is run on a terminal device, the program code is used to cause the terminal device to execute the processing method for knowledge graph completion described in Embodiment 1.

[0126] The program code for executing the present invention can be written in any combination of one or more programming languages. The program code can be executed entirely on the user device, partially on the user device, as a standalone software package, partially on the user device and partially on a remote device, or entirely on a remote device.

[0127] While specific embodiments of the present invention have been described above, those skilled in the art should understand that these are merely illustrative examples, and the scope of protection of the present invention is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but all such changes and modifications fall within the scope of protection of the present invention.

Claims

1. A processing method for knowledge graph completion, characterized in that, Applied to a tourism knowledge graph, the tourism knowledge graph includes triples, each triple including a head entity, a tail entity, and a relation entity; the processing method includes: Acquire tourism data, preprocess the tourism data to obtain first data and second data, the first data includes triples containing two or more types of entities, and the remaining data forms the second data; Construct a tourism knowledge graph based on the first data; The triplet of the missing entity in the first data and the second data are input into a weighted graph convolutional neural network model to calculate the link probability score between the candidate entity in the second data and the triplet of the missing entity. The candidate entity with the highest score is used as the target entity to fill the corresponding triple of the missing entity; The training steps of the weighted graph convolutional neural network model include: Acquire training data, which includes a tourism-related knowledge graph; The training data is input into the W-GCN model to obtain a graph structure representation model, which takes a knowledge graph as input and node embeddings as output. The node embeddings are input into the RotatE model to train the embedding model. The embedding model takes the node embeddings as input and outputs a link probability score between entities.

2. A processing system for knowledge graph completion, characterized in that, An application is made to a tourism knowledge graph, wherein the tourism knowledge graph includes triples, each triple comprising a head entity, a tail entity, and a relation entity, and the processing system includes: The acquisition module is used to acquire tourism data, preprocess the tourism data to obtain first data and second data, wherein the first data includes triples containing two or more entities, and the remaining data forms the second data; The knowledge graph module is used to construct a tourism knowledge graph based on the first data; The weighted graph convolutional neural network module is used to input the triples of missing entities in the first data and the second data into the weighted graph convolutional neural network model to calculate the link probability score between the candidate entity in the second data and the triples of the missing entity. An embedding module is used to fill the corresponding triple of the missing entity with the candidate entity with the highest score as the target entity; The training steps of the weighted graph convolutional neural network model in the weighted graph convolutional neural network module include: Acquire training data, which includes a tourism-related knowledge graph; The training data is input into the W-GCN model to obtain a graph structure representation model, which takes a knowledge graph as input and node embeddings as output. The node embeddings are input into the RotatE model to train the embedding model. The embedding model takes the node embeddings as input and outputs a link probability score between entities.

3. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes a computer program, it implements the processing method for knowledge graph completion as described in claim 1.

4. A computer storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the processing method for knowledge graph completion as described in claim 1.

Images