A commonsense knowledge graph completion method based on graph attention networks and additional information.

Abstract

This invention discloses a method, system, electronic device, and computer-readable storage medium for completing commonsense knowledge graphs based on graph attention networks and additional information, belonging to the field of knowledge graph technology. The method includes acquiring additional information; encoding entity description information and factual information from the additional information; fusing the encoded entity description information, factual information, and entities and relations in the commonsense knowledge graph to obtain a fused embedding representation; inputting the fused embedding representation into a graph attention network layer to aggregate neighborhood information of entities and obtain entity embedding representations and relation embedding representations; and performing feature decoding on the entity embedding representations and relation embedding representations. This significantly enriches the semantic information of entities and solves the problem that current completion techniques, which rely solely on triples or other additional information, cannot address knowledge graph completion in commonsense knowledge graphs with insufficient relation mining and incomplete information.

Description

Technical Field

[0001] This application relates to the field of knowledge graph technology, and in particular to a method for completing common sense knowledge graphs based on graph attention networks and additional information. Background Technology

[0002] The statements in this section merely refer to the background art relevant to this application and do not necessarily constitute prior art.

[0003] In the process of constructing a knowledge graph, a large amount of knowledge information comes from documents and web page information. There are often deviations in the process of extracting knowledge from documents. These deviations come from two aspects: (1) There is a lot of noise information in the documents, that is, useless information. Its generation may come from the knowledge extraction algorithm itself, or it may be related to the effectiveness of the language itself; (2) The amount of information in the documents is limited and will not cover all knowledge, especially a lot of common knowledge.

[0004] All of the above can lead to incomplete knowledge graphs. In summary, most knowledge graphs suffer from incomplete data and require continuous improvement. Therefore, knowledge graph completion is becoming increasingly important in the construction of knowledge graphs.

[0005] Based on the different modeling methods used, knowledge graph completion techniques can be broadly categorized into three types. Translation-based knowledge graph completion methods map entities and relations to a low-dimensional continuous vector space through knowledge representation learning, obtaining a low-dimensional dense vector representation. Then, the relationships between triples are predicted by calculating the distance between vectors. Another type, tensor decomposition-based knowledge graph completion methods, resolve the ambiguity of entities and relations by representing them as tensors. In addition, neural networks are also used in knowledge graph completion tasks. Neural networks have stronger feature extraction capabilities, and some neural networks have functions such as aggregating neighborhood information, which can better represent entities and relations and improve the efficiency of knowledge graph completion.

[0006] Common sense knowledge graphs contain a large amount of common sense such as relationships between people and geographical information, but the relationships between knowledge entities are not fully explored, and there are some incomplete triples. For example, in the Freebase knowledge graph, about 71% of people do not have birthplace information, and 75% do not have nationality information; in the DBpedia knowledge graph, a large number of people's entities do not have birthplace information, with a proportion as high as 66%, and 58% of scientists do not have corresponding research field information.

[0007] Most current knowledge graph completion methods only represent triples. A few methods enhance semantic information by utilizing descriptive information or other information beyond triples, or by aggregating neighborhood information of target nodes. Furthermore, the same entity or relation can have different meanings in different factual contexts, but these information systems are not utilized. Therefore, given these characteristics of commonsense knowledge graphs, existing knowledge graph completion methods cannot accurately and completely complete them. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this application provides a commonsense knowledge graph completion method, system, electronic device, and computer-readable storage medium based on graph attention networks and additional information. By combining neighborhood triples with additional information, factual information and entity description information are explicitly integrated into the learning of entity vector representations. Furthermore, by aggregating neighborhood information through graph attention networks, the feature representation capabilities of entities and relationships are enhanced. This overcomes the problems of inaccuracy and incompleteness in commonsense knowledge graph completion caused by missing information in the commonsense knowledge graph and insufficient mining of relationships between knowledge entities, thereby improving the accuracy and completeness of commonsense knowledge graph completion.

[0009] Firstly, this application provides a commonsense knowledge graph completion method based on graph attention networks and additional information;

[0010] Common sense knowledge graph completion methods based on graph attention networks and additional information include:

[0011] Obtain additional information about entities in the commonsense knowledge graph, including entity description information, factual information, and neighborhood information of the entity; encode the entity description information and factual information.

[0012] The encoded entity description information, factual information, and entities and relationships in the common sense knowledge graph are fused to obtain a fused embedding representation;

[0013] The attention network layer of the input graph is fused to aggregate neighborhood information of entities and obtain entity embedding representations and relation embedding representations.

[0014] Feature decoding is performed on entity embedding representations and relation embedding representations.

[0015] Furthermore, the specific steps for encoding entity description information are as follows:

[0016] Each word in the entity description information is vectorized using a word vector generation tool to obtain the word vector corresponding to the entity description information.

[0017] Word vectors are input into a Bi-LSTM layer for training to obtain entity description embeddings.

[0018] Furthermore, the specific steps for encoding factual information include:

[0019] The TransE model is used to generate the initial entity embedding and initial relation embedding of the commonsense knowledge graph. The initial entity embedding and initial relation embedding are then input into the Bi-LSTM layer to obtain the forward and backward embedding representations of the facts.

[0020] By fusing the forward and backward embedding representations of facts, and extracting features through a one-dimensional convolutional neural network, the embedded representation of factual information is obtained.

[0021] Furthermore, the specific steps for fusing the encoded entity description information, factual information, and entities and relationships in the knowledge graph to obtain the fused embedded representation include:

[0022] The initial entity embeddings and encoded entity description information and factual information of the common sense knowledge graph are concatenated, and then fused and reduced in dimensionality through a fully connected neural network layer to obtain the entity fusion embedding representation.

[0023] The initial relation embeddings and encoded factual information of the common sense knowledge graph are concatenated, and then fused and reduced in dimensionality through a fully connected neural network layer to obtain the relation fusion embedding representation.

[0024] Furthermore, the entity fusion embedding representation is used as the initial node of the graph attention network. The neighborhood information and relations are weighted and summed to calculate the attention value of each neighborhood entity and then normalize it.

[0025] The initial entity embedding representation is obtained by weighting and averaging the attention values ​​through a multi-head attention mechanism.

[0026] Based on the initial entity fusion embedding representation, a linear transformation is performed on the initial entity embedding representation to obtain the entity embedding representation.

[0027] Furthermore, based on the initial relation fusion embedding representation, a linear transformation is performed on the relation embedding representation to obtain the relation embedding representation.

[0028] Furthermore, the specific steps for feature decoding of the entity embedding representation and relation embedding representation are as follows:

[0029] The entity embedding representation and relation embedding representation are input into the ConvKB model for decoding.

[0030] Secondly, this application provides a commonsense knowledge graph completion system based on graph attention networks and additional information;

[0031] A commonsense knowledge graph completion system based on graph attention networks and additional information includes:

[0032] An additional information encoder is configured to: acquire additional information from the commonsense knowledge graph, including entity description information and factual information, and neighborhood information of entities; and encode the entity description information and factual information.

[0033] The additional information fusion encoder is configured to fuse the encoded entity description information, factual information, and entities and relations in the commonsense knowledge graph to obtain a fused embedding representation;

[0034] The encoder is configured to: take the fused embedding representation of the input graph attention network layer, aggregate the neighborhood information of entities, and obtain entity embedding representations and relation embedding representations;

[0035] The decoder is configured to perform feature decoding on entity embedding representations and relation embedding representations.

[0036] Thirdly, this application provides an electronic device;

[0037] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the processor executes the computer instructions, the computer instructions perform the steps of the above-described common sense knowledge graph completion method based on graph attention networks and additional information.

[0038] Fourthly, this application provides a computer-readable storage medium;

[0039] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the aforementioned common-sense knowledge graph completion method based on graph attention networks and additional information.

[0040] Compared with the prior art, the beneficial effects of this application are:

[0041] This invention fully utilizes entity description information and factual information and integrates them into entities and relations. By allocating neighborhood weights through a graph attention network and aggregating neighborhood information, it greatly enriches the semantic information of entities and relations, improves the utilization of the additional information of triples, and adds different semantic expressions to entities and relations. It solves the problems of current knowledge graph completion methods that only consider triple information or only utilize single additional information without considering context, thereby improving the accuracy and efficiency of common sense knowledge graph completion, perfecting the relationship between entities in the common sense knowledge graph, and enhancing the completeness of the common sense knowledge graph. Attached Figure Description

[0042] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an undue limitation of this application.

[0043] Figure 1 A flowchart illustrating an embodiment of this application;

[0044] Figure 2 This is a schematic diagram of the system architecture provided for an embodiment of this application. Detailed Implementation

[0045] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0046] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations according to this application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. Furthermore, it should be understood that the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or apparatus.

[0047] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0048] Example 1

[0049] First, the meanings of the technical terms related to the technical solution provided in this embodiment will be explained:

[0050] An entity refers to a concrete thing in the objective world, such as a person, a city, a plant, or a commodity. Everything in the world is composed of concrete things; this refers to entities. Entities are the most basic elements in a knowledge graph, and different entities have different relationships.

[0051] Relationship: Describes the objective connections between nodes, such as the capital city describing the relationship between Beijing and China.

[0052] Entity description information: Information that explains what the entity is and what characteristics it has.

[0053] Factual information: Knowledge graphs are composed of triples, and each triple is a factual piece of information.

[0054] Bi-LSTM layer: Bidirectional long short-term memory neural network.

[0055] TransE model: Translating Embeddings for Modeling Multi-relational Data, a classic knowledge graph embedding model that can vectorize entities and relations in triples according to "head entity + relation ≈ tail entity".

[0056] Existing common sense knowledge graph completion methods only consider triple information or only utilize single additional information, without considering context, and are therefore unsuitable for completing common sense knowledge graphs with limited information, insufficient mining of relationships between knowledge entities, and some incomplete triples. Therefore, this application provides a common sense knowledge graph completion method based on graph attention networks and additional information.

[0057] Next, combined Figure 1-2 This embodiment discloses a method for completing a commonsense knowledge graph based on graph attention networks and additional information. The method includes the following steps:

[0058] S1. Obtain entity description information from the commonsense knowledge graph and encode the entity description information; specific steps include:

[0059] S101. Vectorize each word in the sentence of the entity description information using the Glove word vector generation tool; where the word vector dimension is 100, and the output dimension is 100; the entity description information of the head entity is represented as follows: The entity description information of the tail entity is represented as in, The word vectors that make up the entity description information, i∈(1,2,3,...,n).

[0060] S102. Through preprocessing, the average length of sentences in the entity description information is set to 69. Sentences longer than 69 are truncated, and sentences shorter than 69 are padded with 0 to obtain the text sequence.

[0061] S103. Feed the text sequence into the Bi-LSTM layer for training to obtain the entity description embedding representation.

[0062] When sentences are long, modeling from only a single direction may result in information loss. Therefore, Bi-LSTM is used. Bi-LSTM can be viewed as two LSTMs with different directions: one LSTM describes the text from left to right, and the other from right to left, capturing entity description information simultaneously in both directions. The outputs of the two LSTMs are then concatenated to obtain the output of Bi-LSTM at time step t, thus yielding the feature representation of each sentence.

[0063] S2. Obtain factual information and encode it; specific steps include:

[0064] S201. Utilize the TransE model to encode and embed entities and relations in the commonsense knowledge graph, obtaining the entity embedding representation and relation embedding representation for each triple; where the dimension is 300, and the embedding representation of the head entity is h. s The embedding of the relation is represented by r, and the embedding of the tail entity is represented by t. s .

[0065] S202. The embedding representation of the triples is input into the Bi-LSTM layer to capture the sequence information features of the triples. Forward and backward inputs of facts are used to obtain a more comprehensive vector representation of the factual information. Specifically, the initial embedding representation of the triples is input into the Bi-LSTM, and finally, features are extracted through a one-dimensional convolutional neural network to obtain the factual information, calculated as follows:

[0066] f=ReLU(Bi-LSTM(h, r, t)*Ω)

[0067] Where ReLU is the activation function, Ω represents the parameters of the convolution kernel, and f represents the fact information embedding representation output by the fact information encoder.

[0068] S3. Merge the encoded entity description information, factual information, and entities and relationships from the commonsense knowledge graph to obtain a fused embedded representation; specific steps include:

[0069] S301. Concatenate the initial entity embedding representations (head entity embedding representations and tail entity embedding representations) of the commonsense knowledge graph with the entity description embedding representations and factual information embedding representations, and then perform fusion and dimensionality reduction through a fully connected neural network layer to obtain the entity fusion embedding representation. The entity fusion embedding representation is as follows:

[0070] v = concat(h) s h h ,f)·W1

[0071] or

[0072] v = concat(t) s , th ,f)·W1

[0073] Here, concat represents the concatenation operation.

[0074] S302. The initial relation embedding representation and fact information embedding representation of the common sense knowledge graph are concatenated and then fused and reduced in dimensionality through a fully connected neural network layer to obtain the relation fusion embedding representation.

[0075] r = concat(r, f) · W2

[0076] Here, concat represents the concatenation operation.

[0077] S4. Input the graph into the attention network layer with the fused embedding representation, aggregate neighborhood information, obtain entity embedding representations, and update relation embedding representations; specific steps include:

[0078] S401. The entity fusion embedding representation is used as the node input feature set of the graph attention network layer, denoted as v = {v1, v2, ..., v...} n}, target node v i (Head entity) adjacent neighboring nodes v j (Tail entity) and relation form a neighborhood triplet set, and neighborhood nodes are represented by u. j This indicates that in order to add relational information to the neighboring node v j In, through u j =ρv j +(1-ρ)r i During the process of adding relations to the aggregated neighborhood, u j This represents the neighboring nodes after adding relational information, where ρ is the weight coefficient and 0 < ρ < 1. In this embodiment, ρ = 0.5, r i This is a representation of relational fusion embedding.

[0079] S402. First, calculate the attention value for each neighboring entity. The calculation formula is expressed as follows:

[0080] e ij =a(Wv i Wu j )

[0081] Where W is the linear transformation weight matrix, and a is the function for calculating the attention coefficients;

[0082] The specific formula is as follows:

[0083] e ij =LeakyRelu(a T [Wv i ]||[Wu j ])

[0084] Here, || represents the concatenation operation. After the concatenation operation, a linear transformation is performed on the weight vector a, and then a nonlinear activation is performed through the LeakyReLU function, where α is set to 0.2.

[0085] Then normalize it to obtain

[0086]

[0087] S403. Weighted summation of the normalized attention values ​​using a multi-head attention mechanism. The initial entity embedding representation v″={v1″,v2″,…,v1″,v2″,v3″,v4″,v5″,v2″,v2″,v3 ... n "},in,

[0088]

[0089] S404. To prevent the loss of initial entity fusion embedding information during the acquisition of new entity embedding representations, this is addressed by adding the initial entity fusion embedding representation of the entities after linear transformation to the final entity embedding representation, represented as follows:

[0090] H′=HW E +H f

[0091] Where H′ is the entity embedding representation, H f H represents the initial entity embedding representation obtained through the graph attention layer, and H represents the initial entity fusion embedding representation.

[0092] This allows the initial embedding of the linearly transformed entity to retain the previously stored information, resulting in the final entity embedding representation.

[0093] S405. After calculating the graph attention, perform a linear transformation on the relation fusion embedding representation to obtain the relation embedding representation, as shown in the following formula:

[0094] R′=RW R

[0095] Where R′ is the relation embedding representation, W R R is a linear transformation matrix of dimension T×T′, where T represents the dimension of the input relation fusion embedding representation, T′ is the dimension of the output relation embedding representation, and R is the relation fusion embedding representation.

[0096] The distance function is

[0097] d (h,r,t) =||h+rt||2,

[0098] The loss function is defined as

[0099]

[0100] The distance function and loss function are used to train the final entity embedding representation and relation embedding representation.

[0101] Where G'={(h',r,t)|h'∈E\h}∪{(h,r,t')|t'∈E\t}, the encoder operation is completed.

[0102] S5. Input the entity embedding representation and relation embedding representation into the ConvKB model for decoding; where the scoring function is... Then, the loss function is used for calculation, which is expressed as follows:

[0103]

[0104] The training was iterated 3000 times, with the learning rate updated every 500 iterations. The initial learning rate was set to 0.001. Each iteration performed batch training on all training data. The Adam optimizer was used to train by minimizing the loss function. L2 regularization was used to prevent overfitting, and the model dropout was set to 0.3.

[0105] Example 2

[0106] This embodiment discloses a commonsense knowledge graph completion system based on graph attention networks and additional information, including:

[0107] An additional information encoder is configured to: acquire additional information from the commonsense knowledge graph, including entity description information, factual information, and entity neighborhood information; encode the entity description information and factual information; the entity description information is encoded by a text encoder, the factual information is encoded by a fact encoder, the entity neighborhood information is encoded by an encoder, and the encoded entity description information and factual information are fused by an additional information fusion encoder.

[0108] The additional information fusion encoder is configured to fuse the encoded additional information with entities and relationships in the commonsense knowledge graph to obtain a fused embedding representation;

[0109] The encoder is configured to: take the fused embedding representation of the input graph attention network layer, aggregate the neighborhood information of entities, and obtain entity embedding representations and relation embedding representations;

[0110] The decoder is configured to perform feature decoding on entity embedding representations and relation embedding representations.

[0111] It should be noted that the aforementioned additional information encoder, additional information fusion encoder, encoder, and decoder correspond to the steps in Embodiment 1. The examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0112] Example 3

[0113] Embodiment 3 of the present invention provides an electronic device, including a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, they complete the steps of the above-described method for completing a common sense knowledge graph based on graph attention networks and additional information.

[0114] Example 4

[0115] Embodiment 4 of the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the above-described common sense knowledge graph completion method based on graph attention networks and additional information.

[0116] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0117] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0118] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment, whereby a series of operational steps are performed to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0119] The descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0120] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A commonsense knowledge graph completion method based on graph attention networks and additional information, characterized by: include: Obtain additional information from the commonsense knowledge graph, including entity description information, factual information, and neighborhood information of entities; encode the entity description information and factual information. The entity description information, factual information and entities in the common sense knowledge graph are fused together, and the factual information and relationships in the common sense knowledge graph are fused together to obtain a fused embedding representation; The attention network layer of the input graph is fused to aggregate neighborhood information of entities and obtain entity embedding representations and relation embedding representations. Feature decoding of entity embedding representations and relation embedding representations; The specific steps for encoding factual information include: The TransE model is used to generate the initial entity embedding and initial relation embedding of the commonsense knowledge graph. The initial entity embedding and initial relation embedding are then input into the Bi-LSTM layer to obtain the forward and backward embedding representations of the facts. By fusing the forward and backward embedding representations of facts, and extracting features through a one-dimensional convolutional neural network, the embedded representation of factual information is obtained.

2. The commonsense knowledge graph completion method based on graph attention networks and additional information as described in claim 1, characterized in that, The specific steps for encoding entity description information are as follows: Each word in the entity description information is vectorized using a word vector generation tool to obtain the word vector corresponding to the entity description information. Word vectors are input into a Bi-LSTM layer for training to obtain entity description embeddings.

3. The commonsense knowledge graph completion method based on graph attention networks and additional information as described in claim 1, characterized in that, The specific steps of fusing the encoded entity description information, factual information, and entities in the common sense knowledge graph, and fusing the encoded factual information and relationships in the common sense knowledge graph to obtain the fused embedded representation include: The initial entity embeddings and the additional information after encoding of the commonsense knowledge graph are concatenated, and then fused and reduced in dimensionality through a fully connected neural network layer to obtain the entity fusion embedding representation. The initial relation embeddings and encoded factual information of the common sense knowledge graph are concatenated, and then fused and reduced in dimensionality through a fully connected neural network layer to obtain the relation fusion embedding representation.

4. The commonsense knowledge graph completion method based on graph attention networks and additional information as described in claim 1, characterized in that, The entity fusion embedding representation is used as the initial node of the graph attention network. The neighborhood information and relations are weighted and summed to calculate the attention value of each neighborhood entity and then normalize it. The initial entity embedding representation is obtained by weighting and averaging the attention values ​​through a multi-head attention mechanism. Based on the initial entity fusion embedding representation, a linear transformation is performed on the initial entity embedding representation to obtain the entity embedding representation.

5. The commonsense knowledge graph completion method based on graph attention networks and additional information as described in claim 1, characterized in that, A linear transformation is performed on the relation fusion embedding representation to obtain the relation embedding representation.

6. The commonsense knowledge graph completion method based on graph attention networks and additional information as described in claim 1, characterized in that, The specific steps for feature decoding of entity embedding representation and relation embedding representation are as follows: The entity embedding representation and relation embedding representation are input into the ConvKB model for decoding.

7. A commonsense knowledge graph completion system based on graph attention networks and additional information, characterized by: include: An additional information encoder is configured to: acquire additional information, including entity description information and factual information, and the entity's neighborhood information; and encode the entity description information and factual information. The additional information fusion encoder is configured to fuse the encoded entity description information, factual information and entities in the common sense knowledge graph, and fuse the encoded factual information and relationships in the common sense knowledge graph to obtain a fused embedding representation. The encoder is configured to: take the fused embedding representation of the input graph attention network layer, aggregate the neighborhood information of entities, and obtain entity embedding representation and relation embedding representation; The decoder is configured to perform feature decoding on entity embedding representations and relation embedding representations; The specific steps for encoding factual information include: The TransE model is used to generate the initial entity embedding and initial relation embedding of the commonsense knowledge graph. The initial entity embedding and initial relation embedding are then input into the Bi-LSTM layer to obtain the forward and backward embedding representations of the facts. By fusing the forward and backward embedding representations of facts, and extracting features through a one-dimensional convolutional neural network, the embedded representation of factual information is obtained.

8. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the steps described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, perform the steps described in any one of claims 1-6.

Images