Method and apparatus for predicting edge using correct edge-predicting model based on positive unlabeled data learning

Abstract

To provide a method and apparatus for predicting an edge adapted to correctly predict an edge having a probability to be coupled in an edge-incomplete graph structure.SOLUTION: A method for predicting an edge which is executed by an apparatus for predicting an edge includes a step for inputting an edge-incomplete graph to an edge predicting model to predict an edge having a probability to be coupled in an edge-incomplete graph structure. The edge predicting model is to process at least one edge observed in an edge-incomplete graph structure as positive data and at least one node pair uncoupled in the edge-incomplete graph structure as unlabeled data, thereby carrying out binary classification.SELECTED DRAWING: Figure 3

Description

Technical Field

[0001] The embodiments disclosed in this specification relate to an edge prediction method and apparatus, and more particularly, to a method and apparatus for learning and utilizing an edge prediction model that accurately predicts edges having a probability of being connected in the future in an edge-incomplete graph.

[0002] This research was conducted as a result of the research projects of the Ministry of Science and ICT and the Institute for Information and Communications Technology Planning and Evaluation (IITP) in Korea, namely, the "XVoice: Multimodal Speech Meta-Learning" project (IITP-2022-0-00641) of the Human-Centered Artificial Intelligence Innovation Source Technology Development Project, the "(SW Star Lab) Development of Flexible and Efficient Model Compression Technology for Supporting Various Applications and Environments" project (IITP-2020-0-00894) of the SW Computing Industry Source Technology Development Project, the "Support for Artificial Intelligence Graduate School (Seoul National University)" project (IITP-2021-0-01343) and the "Artificial Intelligence Innovation Hub Research and Development" project (IITP-2021-0-02068) of the Information and Communication Broadcasting Innovation Personnel Training Project.

Background Art

[0003] In daily life, an edge-incomplete graph can be easily encountered. Examples of edge-incomplete graphs include the friendship relationships in a social network and the citation relationships in papers. In a social network, a user becomes a node, and the friendship relationship between users is regarded as an edge. When a user adds a friend in the social network, it is not necessary to check all users, and it can include relationships that are actually friends but not connected as friendship relationships on the social network. In a paper citation network, papers become nodes, and the citation relationships between papers become edges. When a user cites a paper, it is not necessary to check all published papers, so there may be a situation where a paper that should be cited is missed.

[0004] Techniques for predicting edges in existing edge-incomplete graphs have the drawback that specific edges strongly depend on the incomplete graphs. Existing techniques assume that all edges of a specific edge-incomplete graph are observed (fully-observed) and do not consider unobserved missing links when learning unobserved edges. This makes it impossible to transmit information between nodes that can form edges but are not connected on a specific graph, resulting in the edge prediction model overfitting to a specific edge-incomplete graph.

[0005] Therefore, there is a need for an edge prediction technique that takes into account nodes not connected in an edge-incomplete graph.

[0006] For reference, Patent Document 1 is an invention related to an apparatus and method for predicting node characteristics. However, Patent Document 1 only discloses content for predicting nodes of a graph and does not provide an edge prediction technique that takes into account nodes not connected in the graph.

Prior Art Documents

Patent Documents

[0007]

Patent Document 1

Non-Patent Documents

[0008]

Non-Patent Document 1

Non-Patent Document 2

Non-Patent Document 3

Non-Patent Document 4

Summary of the Invention

Problems to be Solved by the Invention

[0009] The embodiments disclosed in this specification use an edge prediction model that processes at least one observed edge in the structure of an edge-incomplete graph as positive data and processes at least one unconnected node pair in the structure of the edge-incomplete graph as unlabeled data, and aims to accurately predict edges having a probability of being connected in the structure of the edge-incomplete graph.

[0010] Other objects and advantages of the present invention will be understandable from the following description and will be more clearly understood by one example. Also, it will be easily understood that the objects and advantages of the present invention can be realized by the means and combinations thereof disclosed in the claims.

Means for Solving the Problems

[0011] As a technical means for achieving the above-described technical problems, an edge prediction method executed by an edge prediction device includes inputting an edge-incomplete graph into an edge prediction model and predicting an edge having a probability of being connected in the structure of the edge-incomplete graph. The edge prediction model is a model that processes at least one observed edge in the structure of the edge-incomplete graph as positive data, processes at least one unconnected node pair in the structure of the edge-incomplete graph as unlabeled data, and performs binary classification.

[0012] According to another embodiment, the edge prediction device includes a memory for storing an edge incomplete graph and an edge prediction model, and a control unit for inputting the edge incomplete graph into the edge prediction model to predict an edge having a probability of being connected in the structure of the edge incomplete graph. The edge prediction model is a model that processes at least one observed edge in the structure of the edge incomplete graph as positive data, and processes at least one unconnected node pair in the structure of the edge incomplete graph as unlabeled data for binary classification.

[0013] According to still another embodiment, the recording medium is a computer-readable recording medium on which a program for executing an edge prediction method is recorded.

[0014] According to still another embodiment, the computer program is a computer program executed by an edge prediction device and stored on a recording medium for executing an edge prediction method.

Advantages of the Invention

[0015] According to any one of the above-described problem-solving means, it is possible to present an edge prediction method and apparatus capable of classifying unconnected node pairs by processing an edge observed in the structure of an edge incomplete graph as positive data and processing an unconnected node pair in the structure of the edge incomplete graph as unlabeled data.

[0016] Also, according to any one of the above-described problem-solving means, it is possible to present an edge prediction method and apparatus that use the structure of an expected edge incomplete graph to which a probability variable is applied between unconnected node pairs in an edge incomplete graph in the learning process of an edge prediction model so that information can be transmitted between unconnected node pairs.

[0017] Also, according to any one of the above-described problem-solving means, it is possible to present an edge prediction method and apparatus that can convert an expected edge incomplete graph into a line graph, assume it with a Markov network, and efficiently calculate the expected value for the probability variables of the expected edge incomplete graph.

[0018] Also, according to any one of the above-described problem-solving means, it is possible to present an edge prediction method and apparatus that can solve the oversmoothing problem and the training time increasing with the number of nodes by approximating the structure of the expected edge incomplete graph.

[0019] Also, according to any one of the above-described problem-solving means, it is possible to present an edge prediction method and apparatus that can efficiently approximate the expected edge incomplete graph by selecting candidate edges based on the degree of nodes in the structure of the expected edge incomplete graph, selecting edges with high weighted values by probability variables among the candidate edges, and removing the remaining edges.

[0020] Also, according to any one of the above-described problem-solving means, it is possible to present an edge prediction method and apparatus that can gradually improve the quality of the edge prediction model and the quality of the expected edge incomplete graph by repeating the process of updating the parameters of the edge prediction model using the probability variables of the expected edge incomplete graph and updating the probability variables of the expected edge incomplete graph using the prediction probability output by the edge prediction model.

[0021] Also, according to any one of the above-described problem-solving means, by considering the edges newly added in the predicted edge incomplete graph during the learning process of the edge prediction model and applying the randomly sampled edges, the balance between the number of edges connected in the structure of the edge incomplete graph and the number of unconnected edges can be achieved, and an edge prediction method and apparatus that can improve the accuracy of the edge prediction model based on the binary loss function can be presented.

[0022] Also, according to any one of the above-described problem-solving means, by considering the edges newly added in the predicted edge incomplete graph during the learning process of the edge prediction model and applying the corrected loss function with the randomly sampled edges, an edge prediction method and apparatus that can prevent excessive self-reinforcement can be presented.

[0023] The effects obtained from the disclosed embodiments are not limited to the effects mentioned above, and other effects not mentioned will be clearly understandable to those with ordinary knowledge in the technical field to which the disclosed embodiments belong.

Brief Description of the Drawings

[0024] Hereinafter, the attached drawings illustrate preferred embodiments disclosed in this specification, and play a role in better understanding the technical idea disclosed in this specification together with the specific content for implementing the invention. Therefore, the content disclosed in this specification should not be interpreted as being limited only to the matters described in those drawings.

[0025]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

DETAILED DESCRIPTION OF THE INVENTION

[0026] Hereinafter, various embodiments will be described in detail with reference to the accompanying drawings. The embodiments described below can also be implemented in various forms. To more clearly explain the features of the embodiments, a detailed description of matters well-known to those having ordinary knowledge in the technical field to which the following embodiments belong will be omitted. And, parts not related to the description of the embodiments in the drawings are omitted, and similar reference numerals are given to similar parts throughout the specification.

[0027] Throughout the specification, when it is said that a certain configuration is "connected" to another configuration, this includes not only the case where it is "directly connected", but also the case where it is "connected with another configuration interposed therebetween". Also, when it is said that a certain configuration "includes" another configuration, this means that, unless otherwise stated to the contrary, it does not exclude further other configurations, and can further include further other configurations.

[0028] Hereinafter, embodiments will be described in detail with reference to the accompanying drawings.

[0029] "Graph" means a data structure that collects nodes and edges (links) that connect those nodes. Each node has a feature vector that indicates the characteristics of that node. Graph-shaped data can be observed in various ways in the real world. For example, in social network services such as Facebook and Twitter, or streaming services such as Wave and Netflix, and shopping sites such as Coupon and Zimmer Market, the relationships between users can be expressed as graph-shaped data. Also, in the field of chemistry, where the characteristics of compounds such as drugs and proteins are grasped and classified, it can be expressed as graph-shaped data. However, in the real world, due to the practical limitation that not all data can be confirmed, there are cases where some edges are missing. Such a graph with missing edges is called an edge-incomplete graph.

[0030] Table 1 below defines the terms used in this specification.

[0031]

Table 1

[0032] JPEG2025113961000003.jpg62159

[0033] This embodiment learns an edge prediction model based on positive-unlabeled data learning, and accurately predicts edges having a probability of being connected in an edge-incomplete graph. Positive-unlabeled data learning is a type of binary classification. While traditional binary classification aims to learn a binary classification model by utilizing positive and negative learning data instances, positive-unlabeled data learning aims to learn a classification model by utilizing positive-unlabeled learning data instances. That is, positive-unlabeled data learning aims to learn a binary classification model when only some are labeled as positive and the rest are given without being labeled during the model learning process. Positive labeling means classifying the degree (identity or similarity) to which the correct answer (true value) and the predicted value match, and negative labeling means classifying the degree (identity or similarity) to which a value that is not the correct answer and the predicted value match. The identity or similarity can be measured based on various distances between values. Since positive-unlabeled data learning does not give learning instances classified as negative during the model learning process, an existing binary classification model cannot be directly applied.

[0034] This embodiment regards the observed edges of a specific edge-incomplete graph as positive data instances, and processes the remaining unconnected node pairs (node pairs that may be connected in the future) as unlabeled data instances, so that they can be utilized when learning whether unconnected node pairs in a specific graph are connected or not. In particular, after introducing a probability variable indicating a connection relationship for unconnected node pairs, this embodiment utilizes the expected value of the graph instead of the specific graph for learning the edge prediction model to improve the accuracy and efficiency of positive-unlabeled data learning.

[0035] FIG. 1 is a block diagram for explaining the functional configuration of an edge prediction apparatus according to an embodiment.

[0036] Referring to FIG. 1, an edge prediction device 100 according to an embodiment may include an input / output unit 110, a memory 120, a control unit 130, and a communication unit 140.

[0037] The input / output unit 110 may include an input unit for receiving an input from a user and an output unit for displaying information such as an execution result of a task or the state of the edge prediction device 100. That is, the input / output unit 110 is a configuration for receiving data and outputting a result obtained by arithmetic processing of the data. The edge prediction device 100 according to the embodiment can receive an edge prediction request or the like via the input / output unit 110.

[0038] The memory 120 is a configuration capable of storing files and programs, and may be composed of various memories. In particular, the memory 120 can store data and programs that enable the control unit 130 described later to execute operations for edge prediction according to an algorithm presented below.

[0039] The memory 120 can store an edge incomplete graph and an edge prediction model. The memory 120 can store an expected edge incomplete graph and an approximated expected edge incomplete graph. The memory 120 can also store the prediction probability output by the edge prediction model.

[0040] The control unit 130 is a configuration including at least one processor such as a CPU or a GPU, and can control the overall operation of the edge prediction device 100. That is, the control unit 130 can control other components included in the edge prediction device 100 to execute operations for edge prediction. By executing the program stored in the memory 120, the control unit 130 can execute operations for predicting an edge from an edge incomplete graph according to an algorithm presented below.

[0041] The communication unit 140 can perform wired or wireless communication with other devices or networks. For example, when a server that provides services of a specific online platform for collecting or processing data included in a graph is implemented by a separate device, the communication unit 140 can receive an edge-incomplete graph through communication with the server that provides the services of the online platform, and can provide an edge or an edge-incomplete graph with complemented edges generated by a prediction probability based on the received edge-incomplete graph to the server or the user's terminal.

[0042] For this purpose, the communication unit 140 can include a communication module that supports at least one of various wired or wireless communication methods, and the communication module can be implemented in the form of a chipset. The mobile communication or wireless communication supported by the communication unit 140 can be, for example, an N-generation mobile communication protocol, Wi-Fi (Wireless Fidelity), Wi-Fi Direct, Bluetooth (registered trademark), UWB (Ultra-Wide Band), or NFC (Near Field Communication).

[0043] The control unit 130 can input the edge-incomplete graph into an edge prediction model and predict an edge having a probability of being connected in the structure of the edge-incomplete graph. A model based on a graph convolutional network can be applied to the edge prediction model, and other learning models capable of graph processing can also be applied.

[0044] The control unit 130 can classify the structure of the graph into binary classification by processing the edges observed in the structure of the edge-incomplete graph as positive data and the node pairs not connected in the structure of the edge-incomplete graph as unlabeled data by the edge prediction model, and can classify the unconnected node pairs.

[0045] The control unit 130 converts an edge-incomplete graph into another graph to improve the edge prediction accuracy and processing efficiency of the edge prediction model.

[0046] FIG. 2 is a diagram illustrating a graph converted by an edge prediction device according to an embodiment. Referring to FIG. 2, the control unit 130 can convert an edge-incomplete graph into an expected edge-incomplete graph. Here, the expected edge-incomplete graph is a graph to which a probability variable indicating the connection state of node pairs not connected in the structure of the edge-incomplete graph is applied.

[0047] The control unit 130 can convert two adjacent edges in the structure of the edge-incomplete graph into a line graph represented by two connected nodes, and calculate the expected value of the probability variable using a Markov network that models the joint probability distribution of the nodes of the converted line graph.

[0048] The control unit 130 can set the number of edges to be maintained within the structure of the expected edge-incomplete graph, and convert it into an approximated expected edge-incomplete graph by approximating the structure of the expected edge-incomplete graph in such a way that node pairs with a connection probability higher than a reference value are excluded and the remaining node pairs are not connected.

[0049] During the process of learning the edge prediction model, the control unit 130 can use the expected edge-incomplete graph (or the approximated expected edge-incomplete graph) to propagate information in the graph convolutional network of the edge prediction model for learning.

[0050] During the process of learning the edge prediction model, the control unit 130 can update the parameters of the edge prediction model using the expected edge-incomplete graph (or the approximated expected edge-incomplete graph) to which a probability variable indicating the connection state of node pairs not connected in the structure of the edge-incomplete graph is applied by the edge prediction model.

[0051] During the process of learning the edge prediction model, the control unit 130 can update the probability variable of the predicted edge incomplete graph (or the approximated predicted edge incomplete graph) using the prediction probability output by the edge prediction model.

[0052] During the process of learning the edge prediction model, the control unit 130 can learn the edge prediction model by using a binary loss function applied to randomly sampled edges in order to balance the number of edges connected and the number of unconnected edges in the structure of the edge incomplete graph, taking into account the edges added in the predicted edge incomplete graph (or the approximated predicted edge incomplete graph).

[0053] During the process of learning the edge prediction model, the control unit 130 can learn the edge prediction model by using a correction loss function that prevents excessive self-reinforcement based on randomly sampled edges, taking into account the edges added in the predicted edge incomplete graph (or the approximated predicted edge incomplete graph).

[0054] The control unit 130 can predict edges by utilizing the edge prediction model optimized to minimize the binary loss function and the correction loss function.

[0055] FIG. 3 is a flowchart illustrating the operation of the edge prediction apparatus learning the edge prediction model according to an embodiment.

[0056] JPEG2025113961000004.jpg185170

[0057]

Number

[0058]

Number

[0059] JPEG2025113961000007.jpg194170

[0060] By using a random variable of a predicted edge incomplete graph (or an approximated predicted edge incomplete graph) to update the parameters of the edge prediction model, and repeating the process of updating the random variable of the predicted edge incomplete graph (or the approximated predicted edge incomplete graph) by using the prediction probability output by the edge prediction model, the quality of the edge prediction model and the quality of the predicted edge incomplete graph (or the approximated predicted edge incomplete graph) can be gradually improved mutually.

[0061] The edge prediction device is an edge prediction model f θ (i, j) learns to minimize the sum of two loss functions. The binary loss function for positive data and unlabeled data is a loss function that samples edges to balance the imbalance of the number of edges in the actual graph, and can be expressed as in Mathematical Formula 3. In order to solve the problem that if the current parameters of the edge prediction model are not accurate, the quality of the predicted edge incomplete graph will decrease and the parameters will become inaccurate in the next iteration step, the correction loss function that measures the binary cross entropy for edges can be expressed as in Mathematical Formula 4. The sum of the binary loss function and the correction loss function can be expressed as in Mathematical Formula 5.

[0062]

Number

[0063]

Number

[0064]

Number

[0065] The edge prediction device can improve the accuracy of the edge prediction model based on the binary loss function by considering the edges newly added to the predicted edge incomplete graph in the learning process of the edge prediction model and applying the randomly sampled edges, so as to balance the number of connected edges and the number of unconnected edges in the structure of the edge incomplete graph.

[0066] The edge prediction device can improve the accuracy of the edge prediction model by considering the edges newly added to the predicted edge incomplete graph in the learning process of the edge prediction model and applying the corrected loss function with randomly sampled edges to prevent excessive self-reinforcement.

[0067] In step S350, the edge prediction device determines whether the learning end condition is satisfied. When the model converges or reaches the maximum number of iterations, the model learning is interrupted. If the learning end condition is not satisfied, steps S320 to S340 are repeated. If the learning end condition is satisfied, in step S360, the learned edge prediction model is output.

[0068] The algorithm for edge prediction described with reference to FIG. 3 can be expressed in pseudo code as shown in Table 2.

[0069]

Table 2

[0070] Such an algorithm for edge prediction can be called PULL (PU-Learning-based Link predictor).

[0071] FIG. 4 is a flowchart illustrating an edge prediction method according to an embodiment.

[0072] The edge prediction method according to the embodiment shown in FIG. 4 includes the steps that are processed in time series by the edge prediction device shown in FIGS. 1 to 3. Therefore, even if it is said to be the content omitted below, the content described above about the edge prediction device shown in FIGS. 1 to 3 can also be applied to the edge prediction method according to the embodiment shown in FIG. 4.

[0073] Referring to FIG. 4, in step S410, the edge prediction device collects an edge incomplete graph.

[0074] In step S420, the edge prediction device inputs the edge incomplete graph into the edge prediction model and predicts an edge having a probability of being connected in the structure of the edge incomplete graph.

[0075] The edge prediction model applied to the edge prediction method may be a model that processes the edges observed in the structure of the edge incomplete graph as positive data, processes the node pairs not connected in the structure of the edge incomplete graph as unlabeled data, and performs binary classification.

[0076] The edge prediction model applied to the edge prediction method may be a model in which the parameters of the edge prediction model are updated by using an expected edge incomplete graph to which a probability variable indicating the connection state of node pairs not connected in the structure of the edge incomplete graph is applied.

[0077] The edge prediction model applied to the edge prediction method may be a model in which the expected value of the probability variable is calculated by converting the edge incomplete graph into a line graph represented by two nodes connecting two adjacent edges in the structure of the edge incomplete graph and using a Markov network that models the joint probability distribution of the nodes of the converted line graph.

[0078] The edge prediction model applied to the edge prediction method may be a model in which the number of edges to be maintained in the structure of the expected edge incomplete graph is set, and the structure of the expected edge incomplete graph is approximated in a manner of not connecting the remaining node pairs excluding the node pairs whose connection probability is higher than the reference value.

[0079] The edge prediction model applied to the edge prediction method can be a model learned by propagating information in the graph convolutional network of the edge prediction model using an expected edge incomplete graph.

[0080] The edge prediction model applied to the edge prediction method can be a model in which the probability variable of the expected edge incomplete graph is updated by using the prediction probability output by the edge prediction model.

[0081] The edge prediction model applied to the edge prediction method can be a model learned by a binary loss function to which randomly sampled edges are applied to take an equilibrium between the number of edges connected by the structure of the edge incomplete graph and the number of unconnected edges in consideration of the edges added in the expected edge incomplete graph.

[0082] The edge prediction model applied to the edge prediction method can be a model learned by a correction loss function that prevents excessive self-reinforcement based on randomly sampled edges in consideration of the edges added in the expected edge incomplete graph.

[0083] Figures 5 to 7 are diagrams illustrating the edge prediction performance simulated by an embodiment.

[0084] As a result of comparing the existing model with the edge prediction problem performance using a total of five real-life graph data (PubMed, Cora-full, Chameleon, Crocodile, Facebook), PULL of this embodiment showed the highest performance for the performance indicators AUROC (Area Under ROC curve) and AUPRC (Area Under Precision-Recall Curve).

[0085] Referring to Fig. 5 showing the AUROC of PULL by the repetition stage, the dotted line is the actual number of edges, and it can be confirmed that the performance of PULL improves as the repetition stage progresses. That is, PULL improves the quality of the prediction graph as the repetition progresses, and ultimately shows accurate prediction. When the sampled number of edges k for PubMed, Cora-full, and Chameleon exceeds the actual number of edges, the accuracy converges or slightly decreases. This is due to the smoothing problem caused by propagating information through a graph with more edges than the actual graph. Even when the sampled number of edges k for Crocodile and Facebook is more than the actual number of edges, the prediction accuracy improves. This indicates that the actual graph structures of Crocodile and Facebook essentially contain missing edges.

[0086] Referring to Fig. 6 showing the influence of the correction loss function on the edge prediction performance of PULL, PULL-L C represents PULL without the application of the correction loss function L C , and it can be confirmed that PULL shows consistently better performance than PULL-L C . This means that when the predicted graph structure contains a larger number of edges than the actual graph, the correction loss function L C effectively prevents the performance degradation of PULL.

[0087] Referring to Fig. 7 showing the processing time of PULL for the sampled subgraph, it can be confirmed that the time increases linearly as the size increases. This shows the scalability of PULL to large graphs.

[0088] The term "~ part" used in the above embodiments means a software or a hardware component such as an FPGA (field programmable gate array) or an ASIC, and the "~ part" serves a certain role. However, the "~ part" is not meant to be limited to software or hardware. The "~ part" can also be configured to be in an addressable storage medium or configured to cause one or more processors to execute. Thus, as an example, the "~ part" includes components such as software components, object-oriented software components, class components, and task components, and processes, functions, attributes, procedures, subroutines, segments of program patent code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables.

[0089] The components and functions provided within the "~ part" can be combined with a smaller number of components and "~ parts" or separated from additional components and "~ parts".

[0090] Moreover, the components and the "~ part" can also be embodied to cause one or more CPUs within a device or a security multimedia card to execute.

[0091] On the one hand, the edge prediction method according to an embodiment described herein can also be embodied in the form of a computer-readable medium that stores computer-executable instructions and data. Here, the instructions and data can be stored in the form of program code, and when executed by a processor, can generate a predetermined program module and execute a predetermined operation. Further, the computer-readable medium may be any available medium accessible by a computer, including both volatile and non-volatile media, and both removable and non-removable media. Also, the computer-readable medium may be a computer recording medium. The computer recording medium can include any volatile and non-volatile, removable and non-removable media embodied by any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. For example, the computer recording medium may be a magnetic storage medium such as an HDD and an SSD, an optical recording medium such as a CD, a DVD, and a Blu-ray disc, or a memory included in a server accessible via a network.

[0092] Also, the edge prediction method according to an embodiment described herein can also be embodied in a computer program (or computer program product) that includes computer-executable instructions. The computer program includes programmable machine instructions processed by a processor and can be embodied in a high-level programming language, an object-oriented programming language, an assembly language, or a machine language. Also, the computer program can be recorded on a type of computer-readable recording medium (for example, a memory, a hard disk, a magnetic / optical medium, or an SSD (Solid-State Drive), etc.).

[0093] Therefore, the edge prediction method according to an embodiment described in this specification can be implemented by a computer program as described above being executed by a computing device. The computing device can include at least a part of a processor, a memory, a storage device, a high-speed interface connected to the memory and a high-speed expansion port, and a low-speed interface connected to a low-speed bus and the storage device. Each of such components is connected to each other using various buses and can be mounted on a common motherboard or attached in other appropriate ways.

[0094] Here, the processor can process instruction words within the computing device. Such instruction words can have, for example, instruction words stored in the memory or the storage device for displaying graphic information for providing a GUI (Graphic User Interface) on an external input and output device like a display connected to the high-speed interface. As another example, multiple processors and / or multiple buses can be appropriately used together with multiple memories and memory forms. Also, the processor can be implemented as a chipset consisting of chips including multiple independent analog and / or digital processors.

[0095] Also, the memory stores information within the computing device. As an example, the memory can be composed of volatile memory units or a set thereof. As another example, the memory can be composed of non-volatile memory units or a set thereof. Also, the memory can be other forms of computer-readable media such as, for example, magnetic or optical disks.

[0096] The memory device can provide a large-capacity storage space for the computing device. The memory device may be a computer-readable medium or a configuration including such a medium. For example, it can also include devices within a SAN (Storage Area Network) or other configurations, such as a floppy disk device, a hard disk device, an optical disk device, or a tape device, a flash memory, or other semiconductor memory devices or device arrays similar thereto.

[0097] The above-described embodiments are for illustrative purposes. Those having ordinary knowledge in the technical field to which the above-described embodiments belong will be able to understand that they can be easily deformed into other specific forms without changing the technical idea or essential features of the above-described embodiments. Therefore, it must be understood that the foregoing embodiments are illustrative in all respects and not restrictive. For example, each component described as a single type can also be implemented in a distributed manner, and similarly, the components described as being distributed can also be implemented in a combined form.

[0098] The scope to be protected by this specification is determined by the claims described below rather than the foregoing detailed description, and all changes or deformations derived from the meaning and scope of the claims and their equivalent concepts should be construed as being included in the scope of the present invention.

Description of Reference Numerals

[0099] 100 Edge prediction device 110 Input / output unit 120 Memory 130 Control unit 140 Communication unit

Claims

1. An edge prediction method executed by an edge prediction device, comprising: inputting an edge-incomplete graph into an edge prediction model and predicting an edge having a probability of being connected in the structure of the edge-incomplete graph; The edge prediction model is a model that processes at least one observed edge in the structure of the edge-incomplete graph as positive data and processes at least one unconnected node pair in the structure of the edge-incomplete graph as unlabeled data for binary classification, the edge prediction method.

2. The edge prediction model is a model in which the parameters of the edge prediction model are updated by using an expected edge-incomplete graph to which a probability variable indicating the connection state of unconnected node pairs in the structure of the edge-incomplete graph is applied, according to the edge prediction method of claim 1.

3. The edge prediction model is a model in which the expected value of the probability variable is calculated by using a Markov network that converts the edge-incomplete graph into a line graph represented by two nodes connecting two adjacent edges in the structure of the edge-incomplete graph and models the joint probability distribution of the nodes of the converted line graph, according to the edge prediction method of claim 2.

4. The edge prediction model is a model in which the structure of the expected edge-incomplete graph is approximated by setting the number of edges to be maintained in the structure of the expected edge-incomplete graph and connecting the remaining node pairs excluding node pairs whose connection probability is higher than a reference value in a non-connecting manner, according to the edge prediction method of claim 2.

5. The edge prediction model is a model learned by propagating information in the graph convolutional network of the edge prediction model by using the expected edge-incomplete graph, according to the edge prediction method of claim 2.

6. The edge prediction model is a model in which the probability variable of the expected edge-incomplete graph is updated by using the prediction probability output by the edge prediction model, according to the edge prediction method of claim 2.

7. The edge prediction model is (i) a binary loss function that applies randomly sampled edges to balance the number of connected edges and the number of unconnected edges in the structure of the edge-incomplete graph in consideration of the edges added in the expected edge-incomplete graph, and (ii)The edge prediction method according to claim 2, which is a model learned by a correction loss function that prevents excessive self-reinforcement based on the randomly sampled edges, taking into account the edges added to the predicted edge incomplete graph.

8. A memory for storing an edge incomplete graph and an edge prediction model, a control unit that inputs the edge incomplete graph into the edge prediction model to predict an edge having a probability of being connected in the structure of the edge incomplete graph, wherein the edge prediction model is an edge prediction device that processes at least one observed edge in the structure of the edge incomplete graph as positive data, and processes at least one unconnected node pair in the structure of the edge incomplete graph as unlabeled data for binary classification.

9. A computer-readable recording medium on which a program for executing the method according to claim 1 is recorded.

10. A computer program stored on a recording medium for executing the method according to claim 1, which is executed by an edge prediction device.

Images