Method and device for training target estimation ai model

Abstract

According to an embodiment of the present disclosure, provided are a method and electronic device for training a target estimation artificial intelligence (AI) model. The method may comprise the steps of: identifying a target node corresponding to a target estimation AI model; identifying a node related to the target node on the basis of edge weights and edge types in a knowledge graph; selecting a feature of data on the basis of the node related to the target node; and training the target estimation AI model so as to output target information by using selected data including the selected feature.

Description

Method and device for training a target estimation AI model

[0001] The present invention relates to a method and apparatus for training a target estimation AI model, and more specifically, to a method for improving the processing efficiency of an AI model by reducing the dimensionality of data using a knowledge graph.

[0002] The curse of dimensionality refers to a phenomenon where a model's performance actually deteriorates as the dimensionality of the input data increases. As the dimensionality of the data increases, data sparsity may increase, the amount of data required for training may increase rapidly, and the complexity of the model may increase. This can lead to the model overfitting or make training difficult due to inefficiently large computational loads.

[0003] To prevent this, only significant dimensions from the input data can be selected and utilized for training. These selected dimensions can be referred to as features. Through feature selection, it is possible to reduce data dimensions and avoid unnecessary calculations, thereby creating a more efficient model. Feature selection may be necessary when the distribution and quantity of data are limited, as using the original data can lead to degraded model performance or bias in analysis results.

[0004] Efficiency in terms of manpower, infrastructure, and time can be increased when using data corresponding to selected features, compared to building AI models or analyzing data by utilizing all of unrefined, large-scale tabular data. In reality, data is often limited in quantity and variation by dimension, and since training resources are limited, there is an increasing need to efficiently select influential features.

[0005] A method disclosed as a technical means for achieving a technical task may include the steps of: identifying a target node corresponding to a target estimation AI model; identifying a node associated with the target node based on edge weights and edge types in a knowledge graph; selecting features of data based on the nodes associated with the target node; and training a target estimation AI model to output target information using selected data including the selected features.

[0006] A computer-readable recording medium disclosed as a technical means for achieving a technical task may store a program for executing at least one of the embodiments of the disclosed method on a computer.

[0007] A computer program disclosed as a technical means for achieving a technical task may be stored on a recording medium to execute at least one of the embodiments of the disclosed method on a computer.

[0008] An electronic device disclosed as a technical means for achieving a technical task may include a memory in which a program or at least one instruction is stored and at least one processor. By executing the program or at least one instruction stored in the memory by the at least one processor, the electronic device may identify a target node corresponding to a target estimation AI model; identify nodes associated with the target node based on edge weights and edge types in a knowledge graph; select features of data based on nodes associated with the target node; and train a target estimation AI model to output target information using selected data including the selected features.

[0009] FIG. 1 is a drawing for explaining a method of utilizing a knowledge graph according to an embodiment of the present disclosure.

[0010] FIG. 2 is a drawing for explaining modules included in an electronic device according to one embodiment of the present disclosure.

[0011] FIG. 3 is a drawing for explaining a hardware configuration included in an electronic device according to one embodiment of the present disclosure.

[0012] FIG. 4 is a drawing for explaining a method of constructing a knowledge graph according to an embodiment of the present disclosure.

[0013] FIGS. 5 and 6 are drawings for explaining a method of constructing a knowledge graph according to an embodiment of the present disclosure.

[0014] FIG. 7 is a diagram illustrating a method for obtaining selection data using a knowledge graph according to an embodiment of the present disclosure.

[0015] FIG. 8 is a flowchart relating to a method for training a target estimation AI model according to an embodiment of the present disclosure.

[0016] The terms used in this specification will be briefly explained, and the invention will be described in detail.

[0017] The terms used in this specification have been selected to be as widely used as possible, taking into account the functions in the embodiments disclosed herein; however, these terms may vary depending on the intent of those skilled in the art, case law, the emergence of new technologies, etc. Additionally, in specific cases, terms have been arbitrarily selected by the applicant, and in such cases, their meanings will be described in detail in the description section of the relevant embodiments. Therefore, terms used in this disclosure should be defined not merely by their names, but based on their meanings and the overall content of this disclosure.

[0018] A singular expression may include a plural expression unless the context clearly indicates otherwise. Additionally, the expression “at least one of a, b, and c” described throughout the specification may include ‘a alone,’ ‘b alone,’ ‘c alone,’ ‘a and b,’ ‘a and c,’ ‘b and c,’ or ‘a, b, and c all.’

[0019] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art to which the embodiments pertain. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an ideal or overly formal sense unless explicitly defined in this application.

[0020] In this case, any wording mentioned in different embodiments may be used interchangeably, combined, or substituted if the concepts correspond. For example, regarding the same or corresponding concepts, even if the expression 'A' is used in one embodiment and the expression 'B' is used in another embodiment, they may be understood by interchangeably, substituted, or combined.

[0021] Additionally, terms including ordinal numbers, such as 'first' or 'second' as used herein, may be used to describe various components, but said components should not be limited by said terms. Such terms are used solely for the purpose of distinguishing one component from another. For example, the first component may be named the second component, and similarly, the second component may be named the first component.

[0022] When it is stated that a component is "connected" to another component, it should be understood that it may be directly connected to or coupled with that other component, or that there may be other components in between.

[0023] In this specification, terms such as “comprising” or “having” are intended to specify the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should not be understood as precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0024] When a part of a specification is described as "including" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components.

[0025] In one embodiment of the present disclosure, the terms “~ module” or “~ part” used may refer to software or hardware components such as a Field Programmable Gate Array (FPGA) or an Application Specific Integrated Circuit (ASIC), and the “~ part” may perform a specific role. Meanwhile, the meaning of “~ module” or “~ part” is not limited to software or hardware. The “~ module” or “~ part” may be configured to reside in an addressable storage medium or may be configured to run one or more processors. In one embodiment, the “~ module” or “~ part” may include components such as software components, object-oriented software components, class components, and task components, as well as processes, functions, attributes, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. Functions provided through a specific component or a specific “~ module” or “~ part” may be combined or separated into additional components to reduce their number. Additionally, in one embodiment, the '~ module' or '~ part' may include one or more processors.

[0026] Additionally, embodiments of the present disclosure may be represented by functional block configurations and various processing steps. These functional blocks may be implemented by various numbers of hardware and / or software configurations that execute specific functions. For example, embodiments of the present disclosure may employ direct circuit configurations such as memory, processing, logic, etc., which can execute various functions under the control of one or more microprocessors or other control devices.

[0027] Each block of the process flow diagrams attached to this specification and combinations of the flow diagrams may be executed by computer program instructions. Since these computer program instructions may be loaded into the processor of a general-purpose computer, a computer for special purposes, or other programmable data processing equipment, the instructions executed through the processor of the computer or other programmable data processing equipment create means for performing the functions described in the flow diagram block(s).

[0028] These computer program instructions may be stored in computer-available or computer-readable memory that can be directed toward a computer or other programmable data processing equipment to implement a function in a specific way, and the instructions stored in said computer-available or computer-readable memory may also produce a manufactured item containing instruction means that performs the function described in the flowchart block(s).

[0029] Since computer program instructions can be loaded onto a computer or other programmable data processing equipment, instructions that perform a series of operation steps on the computer or other programmable data processing equipment to create a process executed by the computer can also provide steps for executing the functions described in the flowchart block(s).

[0030] Additionally, each block may represent a module, segment, or part of code containing one or more executable instructions for executing a specified logical function(s). Furthermore, in some alternative execution examples, the functions mentioned in the blocks may occur out of order. For instance, two blocks described in succession may actually be executed substantially simultaneously, or the blocks may be executed in reverse order according to their corresponding functions. Alternatively, at least some of the blocks may be omitted during execution.

[0031] Additionally, any / any function or operation described in this disclosure may be processed by a single processor or a combination of processors. The single processor or combination of processors may be a circuitry that performs processing, and may include an application processor (AP, e.g., a central processing unit (CPU)), a communication processor (CP, e.g., a modem), a graphics processing unit (GPU), a neural network processing unit (NPU) (e.g., an artificial intelligence (AI) chip), a Wi-Fi chip, a Bluetooth® chip, a global positioning system (GPS) chip, a near-field communication (NFC) chip, a connectivity chip, a sensor controller, a touch controller, a fingerprint sensor controller, a display driver integrated circuit (IC), an audio codec (CODEC) chip, a universal serial bus (USB) controller, a camera controller, an image processing IC, a microprocessor, a microcontroller, a digital signal processor, an FPGA, an ASIC, a microprocessor unit (MPU), a system-on-chip (SoC), an IC, or similar circuitry. The single processor or combination of processors described above can control the overall operation of an electronic device by executing instructions, such as an operating system, that can be stored in memory. Additionally, the processor or combination of processors can execute other processes or programs residing in memory (e.g., processes related to the present disclosure).

[0032] In the following disclosure, determining the priority between A and B may be referred to in various ways, such as selecting the one with the higher priority according to a predetermined priority rule and performing the corresponding action, or omitting or dropping the action for the one with the lower priority.

[0033] Hereinafter, 'A or B' as described in the present disclosure may be understood as 'A and / or B', which may be understood as including 'A', or 'B', or 'A and B'.

[0034] Additionally, 'at least one of A, B, and C' described in the present disclosure may be understood to include 'A', or 'B', or 'C', or 'any combination of A, B, and C'.

[0035] Additionally, 'at least one of A, B, or C' described in the present disclosure may be understood to include 'A', or 'B', or 'C', or 'any combination of A, B, and C'.

[0036] Additionally, 'A / B' as described in the present disclosure may be understood as 'A and / or B', which may be understood as including 'A', or 'B', or 'A and B'.

[0037] Additionally, 'A, B' described in the present disclosure may be understood as 'A and / or B', which may be understood as including 'A', or 'B', or 'A and B'.

[0038] Additionally, 'A and B' described in the present disclosure may be understood as 'A and / or B', which may be understood as including 'A', or 'B', or 'A and B'.

[0039] Furthermore, the phrase "when conditions A and B are satisfied" as described in the present disclosure is not necessarily limited to cases where both conditions A and B are satisfied, but may be understood to include cases where either condition A or condition B is satisfied individually, cases where both conditions A and B are satisfied, or cases where one or more additional conditions are satisfied together.

[0040] Furthermore, throughout this disclosure, ordinal terms (and similar modifiers) such as 'first', 'second', 'third', etc. are used solely for the purpose of distinguishing various instances, occurrences, configurations, messages, stages, elements, or aspects of elements, operations, or information, as described below. Unless clearly required otherwise by the context, the use of such ordinal terms does not require that the elements, operations, or information distinguished by such terms be structurally different, numerically distinct, or substantially different. For example, 'first signal' and 'second signal' may represent instances of the same signal transmitted at different times, signals containing the same core information even with some variations, or signals having different content or characteristics depending on the specific context. Similarly, 'first value' and 'second value' may represent the same size measured or applied in different situations, or they may represent different sizes. Such interpretation must be determined based on the specific technical context, function, and relationship described in the relevant parts of the disclosure and claims.

[0041] Furthermore, although terms such as "first," "second," etc., as used in this disclosure are used for various elements such as information, objects, actions, and sequences, they are not intended to limit such elements to a specific order. These terms may be understood merely as distinguishing one element from another. For example, a first element may be referred to as a second element, and likewise, a second element may be referred to as a first element.

[0042] Additionally, the terms 'first' and 'second' described in this disclosure may be understood to refer to identical or different elements. For example, if an element is information, the first information and the second information may both be information, and depending on the case, they may be the same information or different information.

[0043] Furthermore, expressions such as "if" and "in case that" as described in the present disclosure or claims may be interpreted, depending on the context, as meaning "when or upon," "in response to," "based on," or "according to," and these expressions may be used interchangeably. In addition, other expressions having substantially the same meaning may be used as substitutes for these expressions, provided that they do not impair the technical features of the present disclosure. Furthermore, if a method step (e.g., a step of transmitting a signal) is performed in relation to such terms (e.g., "in case that" or similar expressions) in accordance with the disclosure of the present specification, this may be interpreted as the method step being performed in response to a prior determination that a specific element has a specific state (e.g., bit length exceeding X).

[0044] Additionally, the term "not perform" as used in this disclosure or claims may be understood, depending on the context, to mean to omit or skip the corresponding step. Such a term may be replaced with other terms having the same or substantially similar meaning.

[0045] The drawings or flowcharts described in this disclosure illustrate exemplary methods that may be implemented according to the principles of this disclosure, and various modifications may be made to the methods illustrated in the flowcharts of this disclosure. For example, although illustrated as a series of steps, the various steps of each drawing or flowchart may overlap, occur in parallel, occur in a different order, or occur multiple times. In other examples, any step may be omitted or replaced with another step.

[0046] Additionally, the process of the flowchart can be performed by an electronic device, and one or more steps of the flowchart can be implemented by one or more processors that execute instructions to perform specific functions.

[0047] The methods and apparatus proposed in the embodiments of the present disclosure may be disclosed together with drawings including flowcharts to illustrate exemplary methods that may be implemented according to the principles of the present disclosure. Such flowcharts may include different branches and / or sub-branches. It should be understood that the principles of the present disclosure are not limited to combinations of all branches and sub-branches disclosed in the embodiments, and may consist of at least one individual branch or individual sub-branch, in particular only a single branch or a single sub-branch.

[0048] The methods and devices proposed in the embodiments of the present disclosure below are not limited to each embodiment and may also be utilized as a combination of all or some embodiments of one or more embodiments proposed in the disclosure.

[0049] Terms used in the following description to identify connection nodes, terms referring to network entities, terms referring to messages, terms referring to interfaces between network entities, terms referring to various identification information, etc., are examples provided for the convenience of explanation. Accordingly, the present disclosure is not limited to the terms described below, and other terms referring to objects having equivalent technical meanings may be used. Furthermore, where appropriate, such terms may be replaced with terms defined in similar technical specifications of standardization organizations such as 3GPP (3rd generation partnership project) Technical Specifications (TS) or ETSI (European Telecommunications Standards Institute).

[0050] Embodiments of the present disclosure are described below with reference to the attached drawings so that those skilled in the art can easily implement them. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein. Furthermore, in order to clearly explain the present disclosure in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification are denoted by similar reference numerals.

[0051] Specific structural or functional descriptions of the embodiments are disclosed for illustrative purposes only and may be modified and implemented in various forms. Accordingly, the embodiments are not limited to the specific disclosed forms, and the scope of this specification includes modifications, equivalents, or substitutions that fall within the technical concept.

[0052] A guide document of the present disclosure refers to a document containing information regarding the relationship between entities. "Entity" may refer to at least one of an object, an attribute, a variable, a class, or a type. A guide document may include a data sheet, a document providing procedures and guidelines necessary to use a product or technology or to perform a specific task, etc. Instead of "guide document," terms such as "data sheet," "technical specification document," "standard specification document," "paper," "patent specification," "report," "policy document," "manual," "news," "reference document," "white paper," "design document," "educational material," or "text data" may be used. For example, in the case of communication data, the guide document may include standard specification documents disclosed in O-RAN (Open-RAN), ETSI (European Telecommunications Standards Institute), IEEE (Institute of Electrical and Electronics Engineers), 3GPP (Third Generation Partnership Project), or 3GPP2, etc.

[0053] The database of the present disclosure may include at least one of a relational database and a graph database. A relational database may store and manage data in a table format, and each table may consist of rows and columns. Rows may represent individual data items (records), and columns may represent attributes or characteristics of the data.

[0054] A graph database may include graph data containing a knowledge graph. A knowledge graph may consist of at least one of nodes, edges, or properties. Nodes may represent entities, and edges may represent relationships between nodes. However, in this case, assistance from experts familiar with the relationships and descriptions of entities is required, as well as a large amount of guide documents and data.

[0055] The system can manage data by constructing a relational database and storing, adding, and deleting data. In various fields, such as wireless communication systems, the unit of data collection in a relational database may be tabular data. In the case of domain data, due to the nature of the data, the relationships between data (100) attributes may be dependent and difficult to understand without specialized knowledge of the relationships, and it may be difficult to create a knowledge graph from the tabular data. Additionally, in the case of a relational database, the amount of data generated may be large in order to prevent data duplication and contradiction.

[0056] According to an embodiment of the present disclosure, an electronic device can obtain a knowledge graph from table data and construct a knowledge graph to effectively represent the relationships between the data. The electronic device can add the knowledge graph obtained from relational data to a database and construct a graph database. A graph database can be more efficient than a relational database in terms of resources because it manages data with new features as separate nodes even when such data is added.

[0057] In addition, according to an embodiment of the present disclosure, a knowledge graph can be constructed by utilizing a small amount of guide documents, data sheet-level documents, fragmented data, or actual data patterns. Nodes may include all types of data actually collected, and the electronic device of the present disclosure may group and design nodes based on definitions or descriptions in guide documents. The electronic device of the present disclosure may express edge weights by inferring causal relationships and associations based on actual data, and design edge types by inferring associations such as dependent, superior, subordinate, or causal relationships.

[0058] Artificial intelligence (AI) models are models that can be used in next-generation wireless communication systems, such as 5G and 6G, which possess higher performance and efficiency by utilizing artificial intelligence (AI) or machine learning (ML). Instead of 'AI model', terms such as 'neural network model', 'AI / ML model', 'ML model', 'deep learning model', 'generative model', 'LLM model', 'Transformer model', 'deep learning model', 'generative model', 'LLM (Large Language Model)', 'MLM (Masked LLM)', 'attention model', 'RNN (Recurrent Neural Networks) based model', 'LSTM (Long Short-Term Memory)', 'GRU (Gated Recurrented Unit)', 'autoencoder model', and 'GNN (Graph Neural Network)' may be used.

[0059] An AI model is characterized by being created through training. Here, being created through training may mean that a basic AI model is trained by a learning algorithm using multiple training data, thereby creating predefined behavioral rules or an artificial intelligence model configured to perform a desired characteristic (or objective).

[0060] Such learning may be performed on the electronic device itself, which includes the AI ​​model according to the present disclosure, or through a separate server and / or system. The trained AI model may be distributed among network functions, servers, and electronic devices. Examples of learning algorithms include, but are not limited to, supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning.

[0061] An AI model can be composed of multiple neural network layers. Each of the multiple neural network layers has multiple weight values, and can perform neural network operations through operations between the results of previous layers and the multiple weights.

[0062] Multiple weights possessed by multiple neural network layers can be optimized based on the learning results of the AI ​​model. For example, multiple weights can be updated so that the loss value or cost value obtained from the AI ​​model during the learning process is reduced or minimized. In the embodiments of the present disclosure, a specific AI model (e.g., a Transformer model) may be described as an example, but the method according to the embodiments of the present disclosure may be applied to various other neural network models.

[0063] FIG. 1 is a drawing for explaining a method of utilizing a knowledge graph according to an embodiment of the present disclosure.

[0064] When data (100) related to the target information (170) to be predicted is manually selected, the optimal features of the data (100) to be selected can be chosen through a process in which 3 to 4 experts analyze the data for about 3 months. When the target information (170) changes, the same method is repeated to select features, and this process takes a lot of time. Accordingly, a method is needed to streamline the process of acquiring the selected data (150).

[0065] The electronic device (1000) can acquire data (100). The data (100) may be loaded from the database of the electronic device (1000), the electronic device (1000) may receive data from another electronic device (e.g., OAM), or the electronic device (1000) may collect data directly.

[0066] The data (100) of the present disclosure may refer to domain data in various fields, such as communication data, manufacturing data, or retail data. Manufacturing data may include attributes such as the pass rate of a defect detection line, process utilization rate, operating line ratio, inventory quantity, worker input quantity, production indicators generated from the production line, and indicators generated from equipment (continuous operating time, error type, current status, part status, etc.). Retail data may include attributes such as customer data that can identify indicators leading to the purchase and repurchase of products and services, such as visitor information, customer dwell time per product in the store, purchase history, repurchase rate, and after-sales service records.

[0067] Communication data may include data related to communication quality and data collected over cellular networks. Communication data may include attributes such as call drop rate, MCS (Modulation and Coding Scheme), CQI (Channel Quality Indicator), HARQ (Hybrid Automatic Repeat and request), and POWER.

[0068] In one embodiment, the attributes of the communication data may include at least one of Performance Management (PM) data, Configuration Management (CM) data, or Key Performance Indicator (KPI). The attributes of the communication data may be further described below if necessary. Instead of 'attribute', terms such as 'feature', 'category', 'field', 'parameter', 'family display name', 'family name', 'family ID', etc. may be used.

[0069] PM data may refer to key indicators for monitoring base station status or base station performance. PM data may include CQI (Channel Quality Indicator), MCS (Modulation and Coding Scheme), indicators representing field conditions, indicators related to signal strength between the base station and the user terminal, SINR (Signal to Interference plus Noise Ratio), terminal transmission and reception volume, indicators related to transmission quality, and indicators related to data transmission speed.

[0070] For example, communication data may be stored in the form of a table containing the MCS of PM data as shown in [Table 1], but this is merely an example for convenience of explanation and is not limited to the examples mentioned.

[0071] TimeCellDL Transmission MCS 0DL Transmission MCS 1...DL Transmission MCS 312024-02-28 07:001366165...62024-02-28 07:051438257...5...................2024-02-28 23:55123350...14

[0072] CM data may refer to base station configuration parameters and can be used to change or optimize network configurations. CM data may include response time limits between a user terminal (UE) and a base station (UE Response Timeout Time), BLER (Block Error Rate), DL target BLER, etc. For example, communication data may be stored in the form of a table including the DL target BLER of the CM data as shown in [Table 2], but this is merely an example for convenience of explanation and is not limited to the example mentioned.

[0073] TimeCellDL Target BLER2024-02-28 07:0011202024-02-28 07:051120.......2024-02-28 23:551120

[0074] KPIs may refer to information regarding important items directly related to service management, and may include data that serve as important indicators for network performance evaluation. KPIs may include RRC (Radio Resource Control) Connection Drop Rate, DL Throughput, IP Throughput, User Throughput, Potential Throughput, Call Drop Rate, etc., but are not limited to the examples mentioned. For instance, communication data may be stored in the form of a table including the DL Throughput of the KPI as shown in [Table 3], but this is merely an example for the convenience of explanation and is not limited to the examples mentioned.

[0075] TimeCellDL Throughput (Mbps)2024-02-28 07:001463.972024-02-28 07:051351.61......2024-02-28 23:551410.10

[0076] The electronic device (1000) can construct a knowledge graph (130) from data (100). The details of how the electronic device (1000) constructs the knowledge graph (130) may be partially omitted if they overlap with the description in FIGS. 4 to 7.

[0077] The electronic device (1000) can identify nodes and relationships between nodes using at least one of a guide document or data (100). The electronic device (1000) can generate a knowledge graph (130) based on the identified nodes and relationships between nodes.

[0078] Relationships between nodes can be represented by at least one of edge types and edge weights. Edge types may represent at least one of associations between nodes, hierarchical structures between nodes such as whether they belong to a parent or child group, dependencies between nodes, or relationships of influence between nodes. However, the mentioned edge types are merely examples for illustrative purposes and are not limited thereto; other types representing relationships between nodes may also be defined.

[0079] Edge weights can quantify and represent the degree of association between data. For example, edge weights can represent at least one of the distance between data, similarity between data, correlation coefficient between data, or strength of the relationship between data.

[0080] The electronic device (1000) can identify nodes and relationships between nodes (e.g., edge types) using a guide document. The electronic device (1000) can define relationships between nodes by inputting the guide document into an AI model (300). A method for identifying relationships between nodes using a guide document will be explained with reference to FIG. 4.

[0081] The electronic device (1000) can identify relationships between nodes (e.g., edge weights) using data. The method by which the electronic device (1000) identifies relationships between nodes using data will be explained in detail with reference to FIGS. 5 and FIGS. For convenience of explanation, parts that overlap with FIGS. 5 and FIGS. 6 may be partially omitted.

[0082] In one embodiment, when the types of nodes and edges are defined and the structure of the graph is formed, the electronic device (1000) may assign edge weights to the relationship between nodes connected by edges or to the edge types. A method for calculating correlation between data, such as correlation, coefficient, covariance, Euclidean distance, mutual information, etc., may be used to assign edge weights.

[0083] For example, an electronic device (1000) can perform clustering by applying multiple thresholds to the distance between nodes mapped in the latent space. Among these, the set clustered by the smallest threshold can be defined as edge type [Grouped], and the groups included in the set clustered by the next threshold can be defined as edge type [Related]. Edge type [Affected] can be defined based on whether the data distribution of each node mapped to the latent space is similar. Distribution matching techniques such as Kullback-Leidler Divergence, JS Divergence, Earth Mover's Distance, Maximum Mean Discrepancy, Hellinger Distance, Cosine Similarity, and Bhattacharyya Distance can be used as criteria to determine similarity.

[0084] In one embodiment, edge weights may be matched by edge type. For example, the edge between node A and node B may be represented as [grouped: 0.8, related: 0.9, affected: 04]. In one embodiment, only a predetermined edge type and edge weight may be represented. For example, the edge between node A and node B may be represented as [grouped: 0.8].

[0085] The electronic device (1000) can simplify the knowledge graph (130) based on at least one of edge weights and edge types. For example, the electronic device (1000) can simplify the knowledge graph (130) by removing an edge in the generated knowledge graph (130) if the edge corresponds to a predetermined edge type or if the edge weight of the edge is below a certain threshold.

[0086] The electronic device (1000) can update the knowledge graph (130) generated using at least one of a guide document or data (100). For example, the electronic device (1000) can adjust at least one of the nodes, edge types, or edge weights of the knowledge graph (130) built based on the previous standard document using the updated standard document. The electronic device (1000) can adjust at least one of the nodes, edge types, or edge weights of the knowledge graph (130) using acquired actual data. Through this, optimization of the knowledge graph (130) can be achieved.

[0087] The electronic device (1000) can obtain selected data (150) based on the knowledge graph (130).

[0088] The method for obtaining selected data (150) will be explained in detail with reference to FIG. 7. Instead of selected data, terms such as feature-selected data, dimensionality-reduced data, subset of attributes data, filtered data, selected feature data, summarized data, and compressed data may be used.

[0089] The electronic device (1000) can identify a target node corresponding to a target estimation AI model (400). The electronic device (1000) can identify nodes associated with the target node based on edge weights and edge types in the knowledge graph (130). The electronic device (1000) can select features of data based on the nodes associated with the target node. The selected data (150) may include selected data features.

[0090] In one embodiment, selected data may be obtained by performing methods such as discarding or ignoring, dropping, zeroing, multiplying by a predetermined weight, filling with an average or median value, maintaining existing values ​​but using a different algorithm, or minimizing the influence of unselected dimensions in the process of handling results after prediction.

[0091] The electronic device (1000) can train a target estimation AI model (400) to output target information (170) using selected data (150). When performing an inference task using the target estimation AI model (400), that is, when executing the target estimation AI model (400), the input data may be processed to include selected data features such as the selected data (150). In one embodiment, regarding the unselected features of the input data, methods such as discarding or ignoring, dropping, zeroing, multiplying by a predetermined weight, filling with an average or median value, maintaining existing values ​​but using a different algorithm, or minimizing the influence of unselected dimensions in the process of handling results after prediction may be performed.

[0092] The target estimation AI model (400) corresponds to the AI ​​model of the present disclosure. Instead of the target estimation AI model (400), terms such as target prediction model, target forecasting AI, objective value inference system, target value prediction model, outcome estimation algorithm, prediction-based model, and goal-oriented model may be used instead. However, the examples mentioned are not limited to other regression models, classification models, and detection models may be used instead.

[0093] The target estimation AI model (400) can be implemented as an 'RNN (Recurrent Neural Networks) based model', 'LSTM (Long Short-Term Memory)', 'GRU (Gated Recurrented Unit)', 'Autoencoder based model', 'GNN (Graph Neural Network) based model', 'Transformer based model', 'CNN based model', etc., but is not limited to the examples mentioned, and the method according to the embodiments of the present disclosure can be applied to various other neural network models.

[0094] For example, an electronic device (1000) can obtain outputs regarding base station parameters, Configuration Management (CM) data, or base station policies, which are target information (170), by executing a target estimation AI model (400). The target estimation AI model (400) can be trained to receive selection data (150) as input and output base station parameters, Configuration Management (CM) data, or base station policies, which are target information (170).

[0095] The electronic device (1000) can obtain an output regarding a network performance-related KPI (Key Performance Indicator), which is target information (170), by executing a target estimation AI model (400). The target estimation AI model (400) can be trained to receive selection data (150) as input and output a KPI (Key Performance Indicator), which is target information (170).

[0096] The electronic device (1000) can obtain an output regarding channel quality-related indicators, which are target information (170), by executing a target estimation AI model (400). The target estimation AI model (400) can be trained to receive selection data (150) as input and output channel quality-related indicators, which are target information (170).

[0097] The electronic device (1000) can obtain an output regarding network traffic-related indicators, which are target information (170), by executing a target estimation AI model (400). The target estimation AI model (400) can be trained to receive selected data (150) as input and output traffic-related indicators, which are target information (170).

[0098] According to an embodiment of the present disclosure, even if at least one of the target information (170) requiring output or the target estimation AI model (400) changes, the electronic device (1000) can select features of the data (100) from the same knowledge graph (130) and obtain selected data (150). Through this, the electronic device (1000) can efficiently select influential features.

[0099] FIG. 2 is a drawing for illustrating modules included in an electronic device according to one embodiment of the present disclosure. Referring to FIG. 2, an electronic device (1000) according to one embodiment of the present disclosure may include a task management module (200), an AI model (300), and a database (500). The electronic device (1000) may further include a target estimation AI model (400).

[0100] The modules (200, 300, 400, 500) included in the electronic device (1000) of FIG. 2 are configurations classified based on function or role. The modules (200, 300, 400, 500) of the electronic device (1000) of FIG. 2 may be software configurations implemented by the processor (1300) of the electronic device (1000), which will be described later with reference to FIG. 3, executing a program stored in memory (1400), and may also be virtual configurations for which no actual matching hardware device exists. In other words, the operations performed by the processor (1300) of the electronic device (1000) by executing a program or instruction stored in memory (1400) may be classified into multiple groups according to function or purpose, and the entities performing the operations included in each classified group may be represented as the modules (200, 300, 400, 500) of FIG. 2. Accordingly, the operations described as being performed by the modules (200, 300, 400, 500) of the electronic device (1000) illustrated in FIG. 2 can actually be seen as being performed by the processor (1300) of the electronic device (1000) executing a program or instruction stored in memory (1400).

[0101] In FIG. 2, one electronic device (1000) is illustrated as including modules (200, 300, 500), but is not limited thereto. One electronic device (1000) may be implemented to include modules (200, 300, 400, 500), at least some of the modules (200, 300, 400, 500) may be implemented to be included in a separate device, and any one of the modules (200, 300, 400, 500) may be implemented to be included in another module.

[0102] In FIG. 2, the task management module (200) is shown to include a knowledge graph construction module (230) and a selection data acquisition module (250), but is not limited thereto. At least some of the knowledge graph construction module (230) and the selection data acquisition module (250) may be implemented to be included in a separate device, and either one of the knowledge graph construction module (230) or the selection data acquisition module (250) may be implemented to be included in another module.

[0103] As such, the modules (200, 300, 400, 500) included in the electronic device (1000) according to one embodiment of the present disclosure may be hardware configurations or software configurations and may be implemented in various forms of electronic devices (e.g., one electronic device or a combination of two or more electronic devices).

[0104] An electronic device (1000) according to one embodiment of the present disclosure may be a server that performs communication with entities of a terminal, base station, or network. For example, the electronic device (1000) may be a server (e.g., an external AI / ML server) that performs communication by being included in a base station, EMS (Element Management System), SON Manager (Self-Organizing Network Manager), OAM (Operation, Administration, and Maintenance), SON Agent (Self-Organizing Network Agent), etc., or included in a separate device from EMS (Element Management System), SON Manager (Self-Organizing Network Manager), OAM (Operation, Administration, and Maintenance), SON Agent (Self-Organizing Network Agent), etc.

[0105] The AI ​​model (300) may be implemented in the form of a transformer and may include at least one of an encoder (310) and a decoder (320). In this case, the encoder (310) and the decoder (320) of the AI ​​model (300) may each include at least one of one or more attention layers and feedforward layers. Alternatively, the AI ​​model (300) may be implemented in the form of an autoencoder and may include at least one of an encoder (310) and a decoder (320). The AI ​​model (300) may be further described below if necessary.

[0106] The job management module (200) can acquire data (100). For example, the data may be loaded from a database (500) of an electronic device (1000), received by the job management module (200) from another electronic device (e.g., OAM), or collected directly by the job management module (200).

[0107] The knowledge graph construction module (230) of the task management module (200) can construct a knowledge graph (130) from data (100).

[0108] The content of the knowledge graph construction module (230) constructing the knowledge graph (130) may be partially omitted if it overlaps with the description in FIGS. 4 to 7.

[0109] The knowledge graph construction module (230) can identify nodes and relationships between nodes using at least one of guide documents or data (100). The knowledge graph construction module (230) can generate a knowledge graph (130) based on the identified nodes and relationships between nodes.

[0110] Relationships between nodes can be represented by at least one of edge types and edge weights. Edge types may represent at least one of associations between nodes, hierarchical structures between nodes such as whether they belong to a parent or child group, dependencies between nodes, or relationships of influence between nodes. However, the mentioned edge types are merely examples for illustrative purposes and are not limited thereto; other types representing relationships between nodes may also be defined.

[0111] Edge weights can quantify and represent the degree of association between data. For example, edge weights can represent at least one of the distance between data, similarity between data, correlation coefficient between data, or strength of the relationship between data.

[0112] The knowledge graph construction module (230) can identify nodes and relationships between nodes (e.g., edge types) using guide documents. The knowledge graph construction module (230) can define relationships between nodes by inputting guide documents into the AI ​​model (300). A method for identifying relationships between nodes using guide documents will be explained with reference to FIG. 4.

[0113] The knowledge graph construction module (230) can identify relationships between nodes (e.g., edge weights) using data. The method by which the knowledge graph construction module (230) identifies relationships between nodes using data will be explained in detail with reference to FIGS. 5 and FIGS. For convenience of explanation, parts that overlap with FIGS. 5 and FIGS. 6 may be partially omitted.

[0114] In one embodiment, when the types of nodes and edges are defined and the structure of the graph is formed, the electronic device (1000) may assign edge weights to the relationship between nodes connected by edges or to the edge types. To assign edge weights, a method for calculating correlation between data, such as correlation, coefficient, covariance, Euclidean distance, mutual information, etc., may be used.

[0115] For example, an electronic device (1000) can perform clustering by applying multiple thresholds to the distance between nodes mapped in the latent space. Among these, the set clustered by the smallest threshold can be defined as edge type [Grouped], and the groups included in the set clustered by the next threshold can be defined as edge type [Related]. Edge type [Affected] can be defined based on whether the data distribution of each node mapped to the latent space is similar. Distribution matching techniques such as Kullback-Leidler Divergence, JS Divergence, Earth Mover's Distance, Maximum Mean Discrepancy, Hellinger Distance, Cosine Similarity, and Bhattacharyya Distance can be used as criteria to determine similarity.

[0116] In one embodiment, edge weights may be matched by edge type. For example, the edge between node A and node B may be represented as [grouped: 0.8, related: 0.9, affected: 04]. In one embodiment, only a predetermined edge type and edge weight may be represented. For example, the edge between node A and node B may be represented as [grouped: 0.8].

[0117] The knowledge graph construction module (230) can simplify the knowledge graph (130) based on at least one of edge weights and edge types. For example, the knowledge graph construction module (230) can simplify the knowledge graph (130) by removing an edge in the generated knowledge graph (130) if the edge corresponds to a predetermined edge type or if the edge weight of the edge is below a certain threshold.

[0118] The knowledge graph construction module (230) can update the knowledge graph (130) created using at least one of a guide document or data (100). For example, the electronic device (1000) can adjust at least one of the nodes, edge types, or edge weights of the knowledge graph (130) built based on the previous standard document using the updated standard document. The electronic device (1000) can adjust at least one of the nodes, edge types, or edge weights of the knowledge graph (130) using acquired actual data. Through this, optimization of the knowledge graph (130) can be achieved.

[0119] The selection data acquisition module (250) of the task management module (200) can acquire selected data (150) based on the knowledge graph (130).

[0120] The method for obtaining selected data (150) will be explained in detail with reference to FIG. 7. Instead of selected data, terms such as feature-selected data, dimensionality-reduced data, subset of attributes data, filtered data, selected feature data, summarized data, and compressed data may be used.

[0121] The selective data acquisition module (250) can identify a target node corresponding to a target estimation AI model (400). The selective data acquisition module (250) can identify nodes associated with the target node based on edge weights and edge types in the knowledge graph (130). The selective data acquisition module (250) can select features of the data based on the nodes associated with the target node. The selective data (150) may include selected data features.

[0122] In one embodiment, selected data may be obtained by performing methods such as discarding or ignoring, dropping, zeroing, multiplying by a predetermined weight, filling with an average or median value, maintaining existing values ​​but using a different algorithm, or minimizing the influence of unselected dimensions in the process of handling results after prediction.

[0123] The electronic device (1000) can train a target estimation AI model (400) to output target information (170) using selected data (150). When performing an inference task using the target estimation AI model (400), that is, when executing the target estimation AI model (400), the input data may be processed to include selected data features such as the selected data (150). In one embodiment, regarding the unselected features of the input data, methods such as discarding or ignoring, dropping, zeroing, multiplying by a predetermined weight, filling with an average or median value, maintaining existing values ​​but using a different algorithm, or minimizing the influence of unselected dimensions in the process of handling results after prediction may be performed.

[0124] According to one embodiment of the present disclosure, the database (500) may include a database having various structures and forms. The database (500) may include at least one of a relational database and a graph database. The database (500) may store at least one of a guide document, data (100), a knowledge graph (130), selection data (150), and target information (170). The database (500) may store at least one mapping information among the guide document, data (100), knowledge graph (130), selection data (150), and target information (170). Information related to tasks performed by the AI ​​model (300) and the target estimation AI model (400) may be stored in the database (500). For example, embedding data and content calculated on the embedding data may be stored in the database (500). Data stored in the database (500) will be further described below if necessary.

[0125] FIG. 3 is a diagram illustrating a hardware configuration included in an electronic device according to one embodiment of the present disclosure. Referring to FIG. 2, an electronic device (1000) according to one embodiment may include a communication interface (1100), an input / output interface (1200), a processor (1300), and a memory (1400). However, the components of the electronic device (1000) are not limited to the examples described above, and the electronic device (1000) may include more components than the components described above, or fewer components. Some or all of the communication interface (1100), the input / output interface (1200), the processor (1300), and the memory (1400) may be implemented in the form of a single chip.

[0126] The communication interface (1100) is configured to transmit and receive signals (control commands and data, etc.) to and from an external device via wired or wireless means, and may be implemented to include a communication chipset that supports various communication protocols. The communication interface (1100) may receive signals from the outside and output them to the processor (1300), or transmit signals output from the processor (1300) to the outside. The electronic device (1000) may communicate with external devices through the communication interface (1100). Instead of the communication interface (1100), terms such as a transceiver unit, which collectively refers to a receiver and a transmitter, may be used.

[0127] The communication interface (1100) can transmit and receive signals with other electronic devices, such as a server, terminal, base station, network element, or network entity. The signals transmitted and received with other electronic devices, such as a server, terminal, base station, network element, or network entity, may include control information and data. To this end, the communication interface (1100) may be composed of an RF transmitter that up-converts and amplifies the frequency of a transmitted signal, and an RF receiver that low-noise amplifies a received signal and down-converts the frequency. However, this is one embodiment of the communication interface (1100), and the components of the communication interface (1100) are not limited to an RF transmitter and an RF receiver.

[0128] Additionally, the communication interface (1100) can perform functions for transmitting and receiving signals through a wireless channel. For example, the communication interface (1100) can receive a signal through a wireless channel and output it to a processor (1300), and transmit the signal output from the processor (1300) through a wireless channel.

[0129] The input / output interface (1200) may include an input interface (e.g., touch screen, keyboard, microphone, etc.) for receiving commands or information from a user, and an output interface (e.g., display panel, speaker, etc.) for displaying the result of an operation according to a user's command or the status of the electronic device (1000). According to one embodiment of the present disclosure, the electronic device (1000) may receive prompts and input data from a user through the input / output interface (1200), and when the operation is completed, it may output the result of the operation (e.g., an answer to a request or question in the prompt, a knowledge graph, selection data, target information, etc. according to the request in the prompt) through the input / output interface (1200).

[0130] A processor (1300) is a configuration that controls a series of processes to enable an electronic device (1000) to operate according to the embodiments described below, and may be composed of one or more processors. One or more processors included in the processor (1300) may be circuitry such as an SoC (System on Chip) or IC (Integrated Circuit). One or more processors included in the processor (1300) may be a general-purpose processor such as a CPU (Central Processing Unit), MPU (Micro Processor Unit), AP (Application Processor), or DSP (Digital Signal Processor); a graphics-dedicated processor such as a GPU (Graphic Processing Unit) or VPU (Vision Processing Unit); an artificial intelligence-dedicated processor such as an NPU (Neural Processing Unit); or a communication-dedicated processor such as a CP (Communication Processor). If one or more processors included in the processor (1300) are artificial intelligence-dedicated processors, said artificial intelligence-dedicated processors may be designed with a hardware structure specialized for processing a specific artificial intelligence model.

[0131] The processor (1300) can write data to memory (1400) or read data stored in memory (1400), and in particular, can process data according to a predefined operation rule or artificial intelligence model by executing a program or at least one instruction stored in memory (1400). Accordingly, the processor (1300) can perform operations described in subsequent embodiments, and operations described as being performed by the electronic device (1000) or modules (200, 300, 400, 500) included in the electronic device (1000) in subsequent embodiments can be seen as being performed by the processor (1300) unless otherwise specified.

[0132] For example, the processor (1300) can receive control signals and data signals through the transceiver (1210) and process the received control signals and data signals. The processor (1300) can transmit the processed control signals and data signals through the communication interface (1100). Additionally, the processor (1300) can write or read data to or from the memory (1400). The processor (1300) can perform the functions of the protocol stack required by the communication standard. To this end, the processor (1300) may include at least one processor or microprocessor.

[0133] Memory (1400) is configured to store various programs or data and may be composed of a storage medium or a combination of storage media such as ROM, RAM, hard disk, CD-ROM, and DVD. Memory (1400) may not exist separately but may be configured to be included in the processor (1300). Memory (1400) may be composed of volatile memory, non-volatile memory, or a combination of volatile and non-volatile memory. Memory (1400) may store a program or at least one instruction for performing operations according to the embodiments described below. Memory (1400) may provide stored data to the processor (1300) upon the request of the processor (1300).

[0134] FIG. 4 is a drawing for explaining a method of constructing a knowledge graph according to an embodiment of the present disclosure.

[0135] Referring to FIG. 4, the electronic device (1000) can construct a knowledge graph (130) using a guide document and an AI model (300).

[0136] A guide document in the present disclosure refers to a document containing information regarding the relationship between entities. A guide document may include data sheets, documents providing procedures and guidelines necessary to use a product or technology or to perform specific tasks, etc. Instead of "guide document," terms such as "data sheet," "technical specification document," "standard specification document," "paper," "patent specification," "report," "policy document," "manual," "news," "reference document," "white paper," "design document," "educational material," and "text data" may be used. For example, in the case of telecommunications data, a guide document may include standard specification documents disclosed in O-RAN (Open-RAN), ETSI (European Telecommunications Standards Institute), IEEE (Institute of Electrical and Electronics Engineers), 3GPP (Third Generation Partnership Project), or 3GPP2, etc.

[0137] In one embodiment, the AI ​​model (300) may be implemented in the form of a transformer and may include at least one of an encoder (310) and a decoder (320). The encoder (310) and the decoder (320) of the AI ​​model (300) may each include at least one of one or more attention layers and one or more feedforward layers. Each attention layer may include one or more attention heads. For example, in the case of multi-head attention, multiple attention heads may operate in parallel, and the outputs obtained from each attention head may be combined to obtain a final output.

[0138] During the process in which the AI ​​model (300) performs operations, the layers included in the encoder (310) and decoder (320) output matrices and pass them to the next layer; the matrices generated in the middle during the operation process are called Hidden State Matrices (HSMs). For example, during the process in which the AI ​​model (300) performs operations, attention layers may output an attention value matrix, and feedforward layers may output an activation matrix. Both the attention value matrix and the activation matrix are included in the HSM. The HSMs output from the layers of the AI ​​model (300) can be stored in the database (500) as intermediate operation results.

[0139] In the embodiments of the present disclosure, a specific AI model (e.g., an LLM graph transformer model) may be described as an example, but the method according to the embodiments of the present disclosure may be applied to various other neural network models.

[0140] The electronic device (1000) can generate a knowledge graph using at least one of a guide document and data. The electronic device (1000) can identify and define relationships between nodes (e.g., edge types) by inputting the guide document into an AI model (300).

[0141] 예를 들어, 텍스트가 [Marie Curie, born in 1867, was a Polish and naturalised-French physicist and chemist who conducted pioneering research on radioactivity. She was the first woman to win a Nobel Prize, the first person to win a Nobel Prize twice, and the only person to win a Nobel Prize in two scientific fields. Her husband, Pierre Curie, was a co-winner of her first Nobel Prize, making them the first-ever married couple to win the Nobel Prize and launching the Curie family legacy of five Nobel Prizes. She was, in 1906, the first woman to become a professor at the University of Paris.] 의 내용을 포함하는 경우, 노드를 [Node(id='Marie Curie', type='Person'), Node(id='Pierre Curie', type='Person'), Node(id='University Of Paris', type='Organization')] Relationships: [Relationship(source=Node(id='Marie Curie', type='Person'), target=Node(id='Pierre Curie', type='Person'), type='SPOUSE')와 같이 전자 장치(1000)는 식별할 수 있다.The relationship between the source node (id='Marie Curie', type='Person') and the target node (id='University Of Paris', type='Organization') can be identified as [type='WORKED_AT')]. Based on this, the electronic device can obtain the knowledge graph shown in FIG. 4.

[0142] [4.1.1.1 Attempted RRC connection establishments a)"This measurement provides the number of RRC connection establishment attempts for each establishment cause." (...) c) Receipt of an RRCConnectionRequest message by the eNodeB / RN from the UE. Each RRCConnectionRequest message received is added to the relevant per establishment cause measurement. The possible causes are included in TS 36.331 [8]. The sum of all supported per cause measurements shall equal the total number of RRCConnectionRequest. In case only a subset of per cause measurements is supported, a sum subcounter will be provided first. (...) e)The measurement name has the form RRC.ConnEstabAtt.Cause where Cause identifies the establishment cause. (...)], [4.1.1.2 Successful RRC connection establishments 1) This measurement provides the number of successful RRC establishments for each establishment cause. (...) 3) Receipt by the eNodeB / RN of an RRCConnectionSetupComplete message following a RRC connection establishment request. Each RRCConnectionSetupComplete message received is added to the relevant per establishment cause measurement. The possible causes are included in TS 36.331 [8]. The sum of all supported per cause measurements shall equal the total number of successful RRC Connection Establishments. In case only a subset of per cause measurements is supported, a sum subcounter will be provided first. (...) 5) Using a guide document containing content such as [...], a relationship can be derived representing the connection success rate between node [RRC.ConnEstabAtt.Cause] and node [RRC.ConnEstabSucc.Cause] = (number of successful connections / number of connection attempts) * 100 = (RRC.ConnEstabSucc.Cause / RRC.ConnEstabAtt.Cause) * 100. Alternatively, it can be expressed as node [RRC.ConnEstabSucc.Cause], [RRC.ConnEstabAtt.Cause], edge [Edge type: "Derived Metric" #Success rate, edge weight = 0.85]. However, this is merely an example for illustrative purposes and is not limited to the examples mentioned.

[0143] The electronic device (1000) can generate a knowledge graph based on identified nodes and relationships between nodes. Relationships between nodes can be represented by at least one of edge types and edge weights. Edge types can represent at least one of associations between nodes, hierarchical structures between nodes such as whether they are upper or lower groups, dependencies between nodes, or influence relationships between nodes. However, the mentioned edge types are merely examples for illustrative purposes and are not limited thereto; other types representing relationships between nodes may be defined.

[0144] The electronic device (1000) can identify edge weights using data. Edge weights can represent the degree of correlation between data by quantifying it. For example, edge weights may represent at least one of the distance between data, the correlation coefficient between data, the strength of the relationship between data, or a weight corresponding to the edge type between data. The process of identifying edge weights using data will be explained in detail with reference to the descriptions in FIGS. 5 and 6. For convenience of explanation, parts that overlap with FIGS. 5 and 6 may be partially omitted.

[0145] The electronic device (1000) can group or layer nodes based on at least one of the edge type and edge weight. For example, the electronic device (1000) can group nodes into representative attributes if the edge type between nodes represents a set type and the edge weight corresponds to a value greater than or equal to a set value.

[0146] For example, nodes [B1, B2, B3, B4] can be grouped into group node B, where the edge type is [grouped] and the edge weight is above a threshold. Nodes [D1, D2, D3, D4] can be grouped into group node D, where the edge type is [grouped] and the edge weight is above a threshold. Grouped nodes B and D can be grouped into group type B / D, where the edge type is [related] and the edge weight is above a threshold.

[0147] In one embodiment, the electronic device (1000) can determine a representative node according to a grouping rule. The grouping rule may be set in advance or defined. For example, the grouping rule may be set at the time of designing the electronic device (1000). Alternatively, the electronic device (1000) may perform grouping by receiving or inputting a setting for the grouping rule.

[0148] The electronic device (1000) can simplify a knowledge graph based on at least one of edge weights and edge types. For example, the knowledge graph can be simplified by removing an edge in the generated knowledge graph if it corresponds to a predetermined edge type or if the edge weight is below a certain level.

[0149] The electronic device (1000) can update a knowledge graph generated using at least one of a guide document or data. For example, the electronic device (1000) can adjust at least one of the nodes, edge types, or edge weights of the knowledge graph using an updated standard specification document. The electronic device (1000) can adjust at least one of the nodes, edge types, or edge weights of the knowledge graph using acquired data. Through this, optimization of the knowledge graph can be achieved.

[0150] FIGS. 5 and 6 are drawings for explaining a method of constructing a knowledge graph according to an embodiment of the present disclosure.

[0151] Referring to FIGS. 5 and 6, the electronic device (1000) can identify nodes and relationships between nodes using data (100). The electronic device (1000) can generate a knowledge graph (130) based on the identified nodes and relationships between nodes.

[0152] Relationships between nodes can be represented by at least one of edge types and edge weights. Edge types may represent at least one of associations between nodes, hierarchical structures between nodes such as whether they belong to a parent or child group, dependencies between nodes, or relationships of influence between nodes. However, the mentioned edge types are merely examples for illustrative purposes and are not limited thereto; other types representing relationships between nodes may also be defined.

[0153] Edge weights can quantify and represent the degree of correlation between data. When the types of nodes and edges are defined and the structure of the graph is formed, the electronic device (1000) can assign edge weights to the relationship between nodes connected by edges or to the edge types. Edge weights can represent at least one of the distance between data, similarity between data, correlation coefficient between data, or strength of the relationship between data. The electronic device (1000) can embed data (100). Embedding data (100) may include mapping the data into a latent space. A latent space refers to a space that represents the relationship between data by compressing the data into an embedding space of a specific dimension. Terms such as embedding space or feature space may be used instead of latent space.

[0154] In one embodiment, an electronic device (1000) can embed data (100) using an AI model (300). The AI ​​model (300) can embed data (100) using an autoencoder-based model or metric learning. Metric learning can be trained to position similar data close together and contrasting data far apart. However, the method of embedding data is not limited to the examples mentioned, and other embedding methods may be used. For example, instead of autoencoder or metric learning, Principal Component Analysis (PCA), t-Distributed Stochastic Neighbor Embedding (t-SNE), Variational Autoencoder (VAE), Self-Organizing Maps (SOMs), Generative Adversarial Network (GAN), Contrastive Predictive Coding (CPC), Transformer-based embedding, etc. may be used instead.

[0155] The electronic device (1000) can identify the degree of correlation between the data of the embedded data (120). The electronic device (1000) can generate a knowledge graph based on the degree of correlation between the identified data.

[0156] The embedded data (120) may be located in a latent space, and the electronic device (1000) may cluster the embedded data (120). For example, the embedded data (120) may be clustered using a clustering algorithm such as K-means or DBSCAN (Density-Based Spatial Clustering of Applications with Noise). The electronic device (1000) may form data groups based on the clustering results and define them as nodes. The electronic device (1000) may establish a hierarchy between clusters by adjusting clustering parameters (e.g., distance threshold). The hierarchy may be utilized in the level structure of a graph. The electronic device (1000) may define the group derived from the clustering results as a single node. The electronic device (1000) may assign a group type by analyzing the characteristics of the data included in each node.

[0157] In one embodiment, the electronic device (1000) can perform clustering by applying multiple thresholds using the distance between nodes mapped to the latent space. Among these, the set clustered with the smallest threshold size can be defined as edge type [Grouped], and the groups included in the set clustered with the next largest threshold size can be defined as edge type [Related]. Edge type [Affected] can be defined based on whether the data distribution of each node mapped to the latent space is similar. A distribution matching technique such as Kullback-Leidler Divergence can be used as a criterion for determining similarity.

[0158] For example, nodes [B1, B2, B3, B4] can be grouped into group node B, where the edge type is [grouped] and the edge weight is above a threshold. Nodes [D1, D2, D3, D4] can be grouped into group node D, where the edge type is [grouped] and the edge weight is above a threshold. Grouped nodes B and D can be grouped into group type B / D, where the edge type is [related] and the edge weight is above a threshold.

[0159] In one embodiment, the electronic device (1000) can simplify a knowledge graph based on at least one of edge weight and edge type. For example, the knowledge graph can be simplified by removing edges in the generated knowledge graph when they are of a predetermined edge type or when the edge weight is below a certain level.

[0160] The electronic device (1000) can update a knowledge graph generated using at least one of a guide document or data. For example, the electronic device (1000) can adjust at least one of the nodes, edge types, or edge weights of the knowledge graph using an updated standard specification document. The electronic device (1000) can adjust at least one of the nodes, edge types, or edge weights of the knowledge graph using acquired data. Through this, optimization of the knowledge graph can be achieved. A method for updating a knowledge graph generated using a guide document is performed with reference to FIG. 4.

[0161] FIG. 7 is a diagram illustrating a method for obtaining selection data using a knowledge graph according to an embodiment of the present disclosure.

[0162] The electronic device (1000) can identify a target node (23) corresponding to a target estimation AI model (400). For example, the electronic device (1000) can determine a node corresponding to the target information that the target estimation AI model (400) intends to output as the target node. Information indicating the target node (23) can be received from a user or another electronic device, and based on the information indicating the target node (23), the electronic device (1000) can identify the target node (23) corresponding to the target estimation AI model (400). However, this is merely an example for convenience of explanation and is not limited to the example mentioned. For example, even if the electronic device (1000) does not receive information directly indicating the target node, it can determine the target node through analysis by receiving target information or user intent as input.

[0163] The electronic device (1000) can identify nodes (25) related to a target node (23) based on edge weights and edge types in a knowledge graph (130). The electronic device (1000) can determine that nodes satisfying edge type conditions and edge weight conditions with the target node (23) are nodes (25) related to the target node (23). For example, the electronic device (1000) can identify one or more nodes satisfying [Related weight > 0.8 and Grouped > 0.0] as related nodes (25).

[0164] In one embodiment, the electronic device (1000) can generate a query representing conditions for edge weights and edge types. The electronic device (1000) can use the query to search for nodes related to a target node in a knowledge graph. For example, the query may represent [Related weight > 0.8 or Grouped > 0.0].

[0165] The electronic device (1000) can select data features based on the nodes (25) associated with the target node (23). The selected data (150) may include the selected data features. For example, if the associated nodes (25) are [a1, a2, a3, a4], the electronic device (1000) can select features corresponding to the nodes [a1, a2, a3, a4]. The selected data (150) may include the selected data features.

[0166] In one embodiment, selected data may be obtained by performing methods such as discarding or ignoring, dropping, zeroing, multiplying by a predetermined weight, filling with an average or median value, maintaining existing values ​​but using a different algorithm, or minimizing the influence of unselected dimensions in the process of handling results after prediction.

[0167] The electronic device (1000) can train a target estimation AI model (400) to output target information (170) using selected data (150). When performing an inference task using the target estimation AI model (400), that is, when executing the target estimation AI model (400), the input data may be processed to include selected data features such as the selected data (150). In one embodiment, regarding the unselected features of the input data, methods such as discarding or ignoring, dropping, zeroing, multiplying by a predetermined weight, filling with an average or median value, maintaining existing values ​​but using a different algorithm, or minimizing the influence of unselected dimensions in the process of handling results after prediction may be performed.

[0168] FIG. 8 is a flowchart relating to a method for training a target estimation AI model according to an embodiment of the present disclosure.

[0169] In operation S810, the electronic device (1000) can identify a target node corresponding to a target estimation AI model.

[0170] The electronic device (1000) can determine a node corresponding to the target information that the target estimation AI model (400) intends to output as a target node. Upon receiving information indicating a target node (23), the electronic device (1000) can identify a target node (23) corresponding to the target estimation AI model (400) based on the information indicating a target node (23).

[0171] In operation S820, the electronic device (1000) can identify nodes associated with a target node based on edge weights and edge types in a knowledge graph.

[0172] The electronic device (1000) can identify nodes and relationships between nodes using at least one of a guide document or data (100). The electronic device (1000) can generate a knowledge graph (130) based on the identified nodes and relationships between nodes.

[0173] Relationships between nodes can be represented by at least one of edge types and edge weights. Edge types may represent at least one of associations between nodes, hierarchical structures between nodes such as whether they belong to a parent or child group, dependencies between nodes, or relationships of influence between nodes. However, the mentioned edge types are merely examples for illustrative purposes and are not limited thereto; other types representing relationships between nodes may also be defined.

[0174] Edge weights can quantify and represent the degree of association between data. For example, edge weights can represent at least one of the distance between data, similarity between data, correlation coefficient between data, or strength of the relationship between data.

[0175] If a knowledge graph has not been constructed, the electronic device (1000) may construct the knowledge graph before identifying nodes associated with a target node based on edge weights and edge types in the knowledge graph.

[0176] If there is a constructed knowledge graph, the electronic device (1000) can update the knowledge graph using at least one of data and guide documents.

[0177] The electronic device (1000) can update the knowledge graph (130) generated using at least one of a guide document or data (100). For example, the electronic device (1000) can adjust at least one of the nodes, edge types, or edge weights of the knowledge graph (130) built based on the previous standard document using the updated standard document. The electronic device (1000) can adjust at least one of the nodes, edge types, or edge weights of the knowledge graph (130) using acquired actual data. Through this, optimization of the knowledge graph (130) can be achieved.

[0178] Some parts of the description of FIGS. 4 to 7 regarding the construction or updating of the knowledge graph (130) may be omitted.

[0179] The electronic device (1000) can simplify a knowledge graph based on at least one of edge weights and edge types. For example, the electronic device (1000) can simplify a knowledge graph by removing an edge in the generated knowledge graph if the edge corresponds to a predetermined edge type or if the edge weight of the edge is below a certain threshold.

[0180] The electronic device (1000) can determine that a node satisfying the target node, edge type condition, and edge weight condition in the knowledge graph is a node associated with the target node.

[0181] In one embodiment, the electronic device (1000) can generate a query representing at least one of the conditions of edge weights and the conditions of edge types. The electronic device (1000) can use the query to search for related nodes in a knowledge graph.

[0182] In operation S830, the electronic device (1000) can select features of the data based on the node associated with the target node.

[0183] Before selecting the characteristics of the data, the electronic device (1000) may acquire the data. However, the acquisition of the data may be performed after operations S810 and S820, before operations S810 and S820, or together with any one of operations S810 and S820. The data (100) may be loaded from the database of the electronic device (1000), the electronic device (1000) may receive data from another electronic device (e.g., OAM), or the electronic device (1000) may directly collect the data.

[0184] The data (100) of the present disclosure may refer to domain data in various fields, such as communication data, manufacturing data, or retail data. Manufacturing data may include attributes such as the pass rate of a defect detection line, process utilization rate, operating line ratio, inventory quantity, worker input quantity, production indicators generated from the production line, and indicators generated from equipment (continuous operating time, error type, current status, part status, etc.). Retail data may include attributes such as customer data that can identify indicators leading to the purchase and repurchase of products and services, such as visitor information, customer dwell time per product in the store, purchase history, repurchase rate, and after-sales service records.

[0185] Communication data may include data related to communication quality and data collected over cellular networks. Communication data may include attributes such as call drop rate, MCS (Modulation and Coding Scheme), CQI (Channel Quality Indicator), HARQ (Hybrid Automatic Repeat and request), and POWER.

[0186] In one embodiment, the attributes of the communication data may include at least one of Performance Management (PM) data, Configuration Management (CM) data, or Key Performance Indicator (KPI).

[0187] In operation S840, the electronic device (1000) can train the target estimation AI model to output target information using selected data including selected features.

[0188] Selected data may include selected data features. For example, if the related node (25) is [a1, a2, a3, a4], the electronic device (1000) may select features corresponding to the node [a1, a2, a3, a4]. Selected data (150) may include selected data features.

[0189] In one embodiment, selected data may be obtained by performing methods such as discarding or ignoring, dropping, zeroing, multiplying by a predetermined weight, filling with an average or median value, maintaining existing values ​​but using a different algorithm, or minimizing the influence of unselected dimensions in the process of handling results after prediction.

[0190] The electronic device (1000) can train a target estimation AI model (400) to output target information (170) using selected data. When performing an inference task using the target estimation AI model (400), that is, when executing the target estimation AI model (400), the input data may be processed to include selected data features such as selected data (150). In one embodiment, regarding the unselected features of the input data, methods such as discarding or ignoring, dropping, zeroing, multiplying by a predetermined weight, filling with an average or median value, maintaining existing values ​​but using a different algorithm, or minimizing the influence of unselected dimensions in the process of handling results after prediction may be performed.

[0191] The electronic device (1000) can obtain output regarding target information, such as base station parameters, CM (Configuration Management) data, or base station policies, by executing a target estimation AI model. The target estimation AI model can be trained to receive selection data as input and output target information, such as base station parameters, CM (Configuration Management) data, or base station policies.

[0192] The electronic device (1000) can obtain an output regarding a KPI (Key Performance Indicator) related to network performance, which is target information, by executing a target estimation AI model. The target estimation AI model can be trained to receive selection data as input and output a KPI (Key Performance Indicator), which is target information.

[0193] The electronic device (1000) can obtain an output regarding channel quality-related indicators, which are target information, by executing a target estimation AI model. The target estimation AI model can be trained to receive selection data (150) as input and output channel quality-related indicators, which are target information (170).

[0194] The electronic device (1000) can obtain an output regarding network traffic-related indicators, which are target information (170), by executing a target estimation AI model. The target estimation AI model can be trained to receive selected data as input and output traffic-related indicators, which are target information.

[0195] The electronic device (1000) can evaluate the performance of a trained target estimation AI model. If the performance of the target estimation AI model reaches the target performance, the electronic device (1000) can determine the features of the selected data that became the input of the target estimation AI model as the input features of the current model. If the performance of the target estimation AI model does not reach the target performance, the electronic device (1000) can search for nodes related to the target node and train the target estimation AI model by adjusting at least one of the edge type condition and the edge weight condition, and the input data composed of features corresponding to the input features during training or inference of the target estimation AI model.

[0196] Various embodiments of the present disclosure may be implemented or supported by one or more computer programs, and computer programs may be formed from computer-readable program code and stored on a computer-readable medium. In the present disclosure, “application” and “program” may represent one or more computer programs, software components, instruction sets, procedures, functions, objects, classes, instances, related data, or parts thereof suitable for implementation in computer-readable program code. “Computer-readable program code” may include various types of computer code, including source code, object code, and executable code. “Computer-readable medium” may include various types of media accessible by a computer, such as read-only memory (ROM), random access memory (RAM), hard disk drive (HDD), compact disc (CD), digital video disc (DVD), or various types of memory.

[0197] Additionally, a device-readable storage medium may be provided in the form of a non-transitory storage medium. Here, a 'non-transitory storage medium' is a tangible device and may exclude wired, wireless, optical, or other communication links that transmit transient electrical or other signals. Meanwhile, this 'non-transitory storage medium' does not distinguish between cases where data is stored semi-permanently and cases where it is stored temporarily. For example, a 'non-transitory storage medium' may include a buffer in which data is stored temporarily. A computer-readable medium may be any available medium accessible by a computer and may include both volatile and non-volatile media, as well as removable and non-removable media. A computer-readable medium includes media in which data can be stored permanently and media in which data can be stored and subsequently overwritten, such as rewritable optical discs or erasable memory devices.

[0198] A non-transient computer-readable storage medium stores one or more computer programs (software modules), said one or more computer programs include computer-executable instructions that operate an electronic device to perform a method according to the present disclosure when executed individually or collectively by one or more processors of an electronic device. Alternatively, said software may additionally be a computer program (or product) that includes instructions that operate an electronic device to perform a method according to the present disclosure when executed individually or collectively by one or more processors of an electronic device.

[0199] The software may be stored in a transient or non-transient storage device, for example, in the form of read-only memory (ROM) (whether or not it is erasable or rewritable), or random access memory (RAM), memory chips, devices, or integrated circuits (ICs). Additionally, the software may be stored in the form of an optically or magnetically readable medium, for example, a compact disc (CD), a digital multifunction disc (DVD), a magnetic disc, or a magnetic tape. It should be understood that the storage device and the storage medium are examples of non-transient machine-readable storage media suitable for storing programs for implementing various embodiments of the present disclosure. Accordingly, various embodiments of the present disclosure may provide a program comprising code for implementing an apparatus or method according to any one of the claims of the present disclosure, and a non-transient machine-readable storage medium storing such program.

[0200] According to an embodiment of the present disclosure, a method for training a target estimation AI (artificial intelligence) model may be provided.

[0201] The method may include a step of identifying a target node corresponding to a target estimation AI model. The method may include a step of identifying nodes associated with the target node based on edge weights and edge types in a knowledge graph.

[0202] The method may include a step of selecting features of data based on nodes associated with the target node. The method may include a step of training the target estimation AI model to output target information using selected data containing the selected features.

[0203] The method may include the step of generating a query representing conditions of edge weights and edge types in identifying nodes related to a target node; and the step of exploring related nodes in a knowledge graph using the query.

[0204] The method may include a step of identifying whether the performance of the target estimation AI model has reached the target performance; a step of adjusting the conditions of the edge weights and the conditions of the edge types if the performance of the target estimation AI model has not reached the target performance; and a step of training the target estimation AI model based on the adjusted conditions of the edge weights and the conditions of the edge types.

[0205] The method of the present disclosure may include the step of constructing a knowledge graph.

[0206] The method may include the step of constructing a knowledge graph using guide documents and an AI model.

[0207] The method may include a step of generating a knowledge graph by running an AI model using a guide document. The method may include a step of updating the generated knowledge graph using data.

[0208] The method may include the step of embedding data; the step of identifying the degree of correlation between the data of the embedded data; and the step of generating a knowledge graph based on the identified degree of correlation between the data.

[0209] According to an embodiment of the present disclosure, a computer-readable recording medium may be provided on which a program for performing any one of the above methods on a computer is recorded.

[0210] According to an embodiment of the present disclosure, an electronic device for training a target estimation AI (artificial intelligence) model may be provided. The electronic device may include a memory in which a program or at least one instruction is stored; and at least one processor.

[0211] By executing a program stored in memory or at least one instruction, the electronic device can identify a target node corresponding to a target estimation AI model.

[0212] By executing a program stored in memory or at least one instruction, the electronic device can identify nodes associated with a target node based on edge weights and edge types in a knowledge graph.

[0213] By executing a program stored in memory or at least one instruction, the electronic device can select data features based on a node associated with a target node.

[0214] By executing a program stored in memory or at least one instruction, the electronic device can train a target estimation AI model to output target information using selected data containing selected features.

[0215] By executing a program stored in memory or at least one instruction, the electronic device can generate a query representing conditions of edge weights and conditions of edge types in identifying nodes related to a target node, and use the query to search for related nodes in a knowledge graph.

[0216] By executing a program stored in memory or at least one instruction, the electronic device can identify whether the performance of the target estimation AI model has reached the target performance, and if the performance of the target estimation AI model has not reached the target performance, adjust the conditions of the edge weights and the conditions of the edge types, and train the target estimation AI model based on the adjusted conditions of the edge weights and the conditions of the edge types.

[0217] An electronic device can construct a knowledge graph by having at least one processor execute a program stored in memory or at least one instruction.

[0218] By having at least one processor execute a program stored in memory or at least one instruction, the electronic device can construct a knowledge graph using guide documents and an AI model.

[0219] By executing a program stored in memory or at least one instruction, the electronic device can generate a knowledge graph by running an AI model using a guide document and update the generated knowledge graph using data.

[0220] By executing a program stored in memory or at least one instruction, an electronic device can embed data, identify the degree of correlation between the data of the embedded data, and generate a knowledge graph based on the identified degree of correlation between the data.

[0221] According to one embodiment, the method according to the various embodiments disclosed herein may be provided by being included in a computer program product. The computer program product may be traded between a seller and a buyer as a product. The computer program product may be distributed in the form of a device-readable storage medium (e.g., compact disc read-only memory (CD-ROM)), or distributed online (e.g., download or upload) through an application store or directly between two user devices (e.g., smartphones). In the case of online distribution, at least a portion of the computer program product (e.g., a downloadable app) may be temporarily stored or temporarily created on a device-readable storage medium, such as the memory of a manufacturer's server, an application store's server, or a relay server.

[0222] The foregoing description of the present disclosure is for illustrative purposes only, and those skilled in the art will understand that modifications can be easily made to other specific forms without altering the technical spirit or essential features of the present disclosure. For example, suitable results may be achieved even if the described techniques are performed in a different order than described, and / or components such as systems, structures, devices, circuits, etc., described are combined or assembled in a form different from described, or replaced or substituted by other components or equivalents. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. For example, each component described as a single unit may be implemented in a distributed manner, and components described as distributed may likewise be implemented in a combined form.

[0223] The scope of the present disclosure is defined by the claims set forth below rather than by the detailed description above, and all modifications or variations derived from the meaning and scope of the claims and equivalent concepts thereof should be interpreted as being included within the scope of the present disclosure.

Claims

1. Regarding the method of training a target estimation AI (artificial intelligence) model, A step of identifying a target node corresponding to the above target estimation AI model; A step of identifying nodes associated with the target node based on edge weights and edge types in a knowledge graph; A step of selecting data features based on nodes associated with the above target node; and A method comprising the step of training the target estimation AI model to output target information using selected data including selected features.

2. In claim 1, the step of identifying a node associated with the target node is: A step of generating a query representing the conditions of the edge weight and the conditions of the edge type; and A method comprising the step of searching for the relevant node in the knowledge graph using the above query.

3. In any one of paragraphs 1 to 2, A step of identifying whether the performance of the above target estimation AI model has reached the target performance; If the performance of the above target estimation AI model does not reach the target performance, a step of adjusting the conditions of the edge weights and the conditions of the edge types; A method further comprising the step of training the target estimation AI model based on conditions of adjusted edge weights and conditions of edge types.

4. A method according to any one of claims 1 to 3, further comprising the step of constructing the knowledge graph.

5. In claim 4, the step of constructing the knowledge graph is, A method comprising the step of constructing the knowledge graph using a guide document and an AI model.

6. In claim 5, the step of constructing the knowledge graph using the guide document and the AI ​​model is, A step of generating the knowledge graph by running the AI ​​model using the above guide document; A method comprising the step of updating the generated knowledge graph using the above data.

7. In any one of claims 4 to 6, the step of constructing the knowledge graph is Step of embedding the above data; A step of identifying the degree of correlation between data of the embedded data; and A method comprising the step of generating a knowledge graph based on the degree of correlation between identified data.

8. A computer-readable recording medium having a program recorded thereon for performing the method of any one of paragraphs 1 through 7 on a computer.

9. In an electronic device for training a target estimation AI (artificial intelligence) model, Memory in which a program or at least one instruction is stored; and It includes at least one processor, By the above at least one processor executing a program stored in the memory or at least one instruction, the electronic device, Identify the target node corresponding to the above target estimation AI model, and Identify nodes associated with the target node based on edge weights and edge types in the knowledge graph, and Select data features based on nodes related to the above target node, and An electronic device that trains the target estimation AI model to output target information using selected data including selected features.

10. In claim 9, the electronic device, by executing a program or at least one instruction stored in the memory by the at least one processor, identifies a node associated with the target node, Generate a query representing the conditions of the above edge weight and the conditions of the above edge type, and An electronic device that searches for the relevant node in the knowledge graph using the above query.

11. In any one of claims 9 to 10, by the at least one processor executing a program or at least one instruction stored in the memory, the electronic device, Identify whether the performance of the above target estimation AI model has reached the target performance, and If the performance of the above target estimation AI model does not reach the target performance, the conditions of the above edge weights and the conditions of the above edge types are adjusted, and An electronic device that trains the target estimation AI model based on conditions of adjusted edge weights and conditions of edge types.

12. An electronic device for constructing the knowledge graph according to claim 9.

13. In claim 12, the electronic device, by executing a program or at least one instruction stored in the memory by the at least one processor, constructs the knowledge graph, An electronic device that constructs the above knowledge graph using guide documents and AI models.

14. In claim 13, the electronic device constructs the knowledge graph using the guide document and the AI ​​model by executing a program or at least one instruction stored in the memory by the at least one processor. By running the AI ​​model using the above guide document, the knowledge graph is generated, and An electronic device that updates the generated knowledge graph using the above data.

15. In claim 12, the electronic device, by executing a program or at least one instruction stored in the memory by the at least one processor, constructs the knowledge graph, Embed the above data, and Identify the degree of correlation between data in the embedded data, and An electronic device that generates a knowledge graph based on the degree of correlation between identified data.

Images