KR102671887B1 - Method For Establishing Database For Global Value Chain For Parts Procurement and System Implementing Same - Google Patents

Abstract

Accordingly, one purpose of the present invention is to continuously collect information on parts that are identical to or replaceable with necessary parts in order to maintain up-to-date information on the supply and demand of necessary parts in the global value chain (GVC) for parts procurement. The goal is to provide a database that is stored in a structured form.
To this end, a system for building a database for a global value chain (GVC) for parts procurement according to an embodiment of the present invention uses artificial intelligence (AI) to collect data from a number of resources on the web based on request information. ) based knowledge mining robot; Unstructured (Raw) data archive for storing the collected data; A data analysis engine that analyzes data in the raw data archive, extracts correlations between two or more parts, sets one or more replacement parts for a specific part, and classifies and generates the data in a standardized form; a parts knowledge database (KDB) for storing standardized data generated by the data analysis engine; and a GVC KDB service that includes a user interface for enabling a user to search for parts and presenting search results from data in the parts KDB.

Description

Method for establishing a database for global value chain for parts procurement and system for implementing the same {Method For Establishing Database For Global Value Chain For Parts Procurement and System Implementing Same}

The present invention relates to a method of building a database for a global value chain (GVC) for parts procurement and a system for implementing the same. More specifically, crawling and data mining ( It relates to a method of building a database on the global value chain that allows easy search of replacement parts and parts supply and demand situations using data mining and scraping functions, and a system for implementing this.

The value chain is where added value is created when a company combines resources such as raw materials, labor, and capital to produce products or services. The global value chain combines the concept of globalization with the existing value chain. In a modern society where globalization has progressed rapidly, companies are expanding their value chains around the world to circumvent trade barriers or reduce manufacturing costs, and no company can do so. It has become difficult to independently produce goods and services. In particular, the automobile industry has a structure in which stable product production is inevitably disrupted when the supply and demand of components such as automotive semiconductors used in automobile manufacturing is not smooth.

Accordingly, the need for a system that can actively respond to changes in the global value chain (GVC), for example, the occurrence of bottlenecks in the global value chain due to parts or material supply risks, is increasing.
For example, a prior document related to the content disclosed in this application is Patent Registration No. 10-2711722.

Accordingly, one purpose of the present invention is to continuously collect information on parts that are identical to or replaceable with necessary parts in order to maintain up-to-date information on the supply and demand of necessary parts in the global value chain (GVC) for parts procurement. and providing a database that is stored in a structured form.

However, the problem to be solved by the present invention is not limited to the above-mentioned, and includes purposes that are not mentioned but can be clearly understood by those skilled in the art from the description below. can do.

Hereinafter, specific means for achieving the purpose of the present invention will be described.

A system for building a database for a global value chain (GVC) for parts procurement according to an embodiment of the present invention uses artificial intelligence (AI) to collect data from a number of resources on the web based on request information. AI) based knowledge mining robot; Unstructured (Raw) data archive for storing the collected data; a data analysis engine that analyzes data in the raw data archive, extracts a correlation between two or more parts, sets one or more replacement parts for a specific part, and classifies the data in a standardized form; a parts knowledge database (KDB) for storing standardized data generated by the data analysis engine; and a GVC KDB service that includes a user interface for enabling a user to search for parts and presenting search results from data in the parts KDB.

In the above-described system, the request information includes at least one of data on a specific part, data on replacement parts for the specific part, and data related to supply and demand of each part.

In the above-described system, the AI-based knowledge mining robot includes a classification model and a crawler, wherein the classification model classifies the specific resource on the web according to whether the specific resource is related to a part to be collected, The crawler is a module for performing data crawling on resources on the web.

In the above-described system, the classification model includes an artificial neural network module, and the artificial neural network module is pre-trained based on a sample web page for pre-learning and the request information.

In the above-described system, the classification model includes an artificial neural network module, the artificial neural network module becomes an agent, the crawler becomes an environment, and the classification model classifies resources. (Action), the collected data collected from the resource by the crawler becomes the state, and the operation of classifying the resource of the classification model, which is the action, in a direction that maximizes the reward as a result of the action. Reinforcement learning for the agent is performed repeatedly.

In the above-described system, the data analysis engine includes an analysis module for analyzing unstructured (Raw) data stored in the raw data archive, and a device for predicting the supply and demand situation for each component based on the unstructured data and past data. Includes prediction module.

In the above-described system, the data analysis engine classifies the data in a standardized form by analyzing the data in the raw data archive and converting the information of each part into the part name, part item code, manufacturing company, detailed specifications, and It includes classification into a form that includes the latest unit price.

In the above-described system, the data analysis engine further operates to cluster parts with similar properties through cluster analysis.

In the above-described system, the GVC KDB service allows data on parts corresponding to input information input by the user through the user interface and data on replacement parts for the parts to be presented on the user interface as search results.

A method for building a database for a global value chain (GVC) for parts procurement according to an embodiment of the present invention includes the steps of collecting data from a number of resources on the web based on requested information, and collecting storing the data in an unstructured (raw) data archive, and extracting a correlation between two or more parts by performing analysis on the data in the raw data archive - selecting a specific part based on the correlation. Setting one or more replacement parts for - classifying the data into a standardized form, storing the standardized data in a parts knowledge database (KDB), and allowing the user to search for parts and presenting a user interface for presenting search results from data of the parts KDB.

As described above, according to the present invention, the following effects are achieved.

According to the database construction system for the global value chain for parts procurement according to an embodiment of the present invention, users such as purchasing managers of automobile manufacturers not only easily search for necessary parts or replacement parts in a short time, By providing the supply and demand status of such parts, it not only reduces the user's search burden, but ultimately has the effect of reducing financial and time costs incurred by manufacturers due to parts supply problems.

However, the effects that can be obtained from the present invention are not limited to the effects mentioned above, and other effects not mentioned above will be clearly understood by those skilled in the art from the description below. You will be able to.

The following drawings attached to this specification illustrate preferred embodiments of the present invention and serve to further understand the technical idea of the present invention along with the detailed description of the invention, so the present invention is limited only to the matters described in such drawings. It should not be interpreted as such.
Figure 1 is a diagram schematically showing a system for building a database (DB) for a global value chain for parts procurement according to an embodiment of the present invention.
Figure 2 is a schematic diagram explaining an AI-based knowledge mining robot according to an embodiment of the present invention.
Figure 3 is an exemplary diagram showing a pre-learning process of a classification model included in an AI-based knowledge mining robot according to an embodiment of the present invention.
Figure 4 is an exemplary diagram showing a reinforcement learning process for an artificial neural network module of a classification model included in an AI-based knowledge mining robot according to an embodiment of the present invention.
Figure 5 is a diagram schematically showing the data analysis engine 30 according to an embodiment of the present invention.
Figure 6 is a diagram schematically showing a method of providing the results of an AI-based knowledge mining robot and data analysis engine to the GVC KDB service according to an embodiment of the present invention.
Figure 7 is a diagram showing an example of a search result screen of the user interface (UI) of the GVC KDB service 50 according to an embodiment of the present invention.
Figure 8 is an exemplary diagram showing a reinforcement learning process for a transformer-based artificial neural network module of a classification model included in an AI-based knowledge mining robot according to another embodiment of the present invention.
Figure 9 is a flowchart of a method for building a DB for GVC for parts procurement according to an embodiment of the present invention.

Hereinafter, with reference to the attached drawings, an embodiment by which a person skilled in the art can easily carry out the present invention will be described in detail. However, the present invention may be implemented in many different forms and is not limited to the examples described herein. Additionally, in explaining in detail the operating principle of a preferred embodiment of the present invention, if a detailed description of a related known function or configuration is judged to unnecessarily obscure the gist of the present invention, the detailed description will be omitted.

In addition, the same reference numerals are used for parts that perform similar functions and actions throughout the drawings. Throughout the specification, when a specific part is said to be connected to or above/below another part, this means not only directly connected or above/below, but also indirectly through other components in between. Also includes cases where it is connected or above/below.

Furthermore, when it is said that a part "includes" a certain element throughout the specification, this does not mean excluding other elements, but may further include other elements, unless specifically stated to the contrary.

Terms such as “first” and “second” may be used to describe various components, but these components should not be limited by these terms. That is, terms are used only for the purpose of distinguishing one component from another. For example, a first component may be named a second component without departing from the scope of the present invention, and similarly, the second component may also be named a first component. The descriptions “and/or” and “/” include a combination of a plurality of related stated items or any of a plurality of related stated items.

Additionally, unless specifically stated otherwise, the expression “singular” is used herein to include one or more entities. Finally, as used herein, “or” means “or” non-exclusively, unless specifically stated otherwise.

In addition, the terms used in the present invention are general terms that are currently widely used as much as possible while considering the functions in the present invention, but this may vary depending on the intention or precedent of a person working in the art, the emergence of new technology, etc. In addition, in certain cases, there are terms arbitrarily selected by the applicant, and in this case, the meaning will be described in detail in the description of the relevant invention. Therefore, the terms used in the present invention should be defined based on the meaning of the term and the overall content of the present invention, rather than simply the name of the term.

Overview of the system for building a database for the parts procurement global value chain

Figure 1 is a diagram schematically showing a system for building a database (DB) for a global value chain for parts procurement according to an embodiment of the present invention.

Referring to FIG. 1, the system 1 for building a database (DB) for the global value chain for parts procurement according to an embodiment of the present invention uses an artificial intelligence (AI)-based knowledge mining robot 10 to request information. Based on (15), data is collected from a number of resources on the Web (20), the collected data is stored in the unstructured (raw) data archive (25), and then the data analysis engine (30) is used to By performing an analysis, the correlation between two or more parts is extracted, one or more replacement parts are set for a specific part, the data is classified and created in a standardized form, and the standardized data is stored in a parts knowledge database (Knowledge Database). , KDB) 40, and providing the GVC KDB service 50 to the user based on the data stored in the part KDB 40. The user searches parts and/or parts-related websites (resources) through the user interface (UI) provided by the GVC KDB service 50 and retrieves search results from data from the parts KDB and/or raw data archive. can be presented.

The AI-based knowledge mining robot 10 classifies resources to determine whether to perform data collection and utilizes data crawling technology to search for resources on the Web 20 (for example, a website). ), hyperlink, data, etc.).

The AI-based knowledge mining robot 10 may be a software module implemented by the processor of a computing device or server device executing computer-executable instructions, may be composed of a separate hardware module, or may be composed of a combination of the two, and is limited thereto. I don't do it.

Data crawling technology is a technology that collects data by directly importing information from multiple resources on the public web. Rather than searching for information on the site for each search, data crawling technology continuously creates new web pages according to a pre-entered method in a computer program. It refers to a technology that collects unstructured data by sequentially performing tasks such as finding, comparing and analyzing existing results, finding new information based on the obtained results, and adding an index to that information. In other words, data crawling involves constantly moving around the web, collecting data from resources on the web, performing primary classification, such as indexing, on the collected data, and then storing it in a database, that is, data collection. , can refer to a series of processes of classification and storage.

According to one embodiment, the AI-based knowledge mining robot 10 continuously performs data crawling on multiple resources on the Web 20, thereby updating data on parts and materials to the latest. . This is because the supply and demand status of each part and material, as well as information on replacement parts or materials, are subject to change at any time.

Because there is a vast amount of information on the Web (20), there is a need to quickly and effectively perform data crawling. Accordingly, according to one embodiment, the AI-based knowledge mining robot 10 may be pre-trained based on artificial intelligence. This will be described in detail later with reference to FIG. 3.

In addition, in order to more clearly set the direction for data crawling of the AI-based knowledge mining robot 10, request information 15 to be obtained through data crawling may be provided as input to the AI-based knowledge mining robot 10. there is. This request information 15 may be preset by the administrator and/or user, or may be requested in real time by the user. The request information 15 may include at least one of data about a specific part (or parts list), data about replacement parts for a specific part, and data related to supply and demand of each part.

The raw data archive 25 may store data collected by the AI-based knowledge mining robot 10. For example, data stored in the raw data archive 25 may be stored after collecting data about a website and performing indexing on the collected data by a data crawler.

The data analysis engine 30 analyzes the data stored in the raw data archive 25 through scraping or data mining to extract meaningful relationships such as correlations and patterns between data, thereby extracting specific parts. You can set replaceable parts or similar parts with similar properties and generate part-related data, which is data about each part, in a standardized form. Part-related data may include not only data on the name, detailed specifications, price, and dimensions of each part, but also supply-demand-related data on the supply and demand situation, for example, the status of supply companies and demand companies, if collectable. And the generated parts-related data can be stored in the parts DB 42 of the parts knowledge database (KDB) 40.

Additionally, the data analysis engine 30 may generate demand prediction data based on past data 27 for parts (eg, past demand data, etc.) and part-related data. Such demand forecast data may be generated based on various data, such as data limited to a specific part, as well as historical data on parts similar to the part or substitute parts (27), and other data related to the parts described below, but are not limited thereto. No. The generated demand prediction data may be stored in the demand prediction DB 43 of the parts KDB 40.

Additionally, the data analysis engine 30 may generate other part-related data related to each part and store it in the other DB 44 of the part KDB 40. Other data related to parts may include, for example, risk data on risk factors that may affect the supply and demand of parts, supply chain disruption scenario data based on risk factors, etc. Other data related to these parts include cases where the supply chain has collapsed or bottlenecks have occurred, and the time taken to recover from such supply chain collapse and bottlenecks, etc., based on historical data on the supply and demand of necessary parts or replacement parts (27). It can be created by considering various factors.

The parts KDB 40 may store data generated by the data analysis engine 30, i.e., part-related data, demand forecast data, and other part-related data, and in some cases from the raw data archive 25. The data can be acquired and stored as is.

The global value chain (GVC) KDB service 50 can provide data about parts to users through a user interface (UI) based on standardized data stored in the parts KDB 40. The GVC KDB service 50 may be provided by a dedicated server or in a cloud form, but is not limited thereto.

Through the user interface (UI) of the GVC KDB service (50), users such as purchasing managers at automobile manufacturers can input key keywords such as the name, image, or purpose of the required part they wish to search for as input information. , GVC KDB service 50 can search for corresponding data for the part KDB 40 based on this input information. This data retrieval process can be performed more effectively by applying artificial neural networks related to image processing or keyword search. Accordingly, even if the user does not know the exact part name, it is possible to search for the necessary part that the user wants.

Through the GVC KDB service (50) established in this way, users such as purchasing managers of automobile manufacturers can easily search for necessary parts, replacement parts, or similar parts in a short time, as well as provide the supply and demand status of such parts. By doing so, it not only reduces the user's search burden, but ultimately reduces the financial and time costs incurred by manufacturers due to parts supply problems.

Hereinafter, each component of the system for building a database (DB) for GVC for parts procurement according to an embodiment of the present invention will be described in detail.

AI-based knowledge mining robot

Figure 2 is a schematic diagram explaining an AI-based knowledge mining robot according to an embodiment of the present invention.

Referring to FIG. 2, the AI-based knowledge mining robot 10 according to an embodiment of the present invention may include a classification model 12 and a crawler 14.

The classification model 12 is a model for classifying a number of resources such as websites, web pages, hyperlinks, and data materials on the Web 20 based on the request information 15. According to one embodiment, the classification model 12 goes through a pre-learning process based on the request information 15 and sample web pages for pre-learning to determine whether a specific resource on the Web 20 is related to the part to be collected. Resources can be classified according to availability. Additionally, the classification model 12 can cause the crawler 14 to perform or not perform data collection according to the resource classification results. In other words, if the resource is classified (i.e. determined) by the classification model 12 as being related to a required part, the crawler 14 is made to collect data from the resource, and if it is classified as an unrelated resource, It may be designed not to cause the crawler 14 to perform data collection on the resource.

The crawler 14 may be a software module that actually performs data crawling. According to one example, when a specific resource is classified by the classification model 12 as a resource related to a required part, the crawler 14 can access the resource, collect data, and index the collected data. there is. It then searches for other resources related to the current resource (e.g., associated URLs, hyperlinks, etc.) and, if the other explored resources are classified by the classification model as a resource associated with the required part, then data is collected for that resource. The task of collecting and indexing can be repeated. In other words, the crawler 14 collects data by crawling through various resources, creates an index of where the data is, and stores it in the raw data archive 25.

According to one embodiment, the crawler 14 may store at least one of the page original text, word set, part image, and classification accuracy of a specific resource in the raw data archive 25, and create an index for this, and this process This can be done by moving around multiple resources. Here, classification accuracy may be a quantification of whether a specific resource is related to a necessary part according to a classification model. If the classification accuracy is higher than the threshold, the specific resource can be classified as related to the necessary part, and if it is less than the threshold, it can be classified as unrelated. there is.

According to one embodiment, when request information 15 is set in advance to build a database, a parts list may be included in the request information and provided to the AI-based knowledge mining robot 10. For example, if a car parts list is included in the request information, data crawling for the car parts list can be continuously performed to ensure that the latest data is maintained in the parts KDB 40, and when necessary parts are added, In this case, the parts KDB 40 can be updated by updating the parts list included in the request information.

Additionally, in some cases, a user may create request information in real time for parts that were not included in the request information. In this case, the AI-based knowledge mining robot 10 can be set to perform data crawling in real time.

Meanwhile, even if only a specific part is input by a user or manager as request information 15 according to one embodiment, the AI-based knowledge mining robot 10 provides not only data about the corresponding part, but also information about replacement parts for the corresponding part. It may be set to perform data crawling on data including at least one of data, data related to supply and demand of the corresponding parts and/or replacement parts (e.g., supply companies, supply and demand situations, and expected demand companies, etc.). Alternatively, in another embodiment, even if only specific parts are entered into the request information 15, the request information 15 is stored in the system 1 to include items such as replacement parts, parts supply companies, supply and demand situations, and expected demand companies. It may be rewritten and provided as input to the AI-based knowledge mining robot 10. That is, the method of creating the request information 15 according to the embodiments of the present invention is not limited, and various modifications may be made.

Meanwhile, after going through the process of collecting and indexing data from resources on the Web 20 by the crawler 14 as described above, the classification model 12 of the AI-based knowledge mining robot 10 is created by the crawler ( The effectiveness of this data collection process can be further improved through reinforcement learning, which evaluates and compensates for the collected data collected by 14). This will be described in detail later in relation to FIG. 4 .

Pre-learning process of AI-based knowledge mining robot

Figure 3 is an exemplary diagram showing a pre-learning process of a classification model included in an AI-based knowledge mining robot according to an embodiment of the present invention.

Referring to FIG. 3, in the pre-learning session of the artificial neural network module of the classification model 12 included in the AI-based knowledge mining robot 10, the artificial neural network module serves as data for pre-learning, including a sample web page for pre-learning 22 and Request information (15) can be input.

The artificial neural network module may be at least one of artificial neural network modules such as Transformer, LSTM (Long-Short Term Memory), RNN (Recurrent Neural Network), CNN (Convolution Neural Network), GAN (Generative Adversarial Network), and AE (AutoEncoder). And machine learning such as multi-layer perceptron, Naive-Bayesian Classification, and Random Forest Classification may be used.

At this time, the sample web page 22 for prior learning is selected for prior learning, and may be selected from among web pages selected in advance by the administrator, web pages created suitable for prior learning, and web pages that previously existed. , web pages containing data corresponding to the request information 15 and web pages without such data may be mixed. However, the configuration of the sample web page 22 for prior learning may vary and is not limited thereto.

Additionally, the request information 15 may correspond to information to be provided through the GVC KDB service 50. The request information 15 may include, for example, data on specific parts, data on replacement parts for specific parts, etc. , It may include at least one of supply-demand related data for each part (e.g., status of supply companies and expected demand companies, demand status of products related to the part, etc.).

In this way, by receiving the sample web page 22 and request information 15 for pre-learning, the artificial neural network module of the classification model 12 can output the collected data.

And the collected data output by the artificial neural network module in the pre-learning session can be compared with ground truth, for example, data that actually corresponds to requested information in a sample web page. This ground truth may be preset by the administrator, and as described above, data corresponding to the request information 15 may or may not exist in the sample web page 22.

Afterwards, the artificial neural network module can be pre-trained so that the difference between the collected data output by the artificial neural network module and the ground truth, that is, the error (-∑ y _i log p _i ), is minimized. That is, the artificial neural network module can adopt a method of updating the weight of the hidden layer based on such errors. For example, sample web pages and request information for pre-learning are input to each node of the input layer such as x1, The collected data predicted based on the function is output to the output layer called y1. And based on the error (-∑ y _i log p _i ) between the collected data and the ground truth, the weight of the artificial neural network module can be updated through back propagation.

Referring again to Figure 2, after pre-training of the artificial neural network module of the classification model 12 is completed as described in relation to Figure 3, when the AI-based knowledge mining robot 10 actually collects data from the resource (i.e. , inference session) determines whether the request information 15 and a specific resource are provided to the AI-based knowledge mining robot 10, and the classification model 12 of the AI-based knowledge mining robot 10 determines whether to proceed with the collection of that resource. is determined by a pre-trained artificial neural network, and when it is decided to proceed with collection, data can be collected by the crawler 14.

Reinforcement learning process for AI-based knowledge mining robot

Figure 4 is an exemplary diagram showing a reinforcement learning process for an artificial neural network module of a classification model included in an AI-based knowledge mining robot according to an embodiment of the present invention.

Reinforcement Learning (RL) is an area of machine learning (ML) in which an agent defined within a given environment recognizes the current state and selects a reward among actions that can be selected. It corresponds to a learning method that selects actions or action sequences that maximize (reward).

An agent refers to a subject that observes the state and selects an action. The environment refers to the state that changes through interaction and rewards are given when the agent acts. The state refers to information that represents the current situation of the environment, and the action refers to the current situation. It refers to what the agent does in the state, and the reward corresponds to information expressing whether the action is good or bad. Here, the reward must be well-matched to the single or multiple goals for improvement, and the state must well represent the environment.

Referring to FIG. 4, in the reinforcement learning process, the artificial neural network module of the classification model 12 becomes the agent, the crawler 14 becomes the environment, and the classification model 12 classifies the resource. The action (i.e., the action of determining whether the resource is associated with a necessary part) becomes an action (a _t ), and the collected data collected from the resource by the crawler 14 is a state (State, S _t ). It can be recognized. In this case, for the artificial neural network module of the classification model 12, which is an agent, the crawler 14 recognizes the collected data collected from the resource as a state (State, _St ) and selectable action (Action) (a _t ), that is, reinforcement learning may be performed to generate the resource classification operation of the classification model 12 to maximize reward (R _t ).

More specifically, the artificial neural network module of the classification model 12, which is an agent, observes the implemented state (i.e., the collected data collected from the resource by the crawler 14) and performs an action (a _t ) is the entity that selects (i.e., the resource classification operation of the classification model 12), and as a result of the action, positive feedback on the data collected by the crawler occurs, for example, a search behavior for data extracted from a specific resource. Alternatively, if explicit feedback such as "Like" occurs through the UI of the GVC KDB service, a relatively large reward is given, and if there is no feedback, a small reward is given, and if there is a negative evaluation or feedback. Compensation can be designed in such a way that compensation is not given in cases where compensation is not given. Reinforcement learning for the agent can be achieved by repeating the process of the agent taking action in a direction that maximizes the reward, that is, classifying the resources of the classification model 12. Accordingly, the classification model 12, which is an agent, can continuously perform optimized resource classification.

According to one embodiment of the present invention, the artificial neural network module of the classification model 12 not only undergoes a pre-learning process based on sample data, but also continuously performs a valid resource classification operation through such reinforcement learning even after pre-training. Since it is designed to do so, it has the effect of being able to efficiently crawl and collect data from the vast resources of the Web (20).

data analysis engine

Figure 5 is a diagram schematically showing the data analysis engine 30 according to an embodiment of the present invention.

Referring to FIG. 5, the data analysis engine 30 includes an analysis module 32 for analyzing unstructured (Raw) data stored in the raw data archive 25, and unstructured (Raw) data stored in the raw data archive 25. and a prediction module 34 for predicting the supply and demand situation for each component based on past data 27. According to one embodiment, the analysis module 32 and the prediction module 34 may extract meaningful relationships such as correlations and patterns between data by analyzing them through scraping or data mining.

More specifically, the analysis module 32 may classify information about each part into a standardized form based on unstructured (raw) data stored in the raw data archive 25. According to one embodiment, the analysis module 32 may classify information on each part into part name, part item code, manufacturing company, detailed specification, and latest unit price. However, the specific information classification form can be modified in various ways.

Additionally, the analysis module 32 may perform correlation analysis to determine the correlation between two or more parts based on unstructured (raw) data stored in the raw data archive 25. In other words, the analysis module 32 can set one or more replacement parts that can be replaced (or are predicted to be replaced) for a specific part through this correlation analysis.

Additionally, the analysis module 32 may cluster parts with similar properties through cluster analysis. Here, the attributes used in cluster analysis may include, but are not limited to, at least one of the names, uses, dimensions, and manufacturers of the parts. By clustering parts with similar properties, similar parts can be set and users can easily search for parts through data on similar parts even if they do not know the exact information about the part they want.

The prediction module 34 obtains historical data on parts from the historical data DB 27, and provides the supply and demand situation for each part based on this historical data and/or unstructured (raw) data stored in the raw data archive 25. By predicting, you can generate demand forecast related data and/or supply forecast related data. As an example, the prediction module 34 can obtain numerical data related to the demand or supply of parts from unstructured (raw) data in the raw data archive 25, and events that affect the supply and demand of parts from past data, etc. By identifying and reflecting this, data related to demand forecast and/or data related to supply forecast can be generated. Furthermore, such supply forecast-related data and demand forecast-related data may be generated based on past data related to not only the corresponding part, but also replacement parts for the corresponding part (or parts similar to the corresponding part), and in this process, the analysis module (32 ) may be used.

Meanwhile, the data analysis engine 30 may generate other data related to a specific part or replacement part, such as other data related to the part, such as the risk of hazards that may affect the supply and demand of the part. It can include factor data, supply chain disruption scenario data based on risk factors, etc. Other data related to these parts is based on historical data on the demand and supply of specific parts or replacement parts, and various factors such as cases where supply chain disruptions or bottlenecks have occurred, and the time taken to recover from such supply chain disruptions and bottlenecks. It can be created with consideration.

Data generated by the data analysis engine 30, i.e., part-related data, demand forecast data (and supply forecast data), and other part-related data, may be stored in the part KDB 40.

GVC KDB Service

Figure 6 is a diagram schematically showing a method of providing the results of an AI-based knowledge mining robot and data analysis engine to the GVC KDB service according to an embodiment of the present invention.

Referring to FIG. 6, the GVC KDB service 50 may provide a user interface (UI) that allows users to search sites and search for parts. The GVC KDB service 50 may be provided by a dedicated server or in a cloud form, but is not limited thereto.

According to one embodiment, the data collected by the AI-based knowledge mining robot 10 using data crawling techniques, etc. and stored in the raw data archive 25 is the resource from which the data was collected (e.g., website, hyper links, etc.), so when a user performs a “site search”, the user interface (UI) of the GVC KDB service 50 is provided with index information from the raw data archive 25. It can be.

In addition, according to one embodiment, the parts KDB 40 stores data on necessary parts and replacement parts in a standardized form according to correlation and similarity between parts through data mining or scraping. When performing a “search,” the standardized data of the part KDB 40 may be provided to the user interface (UI) of the GVC KDB service 50.

Meanwhile, when searching for parts, a user, such as a purchasing manager of an automobile manufacturer, enters key keywords such as the name, image, or purpose of the required part they wish to search through the user interface (UI) of the GVC KDB service 50. It can be entered as information, and the GVC KDB service 50 can search for corresponding data for the part KDB 40 based on this input information and present the results on the user interface (UI). Here, when the user inputs an image as input information, a process of finding the same or similar image from the component KDB 40 using AI-based image processing technology may be involved.

As described above, in the part KDB 40, one or more replacement parts that can be replaced (or are predicted to be replaced) for the necessary parts are set through correlation analysis that determines the correlation between two or more parts, and have similar properties. Since the parts are also clustered through cluster analysis, the user interface (UI) may present data on at least one of necessary parts, replacement parts, and similar parts corresponding to the input information entered by the user, and optionally, Supply-demand related data and demand forecast data on the supply and demand situation of each component can also be presented together or separately.

Meanwhile, in the process of providing search results in response to a user requesting a site search or parts search through the GVC KDB service 50, the data crawling and data mining process is performed in real time or in a pre-built parts KDB The data in (40) and data acquired in real time can also be used in parallel.

Figure 7 is a diagram showing an example of a search result screen of the user interface (UI) of the GVC KDB service 50 according to an embodiment of the present invention.

According to the non-limiting example shown in FIG. 7, when a user searches for a part called “Eccentric Reducer,” the GVC KDB service 50 according to an embodiment of the present invention searches for an eccentric accelerator (Eccentric Reducer). A list of companies, item codes (Code), detailed specifications (Spec.), and recent unit prices of replacement parts or similar parts determined by the data analysis engine 30 as being replaceable or similar to the corresponding parts or eccentric accelerator. It can be presented as Even if the part name (product name), item code, detailed specifications, and unit price are different for each company, the user can easily identify replacement parts and/or similar parts through data mining by the data analysis engine 30, so the user can easily identify replacement parts and/or similar parts. Convenience can be improved. At this time, as illustrated in FIG. 7, the detailed specifications of some parts may contain typos, etc., and the GVC KDB service 50 performs corrections for these obvious data errors and provides them to the user interface (UI). It can also be set.

Furthermore, although not shown in FIG. 7, forecast data such as the supply and demand status of each component can be provided to the user interface (UI), thereby allowing the user to create a production schedule suitable for the supply and demand status of the corresponding part and replacement parts. By enabling planning, a unique effect is created that ultimately reduces the financial and time costs incurred by manufacturers due to parts supply problems.

Reinforcement learning for transformer-based artificial neural network modules for classification models

Figure 8 is an exemplary diagram showing a reinforcement learning process for a transformer-based artificial neural network module of a classification model included in an AI-based knowledge mining robot according to another embodiment of the present invention.

While the example of FIG. 4 explains general artificial neural network-based reinforcement learning, FIG. 8 is different in that it involves performing reinforcement learning through a specific artificial neural network module called a transformer. The transformer-based artificial neural network module shown in Figure 8 is designed to make optimal decisions.

Referring to FIG. 8, the classification model 12 may include a transformer-based artificial neural network module 100, and through reinforcement learning using the transformer-based artificial neural network module 100, the classification model 12 may use resource classification data. (a _t ) can be output.

The transformer-based artificial neural network module 100 includes the future expected reward (R _t ), the collection data (S _t ) collected from the resource by the crawler as the environment, and the classification that is the result previously output by the artificial neural network module 100. Past resource classification data (a _t-1 ) of the model may be input.

Here, the future expected reward (R _t ) represents the sum of rewards expected in the future by taking a specific action in the classification model 12 as Return to go, that is, by the classification model 12 performing a resource classification operation. For example, positive feedback on the crawler's collected data based on the resource classification behavior of the classification model, e.g., search behavior for data extracted from a specific resource, or explicit feedback such as "like", etc. Compensation can be designed in such a way that a relatively large reward is given when it occurs, a small reward is given when there is no feedback, and no reward is given when a negative evaluation or feedback exists.

The past resource classification data (a _t-1 ) of the classification model is the output previously output from the transformer-based artificial neural network module 100 when input and output to the transformer-based artificial neural network module 100 are performed in real time or time series. represents them.

In this way, when the future expected reward (R _t ), the collected data (S _t ), and the past resource classification data (a _t-1 ) of the classification model are input to the transformer-based artificial neural network module 100, the transformer-based artificial neural network module 100 ) can output the optimal resource classification data (a _t ) selected at the current time in the classification model (12).

Hereinafter, a process of processing input data and outputting a result value through the transformer-based artificial neural network module 100 of a classification model according to another embodiment of the present invention will be described.

Referring to Figure 8, according to one embodiment of the present invention, the transformer-based artificial neural network module 100 of the classification model includes a first block 100a (e.g., encoder) and a second block 100b (e.g. For example, a decoder) may be included.

Future expected compensation (R _t ) and collected data (S _t ) may be input to the first block 100a of the artificial neural network module 100. In addition, the future expected compensation (R _t ) and the collected data (S _t ) input in this way are input to the embedding module of the first block 100a and vectorized, so that the embedding information is output and the bundle of output embedding information is divided into different linear layers. Through linear embedding, it is composed of Q (Query feature), K (Key feature), and V (Value feature) including feature dimensions, and then Q (Query feature), K (Key feature), and V (Value feature) are multi-processed. -This can be done with the input data of the head Attention Layer. Q, K, and V can be generated from embedding information through different linear embeddings. Q and K are input as input data of the 1st MatMul operation, calculate the similarity for all K with respect to Q through scale and softmax operation, and output a similarity vector between a series of embedding information, and this similarity vector and V are the 2nd MatMul It can be entered as input data for the operation. Additionally, the Multi-head Attention Layer can output attention information based on the input Q, K, and V values and input it to the Feed Forward Layer. These Multi-head Attention Layers and Feed Forward Layers can be repeated N times, and the repetition can be done in such a way that the encoded information output from the previous Feed Forward Layer becomes the input to the next Multi-head Attention Layer. And the last Feed Forward Layer can output final encoding information, and this final encoding information is K (Key), V to the second Multi-head Attention Layer of the second block (100b) of the transformer-based artificial neural network module (100). (Value) Can be entered as a value.

Additionally, past resource classification data (a _t-1 ) of the classification model may be input to the second block 100b of the transformer-based artificial neural network module 100. Here, the past resource classification data (a _t-1 ) of the classification model is input to the embedding module of the second block 100b and vectorized, so that embedding information is output, and the bundle of output embedding information is linearly embedded through different linear layers. After consisting of Q (Query feature), K (Key feature), and V (Value feature) including feature dimensions, Q (Query feature), K (Key feature), and V (Value feature) are used as the first Multi-head Attention This can be done with the input data of the layer. And the first Multi-head Attention Layer can output a Q (Query) value, and this Q value is input to the second Multi-head Attention Layer. The second multi-head attention layer can receive the final encoding information from the first block 100a as K and V values. And the second multi-head attention layer can output attention information based on the input Q, K, and V values and input it to the feed forward layer. These first Multi-head Attention Layer, second Multi-head Attention Layer, and Feed Forward Layer can be repeated N times, and the decoding information output from the previous Feed Forward Layer is the input of the next first Multi-head Attention Layer. Repetition can be done in this way. And the last Feed Forward Layer can output the final decoding information. This final decoding information passes through the Linear Layer and Softmax Layer, and the transformer artificial neural network module 100 selects the optimal resource classification data (a _t ) can be output.

However, the specific implementation method of the transformer-based artificial neural network module 100 as described above may vary in various ways, and is not limited thereto.

Method for building a database for GVC for parts procurement

Figure 9 is a flowchart of a method for building a DB for GVC for parts procurement according to an embodiment of the present invention.

Referring to FIG. 9, first, in step S10, data can be collected from a number of resources on the web based on request information. This data collection can be performed by AI-based knowledge mining robots using data crawling technology. Data crawling technology is a technology that collects data by directly retrieving information from resources such as multiple websites, hyperlinks, and data on the public web, and provides information about the site for each search. Instead of searching, the process of constantly finding new web pages according to a pre-entered method of a computer program, comparing and analyzing existing results, finding new information based on the obtained results, and adding an index to that information. is performed sequentially to collect unstructured data.

In step S20, the collected data may be stored in an unstructured (raw) data archive. In other words, the data stored in the raw data archive may be stored after collecting data about the website through data crawling and performing indexing on the collected data.

In step S30, analysis of data in the raw data archive may be performed to extract correlations between two or more parts, and one or more replacement parts for a specific part may be set based on the extracted correlations. Additionally, data can be classified into a standardized form. According to one embodiment, the information on each part may be classified into part name, part item code, manufacturing company, detailed specifications, and latest unit price.

In step S40, standardized data may be stored in a parts knowledge database (KDB).

In step S50, a user interface may be presented to enable the user to search for parts and to present search results from data in the parts KDB.

computer-readable recording medium

It is obvious that each step or operation of the method according to the embodiments of the present invention can be performed by a computer including one or more processors by executing a computer program stored in a computer-readable recording medium.

The computer-executable instructions stored in the above-described recording medium can be implemented through a computer program programmed to perform each corresponding step, and such computer program may be stored in a computer-readable recording medium and may be operated by a processor. It is feasible. A computer-readable recording medium may be a non-transitory readable medium. At this time, a non-transitory readable medium refers to a medium that stores data semi-permanently and can be read by a device, rather than a medium that stores data for a short period of time, such as registers, caches, and memories. Specifically, programs for performing the various methods described above include semiconductor memory devices such as erasable programmable read only memory (EPROM), electrically erasable programmable read only memory (EEPROM), and flash memory devices, internal hard disks, and removable memory devices. It may be provided and stored in non-transitory readable media such as magnetic disks such as disks, optical-magnetic disks, and non-volatile memory including CD-ROM and DVD-ROM disks.

Methods according to various examples disclosed in this document may be provided and included in a computer program product. The computer program product may be distributed in the form of a machine-readable storage medium (e.g. compact disc read only memory (CD-ROM)) or online through an application store (e.g. Play Store ^™ ). In the case of online distribution, at least a portion of the computer program product may be at least temporarily stored or created temporarily in a storage medium such as the memory of a manufacturer's server, an application store server, or a relay server.

As described above, a person skilled in the art to which the present invention pertains will understand that the present invention can be implemented in other specific forms without changing its technical idea or essential features. Therefore, the above-described embodiments should be understood in all respects as illustrative and not restrictive. The scope of the present invention is indicated by the claims described later rather than the detailed description, and all changes or modified forms derived from the meaning and scope of the claims and the equivalent concept should be construed as being included in the scope of the present invention.

The features and advantages described herein are not all-inclusive, and many additional features and advantages will become apparent to those skilled in the art upon consideration of the drawings, specification, and claims. Moreover, it should be noted that the language used herein has been selected primarily for readability and teaching purposes and may not be selected to describe or limit the subject matter of the invention.

The foregoing description of embodiments of the invention has been presented for purposes of illustration. It is not intended to be exhaustive or to limit the invention to the precise form disclosed. Those skilled in the art will appreciate that many modifications and variations are possible in light of the above disclosure.

Therefore, the scope of the present invention is not limited by the detailed description, but by any claims of the application based thereon. Accordingly, the disclosure of embodiments of the invention is illustrative and does not limit the scope of the invention as set forth in the claims below.

1: System for building a database (DB) for the global value chain for parts procurement
10: Artificial intelligence (AI)-based knowledge mining robot
15: Request information
20: Web
25: Raw data archive
30: Data analysis engine
40: Parts Knowledge Database (KDB)
42: Part DB
43: Demand forecast DB
44: Other DB
50: GVC KDB service

Claims (10)

As a system for building a database on the Global Value Chain (GVC) for parts procurement,
An artificial intelligence (AI)-based knowledge mining robot to collect data from multiple resources on the web based on requested information;
Unstructured (Raw) data archive for storing the collected data;
By performing analysis on the data in the raw data archive, correlation between two or more parts is extracted, one or more replacement parts that are replaceable or predicted to be replaceable for a specific part are set, and cluster analysis is performed to determine the correlation between two or more parts. a data analysis engine that clusters parts to set similar parts and classifies the data into a standardized form to generate standardized data;
a parts knowledge database (KDB) for storing standardized data generated by the data analysis engine; and
GVC KDB service that includes a user interface to enable users to search for parts and to present search results from data in the parts KDB
Including,
The search results presented through the user interface of the GVC KDB service include at least one of a necessary part corresponding to the input information entered by the user, a replacement part for the necessary part, and a similar part,
The AI-based knowledge mining robot includes a classification model and a crawler,
The classification model classifies a plurality of resources on the web according to whether they are related to the necessary part, and the classification model has a classification accuracy that quantifies whether a specific resource among the plurality of resources on the web is associated with the necessary part. If is higher than the threshold, the specific resource is classified as being associated with the required part,
The crawler is a module for performing data crawling only on resources classified as being related to the necessary parts according to the classification model,
The classification model includes an artificial neural network module, and the artificial neural network module is pre-trained based on a sample web page for pre-learning and the request information,
The data analysis engine is,
An analysis module for analyzing unstructured (raw) data stored in the raw data archive,
A prediction module for predicting the supply and demand situation of the necessary parts and the supply and demand situation for at least one of the replacement parts and the similar parts based on the unstructured data and past data,
The unstructured data includes numerical data on demand or supply of at least one of the necessary part, the replacement part, and the similar part,
The historical data includes at least instances where the supply chain for at least one of the necessary parts, the replacement parts, and the similar parts has collapsed or a bottleneck has occurred and the period of time taken to recover from the supply chain collapse and bottleneck.
Database construction system for GVC.
According to claim 1,
The request information includes at least one of data about a specific part, data about replacement parts for the specific part, and data related to supply and demand of each part.
Database construction system for GVC.
delete delete According to claim 1,
The classification model includes an artificial neural network module, the artificial neural network module becomes an agent, the crawler becomes an environment, and the operation of the classification model to classify resources becomes an action, The collected data collected from the resource by the crawler becomes a state, and the operation of classifying the resource of the classification model, which is the action, is repeated in a direction that maximizes the reward as a result of the action, thereby providing information about the agent. Reinforcement learning is performed
Database construction system for GVC.
delete According to claim 1,
The data analysis engine classifies the data in a standardized form by analyzing the data in the raw data archive and converting the information of each part into a form including the part name, part item code, manufacturing company, detailed specifications, and the latest unit price. including classification as
Database construction system for GVC.
delete delete A method for building a database on a global value chain (GVC) for parts procurement, where the method is performed by a processor,
collecting unstructured data from a plurality of resources on the web based on the requested information;
storing the collected unstructured data in an unstructured (raw) data archive;
performing analysis on the unstructured data in the raw data archive to extract a correlation between two or more parts, and setting one or more replacement parts for a specific part based on the correlation;
Clustering parts with similar properties through cluster analysis of the unstructured data in the raw data archive to set similar parts;
predicting a supply and demand situation for necessary parts and a supply and demand situation for at least one of the replacement parts and the similar parts based on the unstructured data and past data;
Classifying the unstructured data into a standardized form to generate standardized data;
Storing the standardized data in a parts knowledge database (KDB);
Allowing a user to search for parts and presenting a user interface for presenting search results from data of the parts KDB - the search results presented through the user interface correspond to input information entered by the user Includes at least one of the necessary parts and replacement parts and similar parts for the necessary parts -
Including,
The step of collecting unstructured data from a number of resources on the web based on the request information is performed by an AI-based knowledge mining robot,
The AI-based knowledge mining robot includes a classification model and a crawler,
The classification model classifies a plurality of resources on the web according to whether they are related to the necessary part, and the classification model has a classification accuracy that quantifies whether a specific resource among the plurality of resources on the web is associated with the necessary part. If is higher than the threshold, the specific resource is classified as being associated with the required part,
The crawler is a module for performing data crawling only on resources classified as being related to the necessary parts according to the classification model,
The classification model includes an artificial neural network module, and the artificial neural network module is pre-trained based on a sample web page for pre-learning and the request information,
The unstructured data includes numerical data on demand or supply of at least one of the necessary part, the replacement part, and the similar part,
The historical data includes at least instances where the supply chain for at least one of the necessary parts, the replacement parts, and the similar parts has collapsed or a bottleneck has occurred and the period of time taken to recover from the supply chain collapse and bottleneck.
How to build a database for GVC.