Methods and systems for matching public technology and technology users.

Abstract

A method for matching public technologies with technology users will be provided. [Solution] A method for matching public technology and technology consumers, performed by a computing system according to several embodiments, may include the steps of: acquiring data related to public technology; applying the acquired data to a first machine learning model to acquire technology feature data related to the public technology, wherein the first machine learning model is configured to output the technology feature data based on the data related to the public technology; and matching the public technology with one or more technology consumers based on the technology feature data and a plurality of consumer feature data stored in a database.

Description

Technical Field

[0001] The present disclosure relates to a method for matching public technology with technology demanders. More specifically, it relates to a method and system for matching public technology with technology demanders so as to increase the likelihood of technology transactions.

Background Art

[0002] There is a platform for technology transactions between public technology and enterprises that are technology demanders, and technology transactions and technology transfers are carried out on such platforms. However, the productivity of public R&D research and the success rate of technology commercialization with respect to the government's R&D investment rate are at a low level, and it is difficult to connect public technology with technology demanders.

[0003] One of the main reasons why it is difficult to connect public technology with technology demanders is the difference in the information gap between public technology and technology demanders. Existing transaction platforms between public technology and technology demanders either provide only public technology-centered information or rely on an expert-based technology search method for matching between public technology and technology demanders, and there are limitations from the perspective of technology demanders.

[0004] Therefore, there is a need for a method for matching public technology with technology demanders that can increase the success rate of technology transactions.

Prior Art Documents

Patent Documents

[0005] Korean Patent Publication No. 10-2020-0067778 (published on June 12, 2020)

Summary of the Invention

Problems to be Solved by the Invention

[0006] Some technical problems to be solved by the embodiments of the present disclosure are to provide a method and system for improving the success rate of technology transactions between public technology and technology demanders using a machine learning model.

[0007] The technical problem that some embodiments of this disclosure seek to solve is to provide methods and systems that improve speed in matching public technology with technology consumers.

[0008] The technical problem that some embodiments of this disclosure aim to solve is to provide a method and system for lightweighting machine learning models used in matching public technologies with technology consumers.

[0009] The technical challenges of this disclosure are not limited to those mentioned above, and other technical challenges not mentioned above will be clearly understood by a person of ordinary skill in the art of this disclosure from the following description. [Means for solving the problem]

[0010] A method for matching public technology with technology consumers, performed by a computing system according to one embodiment of the present disclosure for solving the aforementioned technical problems, may include the steps of: acquiring data related to public technology; applying the acquired data to a first machine learning model to acquire technology feature data related to the public technology, wherein the first machine learning model is configured to output the technology feature data based on the data related to the public technology; and matching the public technology with one or more technology consumers based on the technology feature data and a plurality of consumer feature data stored in a database.

[0011] Furthermore, the matching method may further include the steps of acquiring multiple data related to multiple technology consumers before the step of acquiring data related to the public technology, applying the multiple data related to the multiple technology consumers to a second machine learning model to acquire the multiple consumer feature data related to the multiple technology consumers, wherein the second machine learning model is configured to output the consumer feature data based on the data related to the technology consumers, and storing the acquired multiple consumer feature data in the database.

[0012] Furthermore, the step of matching the public technology with one or more technology consumers may include the step of applying the technology feature data and the plurality of consumer feature data to a third machine learning model to obtain a score related to tradability for each of the plurality of consumer feature data, wherein the third machine learning model is configured to output a score related to the tradability of technology between the technology feature data and the consumer feature data; the step of extracting one or more consumer feature data from the plurality of consumer feature data based on the obtained scores; and the step of determining one or more technology consumers that match the public technology based on the extracted one or more consumer feature data.

[0013] Furthermore, the step of extracting one or more consumer characteristic data from the plurality of consumer characteristic data may include the step of extracting one or more consumer characteristic data associated with a score that falls within a predetermined rank from among the scores for each of the plurality of consumer characteristic data.

[0014] Furthermore, the step of extracting one or more consumer characteristic data from the plurality of consumer characteristic data may include the step of extracting one or more consumer characteristic data associated with a score equal to or greater than a predetermined critical value among the scores for each of the plurality of consumer characteristic data.

[0015] Furthermore, the step of matching the public technology with one or more technology users includes a step of converting the acquired score into a normalized value within a predetermined range, and the matching method may further include a step of transmitting information related to the one or more technology users matched with the public technology to a user terminal after the step of matching the public technology with one or more technology users. In such a case, the information related to the technology users may include the normalized value.

[0016] A method for matching public technologies with technology consumers, performed by a computing system according to one embodiment of the present disclosure for solving the aforementioned technical problems, may include the steps of: acquiring data related to technology consumers; applying the acquired data to a first machine learning model to acquire consumer feature data related to the technology consumers, wherein the first machine learning model is configured to output the consumer feature data based on the data related to the technology consumers; and matching the technology consumers with one or more public technologies based on the consumer feature data and a plurality of technology feature data stored in a database.

[0017] A method for constructing a database for matching public technologies with technology consumers, performed by a computing system according to one embodiment of the present disclosure to solve the aforementioned technical problems, may include the steps of: acquiring data related to public technologies; applying the data related to public technologies to a first machine learning model to acquire technical feature data related to the public technologies, wherein the first machine learning model is configured to output the technical feature data based on the data related to public technologies; and storing the technical feature data in a database in association with public technology identifiers.

[0018] Furthermore, the first machine learning model may include a preprocessing unit, wherein the data related to the public technology includes at least one text, and the preprocessing unit may be configured to extract at least one keyword from the at least one text, and to reduce a first-size vector based on the extracted keyword to a second-size vector smaller than the first size and output it.

[0019] In addition, the method for constructing the database may further include a step of obtaining technical needer-related data, a step of applying the technical needer-related data to a second machine learning model to obtain needer characteristic data related to the technical needer, where the second machine learning model is configured to output the needer characteristic data based on data related to the technical needer, and a step of associating the needer characteristic data with a technical needer identifier and storing the data in the database.

[0020] In addition, the second machine learning model includes a preprocessing unit, the data related to the common technology includes at least one text, and the preprocessing unit is configured to extract at least one keyword from the at least one text and reduce a vector of a third size based on the extracted one keyword to a vector of a fourth size smaller than the third size and output the vector.

Brief Description of Drawings

[0021] [Figure 1] It is a configuration diagram of a system for matching common technology and technical needers according to an embodiment of the present disclosure. [Figure 2] It is a flowchart for explaining a method for obtaining technical feature data and needer characteristic data according to an embodiment of the present disclosure. [Figure 3] It is a diagram illustrating first load data. [Figure 4] It is a diagram illustrating second load data. [Figure 5] It is a diagram showing an example of common technology data and technical needer data. [Figure 6] It is a diagram illustrating the configuration of a first machine learning model according to an embodiment of the present disclosure. [Figure 7] It is a diagram illustrating the configuration of a second machine learning model according to an embodiment of the present disclosure. [Figure 8] It is a diagram illustrating the configuration of a third machine learning model according to an embodiment of the present disclosure. [Figure 9]A diagram illustrating an artificial neural network model according to an embodiment of the present disclosure. [Figure 10] A flowchart explaining a method for matching public technologies and technology demanders according to an embodiment of the present disclosure. [Figure 11] A diagram illustrating a procedure in which a first machine learning model and a third machine learning model according to an embodiment of the present disclosure are used to calculate a score related to technology transaction possibility. [Figure 12] A flowchart explaining a method for matching public technologies and technology demanders according to another embodiment of the present disclosure. [Figure 13] A diagram illustrating a procedure in which a second machine learning model and a third machine learning model according to an embodiment of the present disclosure are used to calculate a score related to technology transaction possibility. [Figure 14] A hardware configuration diagram of a computing system according to some embodiments of the present disclosure.

Mode for Carrying Out the Invention

[0022] Hereinafter, preferred embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. The advantages and features of the present disclosure, and the method for achieving them, will become clear by referring to the embodiments described in detail later together with the accompanying drawings. However, the technical idea of the present disclosure is not limited to the following embodiments, and can be realized in various different forms. The following embodiments are merely to complete the technical idea of the present disclosure and to fully inform those having ordinary knowledge in the technical field to which the present disclosure belongs of the scope of the present disclosure. The technical idea of the present disclosure is defined only by the scope of the claims.

[0023] It should be noted that when attaching reference numerals to the components of each drawing, the same components are used with the same numerals as much as possible even if they are shown in other drawings. Also, in describing the present disclosure, if a specific description of a related known configuration or function is determined to obscure the gist of the present disclosure, the detailed description thereof will be omitted.

[0024] Unless otherwise defined, all terms used herein (including technical and scientific terms) should be used in a sense that is commonly understood by a person of ordinary skill in the art to which this disclosure pertains. Terms defined in commonly used dictionaries should not be interpreted ideally or excessively unless explicitly defined otherwise. Terms used herein are for illustrative purposes only and are not intended to limit this disclosure. In this specification, singular forms include plural forms unless otherwise specified.

[0025] Furthermore, when describing the components of this disclosure, terms such as 1, 2, A, B, (a), (b), etc., may be used. These terms are merely for distinguishing a component from other components and do not restrict the nature, order, or sequence of the component in question. Where it is stated that a component is “connected,” “joined,” or “connected” to another component, it should be understood that the component may be directly connected to or connected to the other component, but other components may also be “connected,” “joined,” or “connected” between each component.

[0026] As used in this specification, “comprises” and / or “comprising” does not preclude the presence or addition of one or more other components, stages, operations and / or elements that are mentioned.

[0027] Before describing embodiments of this disclosure, we will define the terms used in this disclosure.

[0028] In embodiments of this disclosure, “public technology” may be technology owned by a public institution. For example, public technology may include technologies related to R&D technology challenges at a public institution, technologies related to R&D commercialization challenges, and so on.

[0029] In embodiments of this disclosure, “technology users” may include organizations, individuals, companies, etc., that wish to use public technology. For example, technology users may include companies, individuals, organizations, schools, etc.

[0030] Several embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. Figure 1 is a diagram illustrating the configuration of a system for matching public technology with technology users according to one embodiment of this disclosure.

[0031] Referring to Figure 1, the matching system according to the embodiment of this disclosure may include a user terminal 10, a service server 20, a feature database 30 (hereinafter referred to as the "feature DB"), a public technology database 40 (hereinafter referred to as the "public technology DB"), and a technology demand database 50 (hereinafter referred to as the "technology demand DB"). The service server 20, user terminal 10, public technology DB 40, and technology demand DB 50 can communicate with each other via a network 60. Here, the network 60 may be configured to include a wired internet network, a mobile communication network, etc.

[0032] According to one embodiment, the public technology DB40 stores multiple first raw data sets related to various public technologies held by public institutions. For example, the public technology DB40 can store data on public R&D technology challenges and public R&D commercialization challenges held by public institutions as multiple first raw data sets. Data profiling is performed based on the first raw data, and data related to public technologies (hereinafter referred to as "public technology data") can be extracted and applied to a first machine learning model.

[0033] The Technology Demanders DB50 stores multiple second-row raw data sets related to technology demanders who wish to use public technologies. For example, the Technology Demanders DB50 can store data on technology trading companies, data on technology commercialization companies, etc., as multiple second-row raw data sets. Based on the second-row raw data, data profiling is performed to extract data related to technology demanders (hereinafter referred to as "technology demander data"), which can then be applied to a second machine learning model.

[0034] Examples of the first raw data, second raw data, public technology data, and technology consumer data will be described later with reference to Figures 3 to 5.

[0035] In one embodiment, a first machine learning model is used to extract multiple feature data from multiple public technology data. Here, the feature data represents the features related to public technology as a multidimensional vector. Furthermore, a second machine learning model can be used to extract multiple consumer feature data from multiple technology consumer data. Here, the consumer feature data represents the features related to technology consumers as a multidimensional vector.

[0036] According to one embodiment, the service server 20 applies each of the multiple public technology data to a first machine learning model, obtains multiple technology feature data from the first machine learning model, and stores it in the feature DB 30. The service server 20 can associate public technology identifiers (e.g., technology name, technology ID, etc.) with technology feature data and store them in the feature DB 30. Here, the first machine learning model may be a model that has been trained to perform inference operations based on public technology data and output technology feature data associated with the public technology.

[0037] Furthermore, the service server 20 can apply each of the multiple technology user data sets to a second machine learning model, obtain multiple user feature data from the second machine learning model, and store them in the feature DB 30. The service server 20 can associate the technology user identifier with the user feature data and store it in the feature DB 30. Here, the technology user identifier may be the technology user's login ID, organization name, company name, school name, etc. Here, the second machine learning model may be a model that has been trained to perform inference operations based on the technology user data and output technology feature data associated with the technology user.

[0038] According to one embodiment, when the service server 20 receives technology demander data from the user terminal 10, it generates a list of public technologies with a high probability of technology trading and transmits matching information, including the public technology list, to the user terminal 10. Specifically, the service server 20 can apply the technology demander data received from the user terminal 10 to a second machine learning model and obtain demander feature data from the second machine learning model. Furthermore, the service server 20 can apply the demander feature data and multiple technology feature data stored in the feature DB 30 to a third machine learning model and extract one or more technology feature data that matches the demander feature data based on multiple scores obtained from the third machine learning model. In addition, the service server 20 can generate a list of public technologies based on the extracted one or more technology feature data and transmit matching information, including the generated public technology list, to the user terminal 10.

[0039] Furthermore, when the service server 20 receives public technology data from the user terminal 10, it generates a list of technology demanders with a high probability of technology trading and transmits matching information, including the technology demander list, to the user terminal 10. Specifically, the service server 20 applies the public technology data received from the user terminal 10 to a first machine learning model and obtains technology feature data from the first machine learning model. The service server 20 also applies the obtained technology feature data and multiple technology feature data stored in the feature DB 30 to a third machine learning model and can extract one or more demander feature data that matches the public technology data based on multiple scores obtained from the third machine learning model. In addition, the service server 20 can generate a demander technology list based on the extracted one or more demander feature data and transmit matching information, including the generated demander technology list, to the user terminal 10.

[0040] The user terminal 10 receives matching information from the service server 20, which includes a list of technology demanders or a public technology list with a high probability of technology transactions. For example, if the user is a technology supplier, the user terminal 10 can transmit the public technology data held by the technology supplier to the service server 20 and receive matching information from the service server 20, which includes a list of technology demanders. The technology demander list may include at least one technology demander identifier and a predicted value that numerically represents the probability of a technology transaction being concluded. The predicted value may be a value normalized within a predetermined range based on a score output from a third machine learning model.

[0041] As another example, if the user is a technology user, the user terminal 10 can transmit the technology user and related technology user data to the service server 20 and receive a public technology list from the service server 20. The public technology list may include at least one public technology identifier and a numerically expressed predicted value representing the likelihood of a technology transaction taking place.

[0042] The matching system described above can be implemented as one or more computing devices equipped with processors. For example, each component, such as the service server 20, can be implemented as a single computing device, or the matching system can be implemented as a single computing device. A computing device can include any device equipped with computing capabilities; see Figure 14 for an example of such a device. Because a computing device is a collection of various components (e.g., memory, processors, etc.) that interact with each other, it is sometimes called a "computing system." A computing system also refers to a collection of multiple computing devices that interact with each other.

[0043] A method for pre-extracting and storing consumer characteristic data and technology characteristic data according to one embodiment of this disclosure will be described with reference to Figures 2 to 7. Furthermore, a method for matching public technology with technology consumers will be described with reference to Figures 10 to 13. The method according to this embodiment is performed by one or more computing devices. The methods shown in Figures 2, 10, and 12 are one embodiment for achieving the objectives of this disclosure, and of course, some steps can be added or omitted as needed. For the sake of explanation, the methods shown in Figures 2, 10, and 12 will be described assuming they are performed via the service server shown in Figure 1.

[0044] Figure 2 is a flowchart illustrating a method for obtaining technical feature data and consumer feature data according to one embodiment of the present disclosure.

[0045] The service server retrieves multiple public technology data (S110). For example, the service server receives multiple first-row data associated with multiple public technologies from the public technology database, profiles the received first-row data, and retrieves multiple public technology data. An example of first-row data associated with public technologies is shown in Figure 3.

[0046] Next, the service server applies the acquired public technical data to a first machine learning model to acquire multiple technical feature data (S120). In embodiments of this disclosure, applying data to a machine learning model can be understood as the machine learning model performing inference operations based on the applied data and outputting result data (e.g., feature data).

[0047] Subsequently, the service server stores multiple technical feature data in the feature database (S130). In this case, the service server can associate public technology identifiers with technical feature data and store them in the feature database.

[0048] Furthermore, the service server acquires data on multiple technology users (S140). For example, the service server can receive multiple second-row data associated with multiple technology users from the technology user database, profile the received second-row data, and acquire data on multiple technology users. An example of second-row data associated with technology users is shown in Figure 4.

[0049] Next, the service server applies the acquired data on multiple technology consumers to a second machine learning model to obtain multiple consumer feature data (S150). Subsequently, the service server stores the multiple consumer feature data in the feature database (S160). In this case, the service server can associate the technology consumer identifier with the consumer feature data and store them in the feature database.

[0050] Examples of the first raw data, second raw data, public technology data, and technology consumer data will be explained below with reference to Figures 3 to 5.

[0051] Figure 3 is an example of the first row of data.

[0052] Referring to Figure 3, the first raw data may include Public R&D Technology Transaction Issue Characteristics List 1 and Public R&D Commercialization Issue Characteristics List 2.

[0053] According to one embodiment, the public R&D technology transaction project feature list 1 for extracting the characteristics of public technology may include data / values ​​related to the technology transaction year, region, research period, research implementing body, project leader's specialization, research and development stage, number of patents, number of publications, number of participating researchers, project keywords, total research expenses (sum, cash, in kind), direct costs (cash, in kind), indirect costs, government research expenses, personnel costs (cash, in kind), commissioned research expenses, private research expenses, matching funds, science and technology standards classification (major), science and technology standards classification weighted value, socioeconomic objective code, and application field.

[0054] According to one embodiment, the public R&D commercialization project feature list 2 for extracting the characteristics of public technology may include data / values ​​related to the technology transaction year, region, research period, research implementing body, project leader's specialization, research and development stage, number of patents, number of publications, number of participating researchers, project keywords, whether or not it is a joint research project, total research expenses (sum, cash, in kind), direct costs (cash, in kind), indirect costs, government research expenses, personnel costs (cash, in kind), commissioned research expenses, private research expenses, matching funds, science and technology standard classification (major), science and technology standard classification weighted value, economic and social purpose code, application field and application field weighted value.

[0055] According to one embodiment, data profiling is performed to obtain public technology data based on the public R&D technology trading issue feature list 1 and the public R&D commercialization issue feature list 2. For example, data profiling can be performed based on an algorithm for data profiling.

[0056] Figure 4 is an example of the second row of data.

[0057] Referring to Figure 4, the second raw data may include, as examples of data related to technology consumers, a list of technology trading firm characteristics 3 and a list of technology commercializing firm characteristics 4.

[0058] According to one embodiment, the technology transaction company characteristics list 3 for extracting the characteristics of technology demanders may include data / values ​​related to the year of technology transaction, business history, industry code, 10th industry code, whether or not it is a venture company, whether or not it is an innovation business, company entity classification, company detail classification, company size classification, listed market classification, public listing, main product name, total assets, total capital, total liabilities, tangible assets, intangible assets, current assets, non-operating income, selling and administrative expenses, and return on total assets.

[0059] According to one embodiment, the list of characteristics of technology-commercializing companies for extracting the characteristics of technology users may include data / values ​​related to the year of technology transaction, business history, industry code, 10th industry code, whether or not it is a venture company, whether or not it is an innovation business, company entity classification, company detail classification, company size classification, listed market classification, public listing status, main product name, whether or not it is closed or suspended, sales, net income, total assets, total capital, total liabilities, tangible assets, intangible assets, current assets, operating profit, non-operating income, net income before corporate tax, selling expenses and administrative expenses, operating profit margin, net profit margin, return on total assets, establishment date, and number of employees.

[0060] According to one embodiment, data profiling is performed to obtain technology demander data based on the technology trading company characteristics list 3 and the technology commercialization company characteristics list 4. For example, data profiling can be performed based on an algorithm for data profiling.

[0061] Other data or values ​​besides those exemplified in Figures 3 and 4 may be included in at least one of the following: Public R&D Technology Transaction Issue Characteristics List 1, Public R&D Commercialization Issue Characteristics List 2, Technology Transaction Firm Characteristics List 3, and Technology Commercialization Firm Characteristics List 4.

[0062] Figure 5 shows an example of public technology data and technology consumer data.

[0063] Referring to Figure 5, public technology data 5 may include data values ​​related to the Science and Technology Standard Classification (Major), Science and Technology Standard Classification (Medium), regional code, summary keyword and / or keyword (Korean), R&D stage code, total R&D expenditure, and technology name. Here, the summary keyword and / or keyword (Korean) may be text containing multiple words or text containing one or more sentences. Furthermore, the Science and Technology Standard Classification (Major) is data for broadly classifying technologies, and the Science and Technology Standard Classification (Medium) is data for subclassifying technologies, with each major classification containing at least one subclass.

[0064] Furthermore, the technology user data 6 may include the vendor code, recent number of employees, 10th industry code, main products (in Korean), total capital, business history, number of employees, sales and business objectives, and establishment date. The main products (in Korean) may be a text containing multiple words or a text containing one or more sentences.

[0065] The first and second machine learning models will be described below with reference to Figures 6 and 7.

[0066] Figure 6 illustrates the configuration of a first machine learning model according to one embodiment of the present disclosure.

[0067] Referring to Figure 6, public technology data 110 is applied as input data to the first machine learning model 100, and technology feature data 120 is output from the first machine learning model 100. Figure 6 shows that the public technology data 110 includes Science and Technology Standard Classification 1 (Medium), Science and Technology Standard Classification 1 (Large), regional code, summary keywords, keywords (Korean), R&D stage code, and total R&D expenses. The public technology data 110 is illustrative and can be replaced if more efficient input data exists for determining technology tradability.

[0068] According to one embodiment, the first machine learning model 100 may include a preprocessor in the preceding stage and an autoencoder in the succeeding stage. For example, the preprocessor included in the first machine learning model 100 may be configured to receive a first-size (e.g., 300-size) vector input based on seven keywords extracted from text associated with summary keywords (keyword) and / or keywords (Korean), and reduce it to a second-size (e.g., 70-size) vector smaller than the first size for output. As another example, the preprocessor included in the first machine learning model 100 may be configured to receive a first-size (300-size) vector input based on one keyword extracted from text associated with scientific and technological standards classification 1 (medium), and reduce it to a second-size (e.g., 70-size) vector for output. Here, the type and number of keywords used are not limited to the above examples and can be extracted and changed from a variety of public technology data or technology consumer data. By using seven keywords, a sufficient number of words can be utilized as input, and by inputting in 7*300 vector format, a sufficient number of vectorized values ​​can be utilized. Furthermore, reducing the output size to 7*70 can improve learning efficiency. In the case of an autoencoder in the later stages of the first machine learning model, the Fully Connected Layer (FC) can flatten the tensor learned in the previous layer into a 1*N vector form, and in the case of Batch Normalization (BN), it can reduce the deviation caused by the scale difference of the input variables, preventing the influence of a particular input variable from becoming too large or too small, thereby improving the accuracy of the final prediction result.

[0069] Figure 7 illustrates the configuration of a second machine learning model according to one embodiment of the present disclosure.

[0070] Referring to Figure 7, the technology demander data 210 is applied to the second machine learning model 200, and the demander feature data 220 is output from the second machine learning model 200. In Figure 7, the technology demander data 210 is exemplified to include vendor code, recent number of employees, 10th industry code, main product (in Korean), total capital, establishment date, number of employees, and sales revenue. However, the technology demander data 210 here is illustrative and can be replaced if more efficient input data exists for determining technology tradability.

[0071] According to one embodiment, the second machine learning model 200 may include a preprocessing unit in the preceding stage and an autoencoder in the succeeding stage. For example, the second machine learning model 200 can be configured to receive a third-size (e.g., 300-size) vector input based on seven keywords extracted from text related to the main product (in Korean), and output a fourth-size (e.g., 70-size) vector smaller than the third size. Here, the type and number of keywords used are not limited to the above example. By using seven keywords, a sufficient number of words can be utilized as input, and by inputting in 7*300 vector format, a sufficient number of vectorized values ​​can be utilized. Furthermore, by reducing the output size to 7*70, the efficiency of learning can be improved.

[0072] The autoencoder in the second machine learning model 200 is input with vectorized values ​​output based on the vendor code, recent number of employees, 10th industry code, main product (in Korean), capital statistics, establishment date, number of employees, and sales revenue. The autoencoder in the second machine learning model 200, a Fully Connected Layer (FC), can perform the process of flattening the tensor learned in the previous layer into a 1*N vector form. In the case of Batch Normalization (BN), it can reduce the deviation caused by the scale difference of the input variables and prevent the influence of a particular input variable from becoming too large or too small, thereby improving the accuracy of the final prediction result.

[0073] As described above, text containing multiple words or one or more sentences can be preprocessed so that the vectors are reduced and input to the autoencoder. This improves the computational speed and reduces the computational load of the first machine learning model 100 and the second machine learning model 200.

[0074] Figure 8 illustrates the configuration of a third machine learning model according to one embodiment of the present disclosure.

[0075] Referring to Figure 8, the technology feature data and demand feature data are input into the third machine learning model 300, and the third machine learning model 300 outputs a score indicating the tradability of the technology.

[0076] Technical feature data and consumer feature data can be combined and input into the third machine learning model 300. For example, the combined technical feature data and consumer feature data can be converted into at least one of Q-vectors, K-vectors, and V-vectors, and a dot product operation can be performed on each of the converted Q-vectors and K-vectors to calculate a score for the combined technical feature data and consumer feature data. Then, the V-vector can be added to the score and the score for the combined technical feature data and consumer feature data can be output from the third machine learning model 300.

[0077] As illustrated in Figure 8, the third machine learning model 300 can be composed of an attention module (ATTN), a fully connected layer (FC), and layer normalization (LN). The attention module can perform inference by focusing on one or more predetermined vectors from among the Q vector, K vector, and V vector.

[0078] Furthermore, a Fully Connected Layer (FC) can flatten a tensor learned in the previous layer into a 1*N vector form. Layer Normalization (LN) can normalize data based on its mean and standard deviation.

[0079] According to one embodiment, multiple scores output from the third machine learning model 300 can be converted into percentage values. Furthermore, these converted percentage values ​​can be normalized and converted to values ​​within a predetermined range. For example, an activation function can be used to perform normalization of the percentage values. For instance, the normalized values ​​can be expressed as values ​​between 0 and 1 or between 0% and 100%.

[0080] Figure 9 illustrates an artificial neural network model 900 according to one embodiment of the present disclosure. The artificial neural network model 900, as an example of a machine learning model, may be a statistical learning algorithm or a structure that executes such an algorithm, which is realized based on the structure of a biological neural network in machine learning technology and cognitive science. According to some embodiments, the artificial neural network model 900 may be included in at least one of the first machine learning model, second machine learning model, and third machine learning model described above.

[0081] According to one embodiment, the artificial neural network model 900 can exhibit a machine learning model with problem-solving capabilities by having nodes, which are artificial neurons that form a network by synaptic connections, repeatedly adjust the synaptic weights to learn to reduce the error between the correct output corresponding to a particular input and the inferred output, similar to biological neural networks. For example, the artificial neural network model 900 may include any probabilistic model, neural network model, etc., used in artificial intelligence learning methods such as machine learning and deep learning.

[0082] At least one of the first, second, and third machine learning models described above can be implemented in the form of an artificial neural network model 900. According to one embodiment, the artificial neural network model 900 can be configured to output technology feature data based on public technology data. The artificial neural network model 900 can also be configured to output consumer feature data based on technology consumer data. Furthermore, the artificial neural network model 900 can be configured to output a score indicating the tradability between public technology and technology consumers based on the consumer feature data and the technology feature data.

[0083] The artificial neural network model 900 can be implemented as a multilayer perceptron (MLP) composed of multiple layers of nodes and connections between them. The artificial neural network model 900 according to this embodiment can be implemented using one of various artificial neural network model structures, including MLPs. The artificial neural network model 900 consists of an input layer that receives input signals or data from the outside, an output layer that outputs output signals or data corresponding to the input data, and n hidden layers (where n is a positive integer) located between the input and output layers that receive signals from the input layer, extract characteristics, and transmit them to the output layer.

[0084] In the artificial neural network model 900, multiple input variables and corresponding multiple output variables are matched in the input and output layers, respectively, and the synaptic values ​​between the nodes in the input, hidden, and output layers are adjusted so that the correct output corresponding to a specific input is extracted. As the artificial neural network model 900 is repeatedly trained based on the data in the training dataset, the synaptic values ​​(or weights) between the nodes of the artificial neural network model 900 are adjusted so that the error between the output variable calculated based on the input variable and the target output decreases, and it can converge to an optimal value.

[0085] In the embodiments described above, the first machine learning model, the second machine learning model, and the third machine learning model were implemented independently of each other. However, two or more of the first, second, and third machine learning models can be integrated to form a single machine learning model.

[0086] Each of the first, second, and third machine learning models described above is trained based on a training set. The training set may include labeled data, and each of the first, second, and third machine learning models is trained to output data that approximates the ground truth contained in the labeled data. For example, the third machine learning model can be trained using data labeled based on existing successful technology transaction cases.

[0087] In some embodiments, the training data may include both data from successful and unsuccessful technology transactions. Since unsuccessful technology transactions usually far outnumber successful ones, training with training data that includes both successful and unsuccessful transactions may result in bias in the training results. This allows for a certain degree of similarity in the number of successful and unsuccessful technology transactions during training. For this reason, only a portion of the unsuccessful technology transaction data can be extracted and used, rather than using the entire dataset, and the extraction method is crucial in this case. In this embodiment, the data is extracted exemplified by industry (the 10th industry code of the technology demander data), but the data from unsuccessful technology transactions can also be extracted in a way that corresponds to the industry-specific ratio of successful technology transaction data. Specifically, when analyzing multiple successful technology transaction data obtained from a technology transaction-related database (not shown) by major classification of the 10th industry code (the alphabet and the first or second digit of the preceding number of the 10th industry code), it is found that certain industries have many successful technology transaction cases, while other specific industries have few. For example, in the 10th industry code, C27 Medical, Precision, Optical Instruments and Watchmaking accounts for 30%, while A01 Agriculture accounts for 3%, indicating a concentration of successful technology transaction cases in C27 Medical, Precision, Optical Instruments and Watchmaking. In this case, even for data where technology transactions were unsuccessful, applying the industry-specific ratio of successful technology transaction data to extract the data (as in the example above, extracting data where technology transactions were unsuccessful so that C27 Medical, Precision, Optical Instruments and Watchmaking accounts for 30% and A01 Agriculture accounts for 3%) can reduce bias due to the learning results and enable more accurate learning.

[0088] Figure 10 is a flowchart illustrating a method for matching public technology with technology users according to one embodiment of this disclosure.

[0089] Referring to Figure 10, the service server acquires public technology and related data (S210). For example, the service server can receive first raw data from the user terminal, perform data profiling on the received first raw data, and acquire public technology and related data applicable to the first machine learning model. As another example, the service server can receive public technology and related data from the user terminal that is immediately applicable to the first machine learning model.

[0090] Next, the service server applies the acquired public technology and related data to a first machine learning model to acquire public technology and related technical feature data (S220). Here, the first machine learning model may be a trained model configured to output technical feature data based on public technology and related data.

[0091] Subsequently, the service server matches public technologies with one or more technology consumers based on the technology feature data and multiple consumer feature data stored in the feature database (S230).

[0092] According to one embodiment, a service server can apply technology feature data and a plurality of demand feature data to a third machine learning model to obtain a score associated with tradability for each of the plurality of demand feature data. Here, the third machine learning model may be a trained model configured to output a score associated with technology tradability between the technology feature data and the demand feature data. Subsequently, based on the obtained scores, the service server extracts one or more demand feature data from the plurality of demand feature data and determines one or more technology demanders that match with the public technology based on the extracted one or more demand feature data. For example, the service server can extract one or more demand feature data associated with a score that falls within a predetermined rank from the scores for each of the plurality of demand feature data, and determine the technology demanders associated with the extracted one or more demand feature data as matched with the public technology. As another example, the service server can extract one or more demand feature data associated with a score above a predetermined critical value from the scores for each of the plurality of demand feature data, and determine the technology demanders associated with the extracted one or more demand feature data as matched with the public technology.

[0093] According to some embodiments, the service server can convert the acquired score into a normalized value within a predetermined range. The service server can also transmit information related to one or more technology users matched with public technology to the user terminal. Here, the information related to technology users may include the normalized value.

[0094] Figure 11 illustrates a procedure in which a first machine learning model and a third machine learning model according to one embodiment of the present disclosure are used to calculate a score related to technology tradability.

[0095] The first machine learning model 410 illustrated in Figure 11 corresponds to the first machine learning model 100 in Figure 6, and the third machine learning model 420 illustrated in Figure 11 corresponds to the third machine learning model 300 in Figure 8.

[0096] As illustrated in Figure 11, public technology data 440 is applied to the first machine learning model 410, and technology feature data 450 is output from the first machine learning model 410. Furthermore, the technology feature data 450 and consumer feature data 460 stored in the feature DB 430 are applied to the third machine learning model 420, and a score 470 based on the consumer feature data 460 and technology feature data 450 can be output. Here, the score 470 is a numerical representation of the possibility of technology trading between the public technology associated with the technology feature data 450 and the technology consumer associated with the consumer feature data 460.

[0097] As described above, once public technology data 440 is acquired, the first machine learning model 410 and the third machine learning model 420 are used to match at least one technology user associated with the public technology. Furthermore, since user feature data 460, which is pre-stored in the feature DB 430, is input into the third machine learning model 420, the computing resources required to calculate the user feature data 460 are saved, and the computation time can be reduced. This makes the third machine learning model 420 more lightweight.

[0098] Figure 12 is a flowchart illustrating a method for matching public technology with technology users according to other embodiments of this disclosure.

[0099] Referring to Figure 12, the service server acquires data related to the technology user (S310). For example, the service server can receive second raw data from the user terminal, perform data profiling on the received second raw data, and acquire data related to the technology user that can be applied to the second machine learning model. As another example, the service server can receive data related to the technology user that can be immediately applied to the second machine learning model from the user terminal.

[0100] Next, the service server applies the acquired technology user-related data to a second machine learning model to obtain user feature data associated with the technology user (S320). Here, the second machine learning model may be a trained model configured to output user feature data based on the technology user-related data.

[0101] Subsequently, the service server matches technology consumers with one or more public technologies based on consumer characteristic data and multiple technology characteristic data stored in the feature database (S330).

[0102] According to one embodiment, a service server can apply demand characteristic data and multiple technical characteristic data to a third machine learning model to obtain a score associated with tradability for each of the multiple technical characteristic data. Here, the third machine learning model is configured and trained to output a score associated with the technical tradability between the demand characteristic data and the technical characteristic data. Subsequently, based on the obtained scores, the service server can extract one or more technical characteristic data from the multiple technical characteristic data and determine one or more public technologies that match a technology demander based on the extracted one or more technical characteristic data. For example, the service server can extract one or more technical characteristic data associated with a score that falls within a predetermined rank from the scores for each of the multiple technical characteristic data, and determine that the extracted one or more technical characteristic data and the associated public technologies are matched with a technology demander. As another example, the service server can extract one or more technical characteristic data associated with a score above a predetermined critical value from the scores for each of the multiple technical characteristic data, and determine that the extracted one or more technical characteristic data and the associated public technologies are matched with a technology demander.

[0103] According to some embodiments, the service server can convert the acquired score into a normalized value within a predetermined range. The service server can also transmit information related to one or more public technologies matched with the technology user to the user terminal. Here, the information related to public technologies may include normalized values.

[0104] Figure 13 illustrates a procedure in which a second machine learning model and a third machine learning model according to one embodiment of the present disclosure are used to calculate a score related to technology tradability.

[0105] The second machine learning model 510 illustrated in Figure 13 corresponds to the second machine learning model 200 in Figure 7, and the third machine learning model 520 illustrated in Figure 13 corresponds to the third machine learning model 300 in Figure 8.

[0106] As shown in Figure 13, the technology demander data 540 is applied to the second machine learning model 510, and demander feature data 550 is output from the second machine learning model 510. Furthermore, the demander feature data 550 and the technology feature data 560 stored in the feature DB 530 are applied to the third machine learning model 520, and a score 570 based on the demander feature data 550 and the technology feature data 560 can be output. Here, the score 570 is a numerical representation of the technology tradeability between the technology demander associated with the demander feature data 550 and the public technology associated with the technology feature data 560.

[0107] As described above, once the technology user data 540 is acquired, the second machine learning model 510 and the third machine learning model 520 are used to match the technology user with at least one public technology related to them.

[0108] Figure 14 is a hardware configuration diagram of a computing system according to some embodiments of the present disclosure. The computing system 1000 in Figure 14 may include one or more processors 1100, a system bus 1600, a communication interface 1200, memory 1400 for loading computer programs 1500 performed by the processors 1100, and storage 1300 for storing the computer programs 1500.

[0109] The computing system 1000 in Figure 14 may, for example, represent the hardware structure of a computing system that constitutes the matching system described with reference to Figure 1.

[0110] The processor 1100 controls the overall operation of each configuration of the computing system 1000. The processor 1100 performs calculations on at least one application or program to perform methods / operations according to various embodiments of the disclosure. The memory 1400 stores various data, instructions and / or information. The memory 1400 can load one or more computer programs 1500 from the storage 1300 to perform methods / operations according to various embodiments of the disclosure. The storage 1300 can non-temporarily store one or more computer programs 1500.

[0111] The computer program 1500 may include one or more instructions that implement the methods / operations according to various embodiments of the disclosure. When the computer program 1500 is loaded into memory 1400, the processor 1100 can perform the methods / operations according to various embodiments of the disclosure by executing one or more instructions.

[0112] According to one embodiment, the computer program 1500 may include instructions for operations to acquire data related to public technology, operations to apply the acquired data to a first machine learning model to acquire technology feature data related to public technology, and operations to match public technology with one or more technology consumers based on the technology feature data and a plurality of consumer feature data stored in a database.

[0113] Furthermore, the computer program 1500 may include instructions for operations to acquire data related to technology consumers, operations to apply the acquired data to a second machine learning model to acquire consumer feature data related to technology consumers, and operations to match technology consumers with one or more public technologies based on the consumer feature data and multiple technology feature data stored in a database.

[0114] Furthermore, the computer program 1500 may include instructions for operations to acquire data related to public technology, operations to apply data related to public technology to a first machine learning model to acquire technical feature data related to public technology, and operations to store the technical feature data in a database in association with public technology identifiers.

[0115] In some embodiments, the computing system 1000 described with reference to Figure 14 can be configured using one or more physical servers included in a server farm based on cloud technology such as virtual machines. In this case, at least a portion of the components shown in Figure 14, such as the processor 1100, memory 1400, and storage 1300, may be virtual hardware, and the communication interface 1200 may also be composed of virtualized networking elements such as a virtual switch.

[0116] We have so far described various embodiments of this disclosure and the effects thereof with reference to Figures 1 through 14. The effects of the technical concept of this disclosure are not limited to those described above, and any other effects not mentioned will be clearly understood by a person of the ordinary skill from the following description.

[0117] The methods according to the embodiments of the present invention described herein can be carried out by executing a computer program implemented in computer-readable code. The computer program may be transferred from a first computing device to a second computing device via a network such as the Internet, installed on the second computing device, and thus made available for use on the second computing device. While the drawings show operations in a specific order, it should not be understood that the operations must necessarily be executed in the specific or sequential order shown, or that the desired result cannot be obtained unless all shown operations are performed. In certain situations, multitasking and parallel processing may be advantageous.

[0118] While embodiments of this disclosure have been described above with reference to the attached drawings, a person with ordinary skill in the art to which this disclosure belongs will understand that the present invention can be carried out in other specific forms without altering the technical idea or essential features. Therefore, the above-described embodiment should be understood to be illustrative and not limiting in all respects. The scope of protection of the present invention should be analyzed based on the following claims, and all technical ideas within an equivalent scope should be interpreted as being included in the scope of rights of the technical idea as defined by this disclosure.

Claims

1. In a method of matching public technologies with technology users using computing systems, The stage of acquiring public technology and related data; A step of applying the acquired data to a first machine learning model to acquire technical feature data related to the public technology, wherein the first machine learning model is configured to output the technical feature data based on the data related to the public technology; and A matching method comprising the step of matching the public technology with one or more technology consumers based on the aforementioned technology feature data and a plurality of consumer feature data stored in a database, The first machine learning model includes a preprocessing unit, The aforementioned public technology and related data includes at least one text, The preprocessing unit is configured to extract at least one keyword from the at least one text, and to reduce a first-size vector based on the extracted keyword to a second-size vector smaller than the first size and output it, in a matching method.

2. Before the step of acquiring data related to the aforementioned public technology, The stage of acquiring multiple data points related to multiple technology users; A step of applying a plurality of data associated with the plurality of technology consumers to a second machine learning model to obtain the plurality of consumer feature data associated with the plurality of technology consumers, wherein the second machine learning model is configured to output the consumer feature data based on the data associated with the technology consumers; and The matching method according to claim 1, further comprising the step of storing the acquired plurality of consumer characteristic data in the database.

3. The step of matching the aforementioned public technology with one or more technology users is: A step of applying the aforementioned technical feature data and the plurality of demand feature data to a third machine learning model to obtain a score related to tradability for each of the plurality of demand feature data, wherein the third machine learning model is configured to output a score related to technical tradability between the technical feature data and the demand feature data; Based on the acquired score, the step of extracting one or more consumer characteristic data from the plurality of consumer characteristic data; and The matching method according to claim 1 or 2, further comprising the step of determining one or more technology consumers that match the public technology based on one or more extracted consumer characteristic data.

4. The step of extracting one or more consumer characteristic data from the aforementioned multiple consumer characteristic data is: The matching method according to claim 3, further comprising the step of extracting one or more consumer characteristic data associated with a score that falls within a predetermined rank from among the scores for each of the plurality of consumer characteristic data.

5. The step of extracting one or more consumer characteristic data from the aforementioned multiple consumer characteristic data is: The matching method according to claim 3, further comprising the step of extracting one or more consumer characteristic data from among the scores for each of the plurality of consumer characteristic data that are associated with a score equal to or greater than a predetermined threshold.

6. The step of matching the aforementioned public technology with one or more technology users is: This includes the step of converting the acquired score into a normalized value within a predetermined range. The aforementioned matching method is, After the step of matching the aforementioned public technology with one or more technology users, The further step includes transmitting information related to one or more technology consumers matched with the aforementioned public technology to a user terminal, The matching method according to claim 3, wherein the information related to the technology user includes the normalized value.

7. The second machine learning model includes a preprocessing unit, The data related to the aforementioned technology consumers includes at least one text, The matching method according to claim 2, wherein the preprocessing unit is configured to extract at least one keyword from the at least one text, and to reduce a third-size vector based on the extracted keyword to a fourth-size vector smaller than the third size and output it.

8. In a method of matching public technologies with technology users using computing systems, The stage of acquiring data related to technology users; A step of applying the acquired data to a first machine learning model to acquire consumer characteristic data related to the technology consumer, wherein the first machine learning model is configured to output the consumer characteristic data based on the data related to the technology consumer; and A matching method comprising the step of matching a technology consumer with one or more public technologies based on the aforementioned consumer characteristic data and a plurality of technology characteristic data stored in a database, The first machine learning model includes a preprocessing unit, The data related to the aforementioned technology consumers includes at least one text, The preprocessing unit is configured to extract at least one keyword from the at least one text, and to reduce a first-size vector based on the extracted keyword to a second-size vector smaller than the first size and output it, in a matching method.

9. Before the step of acquiring data related to the aforementioned technology consumers, The stage of acquiring multiple public technologies and related data; A step of applying a plurality of data related to the plurality of public technologies to a second machine learning model to obtain the plurality of technical feature data related to the plurality of public technologies, wherein the second machine learning model is configured to output the technical feature data based on the data related to the public technologies; and The matching method according to claim 8, further comprising the step of storing the acquired plurality of technical feature data in the database.

10. The step of matching the aforementioned technology user with one or more public technologies is: A step of applying the demander characteristic data and the plurality of technical characteristic data to a third machine learning model to obtain a score related to tradability for each of the plurality of technical characteristic data, wherein the third machine learning model is configured to output a score related to technical tradability between the demander characteristic data and the technical characteristic data; The step of extracting one or more technical feature data from the plurality of technical feature data based on the acquired score; and The matching method according to claim 8 or 9, further comprising the step of determining one or more public technologies that match the technology user based on one or more extracted technical feature data.

11. The second machine learning model includes a preprocessing unit, The aforementioned public technology and related data includes at least one text, The matching method according to claim 9, wherein the preprocessing unit is configured to extract at least one keyword from the at least one text, and to reduce a third-size vector based on the extracted keyword to a fourth-size vector smaller than the third size and output it.

Images