KR102545575B1 - Method of providing subscription service for automatically recommending ai model using platform applied with dualized service flow adapted to each characteristic of each customer group and server using the same - Google Patents

Abstract

An AI model automatic recommendation subscription service method through a platform to which a redundant service flow according to characteristics of each customer group is applied is disclosed. That is, (a) the model recommendation management server (MRMS) provides first metadata obtained by providing the type 1 automatic recommendation service to a first customer group subscribing to the type 1 automatic recommendation service whose subscription cost is less than a threshold value. optimizing a reference database of a type 2 automatic recommendation service exceeding the threshold by referring to the reference database; (b) When the MRMS obtains a specific automatic recommendation request from a specific customer included in the second customer group subscribing to the type 2 automatic recommendation service, a specific automatic recommendation request corresponding to the specific automatic recommendation request is obtained by referring to the reference database. A method comprising providing an AI model to the specific customer is disclosed.

Description

AI model automatic recommendation subscription service method and server through a platform to which a redundant service flow is applied according to the characteristics of each customer group USING THE SAME}

The present invention relates to an AI model automatic recommendation subscription service method and server through a platform to which a redundant service flow according to characteristics of each customer group is applied.

According to the prior art, in order to build a server that automatically recommends an AI model, data preparation - data analysis (Exploratory Data Analysis, EDA) - feature engineering - algorithm selection - hyperparameter optimization (HPA) - model evaluation A process that goes through 7 steps of /validation-deployment or more must be established. This is a process that consumes a lot of computing resources, and this problem acts as a major obstacle to the popularization of the AI model automatic recommendation service.

The present invention aims to solve the above problems.

Another object of the present invention is to popularize the automatic recommendation service of an AI model through a platform to which a redundant service flow according to characteristics of each customer group is applied.

In addition, an object of the present invention is to further facilitate the use of the AI model by enabling non-experts to implement the AI model based on GUI without writing code.

The characteristic configuration of the present invention for achieving the object of the present invention as described above and realizing the characteristic effects of the present invention described later is as follows.

According to one aspect of the present invention, in the AI model automatic recommendation subscription service method through a platform to which a redundant service flow according to characteristics of each customer group is applied, (a) a model recommendation management server (MRMS), a first subscription cost is less than the threshold value optimizing a reference database of the type 2 automatic recommendation service exceeding the threshold by referring to first metadata acquired by providing the type 1 automatic recommendation service to a first customer group subscribing to the type 1 automatic recommendation service; (b) When the MRMS obtains a specific automatic recommendation request from a specific customer included in the second customer group subscribing to the type 2 automatic recommendation service, a specific automatic recommendation request corresponding to the specific automatic recommendation request is obtained by referring to the reference database. A method comprising providing an AI model to the specific customer is disclosed.

As an example, in the step (a), the MRMS includes (i) each meta-feature of the analysis target data input by the first customer group and (ii) each of the AI models provided to the first customer group A method characterized by optimizing the reference database with reference to the first metadata, which is pairing data between the liver, is disclosed.

As an example, in the step (b), (b1) when the MRMS obtains specific analysis target data from at least one specific customer included in the second customer group, the specific analysis target data is input to a data analyzer. obtaining a specific meta-feature for; and (b2) the MRMS takes the specific meta feature as an input and applies a model setting operation referring to the reference database through a model and architecture recommendation server (MARS), which is a detailed server that operates in conjunction with the MRMS. A method comprising acquiring the specific AI model optimized for specific analysis target data and then providing the specific AI model to the specific customer is disclosed.

As an example, in the step (b1), the MRMS causes the data analyzer to determine which of the data categories including at least a part of tables, images, and plain texts corresponds to the specific analysis target data, and then the Disclosed is a method characterized in that the specific meta-feature is acquired by applying a meta-feature analysis operation corresponding to the specific data category with specific analysis target data as an input.

As an example, in the step (b1), the MRMS includes (i) a process for causing a chief analyzer included in the data analyzer to perform global statistical calculation of the specific analysis target data, and (ii) the data analyzer By causing the included worker analyzer to perform a process of performing individual statistical calculations of the specific analysis target data in a distributed cluster manner, the data analyzer performs the meta-feature analysis operation with the specific analysis target data as input A method characterized by allowing to apply is disclosed.

As an example, in the step (b2), the MRMS causes the MARS to: (i) a dynamic recommendation algorithm performed by referring to the reference database, (ii) a static recommendation algorithm based on pre-input knowledge And (iii) a method characterized by obtaining the specific AI model by performing at least a part of the genetic algorithm is disclosed.

As an example, in the step (b2), the MRMS takes the specific meta feature as an input and applies a preprocessing setting operation through (i) a preprocessing setting recommendation server (DPRS), which is a detailed server that operates in conjunction with the MRMS By applying a process of obtaining specific preprocessing settings optimized for the specific analysis target data and (ii) a hyperparameter setting operation through a hyperparameter recommendation server (HPRS), specific hyperparameter setting values optimized for the specific analysis target data are obtained. A method characterized by providing the specific preprocessing setting and the specific hyperparameter setting value together with the specific AI model to the specific customer after performing a process of acquiring.

As an example, in the step (b2), the MRMS causes the DPRS to apply a preprocessing setting operation to the specific meta-feature as an input, so that (i) a preprocessing function for the specific analysis target data, (ii) ) a feature extraction method and (iii) a method characterized by obtaining the specific preprocessing settings including whether or not dimensionality reduction is disclosed.

As an example, in the step (b2), the MRMS causes the HPRS to: (i) a dynamic recommendation algorithm performed by referring to the reference database, (ii) a static recommendation algorithm based on pre-input knowledge and (iii) performing at least a part of a genetic algorithm to obtain the specific hyperparameter settings.

As an example, in the step (b2), the MRMS causes the learning cluster to optimize the specific AI model for the specific analysis target data by performing data conversion, learning, and evaluation processes in a distributed processing method, and then the A method characterized by providing to a specific customer is disclosed.

According to another aspect of the present invention, in a model recommendation management server (MRMS) that performs an AI model automatic recommendation subscription service through a platform to which a redundant service flow according to characteristics of each customer group is applied, one or more memories for storing instructions; and one or more processors configured to execute the instructions, wherein the processor is configured to: (I) provide the Type 1 automatic recommendation service to a first customer group subscribing to the Type 1 automatic recommendation service for which a subscription cost is less than a threshold value; a process of optimizing a reference database of a type 2 automatic recommendation service exceeding the threshold by referring to the obtained first metadata; (II) When a specific automatic recommendation request is obtained from a specific customer included in the second customer group subscribing to the type 2 automatic recommendation service, a specific AI model corresponding to the specific automatic recommendation request is recalled by referring to the reference database. A server characterized in that it is configured to perform a process of providing to a specific customer is disclosed.

As an example, the (I) process may include (i) each of the meta-features of the data to be analyzed input by the first customer group and (ii) each of the AI models provided to the first customer group for each of them. A server characterized in that for optimizing the reference database by referring to the first metadata, which is pairing data between the two, is disclosed.

As an example, in the (II) process, (II1) when the processor obtains specific analysis target data from at least one specific customer included in the second customer group, the specific analysis target data is input to a data analyzer. A process of obtaining a specific meta-feature for ; and (II2) the processor takes the specific meta feature as an input and applies a model setting operation referring to the reference database through a model and architecture recommendation server (MARS), which is a detailed server that operates in conjunction with the MRMS. A server comprising a process of obtaining the specific AI model optimized for specific analysis target data and providing it to the specific customer is disclosed.

As an example, in the (II1) process, the processor determines which of the data categories including at least a part of a table, an image, and a plain text corresponds to the specific analysis target data by the data analyzer, and then the Disclosed is a server characterized in that a specific meta-feature is acquired by applying a meta-feature analysis operation corresponding to a corresponding specific data category with specific analysis target data as an input.

As an example, the (II1) process may include (i) a process for causing a chief analyzer included in the data analyzer to perform global statistical calculation of the specific analysis target data, and (ii) the data analyzer By causing the included worker analyzer to perform a process of performing individual statistical calculations of the specific analysis target data in a distributed cluster manner, the data analyzer performs the meta-feature analysis operation with the specific analysis target data as input A server characterized in that it applies is disclosed.

As an example, the (II2) process causes the processor to cause the MARS to (i) a dynamic recommendation algorithm performed by referring to the reference database, (ii) a static recommendation algorithm based on pre-input knowledge And (iii) a server is disclosed, characterized in that for obtaining the specific AI model by performing at least a part of the genetic algorithm.

As an example, in the (II2) process, the processor takes the specific meta feature as an input and applies a preprocessing setting operation through (i) a preprocessing setting recommendation server (DPRS), which is a detailed server that operates in conjunction with the MRMS By applying a process of obtaining specific preprocessing settings optimized for the specific analysis target data and (ii) a hyperparameter setting operation through a hyperparameter recommendation server (HPRS), specific hyperparameter setting values optimized for the specific analysis target data are obtained. A server is disclosed which, after performing a process of obtaining, provides the specific preprocessing setting and the specific hyperparameter setting value to the specific customer together with the specific AI model.

As an example, in the (II2) process, the processor causes the DPRS to apply a preprocessing setting operation to the specific metafeature as an input, such that (i) a preprocessing function for the specific analysis target data, (ii) ) a feature extraction method and (iii) a server characterized in that obtaining the specific preprocessing settings including whether or not dimensionality reduction is disclosed.

As an example, the (II2) process causes the processor to cause the HPRS to (i) a dynamic recommendation algorithm performed by referring to the reference database, (ii) a static recommendation algorithm based on pre-input knowledge and (iii) a server characterized in that the specific hyperparameter setting value is obtained by performing at least a part of a genetic algorithm.

As an example, in the (II2) process, the processor causes the learning cluster to optimize the specific AI model for the specific analysis target data by performing data conversion, learning, and evaluation processes in a distributed processing manner, and then the A server characterized in that it is provided to a specific customer is disclosed.

The present invention has the effect of popularizing the automatic recommendation service of an AI model through a platform to which a redundant service flow according to the characteristics of each customer group is applied.

In addition, the present invention has the effect of making the use of the AI model easier by allowing non-experts to implement the AI model based on the GUI without writing code.

1 is a configuration of a model recommendation management server (MRMS) that performs an AI model automatic recommendation subscription service method through a platform to which a duplication service flow according to characteristics of each customer group according to an embodiment of the present invention is applied, and a server operating in conjunction therewith It is a diagram showing interlocking relationships with
2 is a flowchart illustrating an AI model automatic recommendation subscription service method through a platform to which a duplication service flow according to characteristics of each customer group is applied according to an embodiment of the present invention.
3 is a diagram showing a business model according to an AI model automatic recommendation subscription service method through a platform to which a duplication service flow according to characteristics of each customer group is applied according to an embodiment of the present invention.
4 is a block diagram showing an operation process of an MRMS that performs an AI model automatic recommendation subscription service method through a platform to which a duplication service flow according to characteristics of each customer group is applied according to an embodiment of the present invention and servers operating in conjunction therewith. .

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The detailed description of the present invention which follows refers to the accompanying drawings which illustrate, by way of illustration, specific embodiments in which the present invention may be practiced. These embodiments are described in sufficient detail to enable one skilled in the art to practice the present invention. It should be understood that the various embodiments of the present invention are different from each other but are not necessarily mutually exclusive. For example, specific shapes, structures, and characteristics described herein may be implemented in one embodiment in another embodiment without departing from the spirit and scope of the invention. Additionally, it should be understood that the location or arrangement of individual components within each disclosed embodiment may be changed without departing from the spirit and scope of the invention. Accordingly, the detailed description set forth below is not to be taken in a limiting sense, and the scope of the present invention, if properly described, is limited only by the appended claims, along with all equivalents as claimed by those claims. Like reference numbers in the drawings indicate the same or similar function throughout the various aspects.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily practice the present invention.

1 is a configuration of a model recommendation management server (Model Recommending Management Server, MRMS) that performs an AI model automatic recommendation subscription service method through a platform to which a duplication service flow according to characteristics of each customer group is applied according to an embodiment of the present invention, and this It is a diagram showing an interworking relationship with servers operating in conjunction.

Referring to FIG. 1 , the MRMS 100 may include a communication unit 110, a memory 115, and a processor 120. At this time, input/output and calculation processes of the MRMS 100 may be performed by the communication unit 110 and the processor 120, respectively. However, in FIG. 1, a specific connection relationship between the communication unit 110 and the processor 120 is omitted. Also, the memory 115 may be in a state of storing various instructions to be described later, and the processor 120 may perform the present invention by performing processes to be described later by executing the instructions stored in the memory. Even though the MRMS 100 is described as such, it does not exclude the case where the MRMS 100 includes an integrated processor in which a medium, a processor, and a memory are integrated to implement the present invention.

Here, the MRMS 100 includes a Data Preprocessing Recommendation Server (DPRS) 200, a Model and Architecture Recommendation Server (MARS) 300, which are auxiliary detailed servers, and hyperparameter recommendation It can operate in conjunction with the server (Hyperparameter Recommendation Server, HPRS) 400, and each of the DPRS 200, MARS 300, and HPRS 400 will be able to operate in conjunction with each other. The DPRS 200, MARS 300, and HPRS 400 are sub-servers of the MRMS 100, and their processes may be managed and controlled by the MRMS 100, but are not limited thereto. Also, the DPRS (200), MARS (300), and HPRS (400) may also include a communication unit, memory, and processor similarly to the MRMS (100). A detailed operating method of each of the servers 200, 300 and 400 will be described later.

Having described the configuration of the MRMS 100 according to an embodiment of the present invention and the interworking relationship between it and various servers 200, 300, and 400, the AI model automatic recommendation subscription service method of the present invention will be described in detail. will be. To this end, reference is made to FIG. 2 .

2 is a flowchart illustrating an AI model automatic recommendation subscription service method through a platform to which a duplication service flow according to characteristics of each customer group is applied according to an embodiment of the present invention.

First, the MRMS 100 refers to the first metadata obtained by providing the Type 1 automatic recommendation service to the first customer group subscribing to the Type 1 automatic recommendation service whose subscription cost is less than or equal to the threshold value, and provides a second value that exceeds the threshold value. The reference database of the automatic type recommendation service may be optimized (ie, updated) (S01). Thereafter, when the MRMS 100 obtains a specific automatic recommendation request from a specific customer included in the second customer group subscribing to the type 2 automatic recommendation service, a specific AI model corresponding to the specific automatic recommendation request is obtained by referring to the reference database. can be provided to a specific customer (S02). In order to more clearly describe such a subscription service, reference is made to FIG. 3 .

3 is a diagram showing a business model according to an AI model automatic recommendation subscription service method through a platform to which a duplication service flow according to characteristics of each customer group is applied according to an embodiment of the present invention.

Referring to FIG. 3 , the business model of the subscription service provided by the MRMS 100 can be confirmed. That is, the MRMS (100) provides the first customer group users who subscribe to this service at a somewhat low price (not excluding the concept of free), that is, at a cost below the threshold, or who are provided with a limited range of services. Type 1 automatic recommendation service can be provided. Here, the type 1 automatic recommendation service may be a service that is premised on optimizing the reference database using the first metadata obtained while providing the corresponding service. Due to the nature of the low subscription cost, the size of the first customer group will be quite large, so it will be possible to obtain a large amount of first metadata while providing the first automatic recommendation service to them, and using this, it is possible to optimize the reference database can be expected

The reference database optimized as described above may be used to provide a type 2 automatic recommendation service to the second customer group. That is, the MRMS 100 may provide a specific AI model optimized for a specific customer included in the second customer group by using a reference database updated using such a large amount of first metadata. Here, the second customer group is users who have subscribed to the service at a rather high price, that is, a cost exceeding the threshold, and needs to be provided with a differentiated service compared to the first customer group. Therefore, as described above, the type 2 automatic recommendation service is provided using the optimized reference database, and the second metadata generated while providing the corresponding service may not be used to update the reference database. Instead, each of the metadata acquired while providing the type 2 automatic recommendation service to customers included in the second customer group may be used exclusively for providing the service to each customer. That is, for example, metadata A' obtained while providing service to customer A included in the second customer group may be referred to again only when providing service to customer A again later. Through this method, it is possible to provide a type 2 automatic recommendation service optimized for each of the customers who have subscribed to the type 2 automatic recommendation service without a problem of leaking the data of the recommended AI model. Of course, in order to provide an improved user experience to the first customer group, the updated reference database may also be used when performing the type 1 automatic recommendation service.

Through this dual service flow, it will be possible to provide an AI model automatic recommendation subscription service tailored to the needs of each customer group. The specific technical details of MRMS that can realize the above business model will be explained. To this end, a case in which a specific automatic recommendation service is obtained from a specific customer included in the aforementioned second customer group and a type 2 automatic recommendation service is provided to the specific customer will be described. Here, in the case of the type 1 automatic recommendation service, as will be described later, the process of providing the type 2 automatic recommendation service and It may be the same, and therefore, it is believed that those skilled in the art can fully understand the operation of the MRMS 100 only with the following description. It will be described with reference to FIG. 4 below.

4 is a block diagram showing an operation process of an MRMS that performs an AI model automatic recommendation subscription service method through a platform to which a duplication service flow according to characteristics of each customer group is applied according to an embodiment of the present invention and servers operating in conjunction therewith. .

Referring to FIG. 4 , the MRMS 100 may obtain a specific automatic recommendation request (not shown) from a specific customer as an input. Here, the specific automatic recommendation request may include specific analysis target data that a specific customer wants to analyze, and may also include settings for what kind of analysis the specific customer wants to perform with the specific analysis target data. Such a specific automatic recommendation request may be input to the data analyzer 210 by the MRMS 100 through the DPRS 200. Here, the data analyzer 210 may be included in the DPRS 200 or may be a module driven in conjunction with it. The data analyzer 210 is a module that grasps the overall properties of specific analysis target data, and after first classifying the data categories of specific analysis target data, the meta feature analysis operation obtained for each category is applied to the specific analysis target data. By applying it, a specific meta feature corresponding to a specific analysis target data can be obtained. To explain metafeatures, let's first assume that the data categories include tables, images, and plaintext. At this time, the data analyzer 210 determines which of the above-described data categories the specific analysis target data is included in, and if it is determined to be included in the table data category, each field of the table, that is, a categorical field. , meta-feature analysis operations for each of the integer field and the string field can be applied to specific analysis target data. Examples of meta-features obtained by applying a meta-feature analysis operation on an integer field are as follows.

Meta-Feature Name explanation Mutual_Information interdependence information Skewness_Of_Numeric Skewness (Asymmetry) for Numerical Field Kurtosis_Of_Numeric Kurtosis for Numerical Field StdDev_Of_Numeric Standard Deviation for Numerical Field Attribute_Entropy Entropy by Field Nominal_Distinct_Values Frequency of each value for Nominal Field Noise_To_Signal_Ratio Independence from Labeled Class Correlation Interrelationship by field

Here, the data analyzer 210 that performs such a meta-feature analysis operation may have a Chief-Walker structure. That is, the data analyzer 210 may have a distributed cluster structure, and the chief analyzer performs global statistical calculations on specific analysis target data, distributes the data to the worker analyzer, and basic meta-features, which are some of the meta-features. feature can be obtained. The worker analyzer calculates the distributed data to perform local statistical calculations and obtain more specific meta features.

When a specific meta-feature is obtained through the above process, the MRMS (100) inputs it to the DPRS (200), MARS (300), and HPRS (400) in order and calculates it to obtain specific preprocessing settings, specific AI models, and specific hyperparameters. Set values can be obtained. However, here, an embodiment in which only a specific AI model is obtained using only the MARS 300 and then provided to a specific customer is also possible. Such an embodiment is to be performed by the DPRS 200 and the HPRS 400 in an embodiment in which specific meta features are sequentially input to the DPRS 200, the MARS 300, and the HPRS 400 to be described below. Since the preprocessing setting operation and the hyperparameter setting operation need only be omitted, it is believed that those skilled in the art will be able to understand with reference to the following embodiments.

That is, the MRMS 100 may acquire a specific preprocessing setting optimized for specific analysis target data by applying a preprocessing setting operation through the DPRS 200 with a specific meta feature as an input. Specific preprocessing settings may include a preprocessing function for specific analysis target data, a feature extraction method, and dimensionality reduction. Here, each preprocessing function and feature extraction method may be stored in a predetermined storage, and the preprocessing functions may be stored for each data category of specific analysis target data. For example, preprocessing functions for categorical fields among table data categories can include Ordinary Encode, Dummy Encode, and Label Encode, and preprocessing functions for integer fields include Min-Max, z-Score, and Sequence. can include In addition, in the case of plain text data categories, TF-IDF, Bag of Words, and Tokenize may be included. For each meta-feature value, which of these preprocessing functions to use, which feature extraction method to use, and whether to perform dimensionality reduction may be stored in the reference database 310, and the DPRS 200 may be stored in the reference database A specific preprocessing setting may be obtained by finding a metafeature value most similar to a specific metafeature value among pairing data between metafeatures and preprocessing settings recommended up to now included in step 310 . In a state in which pairing data is not sufficiently accumulated in the reference database 310, a preprocessing setting operation may be performed by referring to developer's empirical information. Here, the reference database 310 may be included in the MARS 300 or may be a DB operated in conjunction with it.

Next, the model setting operation performed by MARS 300 will be described. The MRMS 100 can obtain a specific AI model optimized for specific analysis target data by applying a model setting operation with reference to the reference database 310 through the MARS 300 with specific meta features as input. Here, the MARS 300 basically performs a dynamic recommendation algorithm by referring to the reference database 310, but when the number of pairing data accumulated in the reference database 310 is insufficient or there is no appropriate pairing data, the developer's empirical information At least some of pre-input knowledge-based static recommendation algorithms and genetic algorithms may be performed according to . Here, the reference database 310 is continuously (ie, historically) updated, so it can be seen as a dynamic recommendation algorithm, and the knowledge input according to the developer's empirical information is not automatically updated like the reference database 310, so it can be used The algorithm may be referred to as a static recommendation algorithm. In the case of a genetic algorithm, it will be possible to determine a specific AI model based on the Generation or Population value obtained by performing this, but since this is a prior art, a detailed description will be omitted. It is obvious that an embodiment in which the reference database 310 is not used at all is also possible.

When using the reference database 310, the MARS 300 finds a meta-feature most similar to a specific meta-feature by referring to meta-feature-AI model pairing data recommended so far stored in the reference database 310, and A paired AI model can be acquired as a specific AI model. In addition, the MARS 300 may acquire specific preprocessing settings along with specific analysis target data, which may also be used to find an AI model in the reference database 310.

The AI models stored in the reference database 310 may be classified according to the analysis purpose of the user's data to be analyzed, which is initially input. For example, the classifier may include SVM, Decision Tree, KNN, Clustering can include K-Means, DBSCAN, etc. Neural Network can include CNN, DNN, RNN, etc. Word Embed can include Skip-Gram, CBOW, Fasttext, etc.

Finally, the hyperparameter setting operation performed by the HPRS 400 will be described. The MRMS 100 may acquire a specific hyperparameter setting optimized for specific analysis target data by applying a hyperparameter setting operation through the HPRS 400 with a specific metafeature as an input. Here, the hyperparameter setting operation, similar to the model setting operation, performs a dynamic recommendation algorithm by referring to the reference database 310, but when the number of pairing data accumulated in the reference database 310 is insufficient or there is no suitable pairing data. In this case, it may be performed by performing at least a part of a knowledge-based static recommendation algorithm and a genetic algorithm pre-inputted according to developer's empirical information. Since this can be easily understood by those skilled in the art through the description of the MARS 300, additional description will be omitted.

When a specific preprocessing setting, a specific AI model, and a specific hyperparameter setting value are obtained according to the above process, the MRMS 100 can provide the specific preprocessing setting and specific hyperparameter setting value to a specific customer along with the specific AI model. . Here, specific preprocessing settings before learning, specific AI models, and specific hyperparameter settings may be provided separately, but training may be completed and provided for customer convenience. In this case, the MRMS 100 enables the learning cluster 410 to optimize a specific AI model for specific analysis target data by performing data conversion, learning, and evaluation processes, and then provides it to a specific customer . Here, optimization may be achieved by learning a specific AI model using specific analysis target data. At this time, specific preprocessing settings and specific hyperparameter setting values may also be used. In addition, the above process can be performed in a distributed processing method to achieve efficiency.

As described above, while performing the automatic recommendation service according to the present invention, specific pairing data of specific meta features (specific preprocessing setting, specific AI model, and specific hyperparameter setting value) can be obtained. Since the specific pairing data is obtained while performing the type 2 automatic recommendation service according to the above premise, it may not be used to update the reference database, but may not be used to update the reference database. The first metadata, which is pairing data, may be used to update the reference database.

Embodiments according to the present invention described above may be implemented in the form of program instructions that can be executed through various computer components and recorded on a computer-readable recording medium. The computer readable recording medium may include program instructions, data files, data structures, etc. alone or in combination. Program instructions recorded on the computer-readable recording medium may be specially designed and configured for the present invention, or may be known and usable to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, and magneto-optical media such as floptical disks. media), and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include high-level language codes that can be executed by a computer using an interpreter or the like as well as machine language codes such as those produced by a compiler. The hardware device may be configured to act as one or more software modules to perform processing according to the present invention and vice versa.

In the above, the present invention has been described by specific details such as specific components and limited embodiments and drawings, but these are provided to help a more general understanding of the present invention, and the present invention is not limited to the above embodiments. , Those skilled in the art to which the present invention pertains can make various modifications and variations from these machines.

Therefore, the spirit of the present invention should not be limited to the above-described embodiments and should not be determined, and not only the claims described below but also all modifications equivalent or equivalent to these claims fall within the scope of the spirit of the present invention. will do it

Claims (20)

In the AI model automatic recommendation subscription service method through a platform to which a redundant service flow according to the characteristics of each customer group is applied,
(a) The model recommendation management server (MRMS) refers to first metadata obtained by providing the type 1 automatic recommendation service to a first customer group subscribing to the type 1 automatic recommendation service whose subscription cost is less than a threshold value updating a reference database of type 2 automatic recommendation services exceeding the threshold value; and
(b) When the MRMS obtains a specific automatic recommendation request from a specific customer included in the second customer group subscribing to the type 2 automatic recommendation service, a specific automatic recommendation request corresponding to the specific automatic recommendation request is obtained by referring to the reference database. The step of providing the AI model to the specific customer
Including,
In step (a),
The MRMS receives the first metadata, which is pairing data between (i) each of the meta features of the data to be analyzed input by the first customer group and (ii) each of the AI models provided to the first customer group for each of them update the reference database by referring to;
In step (b),
(b1) obtaining, by the MRMS, a specific meta-feature of the specific analysis target data by inputting the data to a data analyzer when specific analysis target data is obtained from at least one specific customer included in the second customer group; and
(b2) The MRMS takes the specific meta-feature as an input and applies a model setting operation with reference to the reference database through a model and architecture recommendation server (MARS), which is a detailed server that operates in conjunction with the MRMS. Acquiring the specific AI model updated in the analysis target data and providing it to the specific customer,
In the step (b1),
The MRMS causes the data analyzer to determine which of the data categories including at least a part of a table, an image, and plain text corresponds to the specific analysis target data, and then takes the specific analysis target data as an input, Obtaining the specific meta feature by applying a meta feature analysis operation corresponding to a specific data category;
In step (b2),
The MRMS causes the MARS to select at least some of (i) a dynamic recommendation algorithm performed by referring to the reference database, (ii) a static recommendation algorithm based on pre-input knowledge, and (iii) a genetic algorithm. Obtaining the specific AI model by performing
In step (b2),
The MRMS takes the specific meta feature as an input and applies a preprocessing setting operation through (i) a preprocessing setting recommendation server (DPRS), which is a detailed server that operates in conjunction with the MRMS. After performing a process of acquiring preprocessing settings and (ii) a process of obtaining updated specific hyperparameter setting values to the specific analysis target data by applying a hyperparameter setting operation through a hyperparameter recommendation server (HPRS), the specific Together with the AI model, providing the specific preprocessing settings and the specific hyperparameter settings to the specific customer;
In step (b2),
The MRMS causes the HPRS to generate at least some of (i) a dynamic recommendation algorithm performed by referring to the reference database, (ii) a static recommendation algorithm based on pre-input knowledge, and (iii) a genetic algorithm AI model automatic recommendation subscription service method, characterized in that to obtain the specific hyperparameter setting value by performing. delete delete delete According to claim 1,
In step (b1),
The MRMS includes (i) a process of causing a chief analyzer included in the data analyzer to perform global statistical calculations of the specific analysis target data, and (ii) a process of causing a worker analyzer included in the data analyzer to perform the specific analysis AI model automatic, characterized in that the process of performing individual statistical calculations of the target data is performed in a distributed cluster manner so that the data analyzer applies the meta-feature analysis operation with the specific analysis target data as input Referral subscription service method. delete delete According to claim 1,
In step (b2),
(i) preprocessing function, (ii) feature extraction method, and (iii) dimension AI model automatic recommendation subscription service method, characterized in that for obtaining the specific pre-processing settings including whether or not to reduce. delete According to claim 1,
In step (b2),
The MRMS causes the learning cluster to update the specific AI model for the specific analysis target data by performing data conversion, learning and evaluation processes in a distributed processing method, and then providing it to the specific customer AI. Model automatic recommendation subscription service method. In a model recommendation management server (MRMS) that performs an AI model automatic recommendation subscription service through a platform to which a redundant service flow according to characteristics of each customer group is applied,
one or more memories to store instructions; and
and one or more processors configured to execute the instructions, wherein the processor is configured to: (I) obtain a type 1 automatic recommendation service for a first customer group subscribing to the type 1 automatic recommendation service for which a subscription cost is less than a threshold value; a process of updating a reference database of a type 2 automatic recommendation service that exceeds the threshold by referring to the first metadata; and (II) when a specific automatic recommendation request is obtained from a specific customer included in the second customer group subscribing to the type 2 automatic recommendation service, a specific AI model corresponding to the specific automatic recommendation request is obtained by referring to the reference database. Characterized in that it is set to perform the process provided to the specific customer,
The (I) process,
The processor receives the first metadata, which is pairing data between (i) each of the meta features of the data to be analyzed input by the first customer group and (ii) each of the AI models provided to the first customer group for each of them. update the reference database by referring to;
The (II) process,
(II1) a process of acquiring, by the processor, a specific meta-feature of the specific analysis target data by inputting the specific analysis target data to a data analyzer when specific analysis target data is obtained from at least one specific customer included in the second customer group; and
(II2) The processor takes the specific meta-feature as an input and applies a model setting operation with reference to the reference database through a model and architecture recommendation server (MARS), which is a detailed server that operates in conjunction with the MRMS. Including a process of obtaining the specific AI model updated in the analysis target data and providing it to the specific customer,
The (II1) process,
The processor causes the data analyzer to determine which of the data categories including at least a part of a table, an image, and plain text corresponds to the specific analysis target data, and then takes the specific analysis target data as an input, Obtaining the specific meta feature by applying a meta feature analysis operation corresponding to a specific data category;
The (II2) process,
The processor causes the MARS to select at least some of (i) a dynamic recommendation algorithm performed by referring to the reference database, (ii) a static recommendation algorithm based on pre-input knowledge, and (iii) a genetic algorithm. Obtaining the specific AI model by performing
The (II2) process,
The processor takes the specific meta-feature as an input and applies a preprocessing setting operation through (i) a preprocessing setting recommendation server (DPRS), which is a detailed server that operates in conjunction with the MRMS. After performing a process of acquiring preprocessing settings and (ii) a process of obtaining updated specific hyperparameter setting values to the specific analysis target data by applying a hyperparameter setting operation through a hyperparameter recommendation server (HPRS), the specific Together with the AI model, providing the specific preprocessing settings and the specific hyperparameter settings to the specific customer;
The (II2) process,
The processor causes the HPRS to generate at least some of (i) a dynamic recommendation algorithm performed by referring to the reference database, (ii) a static recommendation algorithm based on pre-input knowledge, and (iii) a genetic algorithm. Server, characterized in that to obtain the specific hyperparameter setting value by performing. delete delete delete According to claim 11,
The (II1) process,
The processor performs (i) a process of causing a chief analyzer included in the data analyzer to perform global statistical calculation of the specific analysis target data, and (ii) a worker analyzer included in the data analyzer to perform the specific analysis The server characterized in that by performing a process of performing individual statistical calculations of target data in a distributed cluster manner, the data analyzer applies the meta-feature analysis operation with the specific analysis target data as an input. delete delete According to claim 11,
The (II2) process,
(i) preprocessing function, (ii) feature extraction method, and (iii) dimension Server characterized in that for obtaining the specific pre-processing setting including whether or not to reduce. delete According to claim 11,
The (II2) process,
Server characterized in that the processor updates the specific AI model for the specific analysis target data by performing data conversion, learning, and evaluation processes in a distributed processing method, and then providing the data to the specific customer .

Images