KR101330526B1 - Method, Apparatus and Recording Medium for Migration of Prediction Systems - Google Patents

Abstract

One embodiment of the present invention relates to a method for converting a prediction system, an apparatus thereof, and a recording medium.

According to an embodiment of the present invention, the first training data of the first prediction system is input, the first training data is input to the first prediction system to obtain a first prediction result value, and the first prediction result value is input. A learning data converter configured to generate second learning data in addition to the first learning data; The reliability of the first prediction result added when the second training data is generated is checked, and if the reliability is less than or equal to the threshold value, the second training data is removed. If the reliability exceeds the threshold, the second training data is converted into the second prediction system. A learning data selection unit for selecting as learning data for learning the information; And a training data input unit configured to input the selected second training data to the second prediction system.

According to an embodiment of the present invention, there is an effect of providing a conversion function from one prediction system to another prediction system, and at this time, resolving a shortage of initial learning data that may occur in another prediction system.

Prediction system, learning data

Description

Prediction system conversion method, apparatus and recording medium {Method, Apparatus and Recording Medium for Migration of Prediction Systems}

One embodiment of the present invention relates to a method for converting a prediction system, an apparatus thereof, and a recording medium. More particularly, the present invention relates to a prediction system conversion method, an apparatus and a recording medium for providing a conversion function from one prediction system to another.

In the conventional fields of machine learning, data mining, etc., "Unsupervised Prediction System" predicts the user's intention based on the vast amount of user usage history data. ) "Has been studied a lot. Although this unsupervised prediction system has some advantages of predicting user intention using simple user usage history data, the unsupervised prediction system predicts user intention by using user history data indirectly indicating user intention. Compared with a prediction system that directly receives user preferences as user feedback data and reflects the results, it shows a relatively low prediction performance.

As such, by directly receiving user preferences from the user as user feedback data and learning the results by reflecting the results, a prediction system showing a relatively high prediction performance is compared to an unsupervised prediction system. Supervised Prediction System. "

As a result, the transition from an unsupervised prediction system using simple user usage history data to a supervised prediction system requiring user participation has been made. There is a need for a technique that utilizes an unsupervised prediction system to implement a supervisored prediction system. However, in the unsupervised prediction system constructed from simple learning data such as user usage history data without the existing prediction result value, a supervisored prediction system using learning data such as user feedback data including the prediction result value is used. In the case of performing the conversion, the initial training data of the supervisored prediction system may be insufficient.

Therefore, there is a demand for a technology that provides a function of converting from one prediction system to another prediction system. In this case, a technology for solving a shortage of initial learning data that may occur in another prediction system is required.

Against this background, it is an object of one embodiment of the present invention to provide a conversion function from one prediction system to another.

Another object of one embodiment of the present invention is to solve a lack of initial learning data that may occur in another prediction system when converting from one prediction system to another prediction system.

According to an embodiment of the present invention, in a prediction system transformation apparatus for providing a conversion function between prediction systems, the first learning data of the first prediction system is input, and the first learning data is input to the first prediction system. A learning data converter configured to input a first prediction result value and to generate second learning data by adding the obtained first prediction result value to the input first learning data; Verifying the reliability of the first prediction result value added when the second learning data is generated, removing the second learning data if the confirmed reliability is less than or equal to a threshold, and the verified reliability exceeding the threshold A learning data selector configured to select the second learning data as learning data for learning a second prediction system; And a learning data input unit configured to input the selected second learning data to the second prediction system so as to learn the second prediction system based on the selected second learning data. To provide.

According to an embodiment of the present invention, a prediction system transformation method provided by a prediction system transformation apparatus includes receiving first training data of a first prediction system and converting the input first training data into the first prediction system. A learning data conversion step of obtaining a first prediction result value by inputting the input signal to the second learning data, and generating second learning data by adding the obtained first prediction result value to the input first learning data; Confirming the reliability of the first prediction result value added when the second learning data is generated, removing the second learning data if the confirmed reliability is less than or equal to a threshold, and the verified reliability exceeding the threshold Learning data selection step of selecting the second learning data as learning data for learning a second prediction system; And a learning data input step of inputting the selected second learning data to the second prediction system to train the second prediction system based on the selected second learning data. Provide a method.

In addition, according to an embodiment of the present invention, in a recording medium on which a program for executing a prediction system transformation method is recorded, first learning data of a first prediction system is received, and the first learning data is input to the first learning data. A learning data conversion function for obtaining a first prediction result value by inputting to a prediction system and generating second learning data by adding the obtained first prediction result value to the input first learning data; Verifying the reliability of the first prediction result value added when the second learning data is generated, removing the second learning data if the confirmed reliability is less than or equal to a threshold, and the verified reliability exceeding the threshold A learning data selection function for selecting the second learning data as learning data for learning a second prediction system; and the selected second to learn the second prediction system based on the selected second learning data. A computer readable recording medium having recorded thereon a program for implementing a learning data input function for inputting learning data into the second prediction system is provided.

As described above, according to an embodiment of the present invention, there is an effect of providing a conversion function from one prediction system to another prediction system.

In addition, according to an embodiment of the present invention, there is an effect of resolving a shortage of initial learning data that may occur in another prediction system when converting from one prediction system to another.

Conventionally, a "second prediction system", which is a prediction system using learning data including a prediction result in a "first prediction system", which is a prediction system constructed from simple training data without a prediction result value. In the case of conversion to "," that is, when providing a conversion function between prediction systems, a problem may occur that the initial learning data of the second prediction system is insufficient.

In order to solve this problem, the prediction system transformation method and the apparatus according to an embodiment of the present invention, the learning of the second prediction system from the training data of the first prediction system (hereinafter referred to as "first learning data") Providing a technique for converting to data (hereinafter referred to as " second training data "), and using this training data transformation technique to solve the above-mentioned problems, i.e., an initial lack of initial training data in the second prediction system. Migrate from the first prediction system to the second prediction system. That is, the present invention generates the second learning data necessary for the second prediction system to be newly constructed by using the first learning data of the first prediction system that has been previously constructed, thereby converting the first prediction system to the second prediction system. Provided are a prediction system conversion apparatus, a prediction system conversion method, and a recording medium that provide a technique for conversion.

Hereinafter, some embodiments of the present invention will be described in detail through exemplary drawings. In adding reference numerals to the components of each drawing, it should be noted that the same reference numerals are assigned to the same components as much as possible even though they are shown in different drawings. In the following description of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear.

In addition, in describing the component of this invention, terms, such as 1st, 2nd, A, B, (a), (b), can be used. These terms are intended to distinguish the constituent elements from other constituent elements, and the terms do not limit the nature, order or order of the constituent elements. When a component is described as being "connected", "coupled", or "connected" to another component, the component may be directly connected to or connected to the other component, It should be understood that an element may be "connected," "coupled," or "connected."

1 is a block diagram of a prediction system transformation apparatus 100 according to an embodiment of the present invention that provides a transformation function between prediction systems.

As illustrated in FIG. 1, the prediction system transformation apparatus 100 according to an embodiment of the present invention may transform training data for converting first training data of a first prediction system into second training data of a second prediction system. The unit 110 selects the training data to remove only the second training data whose reliability is less than or equal to a predetermined threshold value from the transformed second training data and selects only highly reliable second training data that exceeds the predefined threshold value. And a learning data input unit 130 for inputting the selected second learning data to the second prediction system.

The first prediction system and the second prediction system mentioned above each perform the prediction function by using the first learning data and the second learning data, respectively, the first learning data of the first prediction system and the second prediction system of the second prediction system. The difference between 2 learning data is the presence or absence of the prediction result value. That is, the first training data of the first prediction system does not have a prediction result value, and the training data of the second prediction system has a prediction result value.

Therefore, in order to convert from the first learning data of the first prediction system to the second learning data of the second prediction system, it is necessary to add the prediction result value. To this end, the above-described learning data converter 110 receives the first learning data of the first prediction system, inputs the input first learning data into the first prediction system, and obtains a first prediction result value. By generating the second learning data by adding the obtained first prediction result value to the input first learning data, the first learning data of the first prediction system is converted into the second learning data of the second prediction system. .

The training data of the prediction system must be useful and meaningful data for the prediction system to learn the training data and obtain meaningful prediction results. Therefore, only useful and meaningful data are selected from the second learning data generated by the above-described learning data converter 110. To this end, the above-described learning data selector 120 calculates the reliability of the first prediction result value added to the first learning data when the second learning data is generated by the learning data converter 110, and calculates the reliability. If the calculated reliability is less than or equal to a predetermined threshold value, the corresponding second training data is removed. If the calculated reliability exceeds the threshold, the second training data is selected as training data for learning the second prediction system.

The above-described learning data input unit 130 may generate the selected second learning data to learn a second prediction system different from the first prediction system based on the second learning data selected by the learning data selection unit 120. 2 Enter into the prediction system.

The above-described learning data selector 120 calculates the reliability of the first prediction result value added to the first learning data when generating the second learning data in the learning data converter 110, based on the calculated reliability. Among the second training data generated by the training data converter 110, the second training data corresponding to the reliability that is less than or equal to the threshold is removed based on the predefined threshold and the second training data corresponding to the reliability that exceeds the threshold In the screening method, such a "confidence calculation method for the first prediction system for the screening of the second learning data" may be used as an example.

The above-described learning data selector 120 may include first learning data corresponding to the first prediction result value added when the second learning data generated by the learning data converter 110 is generated, and the first prediction system. The distance between a reference attribute for a predefined prediction model may be calculated, and the inverse of the calculated distance between the first training data and the reference characteristic may be calculated as "reliability".

In the above-described "method of calculating the reliability for the first prediction system for the selection of the second training data", the prediction model of the first prediction system may be very diverse, among which the "prediction model through regression" The case and the "prediction model through clustering" will be described below.

First, when the prediction model predefined in the first prediction system is a “prediction model through regression,” the above-described learning data selection unit 120 may calculate the reliability in a regression manner. The distance between the characteristic graph as the reference characteristic for the model and the first training data may be calculated, and the inverse of the calculated distance between the first training data and the characteristic graph as the reference characteristic may be calculated with reliability.

The regression method described above is a method of fitting a parameter of a mathematical model to first training data, and a simple example thereof is a linear regression method. For example, if the first training data of the first prediction system consists of (x, y), the first prediction system finds a and b in "y = ax + b" (prediction model), and the relationship between x and y Where "y = ax + b" (prediction model) is correct, the first training data should be close to the "y = ax + b" prediction model. Therefore, the first training data located far from the "y = ax + b" prediction model is the first training data that does not fit the prediction model, and is determined as an outlier. In this case, the inverse of the distance between the "y = ax + b" prediction model and the first training data (x, y) may be calculated as a confidence value. Accordingly, the training data selector 120 may remove the second training data generated from the first training data determined as the outlier since the calculated confidence value is equal to or less than a specific value (threshold value).

Next, when the predictive model defined in the first prediction system is a "prediction model through clustering," the above-described learning data selection unit 120 may calculate the reliability in a clustering manner, and predict through clustering. Compute the distance between the training data cluster center value as a reference characteristic for the model and the first training data that has generated the second training data, and between the first training data and the center of the training data cluster as reference characteristics. The inverse of the calculated distance can be calculated with confidence.

That is, in the aforementioned clustering method, the first training data is clustered (also referred to as 'grouping'), and the closer to the center value of the cluster (also referred to as 'group'), the closer to the center value of the first training data. It can be seen as the first training data that shows the characteristics of the cluster well. Therefore, the distance between the center value of the cluster (group) and the first training data may be calculated, and the inverse of the calculated distance may be calculated as a confidence value. When the calculated confidence value is less than a specific value (threshold value), the corresponding first learning The data may be determined as an outlier. Accordingly, the training data selector 120 may remove the second training data generated from the first training data determined as the outlier because the calculated confidence value is equal to or less than a specific value (threshold value).

2 is a flowchart illustrating a prediction system transformation method provided by the prediction system transformation apparatus 100 according to an embodiment of the present invention.

As illustrated in FIG. 2, the prediction system transformation method according to an embodiment of the present invention includes a training data transformation step of converting first training data of the first prediction system into second training data of the second prediction system (S200). ), The training data selection step (S202) of removing the second training data whose reliability is less than or equal to a predetermined threshold among the converted second training data, and selecting only the reliable second training data that exceeds the predefined threshold. ), And a learning data input step S204 for inputting the selected second learning data to the second prediction system.

In the above-described learning data conversion step (S200), the first learning data of the first prediction system is input, the first learning data is input to the first prediction system to obtain a first prediction result value, and the obtained The second training data is generated by adding the first prediction result value to the input first training data.

In the above-described learning data selection step (S202), the reliability of the first prediction result value added when the second learning data is generated is calculated, and when the calculated reliability is less than or equal to the threshold value, the second learning data is removed, and the calculated reliability If is greater than the threshold value, the second training data is selected as training data for training the second prediction system.

In the above-described learning data input step (S204), the selected second learning data is input to the second prediction system in order to learn the second prediction system based on the selected second learning data.

The prediction system conversion method according to the embodiment of the present invention described above may be implemented in a program and recorded in a computer-readable recording medium.

The program recorded in the recording medium for implementing the prediction system transformation method according to an embodiment of the present invention, receives the first learning data of the first prediction system, the input first learning data to the first prediction system A learning data conversion function for inputting to obtain a first prediction result value and generating second learning data by adding the obtained first prediction result value to the input first learning data; The reliability of the first prediction result value added in the generation of the second training data is calculated, and if the calculated reliability is less than or equal to the threshold, the second training data is removed. Learning data selection function for selecting as learning data for learning the second prediction system; And a learning data input function for inputting the selected second learning data to the second prediction system in order to learn the second prediction system based on the selected second learning data.

Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage, and the like, which are also implemented in the form of a carrier wave (for example, transmission over the Internet). It also includes. The computer readable recording medium may also be distributed over a networked computer system so that computer readable code can be stored and executed in a distributed manner.

In addition, functional programs, codes, and code segments for implementing the present invention can be easily inferred by programmers in the art to which the present invention belongs.

In the above, the apparatus, method and recording medium for providing a migration function from the first prediction system to another second prediction system have been described. The first prediction system and the second prediction system are various methods. It can be implemented as. As an example of the implementation method, Neural Network, Collaborative Filtering, Support Vector Machine, Bayesian Network, Genetic Algorithm, Decision Tree Case Based Reasoning, Pattern Recognition, Reinforcement Learning, and the like may be applied, and there may be various implementation methods without being limited to these examples. In particular, when the first prediction system and the second prediction system are implemented by collaborative filtering, the filter parameters of the collaborative filter of the first prediction system are set to the initial parameters of the second prediction system. Can be used as

The first prediction system referred to herein may be an unsupervised prediction system that predicts user intention based on user usage history data that is first learning data. Such an unsupervised prediction system is a prediction system that has been studied in the technical fields such as machine learning and data mining.

The second prediction system referred to herein is a prediction system different from the first prediction system, and is a supervised prediction system that predicts user intention based on user feedback data, which is second training data. Can be.

Since the user usage history data in the Unsupervised prediction system mentioned above is indirectly representing user intention, the Unsupervised prediction system indicates user preference data from the user. As compared to the supervised prediction system which receives the feedback and reflects the result, the prediction performance is relatively low.

3 is a view for explaining an example of applying the prediction system transformation method according to an embodiment of the present invention.

In FIG. 3, the first prediction system uses a vast amount of user usage history data 3100 as first learning data, and uses the unsupervised prediction system 310 to predict user intention. ), And the second prediction system is a supervised prediction system 320 that predicts user intention based on user feedback data 3200 as second learning data.

Referring to FIG. 3, the prediction system transformation method according to an exemplary embodiment of the present invention uses an unsupervised prediction of user intention based on simple user usage history data 3100 for a user as first learning data. In the prediction system 310, the user's intention is predicted based on the user feedback data 3200 as the second learning data, and the result is converted into a supervisory prediction system 320 that reflects user participation. To this end, the user feedback data, which is the second training data in the supervisored prediction system 320, may be utilized by using the user usage history data 3100, which is the first training data in the unsupervised prediction system 310. By converting to 3200, a migration from the unsupervised prediction system 310 to the supervisored prediction system 320 is possible.

The conversion from the user usage history data 3100 of the unsupervised prediction system 310 to the user feedback data 3200 of the supervisored prediction system 320 converts the user usage history data 3100 into the unsuperbar. The prediction result value is inputted to the predicted prediction system 310 to be output, and the output prediction result value is added to the user usage history data 3100 to the user feedback data 3200 of the super-biased prediction system 320. To convert.

The converted user feedback data 3200 is subjected to a process of removing (generated) user feedback data from the user history data 3100 having low reliability of the corresponding prediction result value, that is, the converted user feedback data 3200. ), Only the user feedback data having high reliability of the corresponding prediction result value is selected and used to train the super biased prediction system 320.

As described above, according to an embodiment of the present invention, a second prediction system such as one prediction system (a first prediction system such as an unsupervised prediction system) and the other prediction system (such as a supervisored prediction system) This has the effect of providing a conversion to).

In addition, according to an embodiment of the present invention, when converting from one prediction system (a first prediction system such as an unsupervised prediction system) to another prediction system (a second prediction system such as a supervised prediction system) In addition, there is an effect of resolving a shortage of initial learning data that may occur in another prediction system (a second prediction system such as a supervised prediction system).

While the present invention has been described in connection with what is presently considered to be the most practical and preferred embodiments, it is to be understood that the invention is not limited to the disclosed embodiments. That is, within the scope of the present invention, all of the components may be selectively coupled to one or more of them. In addition, although all of the components may be implemented as one independent hardware, some or all of the components may be selectively combined to perform a part or all of the functions in one or a plurality of hardware. As shown in FIG. Codes and code segments constituting the computer program may be easily inferred by those skilled in the art. Such a computer program may be stored in a computer readable storage medium and read and executed by a computer, thereby implementing embodiments of the present invention. As the storage medium of the computer program, a magnetic recording medium, an optical recording medium, a carrier wave medium, or the like may be included.

It is also to be understood that the terms such as " comprises, "" comprising," or "having ", as used herein, mean that a component can be implanted unless specifically stated to the contrary. But should be construed as including other elements. All terms, including technical and scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs, unless otherwise defined. Commonly used terms, such as predefined terms, should be interpreted to be consistent with the contextual meanings of the related art, and are not to be construed as ideal or overly formal, unless expressly defined to the contrary.

The foregoing description is merely illustrative of the technical idea of the present invention, and various changes and modifications may be made by those skilled in the art without departing from the essential characteristics of the present invention. Therefore, the embodiments disclosed in the present invention are intended to illustrate rather than limit the scope of the present invention, and the scope of the technical idea of the present invention is not limited by these embodiments. The protection scope of the present invention should be interpreted by the following claims, and all technical ideas within the equivalent scope should be interpreted as being included in the scope of the present invention.

As described above, an embodiment of the present invention is applied to a field of prediction system technology using machine learning, data mining, and the like, and converts a function from one prediction system to another prediction system. It is a very useful invention for generating the effect provided.

1 is a block diagram of a prediction system transform apparatus according to an embodiment of the present invention;

2 is a flowchart illustrating a prediction system transformation method according to an embodiment of the present invention;

3 is a view for explaining an example of applying the prediction system transformation method according to an embodiment of the present invention.

Description of the Related Art

100: prediction system converter

110: training data converter

120: learning data selection unit

130: training data input unit

310: Unsupervised Prediction System

320: Supervised Prediction System

Claims (7)

A prediction system transformation apparatus providing a conversion function between prediction systems, Receiving first training data of a first prediction system, inputting the input first training data to the first prediction system to obtain a first prediction result, and inputting the obtained first prediction result to the input A learning data converter configured to generate second learning data in addition to the first learning data; The reliability of the first prediction result value added when the second training data is generated is calculated, and if the calculated reliability is less than or equal to a threshold value, the second training data is removed, and the calculated reliability exceeds the threshold value. A learning data selector configured to select the second learning data as learning data for learning a second prediction system; And A learning data input unit for inputting the selected second learning data to the second prediction system to train the second prediction system based on the selected second learning data. Predictive system conversion apparatus comprising a. The method of claim 1, The learning data selection unit, When calculating the reliability of the first prediction result value added in the generation of the second training data, Calculating a distance between the first training data corresponding to the first prediction result value and a reference characteristic for a predefined prediction model in the first prediction system, and calculating the inverse of the calculated distance as the reliability; Prediction system conversion apparatus, characterized in that. 3. The method of claim 2, The learning data selection unit, And when the predefined prediction model is a prediction model through regression, the reference characteristic is a characteristic graph. 3. The method of claim 2, The learning data selection unit, And the reference characteristic is a learning data cluster center value when the predefined prediction model is a prediction model through clustering. The method of claim 1, The first prediction system is an unsupervised prediction system that predicts user intention based on user usage history data, which is the first training data, and the second prediction system is the second learning system. Prediction system conversion apparatus, characterized in that the supervised (Supervised) prediction system for predicting the user intention based on the user feedback data that is the data. A prediction system conversion method provided by a prediction system conversion device, Receiving first training data of a first prediction system, inputting the input first training data to the first prediction system to obtain a first prediction result, and inputting the obtained first prediction result to the input A training data transformation step of generating second training data in addition to the first training data; The reliability of the first prediction result value added when the second training data is generated is calculated, and if the calculated reliability is less than or equal to a threshold value, the second training data is removed, and the calculated reliability exceeds the threshold value. Learning data selection step of selecting the second learning data as learning data for learning a second prediction system; And Learning data input step of inputting the selected second learning data to the second prediction system to learn the second prediction system based on the selected second learning data. Prediction system conversion method comprising a. A recording medium having recorded thereon a program for executing a prediction system conversion method, Receiving first training data of a first prediction system, inputting the input first training data to the first prediction system to obtain a first prediction result, and inputting the obtained first prediction result to the input A learning data transformation function for generating second learning data in addition to the first learning data; The reliability of the first prediction result value added when the second training data is generated is calculated, and if the calculated reliability is less than or equal to a threshold value, the second training data is removed, and the calculated reliability exceeds the threshold value. A learning data selection function for selecting the second learning data as learning data for learning a second prediction system; And A learning data input function for inputting the selected second learning data to the second prediction system to learn the second prediction system based on the selected second learning data. A computer-readable recording medium that records a program that implements the program.

Images