JP5946949B1 - DATA ANALYSIS SYSTEM, ITS CONTROL METHOD, PROGRAM, AND RECORDING MEDIUM - Google Patents

Abstract

A data analysis system capable of accurately finding out data related to a predetermined case from an enormous amount of data. A data analysis system extracts a plurality of constituent elements from learning data, and each of the plurality of constituent elements constitutes at least a part of the learning data. From among the components, select the components to be used for the evaluation of the plurality of evaluation data based on the distribution mode in the extracted learning data of the plurality of components, and based on the selected components Evaluate the evaluation data. [Selection] Figure 4

Description

  The present invention relates to a data analysis system for analyzing data, and can be applied to a system having artificial intelligence for analyzing big data, for example.

  As a result of the rapid development of computers and the progress of computerization of society, a huge amount of information (big data) has become widely and closely related to the activities of companies and individuals. Therefore, recently, the necessity of accurately separating desired information from big data has been emphasized.

  As an approach for extracting desired information from big data, a data analysis by a reviewer is applied to a part of the data sampled from the data group, and the remaining data is obtained by using this analysis result. A system capable of automatic analysis is known (for example, JP 2013-182338 A).

JP 2013-182338 A

  According to the data analysis system, data related to a predetermined case can be found out from a huge amount of data. However, even though the degree of data related to a given case is not inherently high, it is evaluated that the data is related to a given case or vice versa. There was a problem. Therefore, an object of the present invention is to provide a system that can accurately find data related to a predetermined case from a vast amount of data.

  The object is a data analysis system for analyzing data, wherein a plurality of evaluation data to be analyzed are stored at least temporarily, and the plurality of evaluation data are evaluated based on learning data. A controller, wherein the controller extracts a plurality of components from the learning data, and each of the plurality of components constitutes at least a part of the learning data and is extracted. The component to be used for the evaluation of the plurality of evaluation data is selected from the plurality of components based on the distribution mode in the learning data of the plurality of extracted components, and the selection is performed. This is achieved by evaluating the evaluation data based on the configured components. Furthermore, a control method of the data analysis system, a program thereof, and a recording medium are provided.

  The above-described disclosure provides a data analysis system and the like that can accurately find data related to a predetermined case from a vast amount of data.

It is a block diagram which shows an example of the hardware constitutions of a data analysis system. It is a figure explaining arrangement | positioning of the component in the data for learning. It is a characteristic view which shows distribution of the evaluation value of each of a some component, and the appearance position in the data for learning of each of a some component. It is a flowchart of the controller of the server apparatus 2 explaining evaluation of the data for evaluation which concerns on 1st Embodiment. It is an operation | movement flowchart of the controller of the server apparatus 2 explaining the evaluation of the data for evaluation which concerns on 2nd Embodiment. It is a control table for the process of integration of component groups.

  Next, an embodiment of a data analysis system will be described based on the accompanying drawings.

[Data analysis system configuration]
FIG. 1 is a block diagram showing an example of a hardware configuration of a data analysis system according to the present embodiment (hereinafter sometimes simply referred to as “system”). The system includes, for example, an arbitrary recording medium (for example, a memory or a hard disk) capable of storing data (including digital data and analog data), and a controller (for example, a control program stored in the recording medium). A computer or a computer system (a system that realizes data analysis by operating a plurality of computers in an integrated manner) that analyzes data stored at least temporarily in the recording medium. Can be realized.

  In the present embodiment, “learning data” (training data) is, for example, data presented to the user as reference data and associated with classification information (classified reference data, a combination of reference data and classification information) ). The learning data may be referred to as “teacher data” or “training data”. “Evaluation data” is data that is not associated with the classification information (unclassified data that is not presented to the user as reference data and is not classified for the user, “unknown data” May be said). Here, the “classification information” may be an identification label used for arbitrarily classifying reference data. For example, the “classification information” widely includes reference data and a predetermined case (the target for which the system evaluates relevance between data). , The range is not limited), and the reference data can be an arbitrary number (for example, two), such as a “Related” label indicating that the two are related to each other and a “Non-Related” label indicating that the two are not related. ) May be classified into groups.

  As illustrated in FIG. 1, the system includes, for example, a server device (server computer) 2 that can execute a main process of data analysis, and one or a plurality of client devices that can execute a related process of the data analysis. (Client computer) 3, a storage system 5 having a database 4 for recording data and evaluation results for the data, and a management computer 6 for providing a management function for data analysis to the client device 3 and the server device 2 And may be provided. Each device is connected as a hardware resource, for example, a memory, a controller, a bus, an input / output interface (for example, a keyboard, a display, etc.), and a communication interface (communication means using a predetermined network) so that the devices can communicate with each other. (Not limited to these examples). The server device 2 includes a (non-temporary) storage medium recording a program and data necessary for data analysis, such as a hard disk, a flash memory, a DVD, a CD, and a BD.

  The client device 3 presents a part of the data to the user as reference data. As a result, the user can input (provide classification information) for evaluation / classification of the reference data via the client device 3. Based on the combination of reference data and classification information (learning data), the server device 2 widely uses patterns (for example, abstract rules, meanings, concepts, styles, distributions, and samples included in the data) from the data. And is not limited to a so-called “specific pattern”), and the relevance between the evaluation data and the predetermined case is evaluated based on the learned pattern.

  The management computer 6 executes predetermined management processing for the client device 3, the server device 2, and the storage system 5. The storage system 5 may be composed of, for example, a disk array system, and may include a database 4 that records data and results of evaluation / classification of the data. The server device 2 and the storage system 5 are communicably connected by a DAS (Direct Attached Storage) method or a SAN (Storage Area Network).

  Note that the hardware configuration shown in FIG. 1 is merely an example, and the above system can be realized by other hardware configurations. For example, a part or all of the processing executed in the server device 2 may be executed in the client device 3, or a part or all of the processing may be executed in the server device 2. Alternatively, the storage system 5 may be built in the server device 2. Further, the user not only performs input for evaluation / classification of sample data via the client device 3 (gives classification information), but also performs the above input via an input device directly connected to the server device 2. You can also. It is understood by those skilled in the art that there can be various hardware configurations capable of realizing the system, and the present invention is not limited to one specific configuration (for example, the configuration illustrated in FIG. 1).

[Data evaluation function]
The system can have a data evaluation function. The data evaluation function evaluates a large number of evaluation data (big data) based on a small number of data (learning data) classified manually. By providing the data evaluation function, the system can, for example, indicate an index indicating the level of relevance between the evaluation data and the predetermined case (for example, a numerical value (for example, a score), a character that enables the evaluation data to be ordered) (Eg, “high”, “medium”, “low”, etc.), symbols (eg, “◎”, “◯”, “△”, “x”, etc.), and / or combinations thereof) The above evaluation can be performed by deriving. The data evaluation function is realized by the controller of the server device 2.

  When the system derives a score as an index for the evaluation, the system can calculate the score by any method. For example, various methods used in the field of machine learning or natural language processing (for example, a K-neighbor method, a method using a support vector machine, a method using a neural network, a method that assumes a statistical model for data (for example, , A method using a Gaussian process, etc.) and / or a method combining these, etc.), or based on various methods used in the field of statistics (eg, configuration) (Based on how often the element appears in the data).

  The “component” may be partial data constituting at least a part of the data, for example, a morpheme, a keyword, a sentence, a paragraph, and / or metadata (eg, e-mail header information) constituting the document. Frame image that constitutes a video, or partial audio that constitutes audio, volume (gain) information, and / or timbre information, partial image that constitutes an image, partial pixel, and / or luminance information , Motion information, and / or 3D information.

  When the system calculates the score based on the frequency at which the component appears in the data, for example, the following calculation method can be considered. First, the system extracts the constituent elements constituting the learning data from the learning data, and evaluates the constituent elements. At this time, for example, the system described above is the degree that a plurality of constituent elements constituting at least a part of the learning data contribute to the combination of data and classification information (in other words, the constituent elements appear according to the classification information). Frequency). As a more specific example, the system evaluates a component using a transmitted information amount (for example, an information amount calculated from a predetermined formula using the appearance probability of the component and the appearance probability of the classification information). By doing so, an evaluation value as evaluation information of the constituent element is calculated according to the following formula 1.

  Here, wgt indicates the initial value of the evaluation value of the i-th component before evaluation. Wgt indicates the evaluation value of the i-th component after the Lth evaluation. γ means an evaluation parameter in the L-th evaluation, and θ means a threshold value in the evaluation. Thereby, for example, the system can be evaluated such that the greater the value of the calculated transmission information amount, the more the component represents the characteristic of the predetermined classification information.

Next, the system associates the component with the evaluation value, and stores both in an arbitrary memory (for example, the storage system 5). Then, the system extracts a component from the evaluation data, inquires whether or not the component is stored in the memory, and if so, an evaluation value associated with the component Is read from the memory, and the evaluation data is evaluated based on the evaluation value. As a more specific example, the system can calculate the score by calculating the following expression using an evaluation value associated with a component constituting at least part of the evaluation data. .
m _j : frequency of occurrence of the i-th component
wgt _i : Evaluation value of the i-th component

  The server device 2 may continue (repeate) the extraction and evaluation of the constituent elements until the recall rate reaches a predetermined target value. The recall is an index indicating the ratio (coverability) of the data to be discovered with respect to a predetermined number of data. For example, when the recall is 80% with respect to 30% of all data, a predetermined case As shown in the figure, 80% of the data to be discovered is included in the data of the top 30% of the index (score). When a person hits the data (linear review) without using a data analysis system, the amount of data to be discovered is proportional to the amount reviewed by the person, so the greater the deviation from this proportionality, the greater the data analysis performance of the system. Will be good.

  The implementation example of the data evaluation function described above is merely an example. In other words, as long as the data evaluation function is a function of “evaluating evaluation data based on learning data”, the specific mode is a specific one configuration (for example, the above-described score calculation method) ) Is not limited.

[Component optimization]
As described above, for evaluation of evaluation data, for example, evaluation values of constituent elements extracted from learning data are used. In this case, if a large number of components with low evaluation values are included in the evaluation data, the evaluation data is high regardless of the true high relevance between the evaluation data and the predetermined case. It may be evaluated.

  Therefore, in the present embodiment, the system uses, for example, the component used for evaluation of the evaluation data from the components extracted from the learning data, and the learning data of the extracted component. The component is optimized by selecting, determining, or extracting based on the distribution mode in Eq., And the evaluation data is evaluated based on the selected component. As a result, the system can accurately determine, determine, classify, etc. the relevance between the evaluation data and the predetermined case. All of the components that are not selected may be used for evaluating the evaluation data, or some components may be used for evaluating the evaluation data and the remaining components may not be used. . For example, the server device 2 may evaluate the evaluation data by using the evaluation value of the selected component as it is, or may evaluate the evaluation data by re-evaluating the selected component, The evaluation data may be evaluated by processing such as increasing the evaluation value of the selected component.

  As described above, the server device 2 uses the distribution form in the learning data of the plurality of extracted constituent elements in order to select the constituent elements. For example, based on the distribution mode, a plurality of components existing in the learning data having a predetermined positional relationship can be selected from a plurality of components extracted from the learning data. Preferably, the distribution of the evaluation values of each of the plurality of constituent elements and the appearance positions of the plurality of constituent elements in the learning data can be used. This will be described in detail below.

  FIG. 2 shows an example of learning data. Each alphabet such as a, b, and c corresponds to a component, and “•” is not extracted as a component, such as particles, adverbs, etc. It is a phrase. FIG. 3 shows the distribution of the evaluation values of each of the plurality of constituent elements and the appearance positions of the plurality of constituent elements in the learning data. The vertical axis is the evaluation value of the component, and the horizontal axis is the appearance position of the component in the learning data. Each bar graph is a component evaluation value. When the smoothing process is performed on the evaluation values of a plurality of constituent elements using, for example, a Gaussian filter, the characteristic indicated by reference numeral 100 is obtained.

  According to this characteristic 100, the superiority or inferiority (for example, the level of the evaluation value) of the constituent elements included in the learning data can be visualized. The component located at the peak (102A to 102I) indicates that it is an element that strongly characterizes the combination of data and classification information (for example, an element that is highly relevant to a predetermined case). At this time, other components (for example, components located in the vicinity of the specific component) having a predetermined positional relationship with the component (herein referred to as “specific component”) are also located at the peak. Under the influence of the constituent element (specific constituent element) (in other words, meaning or meaning related to the specific constituent element), the relevance to the predetermined case may be high.

  Therefore, for example, in the distribution of evaluation values of the component with respect to the appearance position of the component in the learning data, the server device 2 selects the component centering on the peak of the evaluation value. For example, the server device 2 selects the component corresponding to the peak and the components that appear before and after the peak as the “component group”. Here, the component group is, for example, a group of a plurality of components that appear adjacent in the learning data as one group. In FIG. Indicates a group. For example, in the evaluation data, if a, b, and c appear in the order of “a, b, c”, and b has an evaluation value peak, the component group is “a, b , C ″ (words having no meaning between components (the above-mentioned “•”) need not be considered in the component group).

  As can be seen from FIG. 3, since there may be a plurality of peaks, there may be as many component groups as there are peaks. In order to evaluate the evaluation data, the server device 2 may use all the component groups, or may use some component groups based on the magnitude of the peak evaluation value.

  For example, the server device 2 selects a component included in the component group from the components included in the learning data, and evaluates the evaluation data based on the selected component. At that time, for example, in the evaluation data, the server device 2 uses the evaluation value of the evaluation data as compared with the case where the difference (distance) of the appearance position between the constituent elements constituting the constituent element group is small. In addition, when a plurality of components appear in the evaluation data so as to form a group, the evaluation value of the evaluation data may be increased as compared with the case where it is not.

[Evaluation of evaluation data by server device 2]
An evaluation operation of evaluation data by the server device 2 will be described. FIG. 4 is a flowchart of the controller of the server device 2. The controller acquires one or more data as reference data from the evaluation data recorded in the storage system 5 (step S300: reference data acquisition). Each step can be rephrased as a module or means.

  Next, the controller actually reviews the reference data to determine the classification, and acquires the classification information input to the reference data by the user from any input device (S302: acquisition of classification information). The controller composes the learning data by combining the reference data and the classification information, and extracts the constituent elements from the learning data (S304: constituent element extraction).

  Then, the controller evaluates the constituent element (S306: constituent element evaluation), associates the constituent element with the evaluation value, and stores both in the storage system 5 (S308: constituent element storage module). The processing of S300 to S308 corresponds to a “learning phase” (phase in which artificial intelligence learns a pattern). Note that the learning data may be prepared in advance instead of creating from the reference data. For example, when a known document for invalidating a patent related to a certain patent right is found, the learning data is a combination of the description of the scope of claims and the “Related” label.

  The controller creates a distribution (FIG. 2) between the evaluation value of the component and the appearance position of the component for the plurality of components extracted from the learning data (S310: component element distribution creation). As described above, the peak of the evaluation value of the component is determined from the distribution (S312: distribution processing). Then, the controller selects a component group based on the determined peak (S314: component group selection), and records the component belonging to the selected component group and its evaluation value in the storage system 5. .

  Next, the controller acquires evaluation data from the storage system 5 (S316: acquisition of evaluation data). The controller further reads out the constituent element and its evaluation value from the storage system 5, and extracts the constituent element from the evaluation data (S318: constituent element extraction). The controller evaluates the evaluation data based on the evaluation value associated with the constituent element (S320: evaluation data evaluation), and creates a plurality of evaluation data in order information (ranking). The higher the evaluation data, the higher the relevance to the predetermined case. The processing after S310 becomes the evaluation phase with respect to the learning phase. It should be noted that each process included in the above-described flowchart is an example and does not indicate a limited aspect.

  According to the embodiment described above, since the evaluation data can be evaluated by selecting a component having a higher relevance to the predetermined case from among the components extracted from the learning data, Data related to a given case can be found accurately.

<Second Embodiment>
Next, a second embodiment of the data analysis system will be described. A feature of this embodiment is that, by using the evaluation results of the constituent elements included in the learning data, the learning data is divided into a plurality of segments, and each of the plurality of segments is used as a plurality of new learning data. It is to be used for data evaluation. For example, the learning data is divided into a plurality of segments by dividing the constituent elements of the learning data into predetermined patterns based on the distribution of the constituent elements extracted from the learning data in the learning data. be able to. More specifically, a plurality of segments can be set in the learning data by integrating a plurality of constituent element groups selected from the learning data based on the relevance with a predetermined case.

  The operation of the data analysis system according to the second embodiment will be described based on the operation flowchart (FIG. 5) of the controller of the server device 2. Processing (S300 to S314) until the controller selects a component group is the same as that in the first embodiment. In S400, the controller integrates component groups related to each other to create an integrated group (component group integration). The integration of the component group will be specifically described.

  The component groups are arranged without using a word that does not become a component (the above-mentioned “·”), arranged with a small number of the relevant words, or with the last component of the component group When the first component of a component group is a component group related to each other such as the same term, it is expected that the meaning, meaning, etc. of the plurality of component groups are related to each other. The component group is integrated into an integrated group. The server device 2 stores a process of integrating a plurality of component groups in the control table of FIG. 6 and records it in a predetermined area of the memory.

  In FIG. 6, each of the constituent element groups of group numbers (1) to (5) corresponds to an “integrated group” as a single constituent element group, and the constituent element groups of group numbers (6) and (7) are integrated. To become integrated group # 6, as shown in FIG. In FIG. 6, the component group evaluation value is a maximum value as a representative value of the evaluation values of a plurality of component elements belonging to the component group, and the integrated group evaluation value is an evaluation of the component group belonging to the integrated group. It is the maximum value as a representative value.

  Even if the component groups are integrated, the number of integrated groups (# 1 to # 11) may still be large by itself, so the controller further integrates the integrated groups (S402: integrated group integration). The controller obtains the peak of the maximum value of the integrated group (the maximum value distinguished by “*” in FIG. 6) from the distribution of the maximum value of the integrated group, and sets a segment in which the integrated group is integrated for each peak ( Segment setting). FIG. 6 shows that three segments 1, 2, and 3 are set in the learning data. Therefore, as shown in FIG. 2, the controller can divide the learning data into three parts I (segment 1), II (segment 2), and III (segment 3).

  When the controller proceeds to evaluation of evaluation data (S404 (S316 to S320)), the controller refers to the control table (FIG. 6) and evaluates the evaluation data based on the three segments. As the number of learning data data increases, the reproducibility described above can be improved. When evaluating the evaluation data, the controller may use the components of each of the plurality of trainings and their evaluation values, or newly extract the components for each learning data and obtain the evaluation values for use. May be.

[Data format processed by the data analysis system]
In the present embodiment, “data” may be any data expressed in a format that can be processed by a computer. The data may be, for example, unstructured data whose structure definition is incomplete at least in part, and document data (for example, e-mail (attached file header) Information), technical documents (including a wide range of documents explaining technical matters such as academic papers, patent publications, product specifications, design drawings, etc.), presentation materials, spreadsheets, financial statements, meeting materials, Record reports, sales documents, contracts, organization charts, business plans, company analysis information, electronic medical records, web pages, blogs, comments posted on social network services, etc., audio data (eg conversation / music) Data), image data (eg, data composed of a plurality of pixels or vector information), video data (eg, Broadly includes data formed) including a plurality of frame images (not limited to these examples).

  For example, when analyzing document data, the system extracts morphemes contained in document data as learning data as constituent elements, evaluates the constituent elements, and extracts from the document data as evaluation data. Based on the element, the relevance between the document data and the predetermined case can be evaluated. When analyzing voice data, the system may analyze the voice data itself, or convert the voice data into document data by voice recognition, and use the converted document data as an analysis target. Good. In the former case, for example, the system divides the voice data into partial voices of a predetermined length to form components, and uses the voice analysis method (for example, hidden Markov model, Kalman filter, etc.) to convert the partial voices. By identifying, the voice data can be analyzed. In the latter case, speech is recognized using an arbitrary speech recognition algorithm (for example, a recognition method using a hidden Markov model), and the same procedure as described above is performed on the recognized data (document data). Can be analyzed. When analyzing image data, the system, for example, divides the image data into partial images of a predetermined size to form components, and any image recognition method (for example, pattern matching, support vector machine, neural network) Etc.) can be used to identify the partial image. Further, when analyzing video data, the system, for example, divides a plurality of frame images included in the video data into partial images each having a predetermined size to form a component, and an arbitrary image recognition technique (for example, a pattern The video data can be analyzed by identifying the partial image using matching, a support vector machine, a neural network, or the like.

[Example of implementation using software and hardware]
The control block of the above system may be realized by a logic circuit (hardware) formed in an integrated circuit (IC chip) or the like, or may be realized by software using a CPU. In the latter case, the system includes a CPU that executes a program (control program for the data analysis system) that is software that implements each function, and a ROM (in which the program and various data are recorded so as to be readable by the computer (or CPU)). A Read Only Memory) or a storage device (these are referred to as “recording media”), a RAM (Random Access Memory) for developing the program, and the like are provided. And the objective of this invention is achieved when a computer (or CPU) reads the said program from the said recording medium and runs it. As the recording medium, a “non-temporary tangible medium” such as a tape, a disk, a card, a semiconductor memory, a programmable logic circuit, or the like can be used. The program may be supplied to the computer via an arbitrary transmission medium (such as a communication network or a broadcast wave) that can transmit the program. The present invention can also be realized in the form of a data signal embedded in a carrier wave in which the program is embodied by electronic transmission. Note that the above program can be implemented in any programming language. Also, any recording medium that records the above program falls within the scope of the present invention.

[Application example]
Such systems include, for example, discovery support systems, forensic systems, e-mail monitoring systems, medical application systems (eg, pharmacovigilance support systems, clinical trial efficiency systems, medical risk hedging systems, fall prediction (fall prevention) systems, prognosis predictions) System, diagnosis support system, etc.), Internet application system (eg, smart mail system, information aggregation (curation) system, user monitoring system, social media management system, etc.), information leakage detection system, project evaluation system, marketing support system, Artificial intelligence systems that analyze big data, such as intellectual property evaluation systems, fraud monitoring systems, call center escalation systems, credit check systems And may be implemented the association between predetermined cases as any system) can be evaluated. Depending on the field to which the data analysis system of the present invention is applied, in consideration of circumstances peculiar to the field, for example, preprocessing (for example, extracting an important part from the data and extracting only the important part from the data) The analysis target may be applied), or the mode of displaying the data analysis result may be changed. It will be understood by those skilled in the art that a variety of such variations can exist, and all variations fall within the scope of the present invention.

  The present invention is not limited to the above-described embodiments, and various modifications can be made within the scope of the claims, and the technical means disclosed in different embodiments can be appropriately combined. Embodiments to be made are also included in the technical scope of the present invention. Furthermore, a new technical feature can be formed by combining the technical means disclosed in each embodiment.

  1 ... Data analysis system, 2 ... Server device, 3 ... Client device, 4 ... Database, 5 ... Storage system, 6 ... Management computer

Claims (10)

A data analysis system for analyzing data,
A memory for at least temporarily storing a plurality of evaluation data to be analyzed;
A controller that evaluates the plurality of evaluation data based on learning data;
With
The controller is
A plurality of components are extracted from the learning data, and each of the plurality of components constitutes at least a part of the learning data.
From the relationship between the extracted evaluation information of each of the plurality of constituent elements and the position at which each of the plurality of constituent elements appears in the learning data, the distribution mode of the plurality of constituent elements in the learning data is obtained. ,
From among the plurality of components, the components used in the evaluation of the plurality of evaluation data, selected based on the mode of the distribution,
A data analysis system for evaluating the evaluation data based on the selected component. The controller is
The reference data presented to the user and the combination of classification information set in the reference data by the user are used as the learning data,
Based on the degree of contribution to the combination of each of the plurality of constituent elements, generating evaluation information for each of the plurality of constituent elements,
The data analysis system according to claim 1, wherein the evaluation data is evaluated by generating an index for ranking the evaluation data based on the generated evaluation information. The controller is
Based on the distribution mode, among the plurality of components extracted from the learning data, a plurality of components existing in the learning data having a predetermined positional relationship are converted into the plurality of evaluation data. The data analysis system according to claim 1 or 2 , wherein the data analysis system is selected as a component used for evaluation of the data. The controller is
Obtaining at least one peak of evaluation information of a plurality of constituent elements extracted from the learning data based on the distribution mode;
The data analysis system according to claim 3 , wherein a component used for evaluating the plurality of evaluation data is selected from the plurality of components based on the peak. The controller is
Determining a component group including a component corresponding to the position of the peak and another component located in the vicinity of the component of the learning data;
The data analysis system according to claim 4 , wherein a plurality of components belonging to the component group are selected as components used for evaluating the plurality of evaluation data. The controller is
Dividing the learning data into a plurality of segments based on the distribution mode;
The plurality of divided segments are used as new learning data,
Evaluating the plurality of evaluation data based on the new plurality of learning data;
The data analysis system according to any one of claims 1 to 5 . The controller is
Classifying the plurality of selected components into a plurality of groups based on the distribution aspect;
Determining the plurality of segments in the process of integrating the plurality of groups;
The data analysis system according to claim 6 . A method of controlling a data analysis system that evaluates a plurality of evaluation data based on learning data,
The data analysis system is
Extracting a plurality of components from the learning data, each of the plurality of components constituting at least a part of the learning data; and
From the relationship between the extracted evaluation information of each of the plurality of constituent elements and the position at which each of the plurality of constituent elements appears in the learning data, the distribution mode of the plurality of constituent elements in the learning data is obtained. Steps,
From among the plurality of components, the method comprising the components used in the evaluation of the plurality of evaluation data is selected based on the mode of the distribution,
Evaluating the evaluation data based on the selected component;
Control method to execute. A program for causing a computer to perform data analysis for evaluating a plurality of evaluation data based on learning data,
Extracting a plurality of components from the learning data, each of the plurality of components constituting at least a part of the learning data; and
From the relationship between the extracted evaluation information of each of the plurality of constituent elements and the position at which each of the plurality of constituent elements appears in the learning data, the distribution mode of the plurality of constituent elements in the learning data is obtained. Steps,
From among the plurality of components, the method comprising the components used in the evaluation of the plurality of evaluation data is selected based on the mode of the distribution,
Evaluating the evaluation data based on the selected component;
A program that causes a computer to execute. A computer-readable recording medium on which the program according to claim 9 is recorded.