JP2019159539A - Metadata evaluation device, metadata evaluation method, and metadata evaluation program - Google Patents

Abstract

To evaluate the effectiveness of attribute items included in metadata in terms of dividing collected data.SOLUTION: A data set generation unit 21 divides an original data set by attribute items included in metadata attached to data that belongs to the original data set received by an operation unit 13. A feature extraction unit 22 extracts features of the nature of a data group that belongs to each divided data set for the divided data sets. For each combination of two divided data sets, an evaluation unit 23 acquires a partial evaluation based on the features of the nature of the data group in the combination, and further evaluates the attribute items related to the metadata received by an operation unit 13 based on the partial evaluation of each combination.SELECTED DRAWING: Figure 2

Description

  The present invention relates to a technique for evaluating the effectiveness of an attribute item included in metadata added to data.

  Conventionally, utilization of collected data (data sets) has been studied in various fields. In utilizing data, metadata added to the data body is used. The metadata is data indicating attributes of the data body. For example, Patent Document 1 discloses metadata of sensing data sensed by a sensor. Patent Document 2 discloses a technique for ranking a data set using metadata.

JP 2015-226102 A Japanese Patent Laid-Open No. 2006-58082

  However, some of the attribute items included in the metadata include an item for dividing the data set into data sets having different properties, and an item for dividing the data set into data groups having similar properties. The properties of the data group mentioned here include the type of event, the frequency of occurrence of the event, the tendency of occurrence of the event, and the like. Therefore, in terms of dividing the data set, it is said that the attribute item that divides the data set into data groups having similar properties is low in effectiveness.

  Further, as the number of attribute items included in the metadata increases, the data capacity increases. Therefore, from the viewpoint of dividing the data set, the metadata attribute item that divides the data set into data groups having similar properties unnecessarily increases the data capacity of the data set.

  An object of the present invention is to provide a technique for evaluating the effectiveness in terms of dividing a data set with respect to attribute items included in metadata.

  The metadata evaluation apparatus according to the present invention is configured as follows to achieve the above object.

  The attribute accepting unit accepts designation of an attribute item included in metadata added to data belonging to the original data set. The data set generation unit divides the original data set and generates a plurality of divided data sets based on the attribute items received by the attribute reception unit. The feature extraction unit extracts, for each of a plurality of divided data sets, characteristics of the characteristics of the data group belonging to the divided data set. Then, for each combination of two divided data sets, the evaluation unit acquires a partial evaluation based on the characteristics of the characteristics of the data group in the combination, and further receives the meta data received by the attribute reception unit based on the partial evaluation of each combination. Evaluate attribute items related to data.

  In this configuration, the evaluation unit acquires a partial evaluation indicating how similar the properties of the data group are for each combination of two divided data sets. The properties of the data group include the type of event, the frequency of occurrence of events, the tendency of occurrence of events, and the like. When the attribute item included in the metadata is an item for dividing the original data set into data sets having different properties, the properties of the data groups are not similar in the combination of the two divided data sets. On the other hand, if the attribute item included in the metadata is not an item for dividing the original data set into data sets having different properties, the properties of the data groups are similar in the combination of the two divided data sets.

  Therefore, the evaluation unit can evaluate the validity of the attribute item relating to the metadata received by the attribute reception unit from the viewpoint of dividing the original data set into data sets having different properties. In other words, the evaluation unit can also evaluate the validity of the attribute item relating to the metadata received by the attribute reception unit, from the viewpoint of not dividing the original data set into data sets having different properties.

  Further, the evaluation unit may be configured to acquire a partial evaluation for each combination of two divided data sets based on the similarity of the characteristics of the characteristics of the data group in the combination.

  Further, the feature extraction unit may be configured to extract a likelihood function of the data group as a characteristic feature of the data group belonging to the divided data set.

  According to the present invention, it is possible to evaluate the effectiveness in terms of dividing the data set with respect to the attribute item included in the metadata.

It is the schematic which shows an example metadata evaluation system to which the metadata evaluation apparatus concerning this invention is applied. It is a block diagram which shows the structure of the principal part of a metadata evaluation apparatus. It is a flowchart which shows operation | movement of a data evaluation apparatus.

  Embodiments of the present invention will be described below.

<1. Application example>
FIG. 1 is a schematic diagram showing an example of a metadata evaluation system to which the metadata evaluation apparatus according to the present invention is applied. The metadata evaluation system according to this example includes a metadata evaluation device 1, an original data set storage database 2 (original data set storage DB 2), and a divided data set storage unit 3. The divided data set storage unit 3 is provided with a plurality of databases. FIG. 1 shows only the first data set storage database 3a (first data set storage DB 3a) and the second data set storage database 3b (second data set storage DB 3b) provided in the divided data set storage unit 3. ing.

  The original data set is stored in the original data set storage DB 4. The original data set is a collected data group. The data is composed of a data body and metadata indicating attributes of the data body. The data body is, for example, data relating to an N-dimensional real vector, data relating to an N-dimensional real time series vector, and the like. Specifically, for example, the data itself is face image data, and the attribute items indicated by the metadata are gender, age, nationality, shooting date, camera used for shooting, etc. (in this case, the original data A set is a collection of face image data of men and women of various ages.) For example, the data body is image data of the vehicle, and the attribute items indicated by the metadata are the vehicle type, shooting location, shooting date, camera used for shooting, etc. (in this case, the original data set is 2 It is a collection of image data of vehicles of various vehicle types such as wheel cars, light cars, ordinary cars, trucks, buses, etc.). In addition, for example, the data body is audio data, and the attribute items indicated by the metadata are items such as gender, age, nationality, recording date and time, microphone used for recording (in this case, the original data set is This is a collection of audio data concerning the utterances of men and women of various ages.)

  The metadata evaluation apparatus 1 accepts designation of metadata attribute items, and generates a plurality of divided data sets by dividing the original data set stored in the original data set storage DB 2 with the received attribute items. To do. The divided data set is a partial data group of the original data set. The metadata evaluation device 1 stores the divided data sets obtained by dividing the original data set stored in the original data set storage DB 2 in the divided data set storage unit 3 while distinguishing them. Specifically, when the metadata evaluation apparatus 1 divides the original data set stored in the original data set storage DB 2 into two divided data sets, the metadata evaluation apparatus 1 stores one of the divided data sets in the first data set storage DB 3a. The other divided data set is stored in the second data set storage DB 3b. Further, when the metadata evaluation apparatus 1 divides the original data set stored in the original data set storage DB 2 into three divided data sets, the metadata evaluation apparatus 1 stores the first divided data set in the first data set storage DB 3a. The second divided data set is stored in the second data set storage DB 3b, and the third divided data set is stored in a third data set storage DB (not shown).

  For each combination of two divided data sets generated by dividing the original data set stored in the original data set storage DB 2, the metadata evaluation apparatus 1 is similar in the properties of data groups belonging to the two divided data sets. Is estimated. Specifically, feature characteristics are extracted from data groups belonging to two divided data sets, and the similarity of the extracted features is acquired. The properties of the data group mentioned here include the type of event, the frequency of occurrence of the event, the tendency of occurrence of the event, and the like.

  Therefore, with respect to the item of the attribute of the metadata used for the division of the original data set stored in the original data set storage DB 2 this time, the effectiveness in terms of dividing the original data set is shown as two divided data sets. It can be estimated by the similarity of the properties of data groups belonging to. In other words, the effectiveness of the attribute items included in the metadata can be evaluated from the viewpoint of dividing the original data set into data sets having different properties. In other words, the effectiveness of the attribute items included in the metadata can be evaluated from the viewpoint of not dividing the original data set into data sets having different properties.

<2. Configuration example>
FIG. 2 is a block diagram illustrating a configuration of a main part of the metadata evaluation apparatus. The metadata evaluation apparatus 1 includes a control unit 11, a database access unit 12 (DB access unit 12), an operation unit 13, and an output unit 14.

  The control unit 11 controls the operation of each part of the metadata evaluation apparatus 1 main body. The control unit 11 includes a data set generation unit 21, a feature extraction unit 22, and an evaluation unit 23. Details of the data set generation unit 21, the feature extraction unit 22, and the evaluation unit 23 will be described later.

  The DB access unit 12 is a database provided in the original data set storage DB2 and the divided data set storage unit 3 (in the figure, only the first data set storage DB 3a and the second data set storage DB 3b are shown). Interface. The metadata evaluation device 1 reads and writes data to and from the databases provided in the original data set storage DB 2 and the divided data set storage unit 3 via the DB access unit 12.

  The operation unit 13 has an input device such as a keyboard and a mouse, and accepts an operator's input operation to the main body of the metadata evaluation apparatus 1. The operation unit 13 corresponds to the attribute receiving unit referred to in the present invention.

  The output unit 14 outputs an evaluation result evaluated from the viewpoint of dividing the original data set with respect to the metadata attribute item to an external device. The external device may be a display device that displays an evaluation result evaluated from the viewpoint of dividing the original data set, for example, with respect to metadata attribute items, and the original data set is divided based on the evaluation result It may be an information processing apparatus (computer) that generates and outputs aggregated data in which metadata attribute items are classified from the viewpoint of.

  Next, the data set generation unit 21, the feature extraction unit 22, and the evaluation unit 23 included in the control unit 11 will be described.

  The data set generation unit 21 divides the original data set stored in the original data set storage DB 4 by the metadata attribute item specified by the operation unit 13 to generate a plurality of divided data sets. In the data set generation unit 21, for example, the original data set stored in the original data set storage DB 2 is face image data of men and women of various ages, and the specified metadata attribute item is gender. The first divided data set obtained by extracting the female face image data from the original data set and the second divided data set obtained by extracting the male face image data from the original data set are generated.

  The divided data sets generated by dividing the original data set by the data set generation unit 21 are stored in the divided data set storage unit 3 in a distinguished manner. For example, in the above example, the first divided data set obtained by extracting the female face image data from the original data set is stored in the first data set storage DB 3a, and the second divided data set obtained by extracting the male face image data from the original data set. The data set is stored in the second data set storage DB 3b.

  The feature extraction unit 22 extracts features of the characteristics of the data group for each divided data set stored separately in the divided data set storage unit 3. As described above, the characteristics of the data group include the type of event, the occurrence frequency of the event, the occurrence tendency of the event, and the like. Further, the characteristics of the properties of the data group are the distribution of properties such as the type of event, the frequency of occurrence of events, and the tendency of occurrence of events. The feature extraction unit 22 extracts, for example, a probability distribution function, probability density function, or likelihood function of the properties of the data group as features.

  The evaluation unit 23 compares the characteristics of the properties of the data group extracted by the feature extraction unit 22 for each combination of two divided data sets stored separately in the divided data set storage unit 3. The evaluation unit 23 performs partial evaluation on the metadata attribute item used to divide the original data set this time from the viewpoint of dividing the original data set, based on the similarity of the compared features. In terms of dividing the original data set, the metadata attribute item is less effective as the characteristics of the divided data set are more similar (higher in similarity). The evaluation unit 23 evaluates the feature similarity of the data set extracted by the feature extraction unit 22 using a distance between probability distributions or a probability density ratio. In addition, the evaluation unit 23 divides the original data set into data sets having different properties with respect to the metadata attribute items used for dividing the original data set this time using the partial evaluation acquired for each combination of the divided data sets. Efficacy is evaluated from the viewpoint of

  The control unit 11 of the metadata evaluation apparatus 1 includes a hardware CPU, a memory, and other electronic circuits. When the hardware CPU executes the metadata evaluation program according to the present invention, the hardware CPU operates as the data set generation unit 21, the feature extraction unit 22, and the evaluation unit 23. Further, the memory has an area for developing the metadata evaluation program according to the present invention and an area for temporarily storing data generated when the metadata evaluation program is executed. The control unit 11 may be an LSI in which a hardware CPU, a memory, and the like are integrated. The hardware CPU is a computer that executes the metadata evaluation method according to the present invention.

  Further, the original data set storage DB2, the first data set storage DB 3a, the second data set storage DB 3b, and the like included in the divided data set storage unit 3 may be auxiliary storage devices such as a hard disk drive and a solid state drive. Further, the original data set storage DB2, the first data set storage DB3a, the second data set storage DB3b, and the like included in the divided data set storage unit 3 may be configured by one auxiliary storage device or a plurality of auxiliary storages. You may comprise with an apparatus.

<3. Example of operation>
Next, the operation of the metadata evaluation apparatus 1 according to this example will be described. FIG. 3 is a flowchart showing the operation of the metadata evaluation apparatus. The metadata evaluation apparatus 1 executes the process shown in FIG. 3 when an evaluation process for the metadata attribute item is requested. The metadata evaluation device 1 accepts the designation of the attribute item of the metadata to be evaluated this time (s1). The operator operates an input device provided in the operation unit 13 to input the metadata attribute item to be evaluated this time to the metadata evaluation apparatus 1.

  The metadata evaluation apparatus 1 divides the original data set stored in the original data set storage DB 2 by the metadata attribute item received in s1 (s2). In s2, the data set generation unit 21 divides the original data set by the metadata attribute item accepted this time, and generates a plurality of divided data sets. The data set generation unit 21 distinguishes the generated plurality of divided data sets and stores them in the divided data set storage unit 3. For example, when the data set generation unit 21 divides the original data set stored in the original data set storage DB 2 into two divided data sets, the data set generation unit 21 stores one divided data set in the first data set storage DB 3 a, Are stored in the second data set storage DB 3b. In addition, when the data set generation unit 21 divides the original data set stored in the original data set storage DB 2 into three divided data sets, the data set generation unit 21 stores the first divided data set in the first data set storage DB 3 a. The second divided data set is stored in the second data set storage DB 3b, and the third divided data set is stored in a third data set storage DB (not shown).

  The metadata evaluation device 1 extracts features for each divided data set generated in s2 (s3). The feature extraction unit 22 executes the process related to s3. For each divided data set generated in s2, the feature extraction unit 22 uses a probability distribution function and a probability density for properties such as event type, event occurrence frequency, and event occurrence tendency in the data group belonging to the divided data set. Functions or likelihood functions are extracted as features.

  The metadata evaluation apparatus 1 selects a combination of divided data sets (s4). The combination of the divided data sets selected in s4 is a combination of two divided data sets. The metadata evaluation device 1 acquires the similarity of the features extracted in s3 for the divided data set of the combination selected in s4 (s5). In s5, the distance between probability distributions or the ratio of probability density is calculated according to the type of feature extracted in s3, and the degree of similarity corresponding to the calculated value (corresponding to partial evaluation in this invention). To get. The metadata evaluation apparatus 1 determines whether or not there is a combination of divided data sets that have not been subjected to the processing related to s5 (s6), and if there is a combination of unprocessed divided data sets, returns to s4.

  If the metadata evaluation apparatus 1 determines that there is no combination of unprocessed divided data sets in s6, the metadata evaluation apparatus 1 performs an evaluation process for evaluating the metadata attribute item received in s1 (s7). s7 is processing for obtaining an evaluation value for the metadata attribute item to be evaluated this time (the metadata attribute item received in s1). This evaluation value may be, for example, the average value of similarities acquired in s5 for each combination of divided data sets, or the maximum value or the minimum value of similarities acquired in s5 for each combination of divided data sets. Alternatively, the root mean square of the similarities acquired at s5 for each combination of the divided data sets may be used, or the sum of the similarities acquired at s5 may be set for each combination of the divided data sets. The process concerning s4-s7 mentioned above is performed by the evaluation part 23. FIG.

  The metadata evaluation device 1 outputs the metadata attribute item specified in s1 and the evaluation value acquired in s7 as a pair to the external device (s8), and ends this processing.

  Note that if the metadata attribute item received in s1 is a scheme in which the evaluation value acquired in s7 increases as the feature is similar in s5, the property value of the original data set increases as the evaluation value increases. The effectiveness is low in terms of dividing the data sets into different data sets (the smaller the evaluation value, the higher the effectiveness in terms of dividing the original data set into data sets having different properties). Conversely, if the metadata attribute item received in s1 is a method in which the evaluation value acquired in s7 is smaller as the feature is more similar in s5, the original data set is smaller as the evaluation value is smaller. The effectiveness is low in terms of dividing into data sets with different properties (the higher the evaluation value, the higher the effectiveness in dividing the original data set into data sets with different properties).

  Thus, in this example, the effectiveness in terms of dividing the original data set into data sets having different properties can be determined for each metadata attribute item. Therefore, the item of the attribute included in the metadata can be determined with the effectiveness in terms of dividing the original data set into data sets having different properties, and the capacity of the data set can be suppressed.

  In this example, the effectiveness in terms of not dividing the original data set into data sets having different properties can be determined for each metadata attribute item.

<Experimental example 1>
Effectiveness from the viewpoint of dividing the original data set into divided data sets having different properties with respect to the five items of age, sex, nationality, shooting time, and camera used for shooting, the original data set is face image data. Was evaluated by the metadata evaluation apparatus 1.
In this evaluation, it was confirmed that the age and sex were highly effective, and the camera used for photographing was not effective.
Regarding nationality and shooting time, it was not possible to clearly determine whether it was effective.

From this evaluation result, it was confirmed that the capacity of the data set can be reduced by not including in the metadata items related to the camera used for shooting.
Furthermore, by dividing the original data set by age or gender that has been confirmed to be highly effective, it is possible to obtain useful analysis results by analyzing changes in facial image characteristics by age or gender. Was also confirmed.

<Experimental example 2>
The original data set is the image data of the vehicle, and the effectiveness in terms of dividing the original data set into divided data sets having different properties with respect to the four items of vehicle type, shooting location, shooting time, and camera used for shooting. Evaluation was performed using the metadata evaluation apparatus 1.
In this evaluation, it was confirmed that the vehicle type was highly effective, and the shooting time and the camera used for shooting were low.
Regarding the shooting location, it was not possible to clearly determine whether it was effective.

  From this evaluation result, it was confirmed that the capacity of the data set can be reduced by not including the shooting time and the items related to the camera used for shooting in the metadata.

<4. Modification>
In the above example, the metadata evaluation device 1 outputs the evaluation value of the metadata attribute item as a numerical value. For example, the metadata evaluation device 1 ranks it into five or ten predetermined levels, and uses the rank as an evaluation value. It may be configured to output.

  In addition to the evaluation value acquired in s7, the metadata evaluation apparatus 1 may output the similarity for each combination of the divided data sets acquired in s5.

  Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. Further, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, you may combine suitably the component covering different embodiment.

Furthermore, the correspondence between the configuration according to the present invention and the configuration according to the above-described embodiment can be described as in the following supplementary notes.
<Appendix>
An attribute accepting unit (13) for accepting designation of an attribute item included in metadata added to data belonging to the original data set;
A data set generation unit (21) that generates a plurality of divided data sets by dividing the original data set with the attribute items received by the attribute reception unit (13);
For each of the plurality of divided data sets, a feature extraction unit (22) that extracts characteristics of the characteristics of the data group belonging to the divided data set;
For each combination of the two divided data sets, a partial evaluation based on the characteristics of the characteristics of the data group in the combination is acquired, and further, the metadata received by the attribute receiving unit is applied based on the partial evaluation of each combination A metadata evaluation apparatus (1) comprising: an evaluation unit (23) that evaluates attribute items.

DESCRIPTION OF SYMBOLS 1 ... Metadata evaluation apparatus 2 ... Original data set storage database (original data set storage DB)
3 ... Storage unit 3 ... Divided data set storage unit 3a ... First data set storage database (first data set storage DB)
3b ... Second data set storage database (second data set storage DB)
11 ... Control unit 12 ... Access unit 12 ... Database access unit (DB access unit)
DESCRIPTION OF SYMBOLS 13 ... Operation part 14 ... Output part 21 ... Data set production | generation part 22 ... Feature extraction part 23 ... Evaluation part

Claims (5)

An attribute accepting unit that accepts designation of an attribute item included in metadata attached to data belonging to the original data set;
In the attribute item received by the attribute receiving unit, a data set generating unit that generates a plurality of divided data sets by dividing the original data set;
For each of the plurality of divided data sets, a feature extraction unit that extracts characteristics of the characteristics of the data group belonging to the divided data set;
For each combination of the two divided data sets, a partial evaluation based on the characteristics of the characteristics of the data group in the combination is acquired, and further, the metadata received by the attribute receiving unit is applied based on the partial evaluation of each combination A metadata evaluation apparatus comprising: an evaluation unit that evaluates attribute items.   The metadata evaluation apparatus according to claim 1, wherein the evaluation unit acquires a partial evaluation for each combination of the two divided data sets based on similarity of characteristics of data group characteristics in the combination.   The metadata evaluation apparatus according to claim 1, wherein the feature extraction unit extracts a likelihood function of the data group as a characteristic feature of the data group belonging to the divided data set. A data set generation step of generating a plurality of divided data sets by dividing the original data set with the attribute items included in the metadata attached to the data belonging to the original data set received by the attribute receiving unit;
For each of the plurality of divided data sets, a feature extraction step for extracting characteristics of data groups belonging to the divided data sets;
For each combination of the two divided data sets, a partial evaluation based on the characteristics of the characteristics of the data group in the combination is acquired, and further, the metadata received by the attribute receiving unit is applied based on the partial evaluation of each combination A metadata evaluation method in which a computer executes an evaluation step for evaluating an attribute item. A data set generation step of generating a plurality of divided data sets by dividing the original data set with the attribute items included in the metadata attached to the data belonging to the original data set received by the attribute receiving unit;
For each of the plurality of divided data sets, a feature extraction step for extracting characteristics of data groups belonging to the divided data sets;
For each combination of the two divided data sets, a partial evaluation based on the characteristics of the characteristics of the data group in the combination is acquired, and further, the metadata received by the attribute receiving unit is applied based on the partial evaluation of each combination A metadata evaluation program for causing a computer to execute an evaluation step for evaluating an attribute item.

Images