JP2017027495A - Verification device, classification system, verification method, classification method, and computer program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve accuracy and efficiency of text classification processing.SOLUTION: A verification device comprises: a classifier creation section for creating a classifier for verification object classification items of a verification object hierarchy in classification structure data, using verification positive example learning data and verification negative example learning data of the verification object classification items; a goodness-of-fit calculation section for calculating the goodness of fit of the classifier, using the verification positive example learning data or the verification negative example learning data of the verification object classification items of the verification object hierarchy in the classification structure data; and a hierarchy setting section for determining adoption of the verification object classification items in the verification object hierarchy on the basis of the goodness of fit.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a data classification technique.

  For example, Patent Documents 1 and 2 are known as conventional text classification techniques. In the text classification technique disclosed in Patent Document 1, texts are classified into hierarchical categories. In the text classification technique of Patent Document 2, in the hierarchical classification of documents, an information category is configured as a binary tree including nodes of a binary tree including information related to search.

JP 2006-251975 A JP 2006-18829 A

  However, in the conventional text classification technique described above, the classification accuracy may decrease as the hierarchy becomes deeper in the hierarchical category. In addition, when the category hierarchy structure is large, an event that includes an unnecessary category in the actual classification may occur, and the efficiency of the classification process may deteriorate.

  The present invention has been made in view of such circumstances, and provides a verification apparatus, a classification system, a verification method, a classification method, and a computer program capable of improving the accuracy and efficiency of text classification processing. The task is to do.

(1) According to one aspect of the present invention, in a verification apparatus that verifies classification structure data indicating a classification item to which data is classified and a hierarchy of classification items, a correct label is assigned among the labels belonging to the classification structure data. Among the verified verification teacher data, the verification correct example learning data that is the correct verification teacher data to which the correct label including the identification number of the verification target classification item among the classification items of the verification target hierarchy in the classification structure data is given And a verification negative example learning data that is the verification teacher data of the negative example to which the correct answer label not including the identification number of the verification target classification item is given, and generates a classifier for the verification target classification item A classifier generating unit that performs verification of the verification teacher data of the positive example and the verification teacher data different from the verification positive example learning data. Generated by the classifier generator using positive example test data or verification negative example test data that is the verification teacher data of the negative example and the verification teacher data different from the verification negative example learning data A fitness calculation unit that calculates the fitness of the classifier, and a hierarchy setting unit that determines adoption of the verification target classification item in the verification target hierarchy based on the fitness calculated by the fitness calculation unit; , A verification device comprising:
(2) According to one aspect of the present invention, in the verification device according to (1), the verification target hierarchy is verified in order from an upper hierarchy to a lower hierarchy in the classification structure data, and the non-adoption is determined This is a verification device that rejects a lower-level classification item belonging to a verification target classification item without executing the verification procedure.

(3) According to one aspect of the present invention, the verification device according to any one of (1) and (2) above and a classifier of classification structure data in which a classification item verification result by the verification device is reflected are input. And a classification unit that classifies data.
(4) According to one aspect of the present invention, in the classification system of (3), the classification unit starts classification of the input data from the highest hierarchy in the classification structure data, and belongs to the classification item of classification success. It is a classification system that performs only the lower layer classification sequentially.

(5) According to one aspect of the present invention, in the verification method of classification structure data indicating a classification item to which data is classified and a hierarchy of classification items, the verification device has a correct label among the labels belonging to the classification structure data. Verification example learning that is the verification example data of the correct example to which the correct answer label including the identification number of the verification target classification item among the classification items of the verification target hierarchy in the classification structure data is given among the provided verification teacher data A classifier for the verification target classification item using data and verification negative example learning data that is the verification teacher data of the negative example to which the correct answer label not including the identification number of the verification target classification item is assigned A classifier generation step to generate, and the verification device is the verification teacher data of the positive example among the verification teacher data before being different from the verification positive example learning data The classifier using the verification positive example test data that is verification teacher data or the verification negative example test data that is the verification teacher data of the negative example that is different from the verification negative example learning data A fitness calculation step for calculating the fitness of the classifier generated by the generation step; and the verification device in the verification target hierarchy based on the fitness calculated by the fitness calculation step And a hierarchy setting step for determining adoption of a classification item.

(6) In one aspect of the present invention, a verification step of verifying classification structure data by the verification method of (5) above, and a classification unit classifying classification structure data in which a classification item verification result in the verification step is reflected And a classification step of classifying input data using a container.

(7) One aspect of the present invention is a computer program for verifying classification structure data indicating a classification item to which data is classified and a hierarchy of classification items, and among the labels belonging to the classification structure data Verification that is the verification teacher data of the correct example to which the correct answer label including the identification number of the verification target classification item among the classification items of the verification target hierarchy in the classification structure data is provided among the verification teacher data to which the correct label is assigned Using the positive example learning data and the verification negative example learning data that is the verification teacher data of the negative example given the correct answer label that does not include the identification number of the verification target classification item, the verification target classification item A classifier generating step for generating a classifier, and the verification teacher data of the positive example among the verification teacher data, which is different from the verification positive example learning data Using the verification positive example test data that is the verification teacher data or the verification negative example test data that is the verification teacher data of the negative example that is different from the verification negative example learning data A fitness calculation step for calculating the fitness of the classifier generated by the generator generation step, and the adoption of the verification target classification item in the verification target hierarchy based on the fitness calculated by the fitness calculation step A computer program for causing a computer to execute a hierarchy setting step for determining
(8) According to one aspect of the present invention, in the computer program of (7), the classification of input data is performed using a classification structure data classifier that reflects a classification item verification result that is an execution result of the computer program. A computer program for causing a computer to further execute a classification step to be performed.

  According to the present invention, it is possible to improve the accuracy and efficiency of text classification processing.

It is a lineblock diagram showing classification system 1 concerning one embodiment of the present invention. It is a figure which shows the structural example of the classification | category structure data which concern on one Embodiment of this invention. It is a flowchart of the verification method which concerns on one Embodiment of this invention. It is a figure which shows the example of reflection of the classification item verification result with respect to the classification | category structure data which concerns on one Embodiment of this invention. 3 is a flowchart of a classification method according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 is a configuration diagram showing a classification system 1 according to an embodiment of the present invention. The classification system 1 shown in FIG. 1 includes a verification module 11, a classifier generation module 12, a content classification module 13, and a learning database (learning DB) 14.

[Learning database]
The learning DB_14 stores classification structure data and teacher data. The classification structure data is created in advance and stored in the learning DB_14. For creating classification structure data, for example, Grounded Theory Approach technology can be used.

  The classification structure data is data indicating a classification item to which the data is classified and a hierarchy of the classification items. FIG. 2 is a diagram illustrating a configuration example of classification structure data according to the present embodiment. In FIG. 2, the classification structure data has three layers from the first layer to the third layer, and has a classification item for each layer.

  In the classification structure data shown in FIG. 2, the classification item of the first hierarchy is “music”, and the identification number of the classification item “music” of the first hierarchy is “1”. The category items of the second layer, which is the lower layer belonging to the category item “music” of the first layer, are “genre” and “tone”, and the identification number of the category item “genre” of the second layer is “1”. The identification number of the classification item “tone color” in the second hierarchy is “2”. The classification items of the third layer, which is the lower layer belonging to the classification item “genre” of the second layer, are “classic” and “pops”, and the identification number of the classification item “classic” of the third layer is “1”. The identification number of the classification item “pops” in the third hierarchy is “2”. The classification items of the third layer, which is the lower layer belonging to the classification item “timbre” of the second layer, are “about”, “bright” and “warm”, and the identification number of the classification item “about” of the third layer is “ The identification number of the classification item “bright” in the third hierarchy is “2”, and the identification number of the classification item “warm” in the third hierarchy is “3”.

  The classification structure data shown in FIG. 2 is used, for example, when text data is classified by music genre or impression. In the classification structure data shown in FIG. 2, the first hierarchy is a destination for classifying text data whose topic is “music” among the text data to be classified. The second hierarchy is a destination for classifying text data related to “genre” or “tone” among text data whose topic is “music”. The third hierarchy is text data whose topic is “music” and classifies text data related to specific classification items of “genre” or “tone” among text data related to “genre” or “tone”. Become.

  The text data classified by the classification structure data shown in FIG. 2 is given a label composed of the identification numbers of the classified previous classification items. FIG. 2 shows labels “111”, “112”, “121”, “122”, and “123” assigned to the text data classified up to the third hierarchy. For example, a label “111” as a classification result is assigned to text data “My favorite music is classic”. A label “123” as a classification result is assigned to the text data “I like music with a warm tone color”. As text data classified up to the second level, for example, a label “11” as a classification result is assigned to “I like any genre as long as it is music”. As text data classified up to the first level, for example, a label “1” as a classification result is given to “I like music”.

  The teacher data is text data to which the correct label is assigned among the labels belonging to the classification structure data in the learning DB_14. The text data of the teacher data is, for example, text data of a sentence that is freely described as an answer to a questionnaire conducted for the general public. Data to which the correct label belonging to the classification structure data is assigned to the text data is used as teacher data. For example, a human analyzes text data to determine the correct label of the text data, and assigns the correct answer label of the determination result to the text data.

[Classifier generation module]
The classifier generation module 12 generates a classifier using the classification structure data and the teacher data in the learning DB_14. The classifier generation module 12 sorts the teacher data for each label in the classification structure data in the learning DB_14. In the example of the classification structure data shown in FIG. 2, each of the eight labels “1”, “11”, “12”, “111”, “112”, “121”, “122”, and “123” Teacher data is sorted.

  In the classification of the teacher data for each label, the correct label given to the teacher data and the classification target label are compared with the classification target label, and the data is classified into the teacher data in which both match and the teacher data in which they do not match. For example, for the classification target label “111”, among the teacher data in the learning DB_14, the teacher data to which the label “111” is assigned is grouped into a positive example data group of the classification target label “111”, and the label “111” is grouped. Is grouped into the negative example data group of the sorting target label “111”. Further, for example, for the classification target label “11”, among the teacher data in the learning DB_14, the teacher data to which the label “11” is assigned is grouped into a positive data group of the classification target label “11”, and the label “11” is grouped. Teacher data to which “11” is not assigned is grouped into a negative example data group with a sorting target label “11”. Further, for example, for the classification target label “1”, among the teacher data in the learning DB_14, the teacher data to which the label “1” is assigned is grouped into the positive example data group of the classification target label “1”, and the label “ Teacher data to which “1” is not assigned is grouped into a negative example data group with a sorting target label “1”.

  The classifier generation module 12 generates a classifier for the classifier generation target label using the positive example data group and the negative example data group of the classifier generation target label. The classifier for the classifier generation target label has a function of determining whether the text data to be classified corresponds to the classifier generation target label. As the classifier, for example, a classifier such as SVM (Support Vector Machine) or “Bayesian Network” may be used. Alternatively, a positive example vector is generated from the positive example data group of the classifier generation target label, a negative example vector is generated from the negative example data group of the classifier generation target label, and the vector similarity between the positive example vector and the negative example vector is determined. A classifier may be generated based on the basis.

[Verification module]
The verification module 11 verifies the classification structure data in the learning DB_14. The operation of the verification module 11 will be described with reference to FIG. FIG. 3 is a flowchart of the verification method according to the present embodiment. The procedure of the verification method shown in FIG. 3 (steps S111 to S114) is executed in order from the upper hierarchy to the lower hierarchy in the classification structure data. For example, for the classification structure data shown in FIG. 2, first, steps S111 to S114 are executed for the first hierarchy, then steps S111 to S114 are executed for the second hierarchy, and finally the third hierarchy is set. On the other hand, steps S111 to S114 are executed.

  Verification teacher data is used for verification of classification structure data. The verification teacher data includes, for example, the correct label among the labels belonging to the classification structure data with respect to the text data of a sentence that is freely described as an answer to a questionnaire conducted during the most recent fixed period (for example, the most recent half year). It is the given text data. For example, a human analyzes text data to determine the correct label of the text data, and assigns the correct answer label of the determination result to the text data. The verification teacher data is stored in the learning DB_14. The verification teacher data may be all the teacher data among the teacher data stored in the learning DB_14, or may be a part of the teacher data.

(Step S111) The verification module 11 sorts the verification teacher data for the verification target classification items among the verification target hierarchy classification items in the classification structure data. In the classification of verification teacher data, the verification teacher data to which the correct label including the identification number of the verification target classification item is assigned to the verification target classification item and the correct label that does not include the identification number of the verification target classification item are assigned to the verification target classification item. Sorted into verified teacher data. For example, when the classification item “genre” in the second layer of the classification structure data shown in FIG. 2 is the verification target classification item, labels “11” and “111” including the identification number “1” of the classification item “genre”. Alternatively, the verification teacher data to which “112” is assigned and the labels “1”, “12”, “121”, “122” or “123” not including the identification number “1” of the classification item “genre” are given. Into the verified teacher data. The verification teacher data to which the correct label including the identification number of the verification target classification item is assigned is grouped into the verification correct example data group of the verification target classification item. The verification teacher data to which the correct label that does not include the identification number of the verification target classification item is grouped into the verification negative example data group of the verification target classification item.

  The verification module 11 sorts the verification teacher data in the verification example data group of the verification target classification item into verification learning data and verification test data at a predetermined ratio. The verification module 11 also classifies the verification teacher data in the verification negative example data group of the verification target classification item into verification learning data and verification test data at the same ratio. For example, 90% of the verification teacher data among all the verification teacher data in the verification target example data group of the verification target classification item is grouped into the verification positive example learning data group of the verification target classification item, and the remaining 10% verification teacher data Data is grouped into verification example test data groups of verification target classification items. Similarly, out of all the verification teacher data in the verification negative example data group of the verification target classification item, 90% of the verification teacher data is grouped into the verification negative example learning data group of the verification target classification item, and the remaining 10% verification Teacher data is grouped into a verification negative example test data group of verification target classification items.

(Step S112) The verification module 11 uses the verification positive example learning data group and the verification negative example learning data group of the verification target classification item to generate a classifier for the verification target classification item by the classifier generation module 12. . The classifier for the verification target classification item has a function of determining whether the text data to be classified corresponds to the verification target classification item.

(Step S113) The verification module 11 calculates the suitability of the classifier for the verification target classification item using the verification positive example test data group or the verification negative example test data group of the verification target classification item. Examples 1 to 4 of the suitability of the classifier are given below.

(Example 1 of classifier fitness)
An example 1 of the goodness of fit of the classifier is the accuracy rate (Accuracy). The accuracy rate is expressed by the following equation.
Accuracy rate = (TP + TN) / (TP + FP + TN + FN)

(Example of goodness of classifier 2)
Example 2 of the classifier's goodness is a true positive rate. The true positive rate is expressed by the following equation.
True positive rate = TP / (TP + FN)

(Example 3 of classifier fitness)
Example 3 of the goodness of fit of the classifier is a false negative rate. The false negative rate is expressed by the following equation.
False negative rate = FN / (TP + FN)

(Example 4 of classifier fitness)
Example 4 of the goodness of fit of the classifier is Precision. The accuracy is expressed by the following equation.
Accuracy = TP / (TP + FP)

  In Examples 1 to 4 of the suitability of the classifiers described above, TP (True positives) is obtained by applying the verification teacher data of the verification positive example test data group to the classifier, and as a result, the verification teacher data correctly determined as positive examples. It is a number. TN (True negatives) is the number of verification teacher data correctly determined as negative examples as a result of applying the verification teacher data of the verification negative example test data group to the classifier. FP (False positives) is the number of verification teacher data erroneously determined as positive examples as a result of applying the verification teacher data of the verification negative example test data group to the classifier. FN (False negatives) is the number of verification teacher data erroneously determined as negative examples as a result of applying the verification teacher data of the verification positive example test data group to the classifier.

  It should be noted that only one of the above-mentioned classifier suitability examples 1 to 4 may be used for the classifier suitability, or any one of them may be used for the classifier suitability. Also good.

(Step S114) The verification module 11 determines pass / fail of the classifier based on the suitability of the classifier for the verification target classification item. For example, among the examples 1 to 4 of the suitability of the classifier, only one or a plurality of them are evaluated, and when the suitability is good as a result of the evaluation, it is determined to be acceptable. For example, you may determine with a pass when at least any one of the suitable examples 1 to 4 of the classifiers is good. Or you may determine with a pass only when all of the examples 1-4 of the said fitness of a classifier are favorable. As an evaluation method of the fitness level, for example, size comparison between the fitness level and a predetermined threshold value can be given.

  The verification module 11 adopts the verification target classification item when the classifier for the verification target classification item is acceptable. On the other hand, when the classifier for the verification target classification item fails, the verification module 11 rejects the verification target classification item.

  Note that the above-described steps S111 to S114 may not be performed on lower-level classification items belonging to the verification target classification items that have been rejected, and may be rejected.

  The verification module 11 notifies the classifier generation module 12 of verification target classification items to be rejected. When generating the classifier using the classification structure data and the teacher data in the learning DB_14, the classifier generation module 12 deletes the classification item notified as non-adopted from the verification module 11 or cannot be determined. Or

  FIG. 4 is a diagram showing a reflection example of the classification item verification result for the classification structure data according to the present embodiment. FIG. 4 shows an example in which the classification item verification result by the verification module 11 is reflected on the classification structure data shown in FIG. In the example of FIG. 4, the verification module 11 determines that the classification item “timbre” in the second layer is not adopted. Further, since the classification item “timbre” in the second layer is determined not to be adopted, the classification items “about”, “bright”, and “warm” in the third layer, which is a lower layer belonging to the classification item “tone” in the second layer. ”Was decided not to be adopted. Thus, when the classifier generation module 12 generates a classifier for the classification structure data shown in FIG. 2, the classification item “timbre” in the second hierarchy and the classification items “about” and “ Delete “bright” and “warm”. Alternatively, when the classifier generation module 12 generates a classifier for the classification structure data shown in FIG. 2 above, the classification item “timbre” in the second layer and the classification items “about” and “bright” in the third layer ”And“ warm ”are set as undecidable items.

[Content classification module]
The content classification module 13 classifies the text data of the input data using the classifier generated by the classifier generation module 12. The content classification module 13 outputs classification result data indicating the result of the classification. The classification result data is data in which, for example, a label in which classification of input text data has been successfully performed is given to the text data.

  With reference to FIG. 5, the operation of the content classification module 13 according to the present embodiment will be described. FIG. 5 is a flowchart of the classification method according to this embodiment.

(Step S201) The content classification module 13 receives text data (input text data) as input data.

(Step S202) The content classification module 13 classifies the highest hierarchy in the classification structure data with respect to the input text data using the classifier. The classifier determines whether or not the input text data corresponds to the classification item of the highest hierarchy that is the classification target hierarchy. The classifier outputs “applicable” or “non-applicable” as the determination result of the classification item.

  For example, the classifier for the classification structure data reflected in the classification item verification result shown in FIG. 4 determines whether or not the input text data corresponds to the classification item “music” in the first layer. The classifier outputs “applicable” or “not applicable” as the determination result of the classification item “music”. If there are a plurality of classification items in the highest hierarchy, the classifier determines whether or not the input text data corresponds to the classification item for each of the plurality of classification items in the highest hierarchy. The classifier outputs “applicable” or “non-applicable” as the determination result of each classification item.

(Step S203) The content classification module 13 determines whether the classification result of the classification target hierarchy by the classifier is successful. When the determination result of the classification item output from the classifier is “corresponding”, the classification result of the classification item is successful. When the determination result of the classification item output from the classifier is “not applicable”, the classification result of the classification item is a failure.

  The content classification module 13 determines that the classification is successful when the classification result of at least one classification item in the classification target hierarchy is successful. If the classification is successful, the process proceeds to step S204. On the other hand, the content classification module 13 determines that the classification has failed when the classification results of all the classification items in the classification target hierarchy are unsuccessful. If it is a classification failure, the process proceeds to step S206. If the classification is unsuccessful, the lower layer classification belonging to the classification target hierarchy is not executed.

(Step S204) The content classification module 13 determines whether the classification of the lowest hierarchy in the classification structure data has been completed. If the process has been completed up to the classification of the lowest hierarchy, the process proceeds to step S206. On the other hand, if the classification of the lowest hierarchy has not been completed, the process proceeds to step S205.

(Step S205) The content classification module 13 uses the classifier to classify the input text data by one layer belonging to the classification item of the classification success.

  For example, in the case of the classifier for the classification structure data reflecting the classification item verification result shown in FIG. 4, when the classification of the category item “genre” in the second hierarchy is successful, the input text data is Each of the classification items “Classic” and “Pops” in the third layer, which is one lower layer belonging to the classification item “Genre” in the two layers, is performed. The classifier determines whether the input text data corresponds to the classification item “classic” in the third hierarchy, and whether the input text data corresponds to the classification item “pops” in the third hierarchy. The classifier outputs “Applicable” or “Not Applicable” as the determination result of the classification item “Classic” in the third hierarchy and “Applicable” or “Not Applicable” as the determination result of the classification item “Pops” in the third hierarchy. To do.

  It should be noted that the classifier does not perform classification on a classification item that has been deleted or made a non-determinable item according to the classification item verification result. For example, in the classification structure data in which the classification item verification result shown in FIG. 4 is reflected, the classification item “timbre” in the second hierarchy and the classification items “about”, “bright”, and “warm” in the third hierarchy are classified items. This is a classification item that has been deleted according to the verification result or that has been made undecidable. Thereby, the classification by the classifier is not executed for the classification item “tone color” of the second layer and the classification items “about”, “bright”, and “warm” of the third layer.

  After step S205, the process returns to step S203.

(Step S206) The content classification module 13 outputs classification result data indicating a classification result for the input text data. For example, the content classification module 13 outputs, as classification result data, data in which labels of all classification success classification items are added to the input text data.

  According to the classification method shown in FIG. 5 described above, the classification items in the lower layers are further classified with respect to the classification item of the classification success in the higher hierarchy in the classification structure data. As a result, the efficiency of the text classification process can be improved as compared with the case where the classification items of each hierarchy are independently classified without considering the hierarchical structure in the classification structure data.

  According to the present embodiment, for example, when the classifier for the classification structure data in which the classification item verification result shown in FIG. 4 is reflected determines that the input text data corresponds to the classification item “music” in the first layer, Of the classification items “genre” and “tone color” in the second hierarchy, only the classification item “genre” is classified, and the classification item “tone” is not classified. Furthermore, the classification items “about”, “bright”, and “warm” of the third layer belonging to the classification item “tone color” of the second layer are not executed.

  In the classification structure data in which the classification item verification result shown in FIG. 4 is reflected, the classification item “tone color” in the second hierarchy is a classification item that has been rejected from the result of determination of the fitness by the verification module 11. Therefore, by not performing the classification of the classification item “tone color” in the second hierarchy, it is possible to prevent an erroneous classification result from being output and to improve the efficiency of the text classification process. Furthermore, the classification item “about”, “bright” and “warm” in the third layer belonging to the second layer classification item “tone” is also not classified, thereby preventing erroneous classification result output and text. The efficiency of the classification process can be further improved.

  As described above, according to the present embodiment, it is possible to improve the accuracy and efficiency of the text classification process. Further, according to the present embodiment, the following effects can be obtained.

  In general, the number of classification items in the classification structure data increases as it is lower. For this reason, the lower the classification structure data, the less the number of learning data per classification item, and the classification accuracy of the classifier tends to decrease. On the other hand, if the hierarchical structure of classification structure data and the lower limit of the number of learning data are defined in order to prevent a decrease in the number of learning data per classification item, a classifier is automatically generated and text data is classified. Since it becomes easy to occur when it is not possible to perform, it is unusable.

  However, according to the present embodiment, it is possible to generate a classifier that classifies only the classification items that match the verification teacher data among the classification items of the classification structure data. Can keep the classification accuracy high. For example, by using text data similar to the classification target text data that is actually classified by the classifier as the verification teacher data, a classifier that maintains a high classification accuracy of the classification items that match the classification target text data. Can be generated. For example, when classifying text data of a sentence that is freely described as an answer to a questionnaire, the text data collected by the most recently conducted questionnaire is used as verification teacher data as verification teacher data. This makes it possible to generate a highly accurate classifier that matches the text data of the questionnaire response that is actually the classification target.

  Further, for example, if a classifier for classifying text data in a specific field is generated from large-scale classification structure data generated from text data in various fields, it becomes easy to be a classifier with poor classification processing efficiency. For example, it becomes a classifier that performs classification processing of classification items unrelated to a specific field to be classified, and the classification processing becomes inefficient, for example, it takes extra time for the classification processing.

  However, according to the present embodiment, by using text data of a specific field to be classified as verification teacher data, a classifier that performs classification of only classification items that match the text data of the specific field is generated. can do. As a result, it is possible to generate a classifier with high efficiency of the classification process, such that the extra time is not required for the classification process.

  In the above-described embodiment, the verification module 11 corresponds to the fitness calculation unit and the hierarchy setting unit, the classifier generation module 12 corresponds to the classifier generation unit, and the content classification module 13 corresponds to the classification unit.

  As mentioned above, although embodiment of this invention was explained in full detail with reference to drawings, the specific structure is not restricted to this embodiment, The design change etc. of the range which does not deviate from the summary of this invention are included.

Further, a computer program for realizing the functions of the classification system 1 described above may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read into the computer system and executed. . Here, the “computer system” may include an OS and hardware such as peripheral devices.
“Computer-readable recording medium” refers to a flexible disk, a magneto-optical disk, a ROM, a writable nonvolatile memory such as a flash memory, a portable medium such as a DVD (Digital Versatile Disk), and a built-in computer system. A storage device such as a hard disk.

Further, the “computer-readable recording medium” means a volatile memory (for example, DRAM (Dynamic DRAM) in a computer system that becomes a server or a client when a program is transmitted through a network such as the Internet or a communication line such as a telephone line. Random Access Memory)), etc., which hold programs for a certain period of time.
The program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the “transmission medium” for transmitting the program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line.
The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, and what is called a difference file (difference program) may be sufficient.

DESCRIPTION OF SYMBOLS 1 ... Classification system, 11 ... Verification module (fitness calculation part, hierarchy setting part), 12 ... Classifier generation module (classifier generation part), 13 ... Content classification module (classification part), 14 ... Learning database (learning DB) )

Claims (8)

In the verification device for verifying the classification structure data indicating the classification item and the classification item hierarchy to which the data is classified,
Among the verification teacher data to which the correct answer label is assigned among the labels belonging to the classification structure data, the correct answer to which the correct answer label including the identification number of the verification target classification item among the classification items of the verification target hierarchy in the classification structure data is assigned. Using the verification positive example learning data which is the verification teacher data of the example and the verification negative example learning data which is the verification teacher data of the negative example to which the correct label not including the identification number of the verification target classification item is given A classifier generator for generating a classifier for the verification target classification item,
Among the verification teacher data, the verification example data of the positive example and the verification example data that is the verification teacher data different from the verification example learning data, or the verification example data of the negative example, A fitness calculation unit that calculates the fitness of the classifier generated by the classifier generation unit using verification negative example test data that is the verification teacher data different from the verification negative example learning data;
A hierarchy setting unit that determines adoption of the verification target classification item in the verification target hierarchy based on the fitness calculated by the fitness calculation unit;
A verification apparatus comprising: The verification target hierarchy is verified in order from the upper hierarchy to the lower hierarchy in the classification structure data, and the verification procedure is performed for the lower classification items belonging to the verification target classification item determined to be non-adopted. Do not execute, do not adopt,
The verification device according to claim 1. The verification apparatus according to claim 1 or 2,
A classification unit that classifies input data using a classifier of classification structure data in which a classification item verification result by the verification device is reflected;
A classification system comprising: The classification unit starts the classification of the input data from the highest hierarchy in the classification structure data, and sequentially performs only one lower layer classification belonging to the classification item of classification success,
The classification system according to claim 3. In the verification method of the classification structure data indicating the classification item and the classification item hierarchy to which the data is classified,
Among the verification teacher data to which the verification device is assigned the correct label among the labels belonging to the classification structure data, the correct label including the identification number of the verification target classification item among the classification items of the verification target hierarchy in the classification structure data is Verification positive example learning data which is the verification teacher data of the given positive example and verification negative example learning data which is the verification example data of the negative example to which the correct label not including the identification number of the verification target classification item is assigned And a classifier generating step for generating a classifier for the verification target classification item using
The verification apparatus is verification positive example test data which is the verification teacher data of the positive example and is different from the verification positive example learning data among the verification teacher data, or the negative verification teacher A fitness calculation step of calculating a fitness of the classifier generated by the classifier generation step using verification negative example test data which is data and the verification teacher data different from the verification negative example learning data When,
The verification apparatus, based on the fitness calculated in the fitness calculation step, a hierarchy setting step for determining adoption of the verification target classification item in the verification target hierarchy;
Verification method including A verification step of verifying the classification structure data by the verification method according to claim 5;
A classification step in which a classification unit classifies input data using a classifier of classification structure data in which a classification item verification result in the verification step is reflected;
Classification method including A computer program for verifying classification structure data indicating a classification item and a classification item hierarchy to which data is classified,
Among the verification teacher data to which the correct answer label is assigned among the labels belonging to the classification structure data, the correct answer to which the correct answer label including the identification number of the verification target classification item among the classification items of the verification target hierarchy in the classification structure data is assigned. Using the verification positive example learning data which is the verification teacher data of the example and the verification negative example learning data which is the verification teacher data of the negative example to which the correct label not including the identification number of the verification target classification item is given A classifier generating step for generating a classifier for the verification target classification item;
Among the verification teacher data, the verification example data of the positive example and the verification example data that is the verification teacher data different from the verification example learning data, or the verification example data of the negative example, A fitness calculation step of calculating a fitness of the classifier generated by the classifier generation step using verification negative example test data which is the verification teacher data different from the verification negative example learning data;
Hierarchy setting step for determining adoption of the verification target classification item in the verification target hierarchy based on the fitness calculated in the fitness calculation step;
A computer program for causing a computer to execute. A classification step of classifying input data using a classifier of classification structure data reflecting a classification item verification result that is an execution result of the computer program;
The computer program according to claim 7, further causing a computer to execute.

Images