JP2006252333A - Data processing method, data processor and its program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a data processing method capable of enhancing reliability of machine learning when performing the machine learning using two or more pieces of learning data. <P>SOLUTION: In a similar learning data generation part 4, from among n pieces of learning data SDq, a piece of similar learning data SSDq is selected which have high similarity with data to be processed. A machine learning machine 5 uses the similar learning data SSDq to perform the machine learning. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a data processing method, a data processing apparatus, and a program for processing data to be processed using learning data.

For example, the gene analysis system uses a database in which actions that occur between genes are expressed as gene (molecule) names as nodes and the actions as links between nodes.
In order to construct such a database, for example, gene names are extracted from published papers and registered as nodes in the database.
Here, since the number of published papers is enormous, it is too heavy for humans to look at papers and extract gene names.
Therefore, it is conceivable to extract gene names mechanically from paper data using a computer or the like. However, it is difficult to extract a new gene name mechanically.
A similar problem occurs when a unique expression such as a person name, place name, or organization name is extracted from character data in addition to a gene name.
In order to solve such a problem, for example, a pattern in which a desired specific expression appears in a predetermined analysis unit (token) is specified in advance in learned data (training data), such as SVM, and the pattern Is used as learning data to extract the specific expression from the data to be processed.
A conventional machine learning device extracts a desired specific expression from data to be processed using, for example, all of a plurality of learning data held by the machine learning device.
`` Gene / protain recognition using Support Vector Machine after dictionary matching '', Tomohiro Mitsumori, Sevrani Fation, Masaki Murata, Kouichi Doi and Hirohumi Doi BioCreative Workshop: Critical Assessment for Information Extraction in Biology (BioCreative 2004), Granada, Spain, March, 2004 Nakano, Hirai, Use of phrase information in Japanese proper expression extraction, IPSJ Transactions, Vol. 45 No. 3, p934-941, Mar. 2004 Hira, Haruno, attribute selection in text classification by Support Vector Machine, Journal of Information Processing Society, Vol. 45 No. 4, p1113-1123, Apr. 2004

However, since the above-described conventional machine learning device extracts a specific expression from the processed data using all the learning data held by the machine learning device regardless of the attribute of the processed data, the processed data There is a problem that the reliability of extraction of the specific expression is lowered by using learning data having a low degree of similarity with the attribute of.
Similar problems exist in machine learning devices other than the gene analysis system described above.

  In order to solve the above-described problems of the prior art, the present invention provides a data processing method and data processing apparatus capable of improving the reliability of processing when processing data to be processed using a plurality of learning data. And to provide the program.

  In order to solve the above-described problems of the prior art and achieve the above-described object, the data processing method of the first aspect of the invention uses machine learning data generated based on the data to be learned, A data processing method for performing learning processing, for each of a plurality of learning data, generating similarity data indicating similarity between the learning data used to generate the learning data and the processing data The second step of selecting a part of the learning data from the plurality of learning data based on the similarity data generated in the first step, and the second step And a third step of performing machine learning processing on the data to be processed using the selected learning data.

  The data processing method according to the second aspect of the invention uses learning data in which attribute data is attached to each of a plurality of processing unit data constituting the learned data, and adds the attribute data to the processed data by machine learning. A data processing method executed by a data processing apparatus, wherein each of a plurality of pieces of learning data includes the processing unit data constituting the learning data of the learning data and the processing unit data constituting the processing data A first step of generating similarity data indicating the degree of similarity and the processed data among the learned data of the plurality of learning data based on the similarity data generated in the first step A second step of specifying the learning data whose similarity with data satisfies a predetermined criterion, and selecting the learning data corresponding to the specified learning data; Using said learning data selected by the serial second step, a third step of adding the machine learning the attribute data the in the processing unit data constituting the data to be processed.

  A data processing device according to a third aspect of the invention is a data processing device that performs machine learning processing of data to be processed using learning data generated based on the data to be learned, and for each of the plurality of learning data, Similarity data generating means for generating similarity data indicating similarity between the learned data used for generating the learning data and the processed data, and the similarity generated by the similarity data generating means Selection means for selecting a part of the learning data based on the degree data, and processing means for performing machine learning processing on the processed data using the learning data selected by the selection means; Have

The operation of the data processing apparatus according to the third aspect of the invention is as follows.
Similarity data generation means generates similarity data indicating the similarity between the learned data used to generate the learning data and the processed data for each of the plurality of learning data.
Next, the selection unit selects a part of the learning data from the plurality of learning data based on the similarity data generated by the similarity data generation unit.
Next, processing means performs machine learning processing on the processing target data using the learning data selected by the selection means.

  A program according to a fourth aspect of the present invention is a program executed by a data processing device that performs machine learning processing of data to be processed using learning data generated based on the data to be learned, and each of the plurality of learning data The learning data used to generate the learning data, a first procedure for generating similarity data indicating the similarity between the processed data, and the similarity generated in the first procedure Based on the degree data, a second procedure for selecting some of the plurality of learning data and the learning data selected in the second procedure are used to machine-process the processed data. And a third procedure.

  According to the data processing method, the data processing apparatus, and the program thereof of the present invention, when processing data to be processed using a plurality of learning data, the data processing method and data processing that can improve the reliability of the processing A device and its program can be provided.

Hereinafter, a machine learning apparatus according to an embodiment of the present invention will be described.
In the following embodiments, the second embodiment is an embodiment of the present invention in which the machine learning system of the first embodiment is applied to a machine learning system that performs learning processing for papers and the like.
The third embodiment is an embodiment of the present invention in which the machine learning system of the first embodiment is applied to a machine learning system that performs access control to content on the Internet.

<First Embodiment>
FIG. 1 is a configuration diagram of a machine learning system according to the first embodiment of this invention.
As shown in FIG. 1, the machine learning system according to the present embodiment includes, for example, a similar learning data generator 2 and a machine learning machine 3.
The similar learning data generator 2 includes, for example, a similarity calculation unit 3 and a similar learning data generation unit 4.
The machine learning system according to the present embodiment selects a subset (similar learning data SSDq) whose similarity with a problem to be solved (question data TD) satisfies a predetermined condition from a set of correct answer examples (learning data SDq). The learning data for the learning machine is used to improve the learning speed and accuracy.
First, the correspondence between the components of the present embodiment and the components of the present invention will be described.
The problem data TD shown in FIG. 1 etc. corresponds to the processed data of the present invention, and the learning data SDq corresponds to the learning data of the present invention.
Further, the learned data Rq shown in the figure corresponds to the learned data of the present invention.
Moreover, the word of this embodiment respond | corresponds to the processing unit data of this invention.
Further, the similarity data BA (q) of the present embodiment corresponds to the similarity data of the present invention.
The similarity calculation unit 3 shown in FIG. 1 corresponds to the similar data generation unit of the invention of the third aspect, the similar learning data generation unit 4 corresponds to the selection unit of the invention of the third aspect, and the machine learning machine 5 This corresponds to the processing means of the invention of the third aspect.
Further, it is possible to describe the functions of the similarity calculation unit 3, the similar learning data generation unit 4 and the machine learning machine 5 shown in FIG. Corresponds to the program of the invention of this aspect.

[Similarity calculation unit 3]
The similarity calculation unit 3 calculates the similarity between the learned data Rq and the problem data TD for each of the learned data Rq of the n pieces of learning data Sq.
In the present embodiment, the learned data Rq and the problem data TD are POS (Point Of Sale) data, text data, multimedia data, and the like.
These data are configured by combining a plurality of processing unit data.
The similarity calculation unit 3 generates, for each of the plurality of learned data Rq, similarity data indicating the similarity between the processing unit data constituting the learned data Rq and the processing unit data constituting the problem data TD. This is output to the similar learning data generation unit 4.
Specifically, the similarity calculation unit 3 determines the characteristics of each data for each of the learned data Rq and the problem data TD based on the processing unit data constituting each data in a predetermined feature evaluation coordinate system. Generate specified vector data.
Then, the similarity calculator 3 generates similarity data based on the generated vector data.

The similarity calculation unit 3 generates the vector data represented by (X, Y, Z), for example,
A1: d (x, y) ≧ 0
A2: d (x, y) = d (y, x)
A3: The necessary and sufficient condition for d (x, y) = 0 is x = y
A3 ': d (x, x) = 0
A4: d (x, z) ≤ d (x, y) + d (y, z)
age,
B1: A1, A2, A3, A4
B2: A1, A2, A3 ', A4
B3: A1, A2, A3
B4: A1, A2, A3 '
B5: A1, A2
If
Use the function d () that indicates the measure (measured value) that satisfies one of the above B1, B2, B3, B4, or B5, or the similarity that the distance decreases monotonously as the similarity increases The distance is calculated by a calculation formula, and the similarity data indicating the distance is generated.
Here, B1 corresponds to a so-called “distance”. For example, in a three-dimensional space, it is the Euclidean distance, and “d (x, y) = {(x1-y1) ^ 2 + (x2-y2) ^ 2 + (x3-y3) ^ 2} ^ (1 / 2) ".

In addition, the similarity calculation unit 3 sets, for each of a plurality of pieces of learned data Rq, the distance between the processing unit data constituting the learned data Rq and the processing unit data constituting the problem data TD in a predetermined coordinate system. Similarity data shown may be generated.
In this case, the similarity calculation unit 3 uses, as a distance calculation method, an Euclidean distance, a Euclidean square distance, a standardized Euclidean distance, a Minkowski distance, or an evaluation method based on a distance calculation by a kernel method.

In addition, the similarity calculation unit 3 is separately provided with respect to a plurality of distances or similarities obtained for a processing unit data group that is a problem with respect to one learned data Rq or problem data TD. After conversion using the conversion formula to similarity, it is expressed as a similarity vector, converted into a scalar value by a separately defined selection function, and this may be used as similarity data.
Further, the similarity calculation unit 3 performs a calculation for converting a similarity vector having a plurality of similarities as elements into a scalar by summing each element, sum of squares, selecting a maximum value, selecting a minimum value, and the like. May be. Further, the similarity calculation unit 3 may add a positive number that is not zero to the generated distance data, and use the reciprocal as the similarity data.

[Similar learning data generation unit 4]
The similar learning data generation unit 4 uses the learning data SDq of the learning data Rq whose similarity expressed by the similarity data generated by the similarity calculation unit 3 exceeds a predetermined threshold among the n learning data Rq. This is selected and output to the machine learning machine 5 as similar learning data SSDq.
Here, the learning data SDq includes the learned data Rq and its attribute data PD as shown in FIG.
Here, the attribute data PD indicates the attribute of each processing unit data constituting the learned data Rq.
For example, when the learned data Rq and the problem data TD are e-mails, the attribute is information indicating whether or not the e-mail is a junk mail, and the learned data Rq and the problem data TD are document data. Is information indicating the part of speech of the word.

[Machine learning machine 5]
The machine learning machine 5 processes the problem data TD using the similar learning data SSDq input from the similar learning data generation unit 4.
Specifically, the machine learning machine 5 adds the attribute data PD to the processing unit data constituting the problem data TD using the similar learning data SSDq.
The machine learning machine 5 performs supervised learning processing such as SVM (Support Vector Machine), Artificial Neural Network, and genetic algorithm.
The learning rule used for learning by the machine learning machine 5 is a parameter group describing a hyperplane for performing data separation in the Support Vector Machine, and a weight vector for each neuron in the Artificial Neural Network.
The machine learning machine 5 may use a method such as a decision list, a method based on similarity, a simple Bayes method, a maximum entropy method, a decision tree, a neural network, and a discriminant analysis in addition to the above SMV as a machine learning method. Good.

Hereinafter, SVM which the machine learning machine 5 employs as an example will be described.
SVM is disclosed in, for example, Non-Patent Document 3 described above.
When the machine learning machine 5 separates the problem data TD into a positive example set in the superspace, the machine learning machine 5 performs a learning process based on SVM to obtain an optimum separated hyperplane by maximizing the margin.
SVM is a technique based on structural risk minimization that finds hypotheses that guarantee the least generalization error.
In the SVM, for example, when the input vector (problem data TD) is x, the function of the following formula (2) indicates the hypothesis h.

  In the above equation (2), w and b are parameters. Regarding the relationship between the dimension n of the input vector x and the VC dimension λ, the following lemma is known.

Lemma:
As a hypothesis h (x), a hyperplane h (x) = sign {w · x + b} is assumed. There is a sphere having a radius R including all of l pieces of training data (similar learning data SSDq in this embodiment) x = xi (i is an integer from 1 to 1). If || w || is the norm of w, the following equation (4) is established for the VC dimension λ.

From the above equation (4), the VC dimension may depend on || w ||.
The SVM divides the training data into a positive example and a negative example, and specifies a hyperplane in which the margin between the positive and negative examples is the maximum, that is, || w || is the minimum.
The machine learning machine 5 processes the hyperplane specification as a secondary optimization problem using, for example, a Lagrange multiplier.

As described above, according to the machine learning system of the present embodiment, the machine learning machine 5 selects and uses only the n learning data SDq having high similarity to the problem data TD. The learning process of the problem data TD is performed.
Therefore, the learning data SDq having a low similarity with the problem data TD is not used for learning the problem data TD, and the reliability of the processed data TR is increased.
As a result, the reliability of the processed data TR can be improved.

  Further, according to the machine learning system of the present embodiment, in addition to the improvement of processing reliability, the amount of data used for learning is reduced, and the effects of shortening the time required for learning and reducing machine resources can be obtained.

In the present embodiment, text data is input as problem data TD, and a task for extracting a desired word from text data is set for each word, using part-of-speech information, word tail spelling, and word type as attribute data. System presentation. Here, input and assignment setting are not limited to this, and it is obvious that various applications are possible. For example, as shown in FIG. 4, point of sale data, multimedia data such as music, voice, TV program, video image, etc. can be designated as problem data TD, and sales patterns can be used as problem settings. Analysis, filtering of unwanted e-mails and news programs, and extracting video clips desired by the user.
This embodiment can be applied to a system that performs customer trend extraction from POS data, classification of text data and multimedia data, and information extraction.
In addition, the processing unit data of the present embodiment includes point of sales information including merchandise type and the number of units sold, arrival date, sales date, age, gender, family composition, etc., mail text, papers, patents, etc. , HP documents, program guides, lyric documents, etc., or decomposed into sentences and words, score data, time series data such as music, spectrum data such as output results by gas chromatography, news programs, dramas, video images Define a certain structural unit such as video information, etc., configure it as a combination, overlay, composition, sequence, or analyze it as configured, and use some additional procedure for the data The processed data can be added and used.

Second Embodiment
The second embodiment is an embodiment in which the machine learning system of the first embodiment is applied to a machine learning system that performs learning processing of papers and the like.

First, the correspondence between the components of the present embodiment and the components of the present invention will be described.
The problem data TD shown in FIG. 3 etc. corresponds to the processed data of the present invention, and the learning data SDq corresponds to the learning data of the present invention.
Further, the learned data Rq shown in FIG. 4 and the like corresponds to the learned data of the present invention.
Moreover, the word of this embodiment respond | corresponds to the processing unit data of this invention.
Further, the similarity data BA (q) of the present embodiment corresponds to the similarity data of the present invention.
In addition, index data TF (i, j) shown in Expression (6) corresponds to the index data of the present invention.
Step ST2 shown in FIG. 13 corresponds to the first step of the first aspect of the invention, step ST3 corresponds to the second step, and step ST5 corresponds to the third step.
Further, the similarity calculation unit 33 of the similar learning data selection unit 11 shown in FIG. 5 corresponds to the similar data generation unit of the invention of the third aspect, and the learning data selection unit 34 serves as the selection unit of the invention of the third aspect. Correspondingly, the IOB determination unit 72 of the IOB addition unit 22 shown in FIG. 12 corresponds to the processing means of the invention of the third aspect.

FIG. 3 is an overall configuration diagram of the machine learning device 1 according to the embodiment of the present invention.
As illustrated in FIG. 3, the machine learning device 1 includes, for example, a memory 9, a similar learning data selection unit 11, a tag addition unit 13, an IOB determination data generation unit 15, a tag addition unit 21, and an IOB addition unit 22.
Each component of the machine learning device 1 is configured by hardware such as an electronic circuit, for example.
In addition, elements constituting the respective constituent elements described later are also configured using hardware such as an electronic circuit.
The present invention may be realized by a CPU (Central Processing Unit) executing a program for each component of the machine learning device 1 shown in FIG. 3 and part or all of the components.

First, learning data SDq (SD1 to SDn) stored in the memory 9 shown in FIG. 3 and input to the similar learning data selection unit 11 will be described.
FIG. 4 is a diagram for explaining the learning data SDq (SD1 to SDn).
As shown in FIG. 4, the learning data SDq is data in which attribute tag data IOB is associated with each of all words in the learned data Rq in advance.
Here, the attribute tag data IOB indicates “B” for the first word constituting a predetermined term (for example, the name of a protein).
The attribute tag data IOB is a word following the first word, and indicates “I” for the word constituting the predetermined word.
The attribute tag data IOB indicates “O” for a word that does not constitute the predetermined word.
In the present embodiment, the learned data Rq and the problem data TD are, for example, English paper data including the name of the protein. For example, a space is a delimiter, and a period at the end of the sentence and the word immediately preceding it are Dividing is a rule for dividing the English sentence into words.
In the present embodiment, the machine learning device 1 can use n pieces of learning data SDq.

Hereinafter, each component shown in FIG. 3 will be described.
[Similar learning data selection unit 11]
FIG. 5 is a configuration diagram of the similar learning data selection unit 11 shown in FIG.
As illustrated in FIG. 5, the similar learning data selection unit 11 includes an input unit 31, an input unit 32, a similarity calculation unit 33, a learning data selection unit 34, and an output unit 35.
The input unit 31 inputs, for example, n pieces of learning data SDq shown in FIG. 4 from a memory (not shown) included in the machine learning device 1 or from the outside of the machine learning device 1.
Further, the input unit 32 inputs the problem data TD shown in FIG. 6 from the outside of the machine learning device 1.

The similarity calculation unit 33 calculates the similarity between the learned data Rq and the problem data TD for each of the learned data Rq in the n pieces of learning data SDq shown in FIG. 4 input by the input unit 31. To do.
Hereinafter, a method for calculating the similarity will be described.
Here, k is the number of types of words included in the n pieces of learned data Rq and the problem data TD respectively corresponding to the n pieces of learning data SDq.
“I” is an integer from 1 to k, and “j” is an identifier attached to n pieces of learned data Rq and one piece of problem data TD.

  The similarity calculation unit 33 calculates the index data TF (i, j) by the following equation (5).

[Equation 5]
TF (i, j) = (number of times word i appears in learned data Rj (problem data TD)) / (total number of words included in learned data Rj (problem data TD))
... (5)

  Moreover, the similarity calculation part 33 specifies DF (i) by following formula (6).

[Equation 6]
DF (i) = (number of words i appearing among n pieces of learned data Rq and problem data TD)
(6)

  The similarity calculation unit 33 calculates w (i, j) between all the words i for each of the learned data Rq and the problem data TD by the following equations (7) and (8).

[Equation 7]
IDF (i) = log [(N + 1) / DF (i)]
... (7)

[Equation 8]
w (i, j) = TF (i, j) * IDF (i)
... (8)

The IDF (i) decreases its value exponentially as the number of items including the word i among the problem data TD and the learned data Rq increases.
By multiplying such IDF (i) by TF (i, j) to generate w (i, j), it is possible to extract “a”, “the”, “this”, “that”, etc. It is possible to almost eliminate the influence of words that are not expressions but are unrelated to the attribute of the data on the similarity.

  Then, the similarity calculator 33 defines vectors D (q) and D (M) according to the following (9) and (10) for each of the learned data Rq and the problem data TD.

[Equation 9]
D (q) = (w (1, q), w (2, q), ..., w (k, q)
... (9)

[Equation 10]
DM = (w (1, M), w (2, M), ..., w (k, M))
(10)

  Then, the similarity calculation unit 33 calculates the similarity data BA (q) shown in the following (11) for all the learned data Rq.

  The similarity calculation unit 33 outputs the similarity data BA (q) to the learning data selection unit 34.

The learning data selection unit 34 selects only the learning data SDq input from the input unit 31 that has the similarity data BA (q) input from the similarity calculation unit 33 exceeding a predetermined reference value. The similar learning data SSDq is output to the output unit 35.
Note that the learning data selection unit 34 selects a predetermined number of n learning data SDq input from the input unit 31 from those having a high similarity indicated by the similarity data BA (q), and outputs the selected data to the output unit 35. It may be output.
In the example illustrated in FIG. 5, the learning data selection unit 34 outputs the learning data SD1, 3, and 10 to the output unit 35 as the similar learning data SSDq.

  The output unit 35 outputs the similar learning data SSDq input from the learning data selection unit 34 to the tag addition unit 13 shown in FIG.

[Tag addition unit 13]
As shown in FIG. 7, the tag addition unit 13 shown in FIG. 3 has, for each word constituting the learned data Rq of the similar learning data SSDq input from the similar learning data selection unit 11, its part-of-speech data, suffix data, To generate new similar learning data SSDDAq.

FIG. 8 is a block diagram of the tag adding unit 13 shown in FIG.
As illustrated in FIG. 8, the tag adding unit 13 includes, for example, an input unit 41, a part-of-speech tagger unit 42, a Suffix tagger unit 43, and an output unit 44.
The input unit 41 inputs the similar learning data SSDq from the similar learning data selection unit 11 shown in FIG. 3 and outputs it to the part of speech tagger unit 42.
The part-of-speech tagger unit 42 adds part-of-speech data indicating the part of speech to each word in the similar learning data SSDq shown in FIG. 7 input from the input unit 41, and outputs this to the Suffix tagger unit 43.
As shown in FIG. 7, the Suffix tagger unit 43 further adds the suffix (suffix) data to each word of the similar learning data to which the part of speech data input from the part of speech tagger unit 42 is added. Is output to the output unit 44.
In this embodiment, the Suffix tagger unit 43 adds 3 gram suffix.
The output unit 44 outputs the similar learning data SSDAq input from the Suffix tagger unit 43 to the IOB determination data generation unit 15 shown in FIG.

[IOB determination data generation unit 15]
The IOB determination data generation unit 15 generates IOB determination data (feature data) SP used for analysis in the IOB addition unit 22 by using the similar learning data SSDDAq shown in FIG. The data is output to the adding unit 22.

FIG. 9 is a block diagram of the IOB determination data generation unit 15 shown in FIG.
As illustrated in FIG. 9, the IOB determination data generation unit 15 includes, for example, an input unit 51, an SVM learning unit 52, and an output unit 53.
The input unit 51 receives the similar learning data SSDAq from the tag addition unit 13 and outputs it to the SVM learning unit 52.
For example, the SVM learning unit 52 determines whether the attribute tag data IOB of each word is I, O, or B based on the similar learning data SSDDAq shown in FIG. 7 input from the input unit 51. The IOB determination data SP shown in FIG. 10 is generated by the SVM (Support Vector Machines) method using the part of speech data and suffix data of two words before and after each word.
In the learning process by the SVM method, the SVM learning unit 52 uses, for example, a polynomial kernel as a kernel function, uses bear-wise as a multi-value classification extension method, and sets the analysis direction from the beginning to the back of the sentence.
As the learning process by the SVM learning unit 52, for example, the SVM described in the first embodiment is used.
The SVM learning unit 52 outputs the IOB determination data SP to the output unit 53.
The output unit 53 outputs the IOB determination data SP input from the SVM learning unit 52 to the IOB adding unit 22.

  The IOB determination data generation unit 15 is a learning method other than the SVM method, such as a decision list method, a similarity-based method, a simple Bayes method, a maximum entry pea method, a decision tree method, a neural network method, a discriminant analysis method, and the like. May be used.

[Tag addition unit 21]
As shown in FIG. 6, the tag addition unit 21 shown in FIG. 3 adds part-of-speech data and suffix data to each word constituting the question data TD input from the outside of the machine learning device 1 to create a new one. Problem data TDa is generated.

FIG. 11 is a block diagram of the tag adding unit 21 in FIG.
As illustrated in FIG. 11, the tag addition unit 21 includes, for example, an input unit 61, a part-of-speech tagger unit 62, a Suffix tagger unit 63, and an output unit 64.
The input unit 61 inputs problem data TD from the outside of the machine learning device 1 shown in FIG. 3 and outputs it to the part-of-speech tagger unit 62.
The part-of-speech tagger unit 62 adds part-of-speech data indicating the part-of-speech to each word in the question data TD shown in FIG. 6 input from the input unit 61, and outputs this to the Suffix tagger unit 63.
As shown in FIG. 6, the Suffix tagger unit 63 further adds the suffix data to each word of the data to be processed to which the part of speech data input from the part of speech tagger unit 62 is added, and obtains problem data TDa. This is generated and output to the output unit 64.
The output unit 64 outputs the problem data TDa input from the Suffix tagger unit 63 to the IOB addition unit 22 shown in FIG.

[IOB addition unit 22]
FIG. 12 is a configuration diagram of the IOB adding unit 22 shown in FIG.
As illustrated in FIG. 12, the IOB addition unit 22 includes, for example, an input unit 71, an IOB determination unit 72, and an output unit 73.
The input unit 71 outputs the IOB determination data SP input from the IOB determination data generation unit 15 to the IOB determination unit 62.
Based on the IOB determination data SP input from the input unit 61, the IOB determination unit 72 adds attribute tag data IOB to each word of the problem data TDa input from the tag addition unit 21 shown in FIG. The processed data TR shown is generated.
Here, if the problem data TD is a problem to be solved, the processed data TR is a solution of the problem to be solved.
The IOB determination unit 72 outputs the test result data TDa to the output unit 63.
The output unit 73 outputs the processed data TR input from the IOB determination unit 72 to the outside of the machine learning device 1.

Hereinafter, an operation example of the machine learning device 1 illustrated in FIG. 3 will be described.
FIG. 13 is a flowchart for explaining the operation example.
Hereinafter, each step shown in FIG. 13 will be described.
Step ST1:
As shown in FIG. 6, the tag addition unit 21 shown in FIG. 3 adds part-of-speech data and suffix data to each word constituting the question data TD input from the outside of the machine learning device 1 to create a new one. The problem data TDa is generated and output to the IOB adding unit 22.

Step ST2:
The similarity calculation unit 33 of the similar learning data selection unit 11 shown in FIG. 5 performs the learning data Rq for each of the learning data Rq in the n pieces of learning data SDq shown in FIG. Then, the similarity with the problem data TD is calculated to generate similarity data BA (q), which is output to the learning data selector 34.
Step ST3:
The learning data selection unit 34 shown in FIG. 5 selects only n pieces of input learning data SDq that have the similarity data BA (q) input from the similarity calculation unit 33 exceeding a predetermined reference value. It outputs to the tag addition part 13 shown in FIG. 3 as similar learning data SSDq.

Step ST4:
As shown in FIG. 7, the tag addition unit 13 shown in FIG. 3 has, for each word constituting the learned data Rq of the similar learning data SSDq input from the similar learning data selection unit 11, its part-of-speech data, suffix data, Is added to generate new similar learning data SSDAq, which is output to the IOB determination data generation unit 15.
Step ST5:
The IOB determination data generation unit 15 illustrated in FIG. 3 generates IOB determination data (feature data) SP used for analysis in the IOB addition unit 22 using the similar learning data SSDDAq illustrated in FIG. 7 input from the tag addition unit 13. This is output to the IOB adding unit 22.
Step ST16:
The IOB addition unit 22 shown in FIG. 3 adds attribute tag data IOB to each word of the problem data TDa shown in FIG. 6 input from the tag addition unit 21 based on the IOB determination data SP input in step ST5. The processed data TR shown in FIG. 6 is generated.
The machine learning device 1 extracts a specific expression (gene name) in the problem data TD based on the attribute tag data IOB attached to the processed data TR.

As described above, according to the machine learning device 1, only the data having high similarity with the problem data TD among the n pieces of learning data SDq stored in the memory 9 is selected and used. The attribute tag data IOB is added to the problem data TDa in the unit 22.
Therefore, in the addition of the attribute tag data IOB to the problem data TDa, the learning data SDq having a low similarity with the problem data TD is not used, and the reliability of the processed data TR is increased.
As a result, a desired specific expression (gene name) can be extracted from the processed data TR with high reliability.
Further, according to the machine learning 1 of the present embodiment, in addition to the improvement of processing reliability, the amount of data used for learning is reduced, and the effects of shortening the time required for learning and reducing machine resources can be obtained.

<Third Embodiment>
The third embodiment is an embodiment in which the machine learning system of the first embodiment is applied to a machine learning system that controls access to content on the Internet.
FIG. 14 is a diagram for explaining the machine learning system 101 according to the third embodiment of this invention.
In the machine learning system 101, the learning data generation unit 112 downloads a plurality of Web page data W1 stored in a server (not shown) on the Internet 111.
The learning data generation unit 112 generates learning data (teacher data) SDq by adding tag data TG indicating the content classification (attribute) to the downloaded web page data W1 according to a predetermined rule. Is output to the similar learning data selection unit 115.
As the tag data TG, for example, information such as presence / absence of viewing restriction, prohibition under the restriction age, presence of violent expression is shown.

The similar learning data selection unit 115 determines the similarity based on a predetermined criterion based on the similarity between the Web page data W2 that is the processed data downloaded via the Internet 111 and the Web page data W1 of the learning data SDq. The satisfied learning data SDq is selected as the similar learning data SSDq and output to the machine learning machine 116.
The similarity relationship is determined based on similarity data generated using the method described in the first embodiment.

The machine learning machine 116 performs learning processing of the web page data W2 using the similar learning data SSDq input from the similar learning data selection unit 115, and processes the processed web page data W3 to which the tag data TG is attached in the cache memory 118 and Output to the filter 125.
As the learning process by the machine learning machine 116, for example, the SVM described in the first embodiment is used.

The cache memory 118 stores processed web page data W3.
When a browsing request issued by the user is input using the user interface 121 or the like that operates on the computer, the cache search unit 123 reads out the processed Web page data W3 corresponding to the browsing request from the cache memory 118 and filters 125. Output to.
When the processed web page data W3 corresponding to the browsing request is not stored in the cache memory 118, the cache search unit 123 sends the web page data corresponding to the processed web page data W3 to the content loader 131. Output the requested download request.
The content loader 131 transmits the download request to the server via the Internet 111.
Thereby, the Web page data W1 related to the browsing request is downloaded to the learning data generation unit 112.

The filter 125 is incorporated as a function in a predetermined server or a computer used by the user, verifies the tag data TG of the input processed Web page data W3 according to a pre-stored filter rule, and has processed a predetermined condition. The Web page data W3 is output to the user interface 121 after removing the tag data TG.
In the example of FIG. 14, the cache search unit 123 is not particularly essential.

As described above, according to the machine learning system 101, in the similar learning data selection unit 115, only the learning data SDq having an attribute similar to that of the Web page data W2 of the processed data is used as the similar learning data SSDq. 115.
Thereby, in the similar learning data selection unit 115, highly reliable tag data TG can be attached to the Web page data W2, and the filter processing in the filter 125 can be appropriately performed.
Further, according to the machine learning system 101 of the present embodiment, in addition to improving the reliability of processing, the amount of data used for learning can be reduced, the time required for learning can be shortened, and machine resources can be reduced.

The present invention is not limited to the embodiment described above.
In the embodiment described above, the paper (reference) data in the gene field is exemplified as the processed data and learned data Rq of the present invention, but other data may be used.
For example, the present invention can be applied to machine learning processing such as protein expression extraction, proper expression extraction (person name, place name, etc.), modality expression translation, case analysis, case conversion, and ambiguity resolution.

  The present invention is applicable to a data processing system in which attribute data for extracting a predetermined term is added to processing unit data constituting processing data using learning data.

FIG. 1 is a configuration diagram of a machine learning system according to the first embodiment of this invention. FIG. 2 is a diagram for explaining the machine learning system according to the first embodiment of this invention. FIG. 3 is a configuration diagram of a machine learning device according to the second embodiment of the present invention. FIG. 4 is a diagram for explaining learned data Rq and learned data SDq according to the second embodiment of the present invention. FIG. 5 is a block diagram of the similar learning data selection unit shown in FIG. FIG. 6 is a diagram for explaining data to be processed according to the second embodiment of the present invention. FIG. 7 is a diagram for explaining similar learning data according to the second embodiment of this invention. FIG. 8 is a configuration diagram of the tag addition unit of the second embodiment of the present invention. FIG. 9 is a configuration diagram of an IOB determination data generation unit according to the second embodiment of this invention. FIG. 10 is a diagram for explaining IOB determination data according to the second embodiment of this invention. FIG. 11 is a diagram for explaining a tag addition unit according to the second embodiment of this invention. FIG. 12 is a diagram for explaining the IOB addition unit according to the second embodiment of this invention. FIG. 13 is a diagram for explaining an operation example of the machine learning apparatus illustrated in FIG. 3. It is a figure for demonstrating 3rd Embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Machine learning apparatus, 2 ... Similarity learning data generator, 3 ... Similarity calculation part, 4 ... Similarity learning data generation part, 11 ... Similarity learning data selection part, 13 ... Tag addition part, 15 ... IOB determination data generation part , 21 ... tag adding unit, 22 ... IOB adding unit, 31 ... input unit, 32 ... input unit, 33 ... addition circuit, 34 ... learning data selection unit, 35 ... output unit, 41 ... input unit, 42 ... part of speech tagger unit 43 ... Suffix tagger unit, 44 ... output unit, 51 ... input unit, 52 ... SVM learning unit, 53 ... output unit, 61 ... input unit, 62 ... part of speech tagger unit, 63 ... Suffix tagger unit, 64 ... output unit, 71 ... Input unit, 72 ... IOB determination unit, 73 ... Output unit

Claims (15)

A data processing method for machine learning processing of data to be processed using learning data generated based on the data to be learned,
For each of the plurality of learning data, a first step of generating similarity data indicating a similarity between the learning data used to generate the learning data and the processed data;
A second step of selecting a part of the learning data among the plurality of learning data based on the similarity data generated in the first step;
And a third step of performing machine learning processing on the processed data using the learning data selected in the second step. The learning data defines a correspondence relationship between each of a plurality of processing unit data constituting the learned data and attribute data indicating the attribute,
In the third step, based on the correspondence defined by the learning data selected in the second step, the correspondence between a plurality of processing unit data constituting the processing target data and the attribute data The data processing method according to claim 1, wherein a process for defining the data is performed. In the first step, for each of the plurality of learning data, the similarity between the processing unit data constituting the learning data of the learning data and the processing unit data constituting the processing data is determined. Generating said similarity data,
3. The data according to claim 2, wherein the third step performs processing for adding the attribute data to the processing unit data constituting the processing target data, using the learning data selected in the second step. Processing method. In the first step, for each of the data to be learned and the data to be processed, a vector in which the characteristics of each data is predetermined and defined in the feature evaluation coordinate system based on the processing unit data constituting each data. The data processing method according to claim 1, wherein data is generated, and the similarity data is generated based on the vector data. When the learned data and the processed data are document data, and the processing unit data is word data,
5. The first step generates the vector data having the type of the word data appearing in each data as an element of the vector data and the appearance frequency of the word data of the type as the value of the element. The data processing method described in 1. The first step generates the vector data indicated by (X, Y, Z),
A1: d (x, y) ≧ 0
A2: d (x, y) = d (y, x)
A3: The necessary and sufficient condition for d (x, y) = 0 is x = y
A3 ': d (x, x) = 0
A4: d (x, z) ≤ d (x, y) + d (y, z)
age,
B1: A1, A2, A3, A4
B2: A1, A2, A3 ', A4
B3: A1, A2, A3
B4: A1, A2, A3 '
B5: A1, A2
If
The function d () indicating the measure satisfying any one of the above B1, B2, B3, B4, and B5 is used, or the distance is calculated by a similarity calculation formula in which the distance monotonously decreases as the similarity increases. The data processing method according to claim 4, wherein the similarity data indicating the distance is calculated and calculated. A data processing method executed by a data processing apparatus that uses learning data to which attribute data is attached to each of a plurality of processing unit data constituting the learned data, and adds the attribute data to the processed data by machine learning. ,
For each of a plurality of pieces of learning data, a similarity data indicating a degree of similarity between the processing unit data constituting the learned data of the learning data and the processing unit data constituting the processed data is generated. 1 process,
Based on the similarity data generated in the first step, among the learned data of the plurality of learning data, the learned data satisfying a predetermined criterion for similarity to the processed data. A second step of identifying and selecting the learning data corresponding to the identified learned data;
A data processing method comprising: a third step of adding the attribute data to the processing unit data constituting the data to be processed by machine learning using the learning data selected in the second step. The first step includes
Identify the different types of processing unit data included in the processed data and the learned data,
For each of the processed data and the learned data, the number of each of the different types of the processing unit data is specified, and the specified number constitutes each of the processed data and the learned data. Generate index data by dividing by the total number of processing unit data,
The data processing method according to claim 7, wherein the similarity data is generated based on a combination pattern of the index data of the different types of the processing unit data obtained for each of the processed data and the learned data. The first step includes
Generating vector data whose elements are the index data of the processing unit data of the different types obtained for each of the processed data and the learning data;
The data processing method according to claim 8, wherein the similarity data is generated based on a relationship between the vector data of the processed data and the vector data of the plurality of learned data. The first step includes
The index data is subjected to processing for reducing the value of the index data of the processing unit data of the type as the number of the processed data and the learning data including the specified type of processing unit data increases,
The data processing method according to claim 9, wherein the similarity data is generated based on the combination pattern of the index data after the processing is performed. The number of learning data is n, the number of processed data is 1,
The number of types of processing unit data included in the n pieces of learning data corresponding to the n pieces of learning data and the processing data is k,
i is an integer from 1 to k, j is an identifier attached to N pieces of learned data and one piece of processed data,
The index data is expressed as “TF (i, j) = (number of times processing unit data i appears in learned data j (processed data j)) / (processing unit included in learned data j (processed data j)). Total number of data) "
When “DF (i) = (the number of data in which the processing unit data i appears among the n pieces of learned data and the processed data”),
The first step includes
“IDF (i) = log [(N + 1) / DF (i)]”
Calculate
“W (i, j) = TF (i, j) * IDF (i)”
Calculate
A vector D (q) of the n pieces of learned data q (q is an integer of 1 to N) is expressed as D (q) = (w (1, q), w (2, q),. (K, q)), and the vector D (M) of the processed data is DM = (w (1, M), w (2, M),..., W (k, M)),
The data processing method according to claim 10, wherein the similarity data BA (q) is calculated by the following formula (1) for all q of 1 to n.

The first step generates the similarity data using the learned data of the plurality of learning data read from the memory and the processed data input from the outside of the data processing device,
The data processing method according to claim 7, wherein in the second step, the learning data corresponding to the specified learned data is selected from the learning data read from the memory. The data processing device has similar data generation means, selection means, and attribute data addition means,
The similar data generating means performs the first step,
The selection means performs the second step;
The data processing method according to claim 7, wherein the attribute data adding unit performs the third step. A data processing device that performs machine learning processing of processed data using learning data generated based on the learned data,
For each of the plurality of learning data, similarity data generation means for generating similarity data indicating the similarity between the learning data used to generate the learning data and the processed data;
Selection means for selecting a part of the learning data among the plurality of learning data based on the similarity data generated by the similarity data generation means;
A data processing apparatus comprising: processing means for performing machine learning processing on the data to be processed using the learning data selected by the selection means. A program executed by a data processing apparatus that performs machine learning processing of processed data using learning data generated based on the learned data,
For each of the plurality of learning data, a first procedure for generating similarity data indicating a similarity between the learning data used to generate the learning data and the processed data;
A second procedure for selecting a part of the learning data from the plurality of learning data based on the similarity data generated in the first procedure;
A third procedure for performing machine learning processing on the data to be processed using the learning data selected in the second procedure.

Images