JP2019200554A - Generation method, generation program, and generation device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To suppress the influence of no input of a sentence on output. According to a generation method, a computer determines a first word and a first set of intermediate information in a format different from the first sentence according to the first sentence based on learning information as a first set. A sentence generated by generating a first conversion parameter for converting into information according to the information and a second conversion parameter for converting the first word into the first intermediate information, and correlating with the second input item. When the input of the second word associated with the first input item is accepted without accepting the input of, the second word is converted by the second conversion parameter to obtain the second intermediate information of the format. Along with the generation, the second word and the second intermediate information set are converted by the first conversion parameter to generate the second information. [Selection diagram] Fig. 11

Description

  The present invention relates to a generation method, a generation program, and a generation apparatus.

  A technique for generating a system such as a classifier that generates output information based on input information using machine learning is used. However, when the input information is missing, the accuracy of the output information may be reduced, so a technique for predicting the missing information is used.

  As a related technique, a technique has been proposed in which training data is input to a data learning program to create a classifier, and input data generated from data containing missing values is input to the classifier to predict missing values. (For example, see Patent Document 1).

  Further, as a related technique, a technique has been proposed in which when there are a plurality of fields containing missing values, the missing values are predicted from the data of fields other than the fields containing missing values and interpolated (see, for example, Patent Document 2). ).

  Further, as a related technique, a technique for estimating a missing value of incomplete sensor information and acquiring a model by using an iterative inversion method and error back propagation has been proposed (see, for example, Patent Document 3).

  Also, as a related technique, accept the objective variable that you want to predict among the missing values, and calculate the value of the objective variable from the value and coefficient of each explanatory variable of the record including the objective variable based on the result of multiple regression analysis Techniques have been proposed (see, for example, Patent Document 4).

JP 2007-334719 A JP 2000-40079 JP-A-8-212184 JP 2014-63432 A

  For example, when a sentence is missing in the input information, it is conceivable to predict the sentence using other information in the input information. However, since the number of words in the sentence is variable and the order of the words is taken into account, it is difficult to directly predict the missing sentence. And if the prediction precision of the missing sentence falls, the precision of output information will also fall.

  As one aspect, an object of the present invention is to suppress the influence on the output that a sentence is not input.

  In one aspect, the generation method includes: a first word input in association with a first input item; a first sentence input in association with a second input item; Learning information including one word and first information corresponding to the first sentence set, and based on the learning information, the first word and the first sentence according to the first sentence A first conversion parameter for converting a set of first intermediate information in a format different from that of the first sentence into information corresponding to the first information; and the first word is converted into the first intermediate information. The second conversion parameter to be converted into the second input parameter is generated, and the second word associated with the first input item is input without accepting the input of the sentence associated with the second input item. If accepted, the second word is converted by the second conversion parameter, Generating the second intermediate information in the format, and generating the second information by converting the combination of the second word and the second intermediate information by the first conversion parameter To do.

  According to one aspect, it is possible to suppress the influence on the output that a sentence is not input.

It is a figure which shows an example of a production | generation apparatus. It is a figure which shows an example of input information and output information. It is a figure which shows the example of each information and a parameter. It is a figure which shows an example of the parameter production | generation method. It is a figure which shows the example of intermediate information. It is a figure which shows the example of the production | generation method of the parameter converted into intermediate information. It is a figure which shows the example of the conversion method from input information to intermediate information. It is a figure which shows the 1st example of the production | generation method of the parameter converted into other input information. It is a figure which shows the 2nd example of the production | generation method of the parameter converted into other input information. It is a figure which shows the example of the conversion method from input information to other input information. It is a flowchart which shows an example of the process at the time of learning. It is a flowchart which shows an example of the process at the time of operation. It is a figure which shows the example of the determinant which shows input information. It is a figure which shows each information and parameter in an Example. It is a figure which shows the example of the information containing a defect | deletion and an outlier. It is a figure which shows an example of the hardware constitutions of a production | generation apparatus.

  Hereinafter, embodiments will be described with reference to the drawings. FIG. 1 is a diagram illustrating an example of the generation device 1. The generation device 1 is an example of a computer. The generation device 1 is, for example, a server, a personal computer, a tablet terminal, or the like.

  The generation device 1 includes an acquisition unit 11, a first generation unit 12, a second generation unit 13, a determination unit 14, a conversion unit 15, a third generation unit 16, and a storage unit 17.

  The acquisition unit 11 includes a first word input in association with the first input item, a first sentence input in association with the second input item, the first word, and the first sentence. The learning information including the first information corresponding to the set of is acquired.

  The first generation unit 12 sets, based on the learning information, a set of first intermediate information in a format different from the first word and the first sentence corresponding to the first sentence according to the first information. A first conversion parameter to be converted into information is generated.

  The second generation unit 13 generates a second conversion parameter for converting the first word into the first intermediate information. The first generation unit 12 and the second generation unit 13 are an example of a parameter generation unit.

  The determination unit 14 determines whether there is a defect in the input information. Note that the generation device 1 performs the same processing as when there is a defect even when the input information includes information that does not exist in the learning information, that is, when there is an outlier in the input information.

  When the conversion unit 15 receives the input of the second word associated with the first input item without receiving the input of the sentence associated with the second input item, the conversion unit 15 receives the input of the second word associated with the first input item. The second word is converted to generate second intermediate information.

  The third generation unit 16 converts the combination of the second word and the second intermediate information using the first conversion parameter to generate the second information. The conversion unit 15 and the third generation unit 16 are an example of an information generation unit.

  The storage unit 17 stores various generated parameters, intermediate information, and the like.

FIG. 2 is a diagram illustrating an example of input information and output information. The input information and output information shown in FIG. 2 are information regarding movies. The input items of the input information shown in FIG. 2 include a movie genre (x ₁ ), a production company (x ₂ ), and a movie overview (x ₃ ). The movie genre (x ₁ ) and the production company (x ₂ ) are expressed by words, and the movie overview (x ₃ ) is expressed by sentences. The item of output information includes interest (y). In the following description, each item may be expressed using symbols (x ₁ , x ₂ , x ₃ , y).

FIG. 3 is a diagram illustrating examples of information and parameters. As shown in FIG. 3, the input information x ₁ is converted into intermediate information x ₁ ′ by a classifier using the parameter w ₁ . Similarly, the input information x ₂ is converted into intermediate information x ₂ ′ by a classifier using the parameter w ₂ . Further, the input information x ₃ is converted into intermediate information x ₃ ′ by a classifier using the parameter w ₃ . The intermediate information _{x 1 ', x 2',} x 3 ' is converted into y by classifier with parameters _{w h.}

When the generation apparatus 1 acquires x ₁ to x ₃ and y as learning information, the generation apparatus 1 generates each parameter (w _{1 to} w ₃ , w _h ) based on the learning information. Moreover, when the generation apparatus 1 acquires the input information x _{1 to} x ₃ to generate unknown output information, the generation information ₁ includes the input information x _{1 to} x ₃ and the generated parameters (w _{1 to} w ₃ , w _h ). To generate y.

FIG. 4 is a diagram illustrating an example of a parameter generation method. In the example illustrated in FIG. 4, it is assumed that x _{1 to} x ₃ and y are given as learning information. The first generator _12, based on x 1 _~x 3, y, generates the parameters _{(w 1 ~w 3, w h} ). The first generation unit 12 generates a predicted value pred_y of y based on x _{1 to} x ₃ , y and the initial parameter, and adjusts the parameter so that pred_y approaches y. When the predicted value pred_y generated using a certain parameter is closest to y, the first generation unit 12 stores the parameter (w _{1 to} w ₃ , w _h ) in the storage unit 17. The parameters (w _{1 to} w ₃ , w _h ) stored in the storage unit 17 are used for generating unknown output information from input information during operation.

FIG. 5 is a diagram illustrating an example of intermediate information. In the example illustrated in FIG. 5, the intermediate information x ₃ ′ is represented by a matrix. If x ₃ is the sentence, intermediate information x _{3 ',} for example, indicating the meaning of a sentence. It is assumed that the intermediate information x ₃ ′ is a matrix having a smaller number of dimensions than the input information x ₃ . Therefore, the generation device 1 can predict the intermediate information x ₃ ′ more easily than when the input information x ₃ is predicted when the input information x ₃ is missing.

FIG. 6 is a diagram illustrating an example of a method for generating a parameter to be converted into intermediate information. In the example shown in FIG. 6, the second generation unit 13 obtains the _{x 1,} to produce a ₃₁ 'parameter w for generating' intermediate information _{x 3} from _{x 1.} For example, the second generation unit 13 generates pred1_x ₃ ′ that is a predicted value of the intermediate information x ₃ ′ based on x ₁ and the initial parameter w ′ ₃₁ . Then, the second generation unit 13 generates w ′ ₃₁ so that the difference between pred1_x ₃ ′ and x ₃ ′ is the smallest. x ₃ ′ is a value generated in advance by the second generation unit 13 using the parameter w ₃ generated by the first generation unit 12 based on the learning information, the input information, and x ₃ .

Second generating unit 13 similarly acquires _{x 2,} to produce a ₃₂ 'parameter w for generating' intermediate information _{x 3} from _{x 2.}

When the input information x ₃ is missing, the conversion unit 15 generates x ₃ ′ from x ₁ using w ′ ₃₁ . Or, converter 15, when the input information _{x 3} is deficient, 'with _{32, x 3} from _{x 2'} w generates a.

X ₁ or x ₂ in FIG. 6 is an example of the first word. x ₃ is an example of the first sentence. w ′ ₃₁ or w ′ ₃₂ is an example of a second conversion parameter.

FIG. 7 is a diagram illustrating an example of a method for converting input information into intermediate information. If the _{x 3} input information during operation are deficient, converting unit 15, 'with _{31, x 3} from _{x 1'} w the second generator 13 has generated to produce a _1. Furthermore, the conversion unit 15 generates x ₃ ′ _2 from x ₂ using w ′ ₃₂ generated by the second generation unit 13.

Converting unit 15, based on the _2 _{'x 3} is generated from _1 and _{x 2' of x 3} generated from _{x 1,} to generate the _{x 3} '. Conversion unit 15, for example, the average value of the _{x 3} '_1 and _{x 3'} _2, and _{x 3} '. The conversion unit 15 may set x ₃ ′ _1 and x ₃ ′ _2 as x ₃ ′.

In FIG. 7, x ₁ and x ₂ are examples of second words, and x ₃ ′ is an example of second intermediate information. y generated from _{x 1, x 2, x 3} ' is an example of the second information.

FIG. 8 is a diagram illustrating a first example of a method for generating a parameter to be converted into other input information. In the example shown in FIG. 8, the second generation unit 13 obtains the _{x 1,} generates parameters _{w 21} for generating _{x 2} from _{x 1.} Second generating unit 13, for example, based on _{x 1} and the initial parameter _{w 21,} to produce a Pred1_x ₂ is the predicted value of _{x 2.} Then, the second generation unit 13 generates w ₂₁ so that the difference between pred1_x ₂ and x ₂ is minimized.

Second generating unit 13 similarly acquires _{x 3,} it generates parameters _{w 23} for generating _{x 2} from _{x 3.}

  FIG. 9 is a diagram illustrating a second example of a method for generating a parameter to be converted into other input information. The second generation unit 13 generates parameters for converting each input information into other input information by a method similar to the method shown in FIG.

For example, the second generation unit 13 generates a parameter w ₂₁ for converting x ₁ into x ₂ and a parameter w ₃₁ for converting x ₁ into x ₃ . The second generation unit 13 generates a parameter w ₁₂ for converting x ₂ to x ₁ and a parameter w ₃₂ for converting x ₂ to x ₃ . The second generation unit 13 generates a parameter w ₁₃ for converting x ₃ to x ₁ and a parameter w ₂₃ for converting x ₃ to x ₂ .

FIG. 10 is a diagram illustrating an example of a conversion method from input information to other input information. _{If x 2} input information during operation is defective, the converting unit 15 uses the _{w 21} of the first generator 12 has generated, to produce an _x 2 _1 from _{x 1.} Furthermore, the conversion unit 15 generates x ₂ _ ₃ from x ₂ using w ₂₃ generated by the first generation unit 12.

Converting unit 15, based on the _x 2 _3 generated from _x 2 _1 and _{x 3} generated from _{x 1,} to generate the _{x 2.} Conversion unit 15, for _example, the average value of the _x 2 _1 and _x 2 _3, and _{x 2.} For example, the conversion unit 15 may set x ₂ _ 1 and x ₂ _ ₃ as x ₂ .

  FIG. 11 is a flowchart illustrating an example of processing during learning. The acquisition unit 11 acquires learning information for generating a classifier parameter (step S101). The learning information is associated with, for example, a first word input in association with a first input item (for example, genre, production company, etc.) and a second input item (for example, a sentence indicating an outline). Contains the input first sentence. The learning information includes first information (for example, fun) according to the combination of the first word and the first sentence.

The 1st production | generation part 12 produces | generates the parameter which converts the combination of a 1st word and 1st intermediate information into the information according to 1st information (step S102). The parameter generated in step S102 is an example of a first conversion parameter. The parameter generated in step S102 is, for example, a set of W ₁ , W ₂ , and W _h shown in FIG.

  The 2nd generation part 13 starts a repetition process for every input item of input information which acquisition part 11 acquired (Step S103). The second generation unit 13 determines whether the input information of the processing target item is a sentence (step S104).

When the input information is a sentence (YES in step S104), the second generation unit 13 generates a parameter for converting the input information into intermediate information according to other input information (step S105). Second generating unit 13, for example, the _{x 1,} generates the ₃₁ 'parameters W for converting the' _{x 3} is the intermediate information of _{x 3} (see FIG. 6). The parameter generated in step S105 is an example of a second parameter.

If the input information is not a sentence (NO in step S104), the second generation unit 13 generates a parameter for converting the input information into other input information (step S106). For example, the second generation unit 13 generates a parameter W ₂₁ for converting x ₁ into x ₂ (see FIG. 8).

  The 2nd production | generation part 13 complete | finishes a repetition process, when the process of step S104-S106 is performed about all the items of the input information which the acquisition part 11 acquired (step S107).

  FIG. 12 is a flowchart illustrating an example of processing during operation. The acquisition unit 11 receives input information (step S201). The input information is, for example, words and sentences.

  The determination unit 14 determines whether there is a defect in the input information (step S202). The determination unit 14 determines that the input information is missing when some or all of the input information is insufficient.

  When it is determined that the input information is missing (YES in step S202), the conversion unit 15 starts a repetition process for each missing input information (step S203).

The conversion unit 15 determines whether the missing information is a sentence (step S204). When the missing information is a sentence (YES in step S204), the conversion unit 15 converts the input information using the second parameter generated by the second generation unit 13, and intermediate the missing input information. Information is generated (step S205). It is assumed that the intermediate information has a format different from that of the input information. For example, as illustrated in the example of FIG. 7, the conversion unit 15 generates x ₃ ′ _1 from x ₁ using w ′ ₃₁ .

When the missing information is not a sentence (NO in step S204), the conversion unit 15 converts the input information into the missing input information using the parameter generated by the first generation unit 12 (step S206). The conversion unit 15 ends the repetition process (step S207). For example, the conversion unit 15 generates x ₂ _ ₁ from x ₁ using w ₂₁ generated by the first generation unit 12.

The third generation unit 16 determines whether the missing information is a sentence (step S208). If the sentence is missing, intermediate information is generated in the process of step S205. Therefore, in the case of YES in step S208, the third generation unit 16 generates output information corresponding to the set of input information and intermediate information using the first parameter (step S209). For example, the third generation unit 16 generates the output information y based on the input information (x ₁ , x ₂ ), the intermediate information x ₃ ′, and W _h , W ₁ , W ₂ obtained at the time of learning. To do.

In the case of NO in step S202 or step S208, the third generation unit 16 generates output information corresponding to the input information using the first parameter (step S210). For example, the third generation unit 16 generates output information y based on input information (x ₁ , x ₂ , x ₃ ) and W _h , W ₁ , W ₂ , W ₃ obtained at the time of learning.

  In addition, when learning information contains a defect at the time of learning, the production | generation apparatus 1 performs the process shown in FIG. 11 with respect to the learning information except a defect, and produces | generates a parameter. Thereby, the generation device 1 can prevent the loss from becoming noise and reducing the learning accuracy. Further, the generation device 1 performs the same processing as steps S203 to S207 in FIG. 12 for learning information loss, and generates input information or intermediate information. Thereby, since the production | generation apparatus 1 can compensate the defect | deletion of learning information, it can increase the quantity of learning information and can improve the precision of output information.

  As described above, the generation apparatus 1 of the present embodiment generates parameters for generating intermediate information corresponding to sentences from words during learning. And the production | generation apparatus 1 produces | generates intermediate information using the produced | generated parameter, and produces | generates output information using the intermediate information, when a sentence is not input at the time of an operation | movement (it is missing). Therefore, the generation device 1 can generate output information using intermediate information even if information that is difficult to predict, such as a sentence, is missing.

  Further, since the intermediate information is information having a smaller dimension than the input information, the generation device 1 can easily predict the intermediate information using other information.

<Example>
FIG. 13 is a diagram illustrating an example of a determinant indicating input information. The input information shown in FIG. 13 corresponds to the input information shown in FIG. Figure 13 (a) is a determinant indicating the input information _{x 1} showing the "action". The number of rows in FIG. 13A corresponds to the number of movie genres. The second line indicates “action”. Therefore, when the input information x ₁ is “action”, the value in the second row of the determinant is 1.

Figure 13 (b) is a matrix equation representing the input information _{x 2} indicating the "Disney". The number of rows in FIG. 13B corresponds to the number of production companies. Also, assume that the first line indicates “Disney”. Therefore, when the input information _{x 2} is "Disney", the value of the first row of the determinant is 1.

FIG. 13C is a determinant indicating the input information x ₃ indicating “dynamic movie”. The number of rows in FIG. 13C corresponds to the number of registered words, and the number of columns (number of dimensions) corresponds to the number of words in the input information. The second line indicates “dynamic”, the fourth line indicates “na”, and the fifth line indicates “movie”.

  In actual operation, the number of registered words is, for example, about 100,000 words, and the number of words in the input information is about 10. In that case, the number of patterns is 100,000 × 10 = 1 million patterns. Since the number of patterns is greatly different from the number of patterns when the input information is a word, it is difficult to predict a sentence directly from the word.

However, the intermediate information x ₃ ′ is one-dimensional information as shown in FIG. That is, when the input information is converted to the intermediate information x ₃ ′, the order of the included words is lost, but the meaning of the words is not lost.

(1) Example of First Parameter Generation Processing During Learning Hereinafter, an example of parameter generation by the first generation unit 12 during learning will be described. FIG. 14 is a diagram illustrating information and parameters in the embodiment. h represents a set of intermediate information x ₁ ′, x ₂ ′, x ₃ ′. Each information and parameter other than h is the same as the example shown in FIG.

The intermediate information x ₁ ′, x ₂ ′, x ₃ ′ is expressed as the following formulas (1) to (3).

Further, h indicating a set of the intermediate information x ₁ ′, x ₂ ′, and x ₃ ′ is expressed as the following formula (4).

  Also, pred_y, which is the predicted value of the output information y, is expressed as in the following equation (5).

Note that the softmax function of Equation (5) is defined as follows.
softmax (x) = exp (x) / sum (exp (x))

The first generator 12, so as to approach and the pred_y and y, adjusting each parameter _{(w 1 ~w 3, w h} ). The 1st production | generation part 12 learns using back propagation, for example. As the difference (loss) between pred_y and y, SoftMaxCrossEntropy, Mean Squared Error (MSE), or the like is used.

During operation, the third generation unit 16 calculates pred_y using each parameter (w _{1 to} w ₃ , w _h ) generated by the first generation unit 12 and outputs a value predicted from the calculated pred_y as output information. Output as y. For example, in pred_y of Expression (5), when 0.9 in the first line is the largest and the first line indicates “very good”, the third generation unit 16 predicts that y is “very good”. To do.

(2) Example of Second Parameter Generation Processing During Learning Next, an example of a method in which the second generation unit 13 generates a parameter for generating the intermediate information x ₃ ′ during learning will be described. The following description corresponds to the embodiment of the process shown in FIG. pred1_x ₃ ′, which is a predicted value of the intermediate information x ₃ ′ based on x ₁ , is expressed as the following Expression (6).

Further, the difference between pred1_x ₃ ′ and x ₃ ′ is expressed as in the following formula (7) when MSE is used.

The second generation unit 13 generates the parameter w ′ ₃₁ so that the loss shown in Expression (7) is minimized. Similarly, the second generation unit 13 generates a parameter w ′ ₃₂ for generating intermediate information x ₃ ′ from x ₂ .

(3) in the example then during operation of the output information generation process when the sentence (x ₃₎ were missing at the time of operation, illustrating the output information generation process when the sentence (x ₃₎ was missing. The third generation unit 16 generates intermediate information x ₁ ′ based on the input information x ₁ and w ₁ generated at the time of learning as shown in the following formula (8).

The third generator 16, based on the input information x ₂ and w ₂ that is generated at the time of learning, to generate intermediate information x _{2 'as} in the following formula (9).

Then, the conversion unit 15 'based on the _31, as in the following equation (10), the input information _{x 1 x 3'} w the second generator 13 has generated at the time of learning to convert _1.

Then, the conversion unit 15 'based on the _32, as shown in equation (11) in the following equation, the input information _{x 2 x 3'} w the second generator 13 has generated at the time of learning to convert _2.

Then, the conversion unit 15, as in the following equation (12), the average value of the _2 _{'x 3} is generated from _1 and _{x 2' of x 3} generated from _{x 1,} and _{x 3} '.

The set h of intermediate information is represented by the following equation (13) when x ₁ ′, x ₂ ′, and x ₃ ′ generated as in the above equations (10) to (12) are used. .

The third generation unit 16 generates output information y based on W _h generated at the time of learning and _h calculated as described above, as in Expression (14) below. This output information y is an example of second information.

(4) at the time of output information generating process instances then operation when a word (x ₂₎ were missing at the time of operation, illustrating the output information generation process in the case where the word (x ₂₎ was missing. Converting unit 15, based on the _{w 21} generated at the time of learning, into a Pred1_x ₂ is the predicted value of the input information _{x 2} input information _{x 1} as the following equation (15).

The conversion unit 15, based on the _{w 23} generated at the time of learning, into a Pred3_x ₂ is the predicted value of the input information _{x 2} input information _{x 3} as in the following formula (16).

W ₂₃ sum (x ₃ ) in equation (16) is calculated as in equation (17) below.

Converting unit 15, by calculating the Pred1_x _2, and pred3_x average value of ₂ is generated as described above, to generate a _{x 2} as in the following formula (18).

Of the above _{x 2,} for 0.42 in the third row it is the largest value, the converting unit 15 converts 0.42 to 1, to convert the other numbers 0. Conversion unit 15 obtains the _{x 2} as the following equation (19).

The third generation unit 16 uses the input information (x ₁ , x ₂ , x ₃ ) including x ₂ obtained by the above calculation and the W _h , W ₁ , W ₂ , W ₃ obtained at the time of learning. Based on this, output information y is generated.

FIG. 15 is a diagram illustrating an example of information including missing and outliers. FIG. 15A shows an example of input information including a defect. Figure 15 _{x 1} in (a) determines that since all elements are 0, the determination unit 14 is defective.

FIG. 15B is an example of input information including outliers. The set of x ₁ and x ₂ in FIG. 15 (b), and a combination that did not exist in the input information at the time of learning. The determination unit 14 may regard a value included in such input information as an outlier.

FIG. 15C is an example of input information including an outlier. Figure 15 x ₃ in (c) is assumed to be information that did not exist in the input information at the time of learning. The determination unit 14 may regard a value included in such input information as an outlier.

  Next, an example of the hardware configuration of the generation apparatus 1 will be described. FIG. 16 is a diagram illustrating an example of a hardware configuration of the generation apparatus 1. As shown in the example of FIG. 16, in the generation device 1, a processor 111, a memory 112, an auxiliary storage device 113, a communication interface 114, a medium connection unit 115, an input device 116, and an output device 117 are connected to the bus 100. .

  The processor 111 executes the program expanded in the memory 112. A generation program that performs the processing in the embodiment may be applied to the program to be executed.

  The memory 112 is, for example, a random access memory (RAM). The auxiliary storage device 113 is a storage device that stores various information, and for example, a hard disk drive, a semiconductor memory, or the like may be applied. The auxiliary storage device 113 may store a generation program that performs the processing of the embodiment.

  The communication interface 114 is connected to a communication network such as a local area network (LAN) or a wide area network (WAN), and performs data conversion associated with communication.

  The medium connection unit 115 is an interface to which a portable recording medium 118 can be connected. As the portable recording medium 118, an optical disc (for example, Compact Disc (CD) or Digital Versatile Disc (DVD)), a semiconductor memory, or the like may be applied. A generation program for performing the processing of the embodiment may be recorded on the portable recording medium 118.

  The input device 116 is, for example, a keyboard, a pointing device, and the like, and receives input from the user such as instructions and information.

  The output device 117 is, for example, a display device, a printer, a speaker, or the like, and outputs an inquiry or instruction to the user, a processing result, and the like. The generation apparatus 1 may output (display or print) the generated parameters, output information, and the like using the output apparatus 117.

  The storage unit 17 illustrated in FIG. 1 may be realized by the memory 112, the auxiliary storage device 113, the portable recording medium 118, or the like. In the acquisition unit 11, the first generation unit 12, the second generation unit 13, the determination unit 14, the conversion unit 15, and the third generation unit 16 illustrated in FIG. 1, the processor 111 executes the generation program expanded in the memory 112. May be realized.

  The memory 112, the auxiliary storage device 113, and the portable recording medium 118 are computer-readable, non-temporary tangible storage media, and are not temporary media such as signal carriers.

<Others>
The present embodiment is not limited to the embodiment described above, and various modifications, additions, and omissions can be applied without departing from the gist of the present embodiment.

DESCRIPTION OF SYMBOLS 1 Generation apparatus 11 Acquisition part 12 1st generation part 13 2nd generation part 14 Judgment part 15 Conversion part 16 3rd generation part 17 Memory | storage part 100 Bus 111 Processor 112 Memory 113 Auxiliary storage device 114 Communication interface 115 Medium connection part 116 Input device 117 output device 118 portable recording medium

Claims (5)

Computer
A set of a first word input in association with the first input item, a first sentence input in association with the second input item, and the first word and the first sentence. And learning information including the first information in response,
Based on the learning information, a set of first intermediate information in a format different from the first sentence according to the first word and the first sentence is used as information according to the first information. Generating a first conversion parameter for conversion and a second conversion parameter for converting the first word into the first intermediate information;
If an input of a second word associated with the first input item is received without receiving an input of a sentence associated with the second input item, the second conversion parameter causes the second To generate second intermediate information of the format, and to convert a set of the second word and the second intermediate information according to the first conversion parameter, thereby generating second information Generate
A generation method characterized by executing processing. The generation method according to claim 1, wherein the first intermediate information is information having a format different from that of the first sentence, which indicates a meaning of the first sentence. 3. The generation method according to claim 1, wherein the number of dimensions of the matrix indicating the first intermediate information is smaller than the number of dimensions of the matrix indicating the first sentence. A set of a first word input in association with the first input item, a first sentence input in association with the second input item, and the first word and the first sentence. And learning information including the first information in response,
Based on the learning information, a set of first intermediate information in a format different from the first sentence according to the first word and the first sentence is used as information according to the first information. Generating a first conversion parameter for conversion and a second conversion parameter for converting the first word into the first intermediate information;
If an input of a second word associated with the first input item is received without receiving an input of a sentence associated with the second input item, the second conversion parameter causes the second To generate second intermediate information of the format, and to convert a set of the second word and the second intermediate information according to the first conversion parameter, thereby generating second information Generate
A generation program that causes a computer to execute processing. A set of a first word input in association with the first input item, a first sentence input in association with the second input item, and the first word and the first sentence. An acquisition unit for acquiring learning information including the first information in response,
Based on the learning information, a set of first intermediate information in a format different from the first sentence according to the first word and the first sentence is used as information according to the first information. A parameter generation unit for generating a first conversion parameter to be converted and a second conversion parameter for converting the first word into the first intermediate information;
If an input of a second word associated with the first input item is received without receiving an input of a sentence associated with the second input item, the second conversion parameter causes the second To generate second intermediate information of the format, and to convert a set of the second word and the second intermediate information according to the first conversion parameter to obtain second information An information generation unit for generating
A generating apparatus comprising:

Images