JP5993897B2 - Specific apparatus, specific method, and specific program - Google Patents

Description

  The present invention relates to a specific device, a specific method, and a specific program.

  Conventionally, various types of data are labeled to indicate the data category. For example, when a product is listed in an Internet auction, the seller must manually select a category to which the product to be listed belongs from among categories prepared in advance so that the category search or refined search of the product can be performed. Label products with.

JP 2004-062719 A

  However, the above method cannot always provide an appropriate label. For example, in the above method, the exhibitor arbitrarily determines the label of the category to which the product belongs and determines the label of the product. For this reason, in the above method, an inappropriate label may be attached to the product. For this reason, the above method cannot always provide an appropriate label.

  The present application has been made in view of the above, and an object thereof is to provide a specifying device, a specifying method, and a specifying program capable of attaching an appropriate label.

  The specific device according to the present application includes: a receiving unit that receives input data; and a learning device that reproduces feature information of a predetermined classification, and the output data that is output when the input data received by the receiving unit is input. A calculation unit that calculates a reproduction error with respect to input data, and a specifying unit that specifies a classification to which the input data belongs based on the reproduction error calculated by the calculation unit.

  According to one aspect of the embodiment, there is an effect that an appropriate label can be attached.

FIG. 1 is an explanatory diagram illustrating an example of the specifying process performed by the specifying apparatus according to the embodiment. FIG. 2 is an explanatory diagram showing an example of an auto encoder. FIG. 3 is a diagram illustrating a configuration example of the exhibition system according to the embodiment. FIG. 4 is a diagram illustrating a configuration example of the specific device according to the embodiment. FIG. 5 is a flowchart illustrating a generation processing procedure performed by the specific apparatus. FIG. 6 is a flowchart showing a specific processing procedure by the specific device. FIG. 7 is a hardware configuration diagram illustrating an example of a computer that realizes the function of the specific device.

  Hereinafter, a specific apparatus, a specific method, and a specific program (hereinafter referred to as “embodiment”) according to the present application will be described in detail with reference to the drawings. Note that the specific device, the specific method, and the specific program according to the present application are not limited by this embodiment. Moreover, in the following embodiment, the same code | symbol is attached | subjected to the same site | part and the overlapping description is abbreviate | omitted.

[1. Specific processing)
First, an example of the specifying process according to the embodiment will be described with reference to FIG. FIG. 1 is an explanatory diagram illustrating an example of the specifying process performed by the specifying apparatus according to the embodiment. In the example of FIG. 1, an example in which a specifying process for specifying a person who co-occurs with a product is performed by the specifying device 100 is shown.

The identification device 100 is a server device that identifies a classification to which input data belongs by a learning device (for example, an auto encoder) that reproduces feature information of a predetermined classification. This point will be described in detail with reference to FIG. Learner _Le 1 _{~Le n} is a Autoencoder learned characteristics of a given classification. Specifically, the learning unit Le ₁ ~Le _n is input an image in which a plurality of nodes for outputting a calculation result (e.g., neurons) is a DNN connected to (Deep Neural Network) object is rendered to the input data It is an auto encoder that outputs output data (for example, reproduction data) that reproduces input data when input as data. This point will be described with reference to FIG. FIG. 2 is an explanatory diagram showing an example of an auto encoder. Learner Le ₂ is a Autoencoder learned features of laptop. Learner Le _2, when input image NP which laptop is depicted as input data, and outputs the output data RE to reproduce the laptop.

For example, the learning unit Le _2, the pixel value of each pixel of the image NP (e.g., 8-bit values of the gray scale) is input to the corresponding input node, and acquires a value each output node is output. Further, the learning device Le ₂ constructs an image using the values output from the output nodes as pixel values of pixels at positions corresponding to the output nodes, thereby obtaining an image serving as output data. Here, in the auto encoder, since the number of nodes in the central portion is smaller than the number of input nodes, features that are not reproduced among the features included in the input nodes occur. For this reason, when an image in which an object in the auto encoder is drawn is input, if the object is reproduced well (for example, an image without noise is output), the characteristics of the object are displayed. You can judge that you are learning well. On the other hand, the learning unit Le ₃ are the encoder does not learn the features of a notebook personal computer, as shown in FIG. 2, when the image NP which laptop is depicted is input as input data, well image There is a risk of outputting an image in which noise such as missing bits cannot be reproduced.

In the example of FIG. 1, the learning device Le ₁ is an auto encoder that learns the characteristics of the television. Learner Le ₂ is a Autoencoder learned features of laptop. Learner Le _n is a Autoencoder learned characteristics of the smartphone.

In the example of FIG. 1, the learning devices Le _{1 to} Le _n are configured by an input layer, an intermediate layer formed by a multi-stage node group, and an output layer. Here, as shown in FIG. 1, the intermediate layer mainly includes a layer having the smallest number of dimensions.

  First, the identification device 100 receives input data (step S1). Here, the input data is, for example, unknown data to be specified that specifies the classification by the specifying device 100. In the example of FIG. 1, the input data P1 is an image drawn by a notebook computer. In addition to the subject, a background or the like may be drawn on the image.

Subsequently, the specific device 100, depicted by the subject to identify a classification belonging to the input data P1 using a learning device _Le 1 _{~Le n.} Specifically, a particular device 100 first inputs the input data P1 to the learner _Le 1 _{~Le n} (step S2). In FIG. 1, for ease of explanation, three learners Le _1, Le ₂ and learner has learned the features of various different classifications actually shows an example of inputting the input data P1 in the Le _n The input data P1 is input to each learning device prepared in advance. Thus, in the example of FIG. 1, the learning unit Le ₁ outputs the output data OD _1. Learner Le ₂ outputs the output data OD _2. The learning device Le _n outputs output data OD _n .

The specific apparatus 100 calculates relative learner _Le 1 _{~Le n,} a reproduction error with respect to the input data P1 of the output data _OD 1 _{~OD n} to output if you enter the input data P1 (Step S3). Here, the reproduction error is an index indicating the degree of reproduction of output data with respect to input data. For example, the identifying apparatus 100 inputs the grayscale value of each pixel included in the image that is input data to the corresponding input node, and sets the value output by each output node as the grayscale value of each pixel. Further, the identifying apparatus 100 calculates a power value of a difference between a value input to the input node corresponding to each pixel and a value output from the output node. Then, the identifying apparatus 100 sets the sum of the power values calculated for each pixel as the average error value.

That is, the reproduction error becomes lower as the output data accurately reproduces the input data. In the example of FIG. 1, the identifying apparatus 100 calculates the reproduction error of the output data OD ₁ as “50”. Further, the identifying apparatus 100 calculates the reproduction error of the output data OD ₂ as “0”. Further, the identifying apparatus 100 calculates the reproduction error of the output data OD _n as “90”.

Thereafter, the specifying device 100 specifies the classification to which the input data P1 belongs based on the calculated reproduction error (step S4). For example, the specifying apparatus 100 specifies the classification corresponding to the learning device that outputs the output data having the lowest reproduction error among the calculated reproduction errors as the classification to which the input data belongs. In the example of FIG. 1, the classification to which the input data P1 belongs is “notebook personal computer”, which is the classification corresponding to the learning device Le ₂ that outputs the output data OD ₂ that has the lowest reproduction error “0” among the calculated reproduction errors. As specified.

  As described above, the specifying apparatus 100 according to the embodiment receives input data. Further, the identifying apparatus 100 calculates a reproduction error of output data that is output when received input data is input to a learning device that reproduces feature information of a predetermined classification. Further, the specifying device 100 specifies the classification to which the input data belongs based on the calculated reproduction error.

  As a result, the identifying apparatus 100 can identify an appropriate classification to which the input data belongs, and therefore can attach an appropriate label. For example, the specifying device 100 can specify the classification to which the subject drawn in the image belongs when an image in which the subject is drawn is input. For this reason, the specifying device 100 can specify the category to which the product belongs when, for example, an image in which an item newly exhibited by an exhibitor is received in an Internet auction is received.

  Moreover, since the specific apparatus 100 can attach a suitable label with respect to the goods exhibited by an internet auction, for example, the goods with an inappropriate label can be reduced. For this reason, the specific device 100 can improve the convenience of the exhibitor and the user of the auction.

  Moreover, since the identification apparatus 100 can attach an appropriate label to the product simply by inputting the image of the product to the learning device, it is possible to reduce the effort for the seller to review the label. For this reason, the specific device 100 can improve the convenience in the exhibitor's exhibition work.

[2. Configuration of the exhibition system
The configuration of the exhibition system according to the embodiment will be described with reference to FIG. FIG. 3 is an explanatory diagram illustrating an example of an exhibition process performed by the exhibition system according to the embodiment. As shown in FIG. 3, the exhibition system 1 includes a terminal device 10, a server device 50, and a specific device 100. The terminal device 10, the server device 50, and the specific device 100 are connected via a network so as to be communicable by wire or wirelessly. In addition, the exhibition system 1 illustrated in FIG. 3 may include a plurality of terminal devices 10, a plurality of server devices 50, and a plurality of specific devices 100.

  The terminal device 10 is, for example, an information processing device such as a desktop PC, a notebook PC, a tablet terminal, a smartphone, a mobile phone, or a PDA (Personal Digital Assistant). For example, the terminal device 10 is used by a user (for example, a seller) who sells a product in an auction. The terminal device 10 transmits an image prepared by the user to the server device 50. For example, when the terminal device 10 receives a predetermined operation (for example, a click operation) on an exhibition screen that is a web page displayed on the display unit, the terminal device 10 displays an image CP on which a product to be exhibited by the user is depicted. It transmits to the server apparatus 50.

  The server device 50 is a server device that provides various web pages to the terminal device 10. Specifically, the server device 50 provides a web page related to the net auction. For example, the server device 50 provides a management screen (for example, an exhibition screen) that is browsed by an Internet auction site browsed by a consumer or a user who sells products. The server device 50 may provide various web pages such as a technical explanation site, a search engine site, a shopping site, a news site, a finance site (stock price site), etc., as well as the web page of the net auction site.

  The identification device 100 is a server device that identifies a classification to which a product belongs using the learning device Le. Specifically, the identifying apparatus 100 first inputs an image depicting a product as input data to the learning device Le. Subsequently, the identifying apparatus 100 calculates a reproduction error of the output data with respect to the input data. Then, the specifying device 100 specifies the classification to which the product belongs based on the calculated reproduction error.

  Next, the exhibition process by the exhibition system 1 will be described with reference to FIG. First, for example, the terminal device 10 transmits an image CP on which the product C1 to be exhibited on the auction site is drawn to the server device 50 via the exhibition screen provided from the server device 50 (step S21).

  Subsequently, the server device 50 transmits the image CP transmitted from the terminal device 10 to the specific device 100 (step S22). Thereby, the specifying device 100 receives the image CP on which the product C1 is drawn.

  Then, the specifying device 100 specifies the classification to which the product C1 drawn in the image CP belongs (step S23). Specifically, the specifying device 100 inputs the image CP as input data to the learning device Le, and specifies the classification to which the product C1 belongs based on the reproduction error of the output data with respect to the input data.

  Thereafter, the specifying device 100 notifies the server device 50 of the classification to which the specified product C1 belongs (step S24). For example, when specifying the category to which the product C1 depicted in the image CP belongs as “notebook computer”, the specifying device 100 notifies the server device 50 of information related to the category “notebook computer”.

  Then, the server device 50 performs labeling of the product C1 based on the classification notified from the specific device 100 (step S25). For example, when the classification “notebook personal computer” is notified, the server device 50 sets the category of the product C1 to “notebook personal computer”. Accordingly, the server device 50 can appropriately set the label of the product C1 and place it in the auction.

[3. (Specific device configuration)
Next, the configuration of the specifying device 100 according to the embodiment will be described with reference to FIG. FIG. 4 is a diagram illustrating a configuration example of the identification device 100 according to the embodiment. As illustrated in FIG. 4, the identification device 100 includes a communication unit 110, a storage unit 120, and a control unit 130. The specific device 100 includes an input unit (for example, a keyboard and a mouse) that receives various operations from an administrator who uses the specific device 100, and a display unit (for example, a liquid crystal display) that displays various types of information. You may have.

(About the communication unit 110)
The communication unit 110 is realized by a NIC or the like, for example. The communication unit 110 is connected to a network by wire or wirelessly, and transmits / receives information to / from various server devices and terminal devices via the network.

(About the storage unit 120)
The storage unit 120 is realized by, for example, a semiconductor memory device such as a RAM (Random Access Memory) or a flash memory, or a storage device such as a hard disk or an optical disk. The storage unit 120 includes a learning device information storage unit 121.

(About the learning device information storage unit 121)
The learning device information storage unit 121 stores information related to the learning device (for example, DNN). Specifically, the learning device information storage unit 121 stores information on nodes of input layers, intermediate layers, and output layers included in a DNN (for example, an auto encoder) and coupling coefficients between the nodes. For example, the learning device information storage unit 121 stores the learning device Le illustrated in FIG.

(About the control unit 130)
The control unit 130 is configured such that, for example, a CPU (Central Processing Unit), an MPU (Micro Processing Unit), or the like stores various programs (corresponding to an example of the specific program) stored in a storage device inside the specific device 100 in the RAM. This is realized by being executed as The control unit 130 is realized by an integrated circuit such as an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA).

  As illustrated in FIG. 4, the control unit 130 includes a reception unit 131, a generation unit 132, a calculation unit 133, and a specification unit 134, and realizes or executes functions and operations of information processing described below. . Note that the internal configuration of the control unit 130 is not limited to the configuration illustrated in FIG. 4, and may be another configuration as long as the generation processing described later is performed. In addition, the connection relationship between the processing units included in the control unit 130 is not limited to the connection relationship illustrated in FIG. 4, and may be another connection relationship.

(About the reception unit 131)
The accepting unit 131 accepts input data. Specifically, the accepting unit 131 accepts an image depicting a subject. For example, when the seller places a product on the auction site, the accepting unit 131 accepts an image depicting the product to be exhibited from the terminal device 10. In addition, when the learning unit learns, the reception unit 131 receives an image in which subjects of a predetermined classification are drawn. In this case, the reception unit 131 receives a sufficient number of images to learn the characteristics of the predetermined classification.

(About the generator 132)
The generation unit 132 generates a learning device (for example, DNN). Specifically, the generation unit 132 calculates a coupling coefficient between nodes in which output data output when input data is input reproduces feature information of a predetermined classification, and sets the calculated coupling coefficient between the nodes. The auto encoder is generated. As an example, the generation unit 132 calculates a coupling coefficient between nodes included in the DNN by a back-propagation method or supervised learning.

  For example, the generation unit 132 calculates a coupling coefficient between nodes in which output data that is output when the image in which the “notebook computer” is rendered by the reception unit 131 is input reproduces the feature information of the classification of the “notebook computer”. . Then, the generation unit 132 generates an auto encoder by setting the calculated coupling coefficient between nodes. Accordingly, the generation unit 132 can generate a learning device that outputs output data for reproducing a subject when an image depicting a subject belonging to a predetermined category is received. For example, the generation unit 132 generates a learning device that outputs output data, which is an image reproducing a notebook computer, when an image drawn by the notebook computer is input. In addition, the production | generation part 132 produces | generates the learning device corresponding to the various classification | category of the goods used by an online auction, for example. Then, the generation unit 132 stores each generated learning device in the learning device information storage unit 121.

(About the calculation unit 133)
The calculation unit 133 calculates a reproduction error of the output data. Specifically, the calculation unit 133 calculates a reproduction error of output data output when the input data received by the reception unit 131 is input to a learning device that reproduces feature information of a predetermined classification.

  For example, the calculation unit 133 first inputs input data to each learning device that has been stored in the learning device information storage unit 121 and has learned characteristics of various classifications. Subsequently, the calculation unit 133 calculates the reproduction error of the output data output from each learning device. Here, for example, the calculation unit 133 calculates the reproduction error lower as the error between the input data and the output data is smaller. That is, the calculation unit 133 calculates the reproduction error as the output data accurately reproduces the input data. Note that the calculation unit 133 is not limited to the learning device generated by the generation unit 132, and the calculation processing may be performed using a learning device acquired from another.

(About the specific part 134)
The specifying unit 134 specifies the classification to which the input data received by the receiving unit 131 belongs. Specifically, the specifying unit 134 specifies the classification to which the input data belongs based on the reproduction error calculated by the calculating unit 133. For example, the specifying unit 134 specifies the classification corresponding to the learning device that outputs the output data having the lowest reproduction error among the reproduction errors calculated by the calculation unit 133 as the classification to which the input data belongs. This is because an auto encoder having similar characteristics is considered to output output data that reproduces input data with the highest accuracy.

  For example, the specifying unit 134 notifies the server device 50 of the classification specified as the classification to which the input data belongs. Accordingly, the server device 50 can appropriately label the products that the seller sells on the auction site, for example, using the classification notified from the server device 50.

[4. Generation process procedure)
Next, with reference to FIG. 5, a generation process procedure performed by the identification apparatus 100 according to the embodiment will be described. FIG. 5 is a flowchart illustrating a generation processing procedure performed by the specifying apparatus 100 according to the embodiment.

  As illustrated in FIG. 5, the identifying apparatus 100 receives input data (step S101). For example, the specifying apparatus 100 receives a sufficient number of images depicting products belonging to a predetermined category as input data to learn the characteristics of the category to which the product belongs. And when the specific apparatus 100 has not received input data (step S101; No), it waits until input data is received. For example, the specifying apparatus 100 accepts images until a sufficient number is collected to learn the characteristics of the classification to which the product belongs.

  On the other hand, when receiving the input data (step S101; Yes), the specific apparatus 100 inputs the received input data to the learning device (step S102). For example, the specifying apparatus 100 inputs an image in which a product belonging to a predetermined category is drawn as input data to the learning device.

  After that, the identifying apparatus 100 generates a learning device that outputs output data that reproduces the input data when the input data is input based on the received input data. Specifically, the specifying apparatus 100 calculates a coupling coefficient between nodes in which output data output when input data is input reproduces feature information of a predetermined classification (step S103). For example, the identifying apparatus 100 calculates the coupling coefficient by the back propagation method. As an example, the identifying apparatus 100 calculates a coupling coefficient that minimizes an error in output data.

  Then, the identifying apparatus 100 generates a learning device that has learned the characteristics of the predetermined classification (step S104). For example, the specifying apparatus 100 sets a calculated coupling coefficient between nodes, thereby generating a learning device that has learned features of a predetermined classification. Then, the specifying apparatus 100 stores the generated learning device (step S105). For example, the identification device 100 executes the processes of steps S101 to S105 in various classifications of products used in the online auction, and generates learning devices corresponding to the various classifications.

[5. Specific processing procedure)
Next, the procedure of the specifying process by the specifying apparatus 100 according to the embodiment will be described with reference to FIG. FIG. 6 is a flowchart illustrating a specific processing procedure performed by the specific device 100 according to the embodiment.

  As illustrated in FIG. 6, the identifying apparatus 100 receives input data (step S201). For example, the specifying apparatus 100 accepts an image in which a product that a seller sells in an online auction is drawn as input data. And when the specific apparatus 100 has not received input data (step S201; No), it waits until input data is received.

  On the other hand, when receiving the input data (step S201; Yes), the specific device 100 inputs the received input data to each learning device (step S202). For example, the identification device 100 acquires a learning device that has learned various types of characteristics from the learning device information storage unit 121. And specific device 100 inputs into each acquired learning device the picture on which goods which a seller sells for an auction were drawn, respectively.

  Thereafter, the identifying apparatus 100 calculates a reproduction error of the output data (Step S203). Specifically, the identifying apparatus 100 calculates the reproduction error of the output data output from each learning device.

  Then, the specifying device 100 specifies the classification to which the input data belongs based on the calculated reproduction error (Step S204). For example, the specifying apparatus 100 specifies the classification corresponding to the learning device that outputs the output data having the lowest reproduction error among the calculated reproduction errors as the classification to which the input data belongs.

  Then, the specifying device 100 notifies the server device 50 of the specified classification (Step S205). Thereafter, the server device 50 performs labeling of the input data based on the classification notified from the specific device 100. Thereby, the server apparatus 50 can attach a suitable label to the goods put on an auction site, for example.

[6. (Modification)
The identification device 100 according to the above-described embodiment may be implemented in various different forms other than the above-described embodiment. Therefore, in the following, another embodiment of the specific device 100 will be described.

[6-1. (Identify people's classification)
In the above-described embodiment, the identification apparatus 100 uses the learning device that reproduces the feature information of the product classification to identify the classification to which the product depicted in the image that is the input data belongs. Here, the identifying apparatus 100 may identify various classifications using a learning device that has learned various features.

  Specifically, the identifying apparatus 100 calculates a reproduction error for the human motion in the output data output when the human motion is input to the learning device that reproduces the feature information of the classification to which the human belongs, and calculates The classification corresponding to the learning device that outputs the output data having the lowest reproduction error among the reproduced reproduction errors is specified as the classification to which the person belongs.

  First, a process of generating a learning device that reproduces characteristic information of a classification to which a predetermined person belongs by the specific device 100 will be described. For example, the specific apparatus 100 first obtains information on the movement of the person at each time for each business type of the organization to which the person belongs by giving the person a pedometer, a smartphone, or the like. That is, the identifying apparatus 100 acquires a person's movement as time-series data.

  Then, the specifying apparatus 100 generates a learning device (for example, an auto encoder) that learns the characteristics of the person's movement using the acquired information about the person's movement. That is, a learning device that learns human behavior patterns is generated. For example, the identifying apparatus 100 generates a learning device that learns the characteristics of a person's movement for each type of business of the organization to which the person belongs. As an example, the identification device 100 generates a learning device that learns the characteristics of human movement for each type of business such as the restaurant industry, the manufacturing industry, and the fishery industry. Note that the identification device 100 may generate a learning device for each of various user attributes such as gender and job title, not limited to each type of business of an organization to which a person belongs.

  Next, a description will be given of a specifying process in which the specifying device 100 specifies a classification to which a person belongs using a learning device. First, the identifying apparatus 100 accepts the movement of a person to be identified as input data. Then, the identifying device 100 identifies the business type of the organization to which the person belongs using the generated learning device. For example, the identifying apparatus 100 first inputs input data to each learning device. As an example, the identifying apparatus 100 inputs a vector indicating “Mr. A's movement from 10:00 am to 11:00 am” as input data to each learning device.

  Then, the identifying apparatus 100 calculates, for each learning device that reproduces the feature information of the classification to which the person belongs, a reproduction error with respect to the input data of the output data that is output when the accepted person's movement is input. For example, the identifying apparatus 100 calculates a reproduction error based on an error between a vector represented by output data and a vector represented by input data.

  Then, the identifying apparatus 100 identifies the classification corresponding to the learning device that outputs the output data that is the lowest reproduction error among the calculated reproduction errors as the classification to which the person belongs. For example, when the learning device that outputs the output data with the lowest reproduction error corresponds to the restaurant business, the specifying device 100 specifies the classification of the person as “food restaurant”. Note that the identification device 100 may label the user based on the user's behavior, thought, attribute, and the like.

  In this way, the identification device 100 accepts a person's movement as input data. Further, the identification device 100 reproduces an error in the input data of the output data output when the movement of the person accepted by the accepting unit 131 is input to the learning device that reproduces the feature information of the classification to which the predetermined person belongs. Is calculated. In addition, the identifying apparatus 100 identifies the classification corresponding to the learning device that outputs the output data having the lowest reproduction error among the reproduction errors calculated by the calculation unit 133 as the classification to which the person belongs.

  Thereby, since the identifying apparatus 100 can identify an appropriate classification to which the input data belongs with respect to the input data, it can attach an appropriate label to the person. For example, since the specific device 100 can appropriately attach labels classified by business type or job title to a person, it can be used for data analysis in marketing or the like. As an example, the identification apparatus 100 can be used as a sample of big data, so that the accuracy of analysis can be improved.

  The specifying device 100 may specify the commonality of the user friendship characteristics. In this case, the identifying apparatus 100 generates a learning device that has learned the characteristics of friendship. For example, the specifying apparatus 100 generates a learning device that learns features for each user attribute to which a friend belongs. Thereafter, the identification device 100 inputs information about the user's friend as input data to the learning device. Subsequently, the identifying device 100 identifies the characteristics of the user's friendship based on the output data output from the learning device with respect to the input data.

[6-2. (Proposed classification)
In the above embodiment, the identifying apparatus 100 identifies and identifies the classification corresponding to the learning device that outputs the output data that is the lowest reproduction error among the reproduction errors calculated by the calculation unit 133 as the classification to which the input data belongs. An example is shown in which the classification is attached as a product label.

  Here, the identifying apparatus 100 may propose a classification to which the input data belongs according to the reproduction error. For example, the specifying apparatus 100 proposes a category in which the reproduction error calculated by the calculation unit 133 is lower than a predetermined threshold as a category to which the product belongs to a seller who sells the product in an online auction. Subsequently, the seller selects a category most suitable for the product from the proposed categories. Then, the server device 50 puts the category selected by the seller as a product label and places it in the online auction.

  As described above, the specifying apparatus 100 proposes a classification to which the input data belongs according to the reproduction error. Thereby, since the seller can select the category of the product from the mechanically specified classification candidates, the specification apparatus 100 can attach the category reflecting the seller's desire to the product. For this reason, the identification device 100 can reduce the labor for labeling by the seller, and can improve the convenience of the seller.

[6-3. (Auto encoder)
In the above-described embodiment, the identifying apparatus 100 generates a reproduction error with respect to input data of output data that is output when input data received by the receiving unit 131 is input to a learning device that reproduces feature information of a predetermined classification. An example of calculation is shown.

  Here, the identifying apparatus 100 generates a reproduction error of output data output when input data received by the receiving unit 131 is input to a learning device that reproduces feature information after reducing the number of dimensions of the feature information. It may be calculated. For example, the identifying apparatus 100 first generates an auto encoder that has the smallest number of nodes in the central layer among the layers of the node group included in the intermediate layer. That is, the identifying apparatus 100 generates an auto encoder that reduces the number of dimensions of input information of input data to the center of the intermediate layer and then restores the number of dimensions to the original. Specifically, the identifying apparatus 100 generates an auto encoder by correcting the coupling coefficient between nodes so that the output data reproduces the input data with high accuracy. Here, for example, the specifying apparatus 100 does not reproduce the image itself of the input data, but generates an auto encoder that reproduces the subject depicted in the image. As an example, the specifying apparatus 100 generates an auto encoder that reproduces a subject drawn in an image.

  Then, the identifying apparatus 100 calculates a reproduction error for the input data of the output data output when the input data received by the receiving unit 131 is input to the generated auto encoder.

  As described above, the specifying apparatus 100 outputs the input data of the output data that is output when the input data received by the receiving unit 131 is input to the learning device that reproduces the feature information after reducing the number of dimensions of the feature information. The reproduction error for is calculated.

  As a result, the identifying apparatus 100 can generate an auto encoder that reproduces the subject depicted in the image, so that the reproduction accuracy of the output data can be improved. For this reason, the specifying device 100 can improve the accuracy of specifying the classification of the input data.

[6-4. DNN]
In the above-described embodiment, the identification apparatus 100 has shown an example in which a learning device including an input layer, an intermediate layer, and an output layer is generated. Here, the identifying apparatus 100 may generate a learning device including an arbitrary number of nodes in various layers. For example, the specifying apparatus 100 generates a multistage learning device in which a plurality of node groups are included in the intermediate layer. In addition, the node group included in the learning device may be configured by an arbitrary number of nodes.

[6-5. (Applicable)
In the above-described embodiment, the identification apparatus 100 has shown an application example of image recognition that identifies the classification of products to be exhibited in the online auction. Here, the identification device 100 may be applied to various types of recognition without being limited to the classification of products. For example, the specific apparatus 100 is applied to speech recognition that targets speech. As an example, the identifying apparatus 100 generates a learning device that learns speaker classification by inputting speech data as input data. Thereby, the specifying device 100 can specify the classification such as gender, age, nationality, and the like of the speaker who uttered the voice included in the voice data.

[6-6. Device configuration〕
In the above embodiment, the specifying device 100 may be configured integrally with the server device 50. In this case, the identifying apparatus 100 labels the product based on the classification identified using the learning device. Further, the specifying device 100 may be a specifying device that performs only the specifying process by the calculating unit 133 and the specifying unit 134 without performing the generation process for generating the learning device. In this case, the specific device does not have at least the generation unit 132. Then, the specifying device specifies the classification using the learning device generated by the generating device having the generating unit 132, and notifies the server device 50 and the like of the specified classification.

[6-7. Others]
Of the processes described in the above embodiment, all or part of the processes described as being automatically performed can be performed manually, or all of the processes described as being performed manually or A part can be automatically performed by a known method. In addition, the processing procedures, specific names, and information including various data and parameters shown in the document and drawings can be arbitrarily changed unless otherwise specified. For example, the various types of information illustrated in each drawing is not limited to the illustrated information.

  Further, each component of each illustrated apparatus is functionally conceptual, and does not necessarily need to be physically configured as illustrated. In other words, the specific form of distribution / integration of each device is not limited to that shown in the figure, and all or a part thereof may be functionally or physically distributed or arbitrarily distributed in arbitrary units according to various loads or usage conditions. Can be integrated and configured. For example, the reception unit 131 and the generation unit 132 illustrated in FIG. 4 may be integrated.

[6-8. Hardware configuration)
In addition, the specifying apparatus 100 according to the above-described embodiment is realized by a computer 1000 configured as shown in FIG. 7, for example. Hereinafter, the specific device 100 will be described as an example. FIG. 7 is a hardware configuration diagram illustrating an example of a computer 1000 that implements the functions of the specific device 100. The computer 1000 includes a CPU 1100, RAM 1200, ROM 1300, HDD 1400, communication interface (I / F) 1500, input / output interface (I / F) 1600, and media interface (I / F) 1700.

  The CPU 1100 operates based on a program stored in the ROM 1300 or the HDD 1400 and controls each unit. The ROM 1300 stores a boot program executed by the CPU 1100 when the computer 1000 is started up, a program depending on the hardware of the computer 1000, and the like.

  The HDD 1400 stores programs executed by the CPU 1100, data used by the programs, and the like. The communication interface 1500 receives data from other devices via the communication network 500 and sends the data to the CPU 1100, and transmits data generated by the CPU 1100 to other devices via the communication network 500.

  The CPU 1100 controls an output device such as a display and a printer and an input device such as a keyboard and a mouse via the input / output interface 1600. The CPU 1100 acquires data from the input device via the input / output interface 1600. In addition, the CPU 1100 outputs the generated data to the output device via the input / output interface 1600.

  The media interface 1700 reads a program or data stored in the recording medium 1800 and provides it to the CPU 1100 via the RAM 1200. The CPU 1100 loads such a program from the recording medium 1800 onto the RAM 1200 via the media interface 1700, and executes the loaded program. The recording medium 1800 is, for example, an optical recording medium such as a DVD (Digital Versatile Disc) or PD (Phase change rewritable disk), a magneto-optical recording medium such as an MO (Magneto Optical disk), a tape medium, a magnetic recording medium, or a semiconductor memory. It is.

  For example, when the computer 1000 functions as the specific device 100 according to the embodiment, the CPU 1100 of the computer 1000 implements the function of the control unit 130 by executing a program loaded on the RAM 1200. The HDD 1400 stores data in the storage unit 120. The CPU 1100 of the computer 1000 reads these programs from the recording medium 1800 and executes them, but as another example, these programs may be acquired from other devices via the communication network 500.

[7. effect〕
As described above, the specifying device 100 according to the embodiment includes the receiving unit 131, the calculating unit 133, and the specifying unit 134. The accepting unit 131 accepts input data. The calculation unit 133 calculates a reproduction error of output data output when the input data received by the reception unit 131 is input to a learning device that reproduces feature information of a predetermined classification. The specifying unit 134 specifies the classification to which the input data belongs based on the reproduction error calculated by the calculating unit 133.

  In the specifying device 100 according to the embodiment, the specifying unit 134 includes a classification corresponding to a learning device that outputs output data having the lowest reproduction error among the reproduction errors calculated by the calculation unit 133. Specify as a classification.

  In the specifying device 100 according to the embodiment, the reception unit 131 receives an image in which a product is depicted as input data. The calculation unit 133 calculates a reproduction error of output data output when an image received by the reception unit 131 is input to a learning device that reproduces the feature information of the classification to which the predetermined product belongs. The specifying unit 134 specifies the category corresponding to the learning device that outputs the output data having the lowest reproduction error among the reproduction errors calculated by the calculation unit 133 as the category to which the product belongs.

  In the specifying device 100 according to the embodiment, the generation unit 132 calculates a coupling coefficient between nodes in which output data output when input data is input reproduces feature information of a predetermined classification, and the calculated coupling coefficient A learning device is generated by setting between nodes.

  Thereby, since the specific apparatus 100 which concerns on embodiment can identify the appropriate classification to which input data belongs, it can attach an appropriate label. For example, the specifying device 100 can specify the classification to which the subject drawn in the image belongs when an image in which the subject is drawn is input. For this reason, the specifying device 100 can specify the category to which the product belongs when, for example, an image in which an item newly exhibited by an exhibitor is received in an Internet auction is received.

  Moreover, since the specific apparatus 100 can attach a suitable label with respect to the goods exhibited by an internet auction, for example, the goods with an inappropriate label can be reduced. For this reason, the specific device 100 can improve the convenience of the exhibitor and the user of the auction.

  Moreover, since the identification apparatus 100 can attach an appropriate label to the product simply by inputting the image of the product to the learning device, it is possible to reduce the effort for the seller to review the label. For this reason, the specific device 100 can improve the convenience in the exhibitor's exhibition work.

  In the specifying device 100 according to the modification, the receiving unit 131 receives a human movement as input data. The calculation unit 133 calculates a reproduction error with respect to the input data of the output data output when the movement of the person received by the reception unit 131 is input to the learning device that reproduces the feature information of the classification to which the predetermined person belongs. To do. The specifying unit 134 specifies the classification corresponding to the learning device that outputs the output data having the lowest reproduction error among the reproduction errors calculated by the calculation unit 133 as the classification to which the person belongs.

  Thereby, since the specific apparatus 100 which concerns on a modification can specify the suitable classification to which input data belongs, it can attach an appropriate label to a person. For example, since the specific device 100 can appropriately attach labels classified by business type or job title to a person, it can be used for data analysis in marketing or the like. As an example, the identification apparatus 100 can be used as a sample of big data, so that the accuracy of analysis can be improved.

  Further, in the specifying device 100 according to the modification, the calculation unit 133 inputs the input data received by the reception unit 131 to the learning device that reproduces the feature information after reducing the number of dimensions of the feature information. The reproduction error for the output data to be output is calculated.

  As a result, the identification device 100 according to the modification can generate an auto encoder that reproduces the subject depicted in the image, so that the reproduction accuracy of the output data can be improved. For this reason, the specifying device 100 can improve the accuracy of specifying the classification of the input data.

  As described above, some of the embodiments of the present application have been described in detail with reference to the drawings. However, these are merely examples, and various modifications, including the aspects described in the disclosure section of the invention, based on the knowledge of those skilled in the art, It is possible to implement the present invention in other forms with improvements.

DESCRIPTION OF SYMBOLS 1 Exhibition system 10 Terminal apparatus 50 Server apparatus 100 Identification apparatus 121 Learning device information storage part 131 Reception part 132 Generation part 133 Calculation part 134 Identification part

Claims (8)

A reception unit for receiving input data;
For a learning device that reproduces feature information of a predetermined classification, a calculation unit that calculates a reproduction error of the output data output when the input data received by the receiving unit is input;
And a specifying unit that specifies a classification to which the input data belongs based on a reproduction error calculated by the calculating unit. The specific part is:
2. The specification according to claim 1, wherein a classification corresponding to a learning device that outputs output data having the lowest reproduction error among the reproduction errors calculated by the calculation unit is specified as a classification to which the input data belongs. apparatus. The reception unit
Accept an image depicting a product as the input data,
The calculation unit includes:
For a learning device that reproduces characteristic information of a classification to which a predetermined product belongs, a reproduction error between the output data output when the image received by the receiving unit is input and the image is calculated,
The specific part is:
The classification corresponding to the learning device that outputs the output data having the lowest reproduction error among the reproduction errors calculated by the calculation unit is specified as a classification to which the product belongs. Specific device. The reception unit
Accepting human movement as the input data,
The calculation unit includes:
For a learning device that reproduces feature information of a classification to which a predetermined person belongs, calculating a reproduction error for the movement of the person in the output data that is output when the movement of the person accepted by the reception unit is input,
The specific part is:
The classification corresponding to the learning device that outputs the output data having the lowest reproduction error among the reproduction errors calculated by the calculation unit is specified as the classification to which the person belongs. The specific device according to one. Generating that generates the learning device by calculating a coupling coefficient between nodes in which output data output when the input data is input reproduces characteristic information of a predetermined classification, and setting the calculated coupling coefficient between the nodes The specifying device according to claim 1, further comprising: a unit. The calculation unit includes:
The reproduction error of the output data output when the input data received by the receiving unit is input to the learning device that reproduces the feature information after reducing the number of dimensions of the feature information is calculated. The specific device according to claim 1, wherein A specific method performed by a specific device,
A reception process for receiving input data;
A calculation step for calculating a reproduction error for the input data of the output data to be output when the input data received by the reception step is input to a learning device that reproduces feature information of a predetermined classification;
A specifying step of specifying a classification to which the input data belongs based on the reproduction error calculated by the calculating step. Acceptance procedure to accept input data,
A calculation procedure for calculating a reproduction error with respect to the input data of the output data output when the input data received by the reception procedure is input to a learning device that reproduces feature information of a predetermined classification;
A specifying program for causing a computer to execute a specifying procedure for specifying a classification to which the input data belongs based on a reproduction error calculated by the calculating procedure.

Images