JP2020177513A - Information processing device, information processing method, and program - Google Patents

Abstract

To provide an information processing device, an information processing method and a program capable of improving the prediction accuracy of the price of merchandise or a service.SOLUTION: An information processing device includes an acquisition part for acquiring merchandise information on first merchandise or a first service to be sold through a network, a selection part for selecting second merchandise or a second service similar to the first merchandise or the first service from a plurality of second merchandise or second services sold in the past on the basis of the merchandise information acquired by the acquisition part, an estimation part for estimating a probability distribution with the price of the second merchandise or the second service selected by the selection part as at least one probability variable, and a prediction part for predicting the price of the first merchandise or the first service on the basis of the probability distribution estimated by the estimation part.SELECTED DRAWING: Figure 3

Description

The present invention relates to an information processing device, an information processing method, and a program.

Keywords are extracted from the product information of the products to be put up for auction via the network, and the products are referred to by the product category and the successful bid record information for each price range associated with the combination of the extracted keywords. A technique for predicting the market price of a product is known (see, for example, Patent Document 1).

JP-A-2017-116977

However, in the conventional technique, the accuracy of predicting the price of a product or service may be low.

The present invention has been made in view of the above problems, and an object of the present invention is to provide an information processing device, an information processing method, and a program capable of improving the accuracy of predicting the price of a product or service.

One aspect of the present invention has been sold in the past based on an acquisition unit that acquires product information regarding a first product or a first service sold via a network, and the product information acquired by the acquisition unit. A selection unit that selects the second product or the second service similar to the first product or the first service from a plurality of second products or the second service, and the second product selected by the selection unit. Alternatively, the price of the first product or the first service is predicted based on the estimation unit that estimates the probability distribution with the price of the second service as at least one probability variable and the probability distribution estimated by the estimation unit. It is an information processing device including a prediction unit.

According to one aspect of the present invention, the accuracy of predicting the price of goods or services can be improved.

It is a figure which shows an example of the information processing system 1 including the information processing apparatus 100 in 1st Embodiment. It is a figure which shows an example of the web page provided by the service providing apparatus 20. It is a figure which shows an example of the structure of the information processing apparatus 100 in 1st Embodiment. It is a figure which shows an example of the successful bid product information 132. It is a flowchart which shows the flow of a series of processing by the information processing apparatus 100 in 1st Embodiment. It is a figure which shows an example of the probability distribution estimated by kernel density estimation. It is a figure which shows another example of the probability distribution estimated by the kernel density estimation. It is a figure which shows an example of the structure of the information processing apparatus 100A in 2nd Embodiment. It is a flowchart which shows the flow of the series of processing of the runtime by the information processing apparatus 100A in 2nd Embodiment. It is a figure which shows an example of the prediction model 200 in 2nd Embodiment. It is a flowchart which shows the flow of a series of processing of the training by the information processing apparatus 100A in 2nd Embodiment. It is a figure which shows another example of the prediction model 200 in 2nd Embodiment. It is a figure which shows an example of the hardware composition of the information processing apparatus 100 of embodiment.

Hereinafter, an information processing apparatus, an information processing method, and a program to which the present invention is applied will be described with reference to the drawings.

[Overview]
The information processing device is realized by one or more processors. The information processing device acquires product information related to the first product or the first service sold via the network, and based on the acquired product information, among a plurality of second products or the second services sold in the past. Select a second product or second service that is similar to the first product or first service. The information processing device estimates a probability distribution with the price of the selected second product or second service as at least one random variable, and predicts the price of the first product or first service based on the estimated probability distribution. .. As a result, the accuracy of predicting the price of goods or services can be improved. Hereinafter, the goods or services will be collectively referred to as "goods".

<First Embodiment>
[overall structure]
FIG. 1 is a diagram showing an example of an information processing system 1 including an information processing device 100 according to the first embodiment. The information processing system 1 in the first embodiment includes, for example, one or more terminal devices 10, a service providing device 20, and an information processing device 100. These devices are connected via the network NW. Further, a part of these devices may be included in another device as a virtual device. For example, a part or all of the functions of the service providing device 20 are realized by the functions of the information processing device 100. It may be a virtual machine.

Each device shown in FIG. 1 transmits and receives various information via the network NW. The network NW includes, for example, the Internet, WAN (Wide Area Network), LAN (Local Area Network), provider terminal, wireless communication network, wireless base station, dedicated line, and the like. It should be noted that not all combinations of the devices shown in FIG. 1 need to be able to communicate with each other, and the network NW may partially include a local network.

The terminal device 10 is a terminal device including an input device, a display device, a communication device, a storage device, and an arithmetic device, such as a mobile phone such as a smartphone, a tablet terminal, and various personal computers. The communication device includes a network card such as a NIC (Network Interface Card), a wireless communication module, and the like. In the terminal device 10, a UA (User Agent) such as a web browser or an application program is activated, and a request according to the content input by the user is transmitted to the service providing device 20. Further, the terminal device 10 in which the UA is activated causes the display device to display various information based on the information acquired from the service providing device 20.

The service providing device 20 is, for example, a web server that provides a web page to the terminal device 10 in response to a request from a web browser started as a UA. The web page may be, for example, a web page constituting a website such as an auction site or a free market site where products are bought and sold in an auction format on the Internet, or a shopping site where a plurality of stores are opened. Further, the service providing device 20 may provide the terminal device 10 with a web page that provides various services such as a search site, an SNS (Social Networking Service), and a mail service. Further, the service providing device 20 is an application that provides the same service as various websites such as an auction site and a flea market site by providing the content to the terminal device 10 in response to a request from an application started as a UA. It may be a server. Hereinafter, as an example, the service provided by the service providing device 20 will be described as an “auction site”, but this auction site may be read as another service such as a flea market site or a shopping site. In this case, the listing shall be read as sales, and the successful bid shall be read as purchase.

FIG. 2 is a diagram showing an example of a web page provided by the service providing device 20. In the illustrated example, a web page (hereinafter referred to as an input page) for inputting various information when listing a product on an auction site is schematically shown.

For example, on the input page, an input field F1 for inputting the title of the product to be exhibited, an input field F2 for inputting the image of the product, and a button B1 for selecting the product category from a plurality of prepared categories. Button B2 to select whether the item is used or new, button B3 to select whether the seller or the winning bidder will bear the shipping cost for delivering the item to the winning bidder, the price of the item (for example, prompt decision) It includes components such as a display field F3 that displays prices, etc.).

For example, when the seller inputs a title in the input field F of the input page or operates the button B1 to select a category, the provisional product price is automatically displayed in the display field F3. .. Details of the automatic display of this price will be described later.

The information processing device 100 acquires, for example, product information of a product to be exhibited (sold) on the auction site, and based on the acquired product information, from among a plurality of products already sold (purchased) on the auction site. Select one or more items that are similar to the item being listed. The information processing device 100 estimates a probability distribution with the price of the selected product as at least one random variable, and predicts the price of the product to be exhibited based on the estimated probability distribution. The product information includes, for example, the title of the product, the image of the product, the category of the product, the state of the product indicating whether or not it is used, and the shipping charge information indicating whether or not the shipping fee is borne by the seller. Various information such as is included.

[Information processing device configuration]
FIG. 3 is a diagram showing an example of the configuration of the information processing apparatus 100 according to the first embodiment. As shown in the figure, the information processing device 100 includes, for example, a communication unit 102, a control unit 110, and a storage unit 130.

The communication unit 102 includes, for example, a communication interface such as a NIC. The communication unit 102 communicates with the terminal device 10, the service providing device 20, and the like via the network NW.

The control unit 110 includes, for example, an acquisition unit 112, a selection unit 114, an estimation unit 116, a prediction unit 118, and a communication control unit 120.

The components of the control unit 110 are realized, for example, by a processor such as a CPU (Central Processing Unit) executing a program stored in the storage unit 130. In addition, some or all of the constituent elements of the control unit 110 are hardware such as LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), or GPU (Graphics Processing Unit). It may be realized by (circuit unit; circuitry), or it may be realized by the cooperation of software and hardware.

The storage unit 130 is realized by, for example, a storage device such as an HDD (Hard Disc Drive), a flash memory, an EEPROM (Electrically Erasable Programmable Read Only Memory), a ROM (Read Only Memory), and a RAM (Random Access Memory). In addition to various programs such as firmware and application programs, the storage unit 130 stores successful bid product information 132.

FIG. 4 is a diagram showing an example of successful bidder product information 132. As shown in the figure, the successful bid product information 132 includes a product title, a category, an image, and a product ID with respect to a product ID indicating identification information of a product sold on the auction site (hereinafter referred to as a successful bid product). It is data in which the state of the product, whether or not the shipping fee is borne, and the price actually sold are associated with each other.

[Processing flow of information processing device]
Hereinafter, the processing contents of each component of the control unit 110 will be described with reference to the flowchart. FIG. 5 is a flowchart showing a flow of a series of processes by the information processing apparatus 100 according to the first embodiment. The processing of this flowchart may be repeated, for example, at a predetermined cycle.

First, the acquisition unit 112 acquires various information input to the input page of the auction site from the service providing device 20 as product information related to the product to be exhibited (S100) via the communication unit 102. The product to be exhibited is an example of "first product or first service".

Next, the selection unit 114 may use one or a plurality of products similar to the product to be exhibited from among the plurality of successful bid products included in the successful bid product information 132 based on the product information acquired by the acquisition unit 112. Select a successful bid product (S102). A successful bid product is an example of a "second product or second service".

For example, the selection unit 114 has a successful bid product with a product ID in which a title and category that are semantically close to the title and category included in the product information acquired by the acquisition unit 112 are associated with the successful bid product information 132. Is selected as a product similar to the product to be listed. Semantically close means, for example, that when titles and categories are converted into multidimensional vectors by distributed representation, the cosine similarity of those vectors is high. That is, the selection unit 114 converts the title and category included in the product information acquired by the acquisition unit 112 into a vector by the distributed representation, and selects a successful bid product whose similarity with the vector is equal to or higher than the threshold value. When the number of successful bid products whose similarity is equal to or higher than the threshold value is large, the selection unit 114 may select a predetermined number of successful bid products having a high similarity.

Next, the estimation unit 116 estimates a probability distribution using at least one random variable as the price associated with the product ID of the winning bidder selected by the selection unit 114 (S104).

For example, the estimation unit 116 estimates the probability distribution based on the kernel density estimation (KDE). Kernel density estimation is a method of estimating the entire probability distribution (distribution expressed by a probability density function) including the sample data from a finite sample data. It is known that kernel density estimation tends to require less time for calculation than other fitting methods such as linear regression, and it is easy to improve the real-time property of prediction.

FIG. 6 is a diagram showing an example of a probability distribution estimated by kernel density estimation. The horizontal axis represents the winning bid of the successful bid product as a random variable, and the vertical axis represents the probability. Since the probability distribution is obtained by kernel density estimation, the vertical axis may be read as the number of successful bidders at that price, that is, the frequency.

In the illustrated example, the probability distribution D1 is estimated from the prices of the top 200 successful products with the highest similarity among the plurality of successful products similar to the products to be exhibited, and the top 100 successful products with the highest similarity are estimated. The probability distribution D2 is estimated from the prices of the finished products, the probability distribution D3 is estimated from the prices of the top 50 successful products with high similarity, and the probability distribution D4 is estimated from the prices of the top 20 successful products with high similarity. Is estimated. Both probability distributions are multimodal. That is, both probability distributions are multimodal distributions. Multimodality means that there are a plurality of modes (also referred to as modes) that appear most frequently in the probability distribution. The mode value in the present embodiment may be a value that frequently appears in the entire area of the probability distribution, or may be a value that frequently appears in each local region of the probability distribution. That is, when comparing the modes, one mode may be larger than the other mode.

Next, the prediction unit 118 determines whether or not the probability distribution estimated by the estimation unit 116 is a multimodal distribution (S106).

When the probability distribution is a multimodal distribution, the communication control unit 120 requests the service providing device 20 to re-input the product information on the input page via the communication unit 102 (hereinafter, a re-input request). To be referred to) (S108).

Upon receiving the re-entry request, the service providing device 20 displays at least one (preferably both) of the title and category of the product to be exhibited on the input page displayed on the screen of the terminal device 10 used by the seller. Display characters and images that prompt you to re-enter. The combination of the communication unit 102, the communication control unit 120, and the service providing device 20 is an example of the "providing unit".

The fact that the probability distribution is a multimodal distribution means that there is a high probability that the winning bidders selected as similar to the item to be listed include items that are not actually similar. Shown. For example, the random variables that take the mode in the probability distribution D1 that is a multimodal distribution are a winning bid of about 2500 yen and a winning bid of about 8000 yen. In this case, it can be determined that the successful bid products similar to the products to be exhibited include products with a selling price of about 2,500 yen and products with a selling price of about 8,000 yen. it can. Usually, when prices are far apart, they are often different products. For example, when the seller sells the main body of the smartphone, if the model name of the smartphone is "ABCDE", the title of the accessory (for example, a case) that can be attached to the smartphone with the model name "ABCDE". Can also contain the string "ABCDE". Therefore, it is easy to treat different smartphones and accessories as similar products, but the market prices of these products are far apart, and it is possible to reselect similar products for the products to be exhibited from among the successful bid products. preferable.

On the other hand, in the present embodiment, when the probability distribution is a multimodal distribution, the title of the product to be exhibited can be re-entered in the terminal device 10 used by the seller, or the category of the product to be exhibited can be re-entered. In order to display an input page prompting the user to reselect to a more detailed category, the process of S102 can be performed again. That is, the conditions can be changed to reselect successful bid products that are similar to the products to be listed, and as a result, the probability distribution can be re-estimated. Thereby, for example, when the seller sells the main body of the smartphone, the probability distribution can be estimated by excluding products such as accessories of the smartphone and using only the winning bid price of only the smartphone main body as a random variable.

On the other hand, if the probability distribution estimated by the estimation unit 116 is not a multimodal distribution but a monomodal distribution with one mode, the prediction unit 118 will exhibit the product based on the probability distribution. Predict the winning price of the product (S110). For example, the prediction unit 118 predicts the price that takes the mode in the probability distribution, which is a monomodal distribution, as the winning bid of the product to be exhibited.

Next, the communication control unit 120 transmits information on the winning bid price predicted by the prediction unit 118 to the service providing device 20 via the communication unit 102 (S112). In response to this, the service providing device 20 causes the winning bid price of the product to be exhibited to be displayed in the display field F3 on the input page displayed on the screen of the terminal device 10 used by the seller. As a result, the seller can know at the time of listing what price the seller sells is most likely to be awarded, that is, the appropriate price for the successful bid. As a result, it is possible to improve the convenience when the seller uses the auction site.

According to the first embodiment described above, the information processing apparatus 100 acquires product information regarding products sold via the network NW, and based on the acquired product information, a plurality of sold products sold in the past have been sold. From the products, select a sold product that is similar to the product to be sold. The information processing device 100 estimates a probability distribution with the price of the selected sold product as at least one random variable, and predicts the price of the product to be sold based on the estimated probability distribution. As a result, the accuracy of predicting the price of goods or services can be improved.

<Modified example of the first embodiment>
Hereinafter, a modified example of the first embodiment will be described. In the first embodiment described above, the estimation unit 116 has been described as estimating the probability distribution using the kernel density estimation with the price of the winning bidder selected by the selection unit 114 as at least one random variable. Not limited to this. For example, in addition to the price of the successful bid product, the estimation unit 116 further uses the similarity (also referred to as relevancy) with the product to be exhibited, which the selection unit 114 considers when selecting the successful bid product, as a random variable. The distribution may be estimated. That is, the estimation unit 116 may estimate a multidimensional (two-dimensional or higher) probability distribution using a plurality of random variables.

FIG. 7 is a diagram showing another example of the probability distribution estimated by kernel density estimation. The x-axis represents the winning price of the winning product as a random variable, the y-axis represents the similarity between the winning product and the product to be exhibited as a random variable, and the z-axis is the probability or It represents the frequency. In the illustrated example, a three-dimensional probability distribution is estimated.

For example, the prediction unit 118 determines whether or not the three-dimensional probability distribution is a multimodal distribution, and when the three-dimensional probability distribution is a monomodal distribution, takes the mode in the probability distribution. Predict the price as the winning price of the item to be listed. For example, even if the products are different from each other, if the winning bids are close to each other, the products are likely to be treated as the same product. In such a case, as shown in the figure, when the probability distribution is viewed on the x-axis, the probability distribution is a monomodal distribution, but when viewed on the y-axis, the probability distribution is a multimodal distribution.

For example, if the winning bid for the smartphone itself is 50,000 yen and the air purifier is sold for the same 50,000 yen, it is not possible to distinguish between the same product and different products based on the winning bid alone, and the probability distribution. Above, it can be treated as the price of the same item. Therefore, by estimating a three-dimensional probability distribution in consideration of a new random variable called similarity in addition to price, when selecting similar products, it is semantically related to the exhibited products from the population. You can exclude distant items. As a result, the accuracy of predicting the price of goods or services can be further improved.

In addition to the title and category, the selection unit 114 also uses a product code such as a JAN (Japanese article number) code or an EAN (European article number) to select a successful bid product similar to the product to be exhibited. You may. As a result, the accuracy of predicting the price of goods or services can be further improved.

Further, in the first embodiment described above, when the seller or the seller puts up (sells) a product on an auction site, a free market site, a shopping site, or the like, the information processing device 100 predicts an appropriate price of the product. I explained it as something to do, but it is not limited to this. For example, the information processing device 100 may predict an appropriate price of a product when a successful bidder (purchaser) makes a successful bid (purchase) for the product at an auction site, a free market site, a shopping site, or the like.

Further, for example, in the information processing device 100, in addition to information such as a title and a category, information such as the state of the product (used or new) and whether the shipping fee is borne by the seller or the winning bidder is used as a random variable. The probability distribution may be estimated.

<Second Embodiment>
Hereinafter, the second embodiment will be described. In the above-described first embodiment, the probability distribution is estimated with the price of the product purchased in the past as a random variable, and the price of the product is predicted based on the estimated probability distribution. On the other hand, the second embodiment is different from the above-described first embodiment in that the price of the product is predicted by using supervised learning. Hereinafter, the differences from the first embodiment will be mainly described, and the points common to the first embodiment will be omitted. In the description of the second embodiment, the same parts as those of the first embodiment will be described with the same reference numerals.

FIG. 8 is a diagram showing an example of the configuration of the information processing apparatus 100A according to the second embodiment. The control unit 110A of the information processing device 100A according to the second embodiment is a learning unit in addition to the acquisition unit 112, the selection unit 114, the estimation unit 116, the prediction unit 118, and the communication control unit 120 described above. It includes 122. Further, in the storage unit 130A of the information processing apparatus 100A in the second embodiment, model information 134 is stored in addition to various programs such as firmware and application programs and successful product information 132. The successful bidder product information 132 is used as teacher data for the prediction model 200 indicated by the model information 134. Model information 134 will be described later.

[Runtime processing flow]
Hereinafter, a series of run-time processing flows by the information processing apparatus 100A in the second embodiment will be described according to a flowchart. The runtime is a state in which the predicted model 200 that has already been learned is used. FIG. 9 is a flowchart showing a series of run-time processing flows by the information processing apparatus 100A according to the second embodiment. The processing of this flowchart may be repeated, for example, at a predetermined cycle.

First, the acquisition unit 112 acquires various information input to the input page of the auction site from the service providing device 20 as product information related to the product to be exhibited (S200) via the communication unit 102.

Next, the prediction unit 118 inputs the title and category included in the product information acquired by the acquisition unit 112 into the prediction model 200 indicated by the model information 134 (S202).

The model information 134 is information (program or data structure) that defines the prediction model 200. The prediction model 200 is realized by various neural networks such as, for example, a convolutional neural network (CNN).

In the model information 134, for example, the connection of how the neurons (units) contained in each of the input layer, one or more hidden layers (intermediate layers), and the output layers constituting each neural network are connected to each other. It includes various information such as information and the connection coefficient given to data input / output between connected neurons. The connection information includes, for example, the number of neurons contained in each layer, information that specifies the type of neuron to which each neuron is connected, an activation function that realizes each neuron, a gate provided between neurons in the hidden layer, and the like. Contains information. The activation function that realizes the neuron may be, for example, a rectified linear function (ReLU function), a sigmoid function, a step function, or other function. The gate selectively passes or weights the data transmitted between neurons, for example, depending on the value returned by the activation function (eg 1 or 0). The coupling coefficient is a parameter of the activation function. For example, in the hidden layer of a neural network, when data is output from a neuron in one layer to a neuron in a deeper layer, the weight given to the output data is given. including. Further, the coupling coefficient may include a bias component peculiar to each layer.

FIG. 10 is a diagram showing an example of the prediction model 200 in the second embodiment. The prediction model 200 in the second embodiment includes, for example, a first input layer 202, an embedded layer (Embedding) 204, a data deformation layer (Reshape) 206, a plurality of convolution layers 208, and a plurality of pooling layers 210. , Concatenate 212, data transformation layer 214, second input layer 216, embedding layer 218, data transformation layer 220, connection layer 222, fully connected layer 224, and normalization processing layer 226. , The dropout layer 228 and the output layer 230 are included.

Each character included in the character string indicating the title is input to the first input layer 202. The characters input to the first input layer 202 are output to the embedded layer 204.

The embedded layer 204 converts each character input from the first input layer 202 into a multidimensional vector (dense vector) having a fixed number of dimensions, and creates a tensor (for example, a second-order tensor) in which a plurality of converted vectors are arranged. Generate. The embedded layer 204 outputs the generated tensor to the data transformation layer 206.

The data transformation layer 206 transforms the input tensor (including the vector which is the first-order tensor) and outputs the tensor to each of the plurality of convolutional layers 208.

Each convolutional layer 208 convolves the tensor input from the data transformation layer 206 with a weight matrix called a kernel (obtains the inner product), and compresses an area including a plurality of elements overlapping the kernel into one unit area. A univariate value obtained by convolution is associated with this unit area as a feature quantity. Each convolutional layer 208 repeats the convolutional process while shifting the kernel on the tensor to generate a tensor whose element value is the convolutional feature called a convolutional layer formed by a plurality of unit areas. Output to the pooling layer 210.

The pooling layer 210 compresses the number of dimensions of the tensor input from the convolutional layer 208 by pooling (for example, Global Average Pooling) or the like, and outputs the tensor to the connecting layer 212.

The connecting layer 212 connects the tensors input from each of the plurality of pooling layers 210 into one, and outputs the tensors to the data transformation layer 214. It is assumed that the number of dimensions of the tensor input from each of the plurality of pooling layers 210 is the same.

Similar to the data transformation layer 206, the data transformation layer 214 transforms the input tensor and outputs the tensor to the connecting layer 222.

In the second input layer 216, each character included in the ID (Identification Data) for identifying the title and each character included in the character string indicating the title are input. The characters input to the second input layer 216 are output to the embedded layer 218.

Similar to the embedded layer 204, the embedded layer 218 converts each character input from the second input layer 216 into a multidimensional vector (dense vector) having a fixed number of dimensions, and arranges a plurality of converted vectors. Generate a tensor (eg a second-order tensor). The embedded layer 218 outputs the generated tensor to the data transformation layer 220.

Similar to the data transformation layer 206, the data transformation layer 220 transforms the input tensor and outputs the tensor to the connecting layer 222.

The connecting layer 222 connects the tensor input from the data transformation layer 214, that is, the title-based tensor, and the tensor input from the data transformation layer 220, that is, the category-based tensor into one. Is output to the fully connected layer 224. A title-based tensor is an example of a "first feature", and a category-based tensor is an example of a "second feature". A tensor that connects a title-based tensor and a category-based tensor is an example of a "third feature".

The fully connected layer 224 adds the bias component determined by learning to the value returned by the activation function (for example, the ReLU function) when the element of the tensor input from the connecting layer 222 is used as a variable. Generate a tensor replaced with the specified value. The fully connected layer 224 multiplies the tensor by the kernel determined by learning when changing the element value of the tensor. The fully connected layer 224 outputs the generated tensor to the normalization processing layer 226.

The normalization processing layer 226 normalizes the element values of the tensor input from the fully connected layer 224, and outputs the normalized tensor to the dropout layer 228.

In order to prevent overfitting, the dropout layer 228 thins out some elements of the tensor input from the normalization processing layer 226 and outputs them to the output layer 230.

The output layer 230 outputs the tensor input from the dropout layer 228. The tensor output by the output layer 230 is a 0th-order tensor, which is a one-dimensional scalar value. That is, the output layer 230 outputs a numerical value representing the winning bid or the purchase price of the product.

Returning to the description of the flowchart of FIG. Next, the communication control unit 120 acquires the estimated price of the product, which is the output result, from the prediction model 200, and transmits the information indicating the acquired estimated price to the service providing device 20 via the communication unit 102 (S204). ). In response to this, the service providing device 20 causes the winning bid price of the product to be exhibited to be displayed in the display field F3 on the input page displayed on the screen of the terminal device 10 used by the seller. As a result, the seller can know at the time of listing what price the seller sells is most likely to be awarded, that is, the appropriate price for the successful bid. As a result, it is possible to improve the convenience when the seller uses the auction site.

[Training processing flow]
Hereinafter, a series of processing flows of training of the information processing apparatus 100 in the first embodiment will be described according to a flowchart. Training is a state in which the prediction model 200 used at runtime is trained. FIG. 11 is a flowchart showing a series of processing flows of training by the information processing apparatus 100A in the second embodiment.

First, the learning unit 122 acquires the title and category associated with the product ID of each successful bid product from the successful bid product information 132 stored in the storage unit 130A in order to train the prediction model 200. The selected title and category are input to the prediction model 200 (S300).

Specifically, the learning unit 122 inputs each character included in the character string indicating the title of the winning product into the first input layer 202, and each character included in the ID for identifying the title of the winning product, or Each character included in the character string indicating the title is input to the second input layer 216.

In general, as a word embedding expression (Word Embedding (s)), a word or phrase, which is a block of a plurality of characters, is often input to the embedding layer of a neural network. In this case, when words or phrases that are not used for learning are input at runtime, the accuracy of price prediction tends to decrease. On the other hand, in the present embodiment, since the input to the prediction model 200 is made in character units instead of word units or phrase units, all characters estimated to be input at runtime are learned evenly. Can be done. As a result, the accuracy of price prediction can be improved.

Next, the learning unit 122 acquires the output result, that is, the predicted price of the successful bid product from the prediction model 200 in which the title and category of the successful bid product are input (S302).

Next, the learning unit 122 calculates an error between the predicted price acquired from the predicted model 200 and the actual winning bid price associated with the title and category input to the predicted model 200 as a teacher label (S304). For example, the learning unit 122 may calculate the average absolute error between the predicted price and the actual winning bid.

Next, the learning unit 122 determines whether or not the calculated error is within the threshold value (S306), and if the error exceeds the threshold value, the prediction model 200 is based on a gradient method such as an error reverse transmission number. Learn the parameters (S308). This completes the processing of this flowchart.

According to the second embodiment described above, in the information processing apparatus 100A, the successful product information 132 in which the product information (including the title and the category) related to the successful product is associated with the actual successful price of the successful product. The prediction model 200 is learned based on (an example of teacher data), and when a seller puts up a product on an auction site or the like, the product information about the product to be put up is transferred to the learned prediction model 200. By inputting, the winning price of the item to be exhibited is predicted. As a result, the accuracy of predicting the price of the goods or services can be improved as in the first embodiment.

<Modified example of the second embodiment>
Hereinafter, a modified example of the second embodiment will be described. In the second embodiment described above, the information processing apparatus 100A predicts the price output by the prediction model 200 as the appropriate winning bid price of the product to be exhibited at the runtime, but is not limited to this.

For example, the information processing apparatus 100A estimates the price output by the prediction model 200 and the probability distribution using the price of the sold product as at least one probability variable at the runtime, and the price predicted based on the estimated probability distribution ( A statistically calculated value such as an average with the price predicted by the method of the first embodiment may be predicted as an appropriate winning price of the product to be exhibited.

Further, for example, the information processing device 100A predicts, in addition to product information such as title and category, a calculation result of kernel density estimation and a character string such as a product outline and a product description of a successful product similar to the product to be exhibited. By inputting to the model 200, it may be predicted as an appropriate winning price of the product to be exhibited.

FIG. 12 is a diagram showing another example of the prediction model 200 in the second embodiment. In addition to the above-described configuration, the prediction model 200 further includes a third input layer 232, an embedded layer 234, and a data deformation layer 236.

In the third input layer 232, for example, each character of a character string representing a product outline and a product description of a successful product is input, or a numerical value indicating a winning price obtained from a probability distribution estimated by kernel density estimation is input. It is done. The characters input to the third input layer 232 are output to the embedded layer 234.

Similar to the embedded layer 204, the embedded layer 234 converts each character input from the third input layer 232 into a multidimensional vector (dense vector) having a fixed number of dimensions, and arranges a plurality of converted vectors. Generate a tensor. The embedded layer 234 outputs the generated tensor to the data transformation layer 236.

Similar to the data transformation layer 206, the data transformation layer 236 transforms the input tensor and outputs the tensor to the connecting layer 222.

The connecting layer 222 is input from the data transformation layer 214, that is, the title-based tensor, the data transformation layer 220, the category-based tensor, and the data transformation layer 236. The tensor, that is, the tensor based on the product outline, the product description, the estimated price by estimating the kernel density, etc., is concatenated into one, and it is output to the fully connected layer 224. In this way, by learning the prediction model 200 using not only the title and category but also other information such as the product outline, product description, and predicted price by kernel density estimation, the prediction accuracy of the price of the product or service can be further improved. Can be improved.

Further, in the second embodiment described above, in order to predict the appropriate winning bid of the product to be exhibited, the title and category are input to the prediction model 200, that is, the title of the product to be exhibited by the seller on the input page. I explained that it is necessary to enter and the category as product information, but it is not limited to this. For example, the information processing apparatus 100A may predict an appropriate winning bid for the product from the title even when the seller does not select a category and the category of the product to be exhibited is unknown.

For example, the estimation unit 116 uses kernel density estimation to estimate a probability distribution using the ID of the category of the successful bidder product sold in the past as a random variable. When the probability distribution estimated by the estimation unit 116 is a monomodal distribution, the prediction unit 118 predicts the category ID that takes the mode in the probability distribution as the category of the product to be exhibited.

Then, the prediction unit 118 predicts the winning bid price of the product to be exhibited by inputting the title input by the seller and the category predicted from the probability distribution into the trained prediction model 200. In this way, if at least a title is input, the category can be predicted from the title, and the winning bid of the product to be exhibited can be predicted.

<Hardware configuration>
The information processing device 100 of the above-described embodiment is realized by, for example, a hardware configuration as shown in FIG. FIG. 13 is a diagram showing an example of the hardware configuration of the information processing apparatus 100 of the embodiment.

The information processing device 100 includes NIC100-1, CPU100-2, RAM100-3, ROM100-4, secondary storage devices 100-5 such as flash memory and HDD, and drive device 100-6, which are internal buses or dedicated communication lines. It is configured to be interconnected by. A portable storage medium such as an optical disk is mounted on the drive device 100-6. A program stored in a portable storage medium mounted on the secondary storage device 100-5 or the drive device 100-6 is expanded into the RAM 100-3 by a DMA controller (not shown) or the like, and executed by the CPU 100-2. As a result, the control unit 110 is realized. The program referred to by the control unit 110 may be downloaded from another device via the network NW.

Although the embodiments for carrying out the present invention have been described above using the embodiments, the present invention is not limited to these embodiments, and various modifications and substitutions are made without departing from the gist of the present invention. Can be added.

1 ... Information processing system, 10 ... Terminal device, 20 ... Service providing device, 100 ... Information processing device, 102 ... Communication unit, 110 ... Control unit, 112 ... Acquisition unit, 114 ... Selection unit, 116 ... Estimating unit, 118 ... Prediction unit, 120 ... Communication control unit, 122 ... Learning unit, 130 ... Storage unit, 200 ... Prediction model

Claims (14)

The acquisition department that acquires product information related to the first product or first service sold via the network,
Based on the product information acquired by the acquisition unit, the second product or the second service similar to the first product or the first service is selected from the plurality of second products or the second services sold in the past. A selection section for selecting services and
An estimation unit that estimates a probability distribution with the price of the second product or service selected by the selection unit as at least one random variable.
A prediction unit that predicts the price of the first product or the first service based on the probability distribution estimated by the estimation unit.
Information processing device equipped with. The prediction unit predicts the price that takes the mode as the random variable in the probability distribution as the price of the first product or the first service.
The information processing device according to claim 1. The terminal device available to the seller of the first product or the first service is further provided with a providing unit that provides a screen on which the product information can be input.
When the probability distribution has multimodality, the providing unit provides the terminal device with a screen prompting the seller to re-enter the product information.
The information processing device according to claim 1 or 2. The estimation unit estimates the probability distribution based on the kernel density estimation.
The information processing device according to any one of claims 1 to 3. The estimation unit determines the price of the second product or the second service selected by the selection unit and the similarity between the first product or the first service and the second product or the second service. Estimate the probability distribution as the random variable,
The information processing device according to any one of claims 1 to 4. The prediction unit acquires the model learned based on the teacher data in which the price of the second product or the second service is associated with the product information related to the second product or the second service by the acquisition unit. The price of the first product or the first service is predicted based on the output result of the model obtained when the product information of the first product or the first service is input.
The information processing device according to any one of claims 1 to 5. The product information includes at least a character string indicating a title.
The model further includes a learning unit that learns to convert each character included in the character string indicating the title into a multidimensional first feature and output a price based on the first feature.
The information processing device according to claim 6. The product information further includes a character string indicating a category.
The learning unit converts each character included in the character string indicating the category into a multidimensional second feature, and converts the first feature and the second feature into a third feature by combining the first feature and the second feature. , Learn to output a price based on the third feature,
The information processing device according to claim 7. The prediction unit predicts the category of the second product or the second service from the title included in the product information of the second product or the second service.
The learning unit converts each character included in the character string indicating the category predicted by the prediction unit into a multidimensional second feature, and combines the first feature and the second feature. To convert to the third feature and learn to output the price based on the third feature.
The information processing device according to claim 8. The acquisition department that acquires product information related to the first product or first service sold via the network,
The acquisition unit acquires the model learned based on the teacher data in which the price of the second product or the second service is associated with the product information related to the second product or the second service sold in the past. A prediction unit that predicts the price of the first product or the first service based on the output result of the model obtained when the product information of the first product or the first service is input.
Information processing device equipped with. The computer
Acquire product information about the first product or first service sold via the network,
Based on the acquired product information, the second product or second service similar to the first product or first service is selected from a plurality of second products or second services sold in the past.
Estimate the probability distribution with the price of the selected second product or second service as at least one random variable.
Predicting the price of the first product or service based on the estimated probability distribution.
Information processing method. The computer
Acquire product information about the first product or first service sold via the network,
The first acquired model with respect to a model learned based on teacher data in which the prices of the second product or the second service are associated with the product information related to the second product or the second service sold in the past. The price of the first product or the first service is predicted based on the output result of the model obtained when the product information of the product or the first service is input.
Information processing method. On the computer
The process of acquiring product information about the first product or the first service sold via the network,
A process of selecting the second product or the second service similar to the first product or the first service from a plurality of second products or the second services sold in the past based on the acquired product information. When,
The process of estimating the probability distribution with the price of the selected second product or second service as at least one random variable, and
The process of predicting the price of the first product or the first service based on the estimated probability distribution, and
A program to execute. On the computer
The process of acquiring product information about the first product or the first service sold via the network,
The first acquired model with respect to a model learned based on teacher data in which the prices of the second product or the second service are associated with the product information related to the second product or the second service sold in the past. A process of predicting the price of the first product or the first service based on the output result of the model obtained when the product information of the product or the first service is input.
A program to execute.