JP6927899B2 - Predictors, predictors, and programs - Google Patents

Description

The present invention relates to predictors, predictors, and programs.

Conventionally, keywords are extracted from the product information of the products to be put up for auction via the network, and the successful bid record information for each price range associated with the combination of the category and the keyword is referred to to correspond to the product information. A predictor that predicts the market price of a product to be sold is known (see Patent Document 1).

JP-A-2017-116977

"Extension of k-means algorithm that automatically determines the number of clusters", Tsunenori Ishioka, Applied Statistics Vol. 29, No. 3 (2000), p141-149 "Mean Shift: A Robust Approach Toward Feature Space Analysis". Dorin Comaniciu and Peter Meer, IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL.24, NO.5, MAY2002

However, in the device described in Patent Document 1, the distribution of the winning bid amount for each price range is provided as it is, or the average of the maximum value and the minimum value of the price range, the median value of the price range, and the like are provided. In some cases, the information provided was inaccurate.

The present invention has been made in consideration of such circumstances, and one of the objects of the present invention is to provide a prediction device, a prediction method, and a program capable of predicting the winning bid amount more accurately.

In one aspect of the present invention, the winning bid amount corresponds to the acquisition unit for acquiring information on the products or services (hereinafter, products, etc.) to be put up for auction and the information on the products, etc. related to the products, etc. that were sold in the auction. With reference to the attached successful bid record information, the distribution of the successful bid amount that matches the information on the products, etc. acquired by the acquisition unit is obtained, and the successful bid amount corresponding to the peak in the distribution is the product, etc. acquired by the acquisition unit. It is a prediction device including a prediction unit for deriving as a successful bid prediction amount.

According to one aspect of the present invention, the winning bid amount can be predicted more accurately.

It is a figure which shows an example of the structure and use environment of the prediction apparatus 200. It is a figure which shows an example of the contents of the product etc. DB172. It is a figure which shows an example of the contents of an inverted index 174. It is a figure which shows an example of the contents of the successful bid record information 176. It is a figure which shows an example of the interface screen IM1 provided to the terminal apparatus 10 of the seller. It is a flowchart which shows an example of the flow of processing executed by the prediction unit 220. It is a figure which shows typically the result of having quantized the winning bid amount. It is a figure for demonstrating the process from clustering to derivation of a forecast price range. It is a figure which shows an example of the derivation method of the predicted price range in S122.

Hereinafter, embodiments of the prediction device, prediction method, and program of the present invention will be described with reference to the drawings.

[Overview]
The predictor is realized by one or more processors. For example, the prediction device acquires information (product etc. information) of a product or service (hereinafter, product etc.) from a seller who is going to put up an item in an auction (auction) conducted via a network, and refers to the successful bid record information. It is a device that predicts the winning bid amount. As a result, the seller can know about how much the product will be sold before listing, so that the prediction device motivates the seller to participate in the auction and activates economic activities. Can contribute. The prediction device may be a device included in an auction server or the like that manages the auction. That is, the prediction device may be a virtual device.

[overall structure]
FIG. 1 is a diagram showing an example of the configuration and usage environment of the prediction device 200. In this usage environment, devices such as the terminal device 10, the auction server 100, and the prediction device 200 communicate via the network NW. These devices include an interface for connecting to the network NW. The network NW includes the Internet, WAN (Wide Area Network), LAN (Local Area Network) and the like.

[Terminal device]
The terminal device 10 is used by auction sellers and bidders (including users who are about to sell or bid). The terminal device 10 is, for example, a mobile phone such as a smartphone, a personal computer, a tablet terminal, or the like. In the terminal device 10, a UA (User Agent) such as an application program or a browser is activated to request an image, send a query, and perform other processing to the auction server 100.

From the terminal device 10 used by the seller, data regarding the products to be sold and the like is transmitted to the auction server 100. From the terminal device 500 used by the bidder, a query for the bidder to search for a product of interest, information indicating the intention to bid, a bid amount, and the like are transmitted to the auction server 100.

[Auction server]
The auction server 100 includes, for example, an auction management unit (auction master) 110, a data management unit 120, a search execution unit 130, a ranking processing unit 140, a prediction request unit 150, and a storage unit 170. Information such as a product DB (database) 172, an inverted index 174, and successful bid record information 176 is stored in the storage unit 170. The components other than the storage unit 170 are realized by, for example, a hardware processor such as a CPU (Central Processing Unit) executing a program (software). In addition, some or all of these components are hardware (circuits) such as LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), and GPU (Graphics Processing Unit). It may be realized by the part; including circuitry), or it may be realized by the cooperation of software and hardware. The program may be stored in advance in a storage device such as an HDD (Hard Disk Drive) or a flash memory, or is stored in a removable storage medium such as a DVD or a CD-ROM, and the storage medium is stored in the drive device. It may be installed by being attached.

The auction management unit 110 provides a net auction service to the terminal device 10. The auction management unit 110 provides various information about the auction to the data management unit 120 as source data. Source data includes data provided by the seller, as well as data on the progress of the auction. A document ID, which is identification information of the exhibited product, is given to the source data. Further, the auction management unit 110 requests the search execution unit 130 to perform a search using the query received from the bidder's terminal device 10.

The data management unit 120 appropriately processes the source data and registers it in the product or the like DB 172. FIG. 2 is a diagram showing an example of the contents of the product or the like DB 172. Information such as a document ID, a category, a title, a detailed description, an image URL (Uniform Resource Locator), a price, a number of bids, a number of favorites registered, and delayed update data are registered in the search DB 172 in association with each other.

The title is, for example, text information displayed on an introduction page of a product or the like, and is text information displayed at a prominent position such as the upper part of the introduction page. The category is a category of products and the like that are classified hierarchically. The detailed description is text information displayed on an introduction page of a product or the like, and is text information displayed at a position different from the title. The title and detailed description are created, for example, by the seller.

The image URL is a URL (reference information) for referring to an image of a product or the like. The price is the highest bid at the moment. If no bid has been placed, the minimum bid set by the seller is stored. The number of bids is the number of bidders who have bid on the item. The number of favorites registered is the number of users who have registered the auction page of the product or the like as a favorite.

Further, the data management unit 120 generates an inverted index 174 in order to speed up the search by the search execution unit 130. FIG. 3 is a diagram showing an example of the contents of the inverted index 174. The inverted index 174 is information in which a document ID is associated with a word that can be a query. The data management unit 120 registers the document ID of the product or the like including the word or a similar word in the title, or the product or the like estimated to correspond to the word from the image in the inverted index 174. A vector obtained by decentralizing the text may be used to determine similar words.

Further, the data management unit 120 updates the successful bid record information 176 in response to the notification from the auction management unit 110. FIG. 4 is a diagram showing an example of the contents of the successful bid record information 176. In the successful bid record information 176, for example, information such as a document ID, a category, a title, a detailed explanation, and a successful bid price is registered in association with each other.

The search execution unit 130 uses the inverted index 174 to execute the search requested by the bidder's terminal device 10. For example, when the bidder's terminal device 10 requests a search for the query "television", the search execution unit 130 acquires the document IDs = 001, 005, ... From the inverted index 174. Then, the search execution unit 130 extracts information about the product or the like having the document ID = 001, 005, ... From the product or the like DB 172.

The ranking processing unit 140 ranks the products extracted by the search execution unit 130 according to a predetermined criterion, and selects the top k products (k is an arbitrary natural number). Then, the information of the selected k products is provided to the auction management unit 110 together with the ranking information. The auction management unit 110 transmits information for displaying the search result screen to the bidder's terminal device 10 based on the information acquired from the ranking processing unit 140. As a ranking method by the ranking processing unit 140, various methods can be adopted. For example, the ranking processing unit 140 may give a high score to a product having a large number of bids. Further, the ranking processing unit 140 may give a high score to a product having a large number of favorite registrations. Further, the ranking processing unit 140 may give a high score to a product having a low price.

The prediction requesting unit 150 requests the prediction device 200 to predict the winning bid amount when a predetermined operation is performed on the interface screen provided to the seller's terminal device 10. FIG. 5 is a diagram showing an example of the interface screen IM1 provided to the seller's terminal device 10. As shown in the figure, the interface screen IM1 is set with a title input area A1 for inputting a title, an explanation input area A2 for inputting an explanation (detailed description described above), a confirmation transmission button (not shown), and the like. Further, a request button B1 for requesting a successful bid amount prediction is set on the interface screen IM1. When the seller operates the request button B1 with some information entered in the title, the title and the information indicating the request are transmitted to the auction server 100 in association with each other. In response to this, the prediction requesting unit 150 transmits the title to the prediction device 200 and requests the prediction of the winning bid amount. On the interface screen IM1, not only the title and description but also the category of the product or the like may be selected, or the image to be displayed on the introduction screen of the product or the like may be specified. In this case, the prediction by the prediction device 200 may be executed for each category.

[Predictor]
Returning to FIG. 1, the prediction device 200 includes, for example, an acquisition unit 210, a prediction unit 220, and a storage unit 250. One or both of the acquisition unit 210 and the prediction unit 220 is realized by, for example, a hardware processor such as a CPU executing a program (software). In addition, some or all of these components may be realized by hardware such as LSI, ASIC, FPGA, GPU (including circuit section; circuitry), or realized by collaboration between software and hardware. May be done. The program may be stored in a storage device such as an HDD or a flash memory in advance, or is stored in a removable storage medium such as a DVD or a CD-ROM, and the storage medium is attached to the drive device. It may be installed.

The prediction unit 220 includes, for example, a token extraction unit 222, a winning bid distribution derivation unit 224, a clustering processing unit 226, a peak derivation unit 228, a suggestion determination unit 230, and a prediction price range derivation unit 232. Further, the storage unit 250 stores information such as successful bid record information 252 and calculation result 254 for each cluster.

The acquisition unit 210 acquires a part or all of the successful bid record information 176 from the auction server 100 at an arbitrary timing, and stores it in the storage unit 250 as the successful bid record information 252. Also. The acquisition unit 210 acquires the title of the product or the like for which the seller wishes to predict the winning bid amount from the auction server 100 together with the request for prediction. This title is an example of "information on products, etc. related to products or services (hereinafter, products, etc.) to be put up for auction".

Hereinafter, the functions of each part of the prediction unit 220 will be described with reference to the flowchart. FIG. 6 is a flowchart showing an example of the flow of processing executed by the prediction unit 220.

First, the token extraction unit 222 of the prediction unit 220 extracts the token used for the search from the title. The token extraction unit 222 performs morphological analysis using, for example, a morphological analysis engine such as MeCab (or requests another device to perform morphological analysis), and extracts a noun included in the title as a token (S100). Only one token may be extracted, or multiple tokens may be extracted. The token may be given a type such as a proper noun, a brand name, or a product model number obtained as a result of morphological analysis. Further, the token may be given information indicating the position in the title, that is, the number of tokens from the beginning among the tokens included in the title.

Next, the successful bid amount distribution derivation unit 224 refers to the successful bid record information 252 and obtains the distribution of the successful bid amount that matches the title acquired by the acquisition unit 210 (S102 to S110).

First, the successful bid amount distribution derivation unit 224 searches the successful bid record information 252 using each of the tokens extracted by the token extraction unit 222 (S102). The winning bid distribution derivation unit 224 may send a token to the auction server 100 to request a search.

Next, the successful bid amount distribution derivation unit 224 calculates an index value WAND indicating the degree of fitting with the title for each of the records of the successful bid record information 252 that was searched and hit in S102 (S104). WAND is represented by, for example, equation (1). In the formula, m is the record identifier (for example, document ID) of the successful bid record information 252, n is the number of tokens extracted in S102, and i is an argument indicating the number of tokens. Tk _i, as shown in equation (2), i-th token is a function indicating whether a hit in the search. The coefficient α _i is the weight given to the i-th token. The winning bid amount distribution deriving unit 224 determines the _{coefficient α i} based on, for example, the type given to the token, the position in the title, and the like, as described above. Note that α _i may be a fixed value such as 1 (that is, it may be omitted). WAND may be further obtained by dividing the right term of the equation (1) by a value based on n, n, a constant determined by the number of characters in the title, or the like.

Next, the winning bid amount distribution deriving unit 224 extracts a record of the winning bid record information 252 having a large WAND (for example, equal to or higher than a predetermined threshold value), and specifies the winning bid amount included in the extracted record as the target of the following processing. (S106). As a result, a record having a high degree of fitting with the title input by the seller is extracted from the record of the successful bid record information 252. In this way, since the information on the winning bid amount is selected based on WAND having a property intermediate between AND and OR, it is possible to secure the number of data and eliminate data having little relevance. When searching is performed by adopting AND in this technical field, almost matching records are not extracted. However, if OR is adopted and a search is performed, even records with very small relevance will be extracted. In the present embodiment, by adopting WAND, an appropriate number of records can be extracted.

Next, the winning bid distribution derivation unit 224 quantizes (histographs, BINs) the set of winning bids specified in S106 (S108), and derives the distribution of winning bids (S110). Quantization means that, for example, the winning bid amount expressed in 1-yen increments belongs to any of the BINs having a predetermined width. For example, the winning bid distribution derivation unit 224 applies the winning bid amount to BINs set in a predetermined width such as 0 yen to 5 yen, 6 yen to 10 yen, and 11 yen to 15 yen, and determines the number of appearances of each BIN. , Divide by the total number of winning bids and convert to the appearance probability. As a result, the winning bid amount distribution deriving unit 224 derives the winning bid amount distribution. The step size of the BIN may be set so as to increase as the amount of money increases. For example, the BIN may be set so that the logarithm of the winning bid amount is set in predetermined width increments.

FIG. 7 is a diagram schematically showing the result of quantizing the winning bid amount. The vertical axis is the number of appearances of each BIN. The distribution of the winning bid amount may not be uniformly counted as "1" with respect to the number of appearances, but may be derived based on the number of appearances obtained by multiplying the WAND value as a weight and counting. Further, the information on the winning bid amount may be quantized in advance by the winning bid amount distribution deriving unit 224 at a stage where the winning bid amount is stored in the storage unit 250 as the winning bid record information 252, not at the stage where the winning bid amount is predicted.

Next, the clustering processing unit 226 clusters (divides) the distribution of the winning bid amount derived by the winning bid amount distribution deriving unit 224 (S112). The clustering processing unit 226 clusters the quantized distribution of the winning bid amount based on a method such as x-means, meaning, and kernel density estimation (KDE). When kernel density estimation is used, the clustering processing unit 226 divides the cluster at the minimum point of the estimated probability density function.

Next, the peak derivation unit 228 derives a peak (an example of a “local peak”) in each cluster (S114). The peak is a BIN having the highest probability of appearance in the target range. Then, the peak derivation unit 228 selects the peak having the highest appearance probability among the peaks of each selected cluster, and derives the representative value (for example, the median value) of BIN corresponding to the selected peak as the winning bid prediction amount. (S116).

FIG. 8 is a diagram for explaining the process from clustering to derivation of the predicted price range. In the figure, P1 is the peak of cluster 1, and P2 is the peak of cluster 2. The appearance probability of the peak P1 is larger than the appearance probability of the peak P2. Therefore, the peak derivation unit 228 derives the representative value of BIN corresponding to the peak P1 as the winning bid prediction amount.

Returning to FIG. 6, the suggestion determination unit 230 calculates the area S1 of the 1st place cluster and the area S2 of the 2nd place cluster, respectively (S118). The 1st place cluster is the cluster having the highest probability of appearance at the peak, and the 2nd place cluster is the cluster having the 2nd highest probability of appearance at the peak. In the example of FIG. 8, cluster 1 is the 1st place cluster and cluster 2 is the 2nd place cluster. Further, S1 in the figure is the area of the cluster 1, and S2 is the area of the cluster 2. Here, the "area" is a convenient title, and is actually the total value of the appearance probabilities of BINs belonging to the cluster.

At this time, the suggestion determination unit 230 stores at least the area S1 of the area calculated in S118 in the storage unit 250 (memory) as the calculation result 254 for each cluster.

Then, the suggestion determination unit 230 determines whether or not the value obtained by dividing the area S2 by the area S1 exceeds the threshold Th1 (S120). When the value obtained by dividing the area S2 by the area S1 exceeds the threshold value Th1, the suggestion determination unit 230 determines that the reliability of the predicted winning bid amount is low because the distribution of the winning bid amount shows a multimodality with relatively no superiority or inferiority. Then, it is determined that there is no answer to this prediction request, and the processing of this flowchart is terminated. In this way, the suggestion determination unit 230 determines not to suggest the predicted bid amount when the area S1 and the area S2 are closer than the reference. As a result, it is possible to prevent a situation in which the seller is misleading by answering the expected bid amount with low reliability. The threshold value Th1 is, for example, a value of about 0.2 to 0.6.

When the category of the target product or the like is obtained, the suggestion determination unit 230 may set the threshold value Th1 according to the category. Further, when the prediction device 200 obtains a positive judgment in S120, the prediction device 200 may add information recommending "do not suggest" to the successful bid prediction amount and transmit it to the auction server 100. Further, the determination process of S120 may be replaced with, for example, "whether or not the value obtained by dividing the area S1 by the area S2 exceeds the threshold value Th2", and if it exceeds, the process may proceed to S122. The threshold value Th2 in this case is, for example, a value of about 1.5 to 5.

When the value obtained by dividing the area S2 by the area S1 does not exceed the threshold value Th1, the suggestion determination unit 230 determines that there is an answer to the prediction request this time. In this case, the predicted price range deriving unit 232 derives a range in which the ratio of the partial area centered on the peak to the area of the entire cluster approaches a predetermined ratio in the first-ranked cluster as the predicted price range (S122).

FIG. 9 is a diagram showing an example of a method for deriving the predicted price range in S122. In this example, the partial area Spa centered on the peak is the sum of the appearance probabilities of the BIN corresponding to the peak and the BIN corresponding to the peak and the BINs shifted by x from the BIN corresponding to the winning bid to the higher side and the lower side of the winning bid. .. For example, the predicted price range derivation unit 232 compares Spa / S1 with the threshold value Th3 while incrementing x by 1 in order from 0, and predicts the prices Pl and Ph corresponding to x when the threshold value Th3 is exceeded for the first time. The lower and upper limits of the price range. In the figure, Pe is the predicted successful bid amount. The predicted price range derivation unit 232 reads out the area S1 stored in the storage unit 250, and performs the above processing using the read area S1. This eliminates the need to obtain the area of the first-ranked cluster again, so that high-speed processing can be realized.

Then, the prediction unit 220 transmits the successful bid prediction amount Pe, the lower limit Pl and the upper limit Ph of the predicted price range to the auction server 100 (S124). The auction server 100 transmits these information to the seller's terminal device 10 and causes the seller's terminal device 10 to display the information.

In addition, when all the winning bids fit in one cluster in S112 (when the winning bids do not show multimodality), the processing of S114 and S116 is "the representative value of BIN corresponding to one peak is the predicted winning bid amount". The process of "deriving as" is performed, the processes of S118 and S120 are skipped, and the processing target of S122 is not the first-ranked cluster but the entire winning bid amount.

According to the prediction device of the embodiment described above, the winning bid amount is associated with the acquisition unit 210 that acquires the product information related to the products to be put up for auction and the product information related to the products sold at the auction. With reference to the successful bid record information 252, the distribution of the successful bid amount that matches the information on the products, etc. acquired by the acquisition unit 210 is obtained, and the successful bid amount corresponding to the peak in the distribution is the successful bid for the product, etc. acquired by the acquisition unit 210. By providing the prediction unit 220 derived as the predicted amount, the winning bid amount can be predicted more accurately.

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.

100 Auction server 200 Predictor 210 Acquisition unit 220 Prediction unit 222 Token extraction unit 224 Successful bid amount distribution derivation unit 226 Clustering processing unit 228 Peak derivation unit 230 Suggestion judgment unit 232 Predicted price range derivation unit

Claims (9)

An acquisition department that acquires product information related to products or services (hereinafter referred to as products, etc.) to be put up for auction.
In the token extracted from the product etc. information acquired by the acquisition unit by referring to the successful bid record information in which the successful bid amount is associated with the product etc. information related to the product etc. that was sold in the auction, and the successful bid record information. An index value indicating the degree of fitting with the token extracted from the product information is derived, and the distribution of the successful bid amount is obtained based on the successful bid amount corresponding to the product etc. information in the successful bid record information in which the index value is equal to or more than the threshold value. , A prediction unit that derives the winning bid amount corresponding to the peak in the distribution as the winning bid prediction amount of the product or the like acquired by the acquisition unit.
Predictor device. An acquisition department that acquires product information related to products or services (hereinafter referred to as products, etc.) to be put up for auction.
With reference to the successful bid record information in which the successful bid amount is associated with the product information related to the product or the like that was sold in the auction, the distribution of the successful bid amount that matches the product or the like information acquired by the acquisition unit is obtained, and the distribution is described. It is provided with a prediction unit that derives the winning bid amount corresponding to the peak in the above as the winning bid prediction amount of the product or the like acquired by the acquisition unit.
When the distribution of the winning bid amount is multimodal, the prediction unit selects one of the plurality of local peaks and derives the winning bid amount corresponding to the selected one peak as the winning bid prediction amount. ,
Predictor. The prediction unit divides the distribution of the winning bid amount and derives the maximum value in the divided distribution as the plurality of local peaks.
The prediction device according to claim 2. The prediction unit compares the total value of the appearance probabilities in each of the distribution having the largest local peak and the distribution having the second largest local peak among the divided distributions, and when the total values are closer than the reference, the prediction unit compares them. Decide not to suggest the estimated winning bid amount,
The prediction device according to claim 3. The prediction unit derives a predicted price range based on the distribution having the largest local peak among the divided distributions.
The prediction device according to claim 3 or 4. In the distribution having the largest local peak among the divided distributions, the prediction unit corresponds to a range in which the ratio of the total value of the appearance probabilities to the entire distribution approaches a predetermined ratio, centering on the winning bid amount corresponding to the local peak. Derivation of the forecast price range based on the winning bid amount,
The prediction device according to claim 5. The prediction unit
Among the divided distributions, the total value of the appearance probabilities in each of the distribution having the largest local peak and the distribution having the second largest local peak is compared, and if the total values are closer than the standard, the predicted bid amount is calculated. Decided not to suggest,
In the distribution with the largest local peak among the divided distributions, based on the winning bid amount corresponding to the range in which the ratio of the total value of the appearance probabilities to the entire distribution approaches a predetermined ratio, centering on the winning bid amount corresponding to the local peak. To derive the forecast price range,
The total value of the appearance probabilities in the distribution having the largest local peak calculated in the process of determining whether or not to suggest is stored in the memory, and is read from the memory in the process of deriving the predicted price range. use,
The prediction device according to claim 3. The computer
Obtain information on products, etc. related to products or services (hereinafter referred to as products, etc.) to be put up for auction,
The token extracted from the acquired product information and the product information in the successful bid information are obtained by referring to the successful bid record information in which the successful bid amount is associated with the product information related to the successful bid in the auction. An index value indicating the degree of fit with the token extracted from is derived, and the distribution of the winning bid amount is obtained based on the winning bid amount corresponding to the product etc. information in the winning bid record information in which the index value is equal to or more than the threshold value. The winning bid amount corresponding to the peak in
Prediction method. On the computer
Obtain information on products, etc. related to products or services (hereinafter referred to as products, etc.) to be put up for auction.
The token extracted from the acquired product information and the product information in the successful bid information by referring to the successful bid record information in which the successful bid amount is associated with the product information related to the successful bid in the auction. An index value indicating the degree of fit with the token extracted from is derived, and the distribution of the winning bid amount is obtained based on the winning bid amount corresponding to the product etc. information in the winning bid record information in which the index value is equal to or more than the threshold value. The winning bid amount corresponding to the peak in the distribution is derived as the winning bid predicted amount of the acquired product or the like.
program.

Images