KR20210081161A - Data processing method and apparatus - Google Patents

Abstract

In a method for processing information in accordance with one embodiment, provided is the method including: a step of receiving a first information expressed in a one-dimensional format and including a categorical variable; a step of determining one or more criteria to be used for rearranging the first information based on a characteristic of the categorical variable included in the first information; a step of converting the first information into a two-dimensional format based on the one or more criteria to obtain a second information; and a step of determining a characteristic of the first information by applying a convolutional neural network (CNN) to the second information converted into the two-dimensional format. Therefore, the present invention is capable of improving a decision accuracy of a model.

Description

Information processing method and apparatus {DATA PROCESSING METHOD AND APPARATUS}

The present invention relates to an information processing method and apparatus, and more particularly, by converting first information in a one-dimensional format including a categorical variable into second information in a two-dimensional format and applying it to a CNN, correlation between adjacent bits The present invention relates to a method and apparatus capable of efficiently performing meaningful data analysis on numerous categorical variables by being able to analyze local correlations even for categorical variables of randomness without randomness.

The conventional technology to analyze and utilize customer behavior data uses a machine learning model by extracting variables or a method that fits a linear model by converting it into product preference. In the former case, there is a disadvantage that a lot of work is required in the variable extraction process and the bias is reflected by reflecting the intuition of the field expert. In the latter case, there is a disadvantage that the relationship between the customer and the product is modeled linearly.

These conventional techniques mainly use structured data, which are generally regarded as data that is easy to generate and analyze statistical variables, but when the amount of variables is large and the analysis task is complicated, how feature engineering is performed on the data? has a significant impact on model performance. Accordingly, various technologies related to machine learning models for analyzing customer behavior data by extracting characteristics from structured data are being developed.

In general, the number of cases in which characteristics are extracted from customer behavior data can range from thousands to tens of thousands, but it is difficult to quantify or simply express in terms of categorical variables rather than continuous variables or sequence scales. , the number of variables increases exponentially in the process for analysis.

In the conventional technology, in terms of marketing that provides promotions by extracting customers that meet daily conditions, it is difficult to generate all these variables every time, and it is practically impossible to store the customers' states that change every day. As a method for solving this problem, the prior art uses a method of extracting characteristics by selecting only some of the exploding variables, but since this method is a structure in which the analyst's intuition, common sense, bias, etc. are inevitably reflected, there is a limitation as a result There is a disadvantage in that the accuracy of the results is lowered as only the necessary information is used.

Accordingly, the demand for technology that can obtain good learning results by reflecting a large number of variables generated in the pre-processing of customer behavior data for analyzing customer life and purchase patterns in the analysis is steadily increasing.

Korean Patent Application Laid-Open No. 10-2017-0096298 relates to a deep learning system using image patterning based on a convolutional neural network and an image learning method using the same, comprising: an image input unit for inputting an input image; a patterning module that generates a plurality of patterned patterned image input images received from the image input unit; a CNN learning unit based on a convolutional neural network (CNN) that learns the input image received from the image input unit and the pattern image received from the patterning module; a CNN execution unit receiving learning information from the CNN learning unit and an input image received from the image input unit; and a final classification unit that receives the image information from the CNN execution unit and classifies the object of the image information by type.

An embodiment of the present invention can solve the problems of the prior art by providing an information processing method and apparatus, and more specifically, converts first information in a one-dimensional format including a categorical variable into a second information in a two-dimensional format. By converting into 2 information and applying it to CNN, it is possible to analyze local correlation even for categorical variables of randomness that are not related between adjacent bits, so that meaningful data analysis for numerous categorical variables can be efficiently performed. Even if the number of is increased, it is possible to provide an efficient effect in terms of storage capacity.

Objects of the present invention are not limited to the objects mentioned above, and other objects not mentioned will be clearly understood from the description below.

An information processing method according to a first aspect of the present disclosure includes: receiving first information expressed in a one-dimensional format and including a categorical variable; determining one or more criteria used for rearranging the first information based on a characteristic of the categorical variable included in the first information; converting the first information into a two-dimensional format based on the one or more criteria to obtain second information; and determining a characteristic of the first information by applying a Convolutional Neural Network (CNN) to the second information converted into the two-dimensional format.

In addition, the determining of the characteristic of the first information by applying the CNN may determine the attribute of the first information based on local characteristics between locations of the information included in the second information.

Also, the categorical variable may include a nominal variable, and the first information may include a bitstream in which information on variables indicated by the nominal variable is sequentially provided.

In addition, the categorical variable may further include an ordinal variable, and the one or more criteria may include an order in which the ordinal variable is based.

Also, the first information may include purchase information, the categorical variable may include category information about a product for which purchase occurs, and the one or more criteria may include the category.

In addition, the first information may include purchase information, the categorical variable may include information on a commercial area or industry in which a purchase occurs, and the one or more criteria may include the commercial area or the industry.

In addition, the second information includes the first information, and applying the CNN to determine the characteristics of the first information includes: filtering the second information to obtain main information; and determining a characteristic of the first information based on the main information.

The information processing method may further include: classifying the first information into a first matrix, a second matrix and a third matrix according to the one or more criteria; obtaining an image in which the first matrix, the second matrix, and the third matrix are represented by R, G, and B, respectively; and determining the characteristic of the first information by applying the CNN to the image.

In addition, the first information further includes a numeric variable, and the information processing method obtains from the first information 1-1 information consisting of the categorical variable and 1-2 information consisting of the numerical variable to do; obtaining 2-1 information by converting the 1-1 information into a two-dimensional format based on the characteristics of the categorical variable; obtaining 2-2 information by converting the 1-2 information into a two-dimensional format based on the characteristics of the numeric variable; Based on the inter-location local characteristics of the information included in the 2-1 information and the inter-location local characteristics of the information included in the 2-2 information, the 2-1 information and the 2-2 information applying each of the CNNs; and determining a characteristic of the first information using a result of applying the CNN to the 2-1 information and the 2-2 information.

The information processing method includes: obtaining, from the 1-1 information, 1-3 th information composed of the nominal variable and 1-4 th information composed of the ordinal variable; converting the 1-3 th information into a two-dimensional format to obtain 2-3 th information based on the characteristic of the nominal variable; converting the 1-4th information into a two-dimensional format to obtain 2-4th information based on the characteristics of the ordinal variable; Based on the inter-location local characteristics of the information included in the 2-3 information and the inter-location local characteristics of the information included in the 2-4 information, the 2-3 information and the 2-4 information are applying each of the CNNs; and determining a characteristic of the first information using a result of applying the CNN to the 2-3 th information and the 2-4 information.

An information processing apparatus according to a second aspect of the present disclosure includes: a communication unit for receiving first information expressed in a one-dimensional format and including a categorical variable; and determining one or more criteria used for rearranging the first information based on a characteristic of the categorical variable included in the first information, and converting the first information into a two-dimensional format based on the one or more criteria to obtain the second information, and a processor for determining the characteristic of the first information by applying a Convolutional Neural Network (CNN) to the second information converted into the two-dimensional format.

Also, the processor may determine the attribute of the first information based on a local characteristic between positions of the information included in the second information.

Also, the categorical variable may include a nominal variable, and the first information may include a bitstream in which information on variables indicated by the nominal variable is sequentially provided.

In addition, the categorical variable may further include an ordinal variable, and the one or more criteria may include an order in which the ordinal variable is based.

Also, the first information may include purchase information, the categorical variable may include category information about a product for which purchase occurs, and the one or more criteria may include the category.

In addition, the first information may include purchase information, the categorical variable may include information on a commercial area or industry in which a purchase occurs, and the one or more criteria may include the commercial area or the industry.

In addition, the second information may include the first information, the processor may filter the second information to obtain main information, and determine a characteristic of the first information based on the main information.

Further, the processor classifies the first information into a first matrix, a second matrix and a third matrix according to the one or more criteria, wherein the first matrix, the second matrix and the third matrix are R, G , B may be obtained, and the characteristics of the first information may be determined by applying the CNN to the image.

In addition, the first information further includes a numeric variable, and the processor obtains, from the first information, 1-1 information consisting of the categorical variable and 1-2 information consisting of the numerical variable, The 1-1 information is converted into a two-dimensional format based on the characteristics of the categorical variable to obtain the 2-1 information, and the 1-2 information is converted into a two-dimensional format based on the characteristics of the numeric variable. to obtain 2-2 information, and based on the inter-location local characteristics of the information included in the 2-1 information and the inter-location local characteristics of the information included in the 2-2 information, the second information is obtained. The CNN is applied to the 2-1 information and the 2-2 information, respectively, and the characteristic of the first information is determined using the result of applying the CNN to the 2-1 information and the 2-2 information can

In addition, the processor obtains, from the 1-1 information, 1-3 th information including the nominal variable and 1-4 th information including the ordinal variable, and based on the characteristic of the nominal variable, the 1-3 information is converted into a two-dimensional format to obtain 2-3 information, and information 1-4 is converted into a two-dimensional format based on the characteristics of the ordinal variable to obtain information 2-4; , the 2-3th information and the 2-4 information based on the inter-location local characteristics of the information included in the 2-3 information and the inter-location local characteristics of the information included in the 2-4 information Each of the CNNs may be applied to , and the characteristics of the first information may be determined using a result of applying the CNN to the 2-3 th information and the 2-4 information.

A third aspect of the present disclosure may provide a computer-readable recording medium recording a program for executing the method according to the first aspect on a computer. Alternatively, the fourth aspect of the present disclosure may provide a computer program stored in a recording medium to implement the method according to the first aspect.

According to an embodiment of the present invention, it is possible to improve the learning efficiency of a model for determining a customer's purchase attribute and improve the decision accuracy of the model by compressing and reflecting numerous variables in one image.

In addition, through image-based CNN learning, it is possible to read local correlations even for categorical variables of randomness that are not related between adjacent bits because there is no vertical, horizontal, left-right relationship, so that meaningful data analysis can be performed on numerous categorical variables. .

In addition, many major variables can be applied to data analysis even if the user does not edit and insert only good variables according to his or her own bias.

In addition, instead of creating numerous variables, it is possible to compress and store one matrix information, so that even if the number of variables increases, there is an efficient effect in terms of storage capacity.

It should be understood that the effects of the present invention are not limited to the above-described effects, and include all effects that can be inferred from the configuration of the invention described in the detailed description or claims of the present invention.

1 is a block diagram illustrating an example of a configuration of an information processing apparatus according to an embodiment.
FIG. 2 is a flowchart illustrating an example of a method in which the information processing apparatus in FIG. 1 performs information processing.
3 is an operation in which the information processing device in FIG. 1 converts first information expressed in a one-dimensional format into second information expressed in a two-dimensional format, and applies the second information to CNN to determine the characteristics of the first information It is a diagram for explaining an example of.
FIG. 4 is a diagram for explaining an example of an operation in which the information processing device of FIG. 1 evaluates a result of determining a characteristic of first information by applying a plurality of learning models.
FIG. 5 is a diagram for explaining an example of an operation in which the information processing apparatus of FIG. 1 determines one or more criteria used for rearranging first information.
FIG. 6 is a view for explaining various embodiments of an operation in which the information processing apparatus of FIG. 1 obtains second information by converting first information into a two-dimensional format based on one or more criteria;

Hereinafter, the present invention will be described with reference to the accompanying drawings. However, the present invention may be embodied in several different forms, and thus is not limited to the embodiments described herein. And in order to clearly explain the present invention in the drawings, parts irrelevant to the description are omitted, and similar reference numerals are attached to similar parts throughout the specification.

Throughout the specification, when a part is said to be "connected" with another part, it includes not only the case where it is "directly connected" but also the case where it is "indirectly connected" with another member interposed therebetween. . In addition, when a part "includes" a certain component, this means that other components may be further provided without excluding other components unless otherwise stated.

Hereinafter, an embodiment will be described in detail with reference to the accompanying drawings.

1 is a block diagram illustrating an example of a configuration of an information processing apparatus 100 according to an embodiment.

Referring to FIG. 1 , an information processing apparatus 100 according to an embodiment may include a communication unit 110 , a processor 120 , and a storage unit 130 .

The communication unit 110 according to an embodiment may be connected to other devices (eg, terminals, servers) through a network, and may include, for example, all types of wired/wireless communication devices capable of communicating with other devices. Here, the network may be configured through various communication networks such as wired and wireless, for example, a local area network (LAN), a metropolitan area network (MAN), and a wide area network (WAN). Network), etc., may be composed of various communication networks.

The communication unit 110 according to an embodiment may receive first information expressed in a one-dimensional format. For example, the communication unit 110 may collect first information corresponding to structured data expressed as a bit stream from one or more terminals (eg, affiliated store terminals, etc.) or servers (eg, payment server).

In an embodiment, the first information includes the customer's personal information (eg, gender, age, annual income, etc.), purchase information (eg, whether to purchase, purchase amount, purchased product/service, purchased product/service category, purchasing merchant, purchase Industry, type of purchase, purchase location, type of payment method, etc.), behavioral information (e.g., site access, product information inquiry, phone call, etc.) and location information (e.g. customer movement history, purchase location history, etc.) may contain one.

In an embodiment, the customer behavior information includes whether the customer clicks on a specific category (or product) (eg, whether product/service information is provided according to the customer's inquiry input), the number of clicks (eg, the customer's inquiry input) may include at least one of the number of times the product/service information is provided according to the number of times the product/service information is provided) and a click order (eg, a time-series order of categories in which product/service information is provided according to a customer's inquiry input).

Here, the category indicates a classification category for a product or service, and in an embodiment, the type and level of refinement of the category may be determined according to a plurality of preset grades. For example, the category corresponding to the 2nd grade among grades 1 to 6, in which the category system is defined to indicate a relatively lower category as the value of the grade increases, is included in the above-mentioned customer's first information or as one or more criteria to be described later. can be used

In an embodiment, the first information may include a categorical variable. Here, the categorical variable refers to a variable that can be classified into a class or range having several identical properties. For example, when the first information includes purchase information of a customer, the categorical variable included in the first information may include information on a category, commercial area, or industry for a product in which purchase occurs.

In an embodiment, the categorical variable included in the first information may include a nominal variable, and the first information may include a bitstream in which information on variables indicated by the nominal variable is sequentially provided. . Here, the nominal variable is a variable whose order cannot be determined between the values of the variable, and represents a variable that does not correlate with the order of information because the order or scale between items is meaningless. For example, when the first information includes customer purchase information, the nominal variable included in the first information is the category (eg, women's fashion, air purifier, etc.) of the product in which the purchase occurs, and the commercial area (eg: Myeongdong Street, Hongdae, City Hall, etc.) and information on business types (eg, restaurants, finance, retail, etc.) may be included.

In an embodiment, the categorical variable included in the first information may further include an ordinal variable. Here, the ordinal variable is a variable in which there is a sequence or order between the values of the variable, and represents a finite variable in which the order or scale between items is significant and has a back-and-forth correlation with the order of information. For example, if the first information includes personal information and purchase information of a customer, the ordinal variable included in the first information may be the customer's purchasing grade (eg, A to D grade) or age group (eg, 20s, 30's, etc.).

In an embodiment, the first information may further include a numeric variable. Here, the numerical variable represents a variable that has a countable or continuous property and thus has a clear correlation between the order of information and the sequence. For example, when the first information includes purchase information or behavior information of a customer, the numeric variable included in the first information is the number of purchases (eg, 10 times) or clicks of the customer for a specific product (or category). It can include information about the number of times (eg, 5 times).

The processor 120 according to an embodiment may store and manage the received first information in the storage unit 130, and through statistical analysis of the first information, the first information may be stored for a preset period (eg, the last 14 days). Statistical information for the first information may be generated and included in the first information.

The processor 120 according to an embodiment may determine one or more criteria used for rearranging the first information based on the characteristics of the categorical variable included in the first information. In an embodiment, the one or more criteria may include, but are not limited to, information on one or more of a category, a commercial district, and an industry for a product for which a purchase occurs, and, in addition, a purchase or inquiry for a product (or category) It may further include a date on which is generated, a purchase amount, the number of purchases, a purchase order, whether to purchase, the number of clicks, and whether or not to click.

For example, when the first information includes purchase information and the categorical variable included in the first information includes categorical information about a product for which purchase occurs, the processor 120 determines the characteristics of the corresponding categorical variable (eg, : When the category information for a product for which a purchase occurs) satisfies a specific preset condition (eg, including customer ID, category information, and the number of purchases by a customer for a category), a plurality of criteria corresponding to the condition (eg: It may be determined to use a customer ID, a category for a product in which a purchase occurs, and the number of times a customer purchases a category) for reordering the first information.

The processor 120 according to an embodiment may obtain the second information by converting the first information into a two-dimensional format based on one or more determined criteria. For example, the processor 120 applies each of the determined one or more criteria to an axis (eg, X-axis, Y-axis, Z-axis) or value (eg, a value of each matrix) of a matrix of a preset size to each criterion. According to the method of arranging the first information, matrix information in a two-dimensional format indicating the number of purchases for each category for each customer may be generated and generated as the second information.

In an embodiment, the second information may include the first information, and may be matrix information in a two-dimensional format or an image in a two-dimensional format. For example, the processor 120 may generate matrix information in a two-dimensional format by rearranging the first information based on one or more criteria described above, and apply the obtained matrix information to a pre-stored image algorithm to generate the matrix information. Image conversion may be performed by allocating an image output value (eg, an integer greater than or equal to 0 and less than or equal to 255) corresponding to a value of a matrix to pixels of an image corresponding to a position, and thus, at least one of R, G, and B channels An image expressed through one or more may be generated as the second information. The generated second information may include the first information in a format (eg, an image format) different from the format of the first information (eg, a format of fixed data).

In an embodiment, when the first information includes a categorical variable, the information processing apparatus 100 uses one or more criteria including one or more of a category for a product for which a purchase occurs, a commercial area, and a business type to first The information may be converted into second information in a two-dimensional format. For example, when the first information includes each customer's personal information (eg, customer ID), category information about a product, and purchase information (eg, whether or not to purchase), the information processing device 100 may display the first By applying the criteria of customer's personal information to the axis (eg X-axis), applying the criteria for category information to the second axis (eg, Y-axis), and applying the customer's purchase or noting criteria to each matrix value, Matrix information may be obtained by sequentially arranging information on whether or not a customer purchases by category for each customer in a matrix, and the matrix may be converted into an image expressed through one of the RGB channels to generate second information. The second information generated in this way may be used in an information characteristic determination model for predicting the purchase possibility of a customer by category by analyzing which category of product each customer mainly purchases.

In another embodiment, when the categorical variable included in the first information further includes an ordinal variable, the information processing device 100 may use one or more criteria including an order that serves as a criterion for the ordinal variable. One information may be converted into second information in a two-dimensional format. For example, the information processing device 100 includes, as a categorical variable, each customer's personal information (eg, customer ID) and category information about a product in the first information, and the customer's click on the product as an ordinal variable When time is included, the criteria of category information for products is applied to the first axis (eg, X-axis) of the matrix, and the criterion of the order that serves as the basis for the customer's click time on the second axis (eg, Y-axis) By applying the criteria of customer's personal information (eg, customer ID) to each matrix value, the customer's personal information (eg, customer ID) is sequentially arranged in the matrix according to the click order by category to obtain matrix information The second information may be generated by converting the matrix into an image represented through one of the RGB channels. The second information generated in this way may be used in an information characteristic determination model for predicting the searchability of a customer by category by analyzing which category of product is preferentially clicked by each customer or which customer mainly clicks by category.

In another embodiment, the information processing device 100 includes a category when the first information includes purchase information and the categorical variable included in the first information includes category information about a product for which purchase occurs. The first information may be converted into the second information in a two-dimensional format using one or more criteria. For example, when the first information includes the number of purchases of each customer for a category and category information associated with the purchase as a categorical variable is included in the first information, the first axis of the matrix (eg, X axis), apply the criteria of customer personal information (eg, customer ID) to the second axis (eg, Y-axis), and apply the criteria of the number of customer purchases to each matrix value. , matrix information may be obtained by sequentially arranging the number of purchases for each customer by category in the matrix, and the matrix may be converted into an image expressed through one of the RGB channels to generate second information. The second information generated in this way may be used in an information characteristic determination model for predicting the purchase possibility of a customer by category by analyzing which category of product the customer purchases more of.

In another embodiment, the information processing device 100 is the first information includes purchase information, and when the categorical variable included in the first information includes information on the commercial district or industry in which the purchase occurs, the commercial district or The first information may be converted into the second information in a two-dimensional format by using one or more criteria including the industry type. For example, when the first information includes the number of purchases of each customer for a category in the first information and information on a commercial area and industry associated with the purchase as a categorical variable is included, the first axis of the matrix (Example: X-axis) Apply the standard of the business district, apply the standard of the industry to the 2nd axis (e.g., the Y-axis), and apply the customer's purchase count to each matrix value. Matrix information may be obtained by sequentially arranging the number of purchases of the customer, and the matrix may be converted into an image expressed through one of the RGB channels to generate second information. The second information generated in this way may be used in an information characteristic determination model for predicting the purchase possibility by industry by industry by analyzing which type of product the customer mainly purchases in which commercial area.

Various embodiments of the operation of the information processing apparatus 100 converting the first information into the second information in a two-dimensional format based on one or more criteria according to an embodiment will be described below with reference to FIG. 6 and will be additionally described later. .

The processor 120 according to an embodiment may determine a characteristic of the first information by applying a Convolutional Neural Network (CNN) to the second information converted into a two-dimensional format. In an embodiment, the characteristic of the first information may include a purchasing attribute of the customer, and the purchasing attribute of the customer may include a probability of purchasing a particular product of the customer over time, for example, a specific period or It may include a purchase probability that quantifies whether a customer will actually purchase a particular product at a particular time period.

Here, CNN corresponds to a type of deep neural network that extracts variable features using a convolutional layer and then outputs a desired output value, and in one embodiment, CNN is a deep neural network (DNN) that does not use a convolutional layer. In contrast to Neural Network), it is possible to learn local characteristics based on the combination of adjacent variables, and according to the design of the convolutional layer, it is possible to learn information in chronological order, and it uses less memory than RNN (Recurrent Neural Networks) It has the characteristics of fast learning and reasoning speed.

For example, when an image expressed in RGB in a two-dimensional format is obtained through transformation of the first information, the processor 120 applies the obtained image to a pre-stored CNN algorithm to generate an information characteristic determination model, and generate By inputting specific category information into the information characteristic determination model, the probability of purchasing a specific product or category of the customer according to time included in the characteristic of the first information may be determined. In addition, the processor 120 may improve the performance of the information characteristic determination model through CNN-based deep learning by using a plurality of images converted from the collected information as a training dataset.

The processor 120 according to an embodiment may determine the attribute of the first information based on local characteristics between positions of the information included in the second information. For example, the processor 120 can read the local correlation by analyzing the combination of adjacent variables reflected in the image through the CNN by applying the CNN to the second information converted into the two-dimensional image and performing learning. have.

In this case, the information on the categorical variables included in the second information may have a random characteristic, where random may mean that there is no relation between adjacent bits because there is no vertical, horizontal, left-right relationship. That is, the processor 120 may compress and reflect numerous variables in the second information, and even though numerous categorical variables included in the second information have random characteristics, the second information expressed as an image is adjacent to the second information vertically, horizontally, horizontally and vertically. It can be used meaningfully by analyzing the combination between pixels to analyze local characteristics between positions.

Accordingly, the processor 120 can effectively extract the characteristics of the information on the categorical variables included in the image through the convolutional layer of the CNN, and perform CNN-based deep learning using the image to generate numerous variables. By efficiently reflecting it, the performance of the model that determines the purchasing attributes of customers can be greatly improved.

The processor 120 according to an embodiment may obtain main information by filtering the second information, and may determine a characteristic of the first information based on the obtained main information. For example, the processor 120 may extract information about a plurality of variables corresponding to a nominal variable, an ordinal variable, or a numerical variable from the second information through the CNN, and perform filtering on the extracted information. It is possible to obtain main information including information on one or more preset main variables (eg, categories of products purchased by the customer, customer ID, purchase amount, purchase time, etc.), and information on filtered main variables These factors can be input into the information characterization model to determine the purchasing attributes of customers.

The processor 120 according to an embodiment classifies the first information into a first matrix, a second matrix, and a third matrix according to one or more determined criteria, and the first matrix, the second matrix, and the third matrix are R, An image represented by G and B may be obtained, and a characteristic of the first information may be determined by applying a CNN to the obtained image.

For example, the processor 120 generates a matrix in which information on whether or not the customer purchases by category for each customer is sequentially arranged according to the above-described example, but the information on the category purchased by the customer can be retrieved in the same period (eg, by week). In total, it is possible to create a first matrix reflecting the categories purchased in the period of the last 1 week, a second matrix reflecting the categories purchased in the period of the last 2 weeks, and a third matrix reflecting the categories purchased in the period of the last 3 weeks. have. Also, the processor 120 may generate an image expressed in RGB by converting the generated first matrix, the second matrix, and the third matrix through the R channel, the G channel, and the B channel, respectively.

When the first information further includes a numeric variable, the processor 120 according to an embodiment obtains, from the first information, the 1-1 information consisting of the categorical variable and the 1-2 information consisting of the numeric variable. can do. In addition, the processor 120 converts the 1-1 information into a two-dimensional format based on the characteristics of the categorical variable to obtain the 2-1 information, and generates the 1-2 information based on the characteristics of the numeric variable. 2-2 information may be obtained by converting it into a two-dimensional format. In addition, the processor 120 is configured to perform the 2-1 information and the 2-th information based on the inter-location local characteristics of the information included in the 2-1 information and the inter-location local characteristics of the information included in the 2-2 information. The CNN may be applied to the second information, respectively, and the characteristics of the first information may be determined using the results of applying the CNN to the 2-1 information and the 2-2 information.

For example, the processor 120 transmits the first information to the first information consisting of the customer's ID, category information about the product, and 1-1 information including whether the customer has purchased the product, and the first information consisting of the number of purchases and the purchase time of the customer for the category. Classify into -2 information, generate a 2-1 image indicating whether each customer purchases by category and a 2-2 image indicating the number of purchases by category over time, and generate 2-1 and 2-2 images It can be applied to CNN to analyze the local characteristics of each image to determine the probability of a customer's purchase, which indicates whether the customer will purchase a product in a particular category at a particular time.

The processor 120 according to an embodiment obtains, from the 1-1 information, 1-3 th information composed of nominal variables and 1-4 th information composed of ordinal variables, and based on the characteristics of the nominal variables, 1-3 information is converted into a two-dimensional format to obtain 2-3 information, and information 2-4 is obtained by converting information 1-4 into a two-dimensional format based on the characteristics of an ordinal variable; CNN is applied to the 2-3 information and the 2-4 information, respectively, based on the local characteristics between the positions of the information included in the 2-3 information and the local characteristics between the positions of the information included in the 2-4 information, , it is possible to determine the characteristics of the first information by using the results of CNN applied to the 2-3 information and the 2-4 information.

For example, the processor 120 transmits the 1-1 information to the 1-3 information consisting of category information about the product and whether the customer has purchased the product, the customer's purchase grade (eg, grades 1 to 6) and time zone (eg, : morning, afternoon, evening, night), and generating 2-3 images indicating whether to purchase by category by customer and 2-4 images indicating whether to purchase by purchase grade by time period, By applying the 2-3 and 2-4 images to the CNN, the local characteristics of each image can be analyzed to determine the purchasing probability of a customer, which indicates whether a customer of a specific purchasing class will purchase a product in a specific category at a specific time. have.

The processor 120 according to an embodiment may include a first local characteristic indicating a relationship between positions of information included in the 2-2 information (eg, a numeric variable), and the 2-3 information (eg, a nominal variable) obtain a second local characteristic indicating the relation between the positions of the information included in , and a third local characteristic indicating the correlation between the positions of the information included in the 2-4 information (eg, ordinal variable), and The characteristic of the first information may be determined by applying different weights to the local characteristic, the second local characteristic, and the third local characteristic.

In an embodiment, the processor 120 may configure a first weight that is a weight applied to the first local characteristic, a third weight that is a weight applied to the third local characteristic, and a second weight that is a weight applied to the second local characteristic. Larger values may be assigned in the order of weights (eg, first weight > third weight > second weight).

Accordingly, the processor 120 assigns the highest weight to the numeric variable and the lowest weight to the nominal variable, thereby reflecting the correlation between positions included in the numeric variable more to reflect the characteristics of the first information. can decide In general, since local features of numerical variables are more significant than local features of ordinal variables, and local features of ordinal variables are more significant than local features of nominal variables, processor 120 calculates the weight Based on this, it is possible to determine the reflection ratio of meaningful information, so that the customer's purchase probability can be predicted with higher accuracy.

In an embodiment, the second weight is determined according to the number of categories included in the categorical variable, and when the number of categories is less than or equal to 2, the first weight, the second weight, and the third weight are the largest in the order. may be given (eg, first weight > second weight > third weight). For example, if the nominal variable has one or two categories, the relevance of values within the same category may be more significant than the relevance in the ordinal variable, so the processor 120 determines the number of meaningful information based on the number of categories. You can adjust the reflection ratio.

The processor 120 according to an embodiment may be implemented as a central processor unit (CPU) that controls the overall operation of the information processing device 100 , and is electrically connected to the communication unit 110 and the storage unit 130 to make these You can control the flow of data between them.

The storage unit 130 according to an embodiment may store the first information and the second information, and other data necessary for the overall operation of the information processing apparatus 100 . In one embodiment, the storage unit 130 is a main memory implemented as a volatile memory such as a secondary storage device or RAM (Random Access Memory) implemented as a non-volatile memory such as a solid state disk (SSD) or HDD (hard disk drive). device may be included. In addition, in an embodiment, the storage unit 130 may be implemented as a database or a cloud or a separate storage server to provide data and storage space necessary for the information processing apparatus 100 through a wired/wireless communication network.

In addition, those of ordinary skill in the art can understand that other general-purpose components other than those shown in FIG. 1 may be further included in the information processing apparatus 100 . For example, the information processing apparatus 100 may include an algorithm for implementing a CNN-based learning model, and in addition, an input/output interface for receiving or outputting a user input, a display for visualizing output information, etc. can do. In addition, according to another embodiment, some of the components shown in FIG. 1 may be omitted. Also, the information processing apparatus 100 according to an embodiment may be implemented as one or more servers.

FIG. 2 is a flowchart illustrating an example of a method in which the information processing apparatus 100 of FIG. 1 performs information processing.

In operation S210 , the information processing apparatus 100 according to an embodiment may receive first information expressed in a one-dimensional format and including a categorical variable. In an embodiment, the first information may include at least one of customer's personal information, purchase information, behavior information, and location information, and may include at least one of a nominal variable, an ordinal variable, and a numeric variable. .

In operation S220, the information processing apparatus 100 according to an embodiment may determine one or more criteria used for rearranging the first information based on the characteristics of the categorical variable included in the first information. In an embodiment, the one or more criteria may include at least one of category information on a product for which purchase occurs, and information on a commercial area or industry.

In operation S230, the information processing apparatus 100 according to an embodiment may obtain the second information by converting the first information into a two-dimensional format based on one or more criteria. In an embodiment, the categorical variable may include a nominal variable, and the first information may include a bitstream in which information on variables indicated by the nominal variable is sequentially provided. In addition, the categorical variable may further include an ordinal variable, and the one or more criteria may include an order in which the ordinal variable is based. In addition, the categorical variable may further include an ordinal variable, and the one or more criteria may include an order in which the ordinal variable is based. In addition, the first information may include purchase information, the categorical variable may include category information about a product for which purchase occurs, and the one or more criteria may include a category. In addition, the first information may include purchase information, the categorical variable may include information on a commercial area or industry in which purchase occurs, and one or more criteria may include a commercial area or industry type.

In step S240, the information processing apparatus 100 according to an embodiment may determine the characteristic of the first information by applying the CNN to the second information converted into a two-dimensional format. In an embodiment, the information processing apparatus 100 may determine the attribute of the first information based on local characteristics between positions of information included in the second information. Also, the information processing apparatus 100 may obtain main information by filtering the second information, and may determine a characteristic of the first information based on the main information.

According to an embodiment of the present invention, even if the user does not edit and put good variables according to his or her bias, numerous major variables can be compressed into an image and reflected in CNN-based data analysis, so even if the number of variables increases, the storage capacity is reduced. in terms of effectiveness.

According to an embodiment of the present invention, by converting categorical variables expressed in one-dimensional format into two-dimensional images and learning the image based on CNN, a number of variables are compressed and reflected in one image to learn efficiency can be improved.

3 is a diagram in which the information processing device 100 in FIG. 1 converts first information expressed in a one-dimensional format into second information expressed in a two-dimensional format, and applies CNN to the second information to determine the characteristics of the first information It is a diagram for explaining an example of an operation of determining.

Referring to FIG. 3 , the information processing apparatus 100 according to an embodiment manages the customer's personal information and the customer's purchase information (eg, refer to the identification number 310 ), as in an example shown in identification numbers 310 to 320 . may receive first information (eg, refer to identification numbers 310 to 320) expressed in a one-dimensional format and including categorical variables from a server (eg, a customer information management server).

In addition, the information processing device 100 determines one or more criteria used for rearranging the first information based on the characteristics of the categorical variable included in the first information, as in an example shown in identification number 330, and the determined The second information may be obtained by converting the first information into matrix information in a two-dimensional format (eg, refer to identification number 330) based on one or more criteria.

In addition, the information processing apparatus 100 may apply the CNN to the second information converted into matrix information in a two-dimensional format, as in an example shown in identification number 340 (eg, refer to identification number 340), CNN-based Through deep learning, information characteristic determination models can be created and improved by learning.

In addition, the information processing apparatus 100 may determine the attribute of the first information based on the local characteristics between the positions of the information included in the second information, as in an example shown in the identification number 340, for example, It is possible to determine the probability of a customer's purchase of a particular product (eg, see identification number 350).

FIG. 4 is a diagram for explaining an example of an operation in which the information processing apparatus 100 of FIG. 1 evaluates a result of determining a characteristic of first information by applying a plurality of learning models.

Referring to FIG. 4 , the information processing apparatus 100 applies the customer's purchase information to a plurality of learning models, and predicts the customer's purchase probability for each category using each learning model whose performance is improved through learning. It is determined, and a performance index for each learning model can be calculated by comparing the predicted value with the actual value.

In an embodiment, the plurality of learning models may include at least one of RF, LighGBM, MLP, DNN, one-dimensional CNN, and two-dimensional CNN. For example, when the learning model is CNN-based, the information processing apparatus 100 converts first information expressed in a one-dimensional format to second information expressed in a matrix or image in a two-dimensional format according to the above-described method. By applying the CNN to the transformed second information, it is possible to determine the predicted value of the purchase probability of the customer for each category. In addition, when the learning model is based on a learning algorithm other than CNN, the information processing device 100 applies the learning algorithm to the first information expressed in a one-dimensional format to predict the customer's purchase probability for each category can be decided

In an embodiment, the information processing apparatus 100 may calculate a first performance indicator based on a mean squared error (MSE). Here, the MSE is an average of the squares of the difference between the actual value and the predicted value, and the smaller the value, the better the prediction of the actual value.

In an embodiment, the information processing apparatus 100 may calculate a lift indicating the second performance indicator according to Equation 1 below. For example, the lift represents the ratio of how many more products in the category were actually purchased by the top customers based on the purchase score predicted by each model compared to the randomly selected customers, and a larger value indicates a better prediction of the actual value. can Accordingly, the information processing apparatus 100 may calculate a performance index with higher precision by evaluating whether a customer with a high number of predicted products purchases a product.

[Equation 1]

(Where, N _T represents the total number of customers, N _C represents the number of customers who purchased a product in the corresponding category, N _X represents the number of customers corresponding to the top X% of the purchasing score predicted by the model, N _XC represents the number of customers who actually purchased the product in the category among the top customers, and X can be set by the user)

As an example of the test result shown in FIG. 4 , in accordance with an embodiment of the present invention, when the first information in a one-dimensional format including a categorical variable is converted into second information in a two-dimensional format and applied to the CNN, high It can be seen that the lift value and the low MSE value are calculated, showing high performance.

That is, according to an embodiment of the present invention, by efficiently compressing numerous variables including categorical variables in a two-dimensional format and reflecting them in CNN, it is efficient in terms of storage capacity and high performance improvement of the learning model can also be achieved. , it is possible to read local correlations even for categorical variables with randomness that are not related between adjacent bits because there is no vertical, horizontal, and vertical relationship, so that meaningful data analysis can be efficiently performed on numerous categorical variables.

FIG. 5 is a diagram for explaining an example of an operation in which the information processing apparatus 100 of FIG. 1 determines one or more criteria used for rearranging first information.

Referring to FIG. 5 , when the type of information included in the first information satisfies any one of a plurality of pre-stored conditions, the information processing apparatus 100 according to an embodiment uses a conversion method corresponding to the condition. It may be decided to rearrange the first information. Here, the plurality of conditions means the number of cases according to combinations of types of information included in the first information.

In one embodiment, each transformation method includes one or more criteria for reordering the first information, for example, as shown in FIG. 5 , whether to present time information, presenting demographics data of the customer It may include information on at least one of a method, an overall information presentation method, a neural network, and a priority.

In one embodiment, whether to present time information indicates information on a method of presenting time information (eg, by inquiry order, by date, high density, low density, etc.) in the process of rearranging the first information in a two-dimensional format, and the customer's The characteristic information presentation method indicates information on a method of presenting the first information (eg, separate reflection, mixing, use of a separate channel among RGB channels), and the total information presentation method indicates a method of arranging the entire information in a matrix (eg: order presentation, random presentation), the neural network indicates information about the CNN neural network structure (eg, one-dimensional, two-dimensional) to be used for deep learning, and the priority is the weight given to the transformation method to be determined preferentially shows information about

For example, when the number of clicks of a customer by category is included among the customer's personal information and the customer's behavior information included in the first information, the processor 120 sets whether to present the time information as 'non-present', and the customer's characteristic information According to the first transformation method (eg BCL, see Fig. 3a) which determines the presentation method as 'shuffle', the overall information presentation method as 'order presentation', and the neural network as 'two-dimensional', the customer ID and the last 14 Information on the number of clicks of customers by category during the day can be sequentially arranged in a two-dimensional matrix. That is, the first information can be rearranged in a two-dimensional format by determining the criteria so that the information on the number of clicks by customer ID and the number of clicks by category is mixed and presented in one matrix without being based on time, and each is presented sequentially. .

The information processing apparatus 100 according to an embodiment may preferentially determine a conversion method having a higher preset priority when the type of information included in the first information satisfies two or more of a plurality of pre-stored conditions.

In one embodiment, the name (eg, ACL, BCR, etc.) of the conversion method shown in FIG. 2 is A if time information is presented, B if time information is not presented, C if the number of channels used among RGB channels is one, channel If the number is three, it can be managed by adding D, if there is correlation information in the order of information presented according to the order presentation, L if there is no correlation information, and R is added. We can assist you in making it easier to check.

6 is a view for explaining various embodiments of an operation in which the information processing apparatus 100 of FIG. 1 converts first information into a two-dimensional format based on one or more criteria to obtain second information.

Referring to FIG. 6A , the information processing apparatus 100 according to the first embodiment converts the first information into a two-dimensional format based on the customer's personal information and information on the number of clicks for each category during the customer's preset period. can be converted to

In the first embodiment, the first information includes the customer's personal information (eg, customer ID) and information on the number of clicks for each category during the customer's preset period (eg, the last 14 days), information processing The device 100 may obtain matrix information in which the first information is arranged by distinguishing the personal information of the customer and the information on the number of clicks for each category during the preset period of the customer.

For example, the information processing apparatus 100 inputs each customer's personal information, usage information, and preference information for each category included in the first information as dots according to a specific location in a predetermined order to form a scatter plot. Conversion may be performed according to a first conversion method (eg, BCL) indicated in the form, specifically, each customer's personal information (eg, customer ID) and site usage information are dummy, and the 14 It is possible to obtain matrix information in a two-dimensional format by sequentially arranging the elements of a vector that sums up 2nd-grade category information of recently viewed products that have occurred on a daily basis in a matrix of a preset size (refer to identification number 601).

For example, the information processing device 100 first sequentially arranges a predetermined number (eg, number of customers) of personal information (eg, customer ID) for each customer in a matrix order, and then a predetermined number (eg, 2) The number of clicks for each category of rating categories) may be sequentially arranged.

Referring to FIG. 6B , the information processing apparatus 100 according to the second embodiment converts the first information into a two-dimensional format based on the customer's personal information and information on the number of clicks for each category during the customer's preset period. can be converted to

In the second embodiment, the first information includes the customer's personal information (eg, customer ID) and information on the number of clicks for each category during the customer's preset period (eg, the last 14 days), information processing The apparatus 100 may obtain first information matrix information in which personal information of the customer and information on the number of clicks for each category during a preset period of the customer are randomly arranged.

For example, the information processing apparatus 100 inputs the personal information of each customer, usage information, and preference information for each category included in the first information as dots according to a specific location at random without a set order, and displays the first information in the form of a scatter diagram. 2 Transformation is performed according to a transformation method (eg, BCR), and specifically, matrix information in a two-dimensional format is obtained by randomly arranging elements of the matrix generated according to the first embodiment in a matrix of a preset size. may be obtained (see identification number 602).

For example, the information processing device 100 may store personal information (eg, customer ID) of a preset number (eg, number of customers) of each customer in matrix order and a preset number (eg, number of second-class categories) in the matrix. The number of clicks per category can be randomly arranged.

Accordingly, the information processing apparatus 100 arranges the elements of the matrix at random positions and uses the matrix information from which the correlation information between adjacent pixels reflected in the order of variables, etc. is removed as a learning data set to determine the customer's purchase attribute. Determination accuracy can be further improved.

Referring to FIG. 6C , the information processing apparatus 100 according to the third embodiment may convert the first information into a two-dimensional format based on a click order and a product category.

In a third embodiment, the one or more criteria described above may include a click order and a category for products, and the matrix information may include information indicating a customer's click order for each category.

For example, the information processing device 100 records each customer's personal information, usage information, and category preference information included in the first information at each customer's product inquiry (or site use) occurrence time to indicate the order. Conversion is performed according to the third conversion method (eg, ACL), and specifically, for each customer recently viewed product, from the last inquiry date of the category to the most recent inquiry date, the customer's personal information and use of the site Information can be entered in every row of hits, and the total image length can be set to the maximum number of recently viewed products of 54. In this way, category information is displayed over time for the click order of customers for each category. By disposing, matrix information in a two-dimensional format can be obtained (refer to identification number 603).

For example, the information processing device 100 applies a time criterion for indicating a click order to a first axis (eg, X-axis) of the matrix, and a criterion of category information to a second axis (eg, Y-axis) , and by applying the criterion of whether or not the customer clicks to each matrix value, information on whether the customer clicks by category by time can be sequentially arranged in the matrix. In an embodiment, the type and arrangement order of the category may be initially randomly determined and then be fixedly used.

Referring to FIG. 6D , the information processing apparatus 100 according to the fourth embodiment may convert the first information into a 2D format based on a date and a category for a product. In the fourth embodiment, the above-described one or more criteria may include categories for dates and products, and the matrix information may include information on the number of clicks of customers for each category corresponding to each date.

For example, the information processing apparatus 100 adds up each customer's personal information, usage information, and category-specific preference information included in the first information by day, and records and displays the customer's information change or occurrence history by date. The conversion is performed according to the conversion method (eg, A2CL), and specifically, the category information of the matrix generated according to the third embodiment is summed for each product generated on the same day, and a preset number (eg, 14) It is possible to obtain matrix information in a two-dimensional format by arranging it in a column of , and inputting personal information into all columns (refer to identification number 604).

For example, the information processing device 100 applies the criterion of the date to the first axis (eg, the X-axis) of the matrix, applies the criterion of the category information to the second axis (eg, the Y-axis), and each matrix By applying the criterion of the number of clicks of the customer to the value, information on the number of clicks of the customer by date and category can be sequentially arranged in the matrix.

Referring to FIG. 6E , the information processing apparatus 100 according to the fifth embodiment may convert the first information into a two-dimensional format based on a business type for time and purchase history. In the fifth embodiment, the above-mentioned one or more criteria may include a business type for time and purchase history, and the matrix information may include information on a purchase amount of a customer for each type of business corresponding to each time zone.

For example, the information processing device 100 divides and sums up the amount of use of affiliated stores for each customer and each industry by month, included in the first information, by day and time period, and sums up each customer's recent industry usage and usage amount by industry in one matrix. According to the fifth transformation method of arranging, matrix information in a two-dimensional format may be obtained (refer to identification number 605).

For example, the information processing device 100 applies the standard of the industry to the first axis (eg, the X-axis) of the matrix, and applies the standard of the day (or time zone) to the second axis (eg, the Y-axis), , by applying the standard of the customer's purchase amount to each matrix value, information on the customer's purchase amount for each business day (or time zone) can be sequentially arranged in the matrix.

Referring to FIG. 6F , the information processing apparatus 100 according to the sixth embodiment may convert the first information into a two-dimensional format based on a business type for time and purchase history. In the sixth embodiment, the above-mentioned one or more criteria may include industry types for time and purchase history, and the matrix information may include information on the number of purchases by customers for each industry corresponding to each time zone.

For example, the information processing device 100 divides and sums the number of times of use of affiliated stores by business type by month for each customer included in the first information by day and time period and sums up each customer's recent business use and frequency of use by business type in one matrix. According to the sixth transformation method of arranging, matrix information in a two-dimensional format may be obtained (refer to identification number 606).

For example, the information processing device 100 applies the standard of the industry to the first axis (eg, the X-axis) of the matrix, and applies the standard of the day (or time zone) to the second axis (eg, the Y-axis), , by applying the criterion of the number of purchases of customers to each matrix value, information on the number of purchases by customers by day (or time zone) for each industry type may be sequentially arranged in the matrix.

Mattress information obtained according to the fifth and sixth embodiments may be used in an information characteristic determination model for predicting the availability of each industry by customer time period, and multiple matrix information using one or more image channel characteristics among RGB channels. may be displayed overlaid on the image of , and the above-described time zone and the number of days may be adjusted according to the information summation criteria.

Referring to FIG. 6G , the information processing apparatus 100 according to the seventh embodiment may convert the first information into a two-dimensional format based on a business type for time and purchase history. In the seventh embodiment, the above-described one or more criteria include industry for time and purchase history, and the matrix information includes information on the customer's purchase amount and number of purchases for each industry corresponding to each time zone. can

For example, the information processing device 100 collects purchase information for each customer included in the first information in units of a preset period (eg, in units of one month or one week), and collects them in two or more channels (eg, in the third dimensional direction) to obtain matrix information in a two-dimensional format according to the seventh transformation method representing a customer's monthly or weekly purchasing pattern (refer to identification number 607).

For example, the information processing device 100 applies the standard of the industry to the first axis (eg, the X-axis) of the matrix, and applies the standard of the day (or time zone) to the second axis (eg, the Y-axis), , by applying the criterion of period (eg, one week) to the third axis (eg, Z-axis), and applying the criterion of the number of customers' purchases to each matrix value, by day (or time period) by business type in each matrix By sequentially arranging information on the number of purchases (or purchase amount) of the customer, and arranging a plurality of matrices along the third axis, a weekly (or monthly) purchase pattern may appear in the matrix information.

The mattress information obtained according to the seventh embodiment may be used in an information characteristic determination model for predicting the purchase possibility by time of the next day of the week of the customer, and the above-described time period and the number of days may be adjusted according to the information summation criteria. .

Referring to FIG. 6H , the information processing apparatus 100 according to the eighth embodiment may convert the first information into a two-dimensional format based on a commercial area for time and purchase history. In the eighth embodiment, the above-mentioned one or more criteria may include commercial districts for time and purchase history, and the matrix information may include information on whether the customer purchases each commercial district corresponding to each state.

For example, the information processing device 100 codes whether the customer purchases for each commercial district in which the affiliate store where the customer's purchase occurs for each preset period (eg, each state) included in the first information, and provides location information for one year. , and it is possible to obtain matrix information in a two-dimensional format according to the eighth transformation method of arranging whether customers purchase or not for each state in a matrix (refer to identification number 608).

For example, the information processing device 100 applies the criterion of the commercial district to the first axis (eg, X-axis) of the matrix, and applies the criterion of attention to the second axis (eg, the Y-axis), and applies the criterion of each business district to the matrix. Information on whether a customer has purchased by week can be sequentially arranged.

Referring to FIG. 6I , the information processing apparatus 100 according to the ninth embodiment may convert the first information into a two-dimensional format based on time and a commercial district for a purchase history. In the ninth embodiment, the one or more criteria described above may include commercial districts for time and purchase history, and the matrix information may include information on the number of purchases of the customer for each commercial district corresponding to each state. .

For example, the information processing device 100 codes the number of purchases of the customer for each commercial district in which the affiliate store where the customer's purchase occurred for each preset period (eg, each state) included in the first information, thereby providing location information for one year. can be analyzed, and matrix information in a two-dimensional format can be obtained according to the ninth transformation method of arranging the number of purchases by customers for each state in a matrix (refer to identification number 609).

For example, the information processing device 100 applies the criterion of the commercial district to the first axis (eg, X-axis) of the matrix, and applies the criterion of attention to the second axis (eg, the Y-axis), and applies the criterion of each business district to the matrix. Information on the number of purchases by customers per week can be sequentially arranged.

Mattress information obtained according to the eighth and ninth embodiments may be used in an information characteristic determination model for predicting where the customer will be by day of the week and by reflecting this in the learning of the customer's preference history in the business district. The characterization model can be improved. In addition, information characteristic determination model for predicting the purchase possibility considering the customer's expected location and past location history, the information characteristic determination model for predicting the visiting customer group by commercial area, and information for predicting the visitability by location history by customer It can be used in the characteristic determination model, and the accuracy can be improved by reflecting the expected location for each customer TPO in the targeting.

Referring to FIG. 6J , the information processing apparatus 100 according to the tenth embodiment may convert the first information into a two-dimensional format based on the day of the week and the time zone. In the tenth embodiment, the above-described one or more criteria may include a day and time zone, and the matrix information may include information on the number of purchases of customers corresponding to each day and each time zone.

For example, the information processing device 100 for each customer included in the first information for each preset period (eg, monthly or weekly) for purchases that occurred within the period for each type of business, all purchases, or major business types are arranged according to time series According to the tenth transformation method of summing the number of purchases in the form of , and arranging them in a matrix, matrix information in a two-dimensional format can be obtained (refer to identification number 610).

For example, the information processing device 100 applies the time zone standard to the first axis (eg, X-axis) of the matrix, and applies the reference of the day of the week to the second axis (eg, Y-axis), and displays the time in the matrix. Information on the sum of the number of purchases by customers for each unit and day of the week may be sequentially arranged.

The mattress information obtained according to the tenth embodiment may be used in an information characteristic determination model for predicting a customer's purchase occurrence pattern, time and day usage pattern. In addition, the information processing device 100 separates several industries (eg, frequent industries such as convenience stores, online, restaurants, etc.) selected by a user (eg, an administrator) by channels of R, G, and B and arranges them in respective matrices You may.

Referring to FIG. 6K , the information processing apparatus 100 according to the eleventh embodiment may convert the first information into a two-dimensional format based on a commercial district and a business type for a purchase history. In the eleventh embodiment, the above-described one or more criteria may include commercial districts and industries for the purchase history, and the matrix information may include information on the number of purchases of the customer corresponding to each commercial district and each industry.

For example, the information processing device 100 sums up and codes the number of purchases of customers for each business district for each industry with respect to the purchase history that has occurred in the last three months for each customer included in the first information, and codes the number of purchases of the customer by industry according to the major commercial districts. According to the eleventh transformation method indicating the number of purchases, matrix information in a two-dimensional format may be obtained (refer to identification number 612).

For example, the information processing device 100 applies the standard of the business district to the first axis (eg, the X-axis) of the matrix, and applies the standard of the industry to the second axis (eg, the Y-axis), and applies the standard of the business district to the matrix. Information on the sum of the number of purchases by customers for each industry may be sequentially arranged. In an embodiment, the commercial district may include an administrative district.

Accordingly, the information processing device 100 derives matrix information on which type of business the customer mainly consumes in which commercial area, and selects several types of business (eg, convenience store, online, restaurant, etc.) by the user (eg, manager). Frequent industries) may be separated by channels of R, G, and B and arranged in each matrix.

The mattress information obtained according to the eleventh embodiment may be used in an information characteristic determination model for analyzing which type of business the customer mainly consumes in which commercial area and predicting the purchase possibility of the customer by each type of business area.

Referring to FIG. 6L , the information processing apparatus 100 according to the twelfth embodiment may convert the first information into a two-dimensional format based on time, a commercial district and a business type for purchase history.

In the twelfth embodiment, the above-described one or more criteria include time, a commercial district and a business type for purchase history, and the matrix information is information about the customer's purchase amount or number of purchases for each industry corresponding to each time zone. First matrix information including, the second matrix information including information on whether or not the customer purchases or the number of purchases for each commercial district corresponding to each state, and the customer corresponding to each commercial district and each industry may include third matrix information including information on the number of purchases.

For example, the information processing device 100 may determine the type of business, day, time, and date of the customer's purchase history that occurred during the period according to the period division for each preset period (eg, monthly, weekly, yearly, etc.) included in the first information. The commercial district is divided into two pairs, and the number of purchases that occur in each area is summed and coded, and the generated three matrices are converted to 2 according to the twelfth transformation method indicating the number of purchases by customers along the axes of time, commercial district, and industry. It is possible to obtain matrix information in a dimension format (refer to identification number 612).

For example, the information processing device 100 applies the standard of the day (or time) to the first axis (eg, the X-axis) of the matrix, and applies the standard of the business district to the second axis (eg, the Y-axis), , by applying the industry standard to the third axis (eg, Z axis), sequentially arranging information about the sum of the number of purchases by customers along the corresponding axis in the matrix, and arranging multiple matrices along multiple axes have.

The mattress information obtained according to the twelfth embodiment may be used in an information characteristic determination model for analyzing which type of business and at which time period the customer mainly consumes in which commercial district and predicting the purchase possibility of the customer for each time period by business district and industry.

Referring to FIG. 6M , the information processing apparatus 100 according to the thirteenth embodiment may convert the first information based on time, a commercial district and a business type for purchase history.

For example, the information processing apparatus 100 obtains the first matrix information, the second matrix information, and the third matrix information in a two-dimensional format according to each of the above-described fifth transformation method, ninth transformation method, and eleventh transformation method. You can, apply the standard of the day (or time) to the first axis (eg, X-axis), apply the standard of the business district to the second axis (eg, Y-axis), and apply the standard of the third axis (eg, Z-axis) ) by applying the industry standard, the first matrix information, the second matrix information, and the third matrix information may be aligned along each axis.

Referring to FIG. 6N , the information processing apparatus 100 according to the fourteenth embodiment may convert the first information based on time, a commercial district and a business type for purchase history.

For example, the information processing apparatus 100 obtains three pieces of matrix information in a two-dimensional format for each period having the same time section according to any one of the above-described first to fourth transformation methods, R, G, and B can be sorted into multiple channels of

For example, the purchase attribute determination device 100 collects category information purchased by the customer for each customer in the same period (eg, daily, weekly, monthly, etc.), and collects the purchase category information in the first week of the R channel , the category information purchased in the 2nd week can be sorted into the G channel, and the category information purchased in the 3rd week can be sorted by the B channel.

Accordingly, the purchase attribute determining apparatus 100 may provide an effect of reducing the abbreviation by dividing values abbreviated to one value regardless of the viewpoint information based on the viewpoint information. That is, although the amount of information to be stored increases, the storage capacity may be maintained or reduced, so that an image may be generated in a more efficient manner.

The mattress information obtained according to the fourteenth embodiment may be used in an information characteristic determination model for predicting category information to be purchased by the borrower based on category information purchased by the customer during a specific period (eg, the last three weeks).

Meanwhile, the above-described method can be written as a program that can be executed on a computer, and can be implemented in a general-purpose digital computer that operates the program using a computer-readable recording medium. In addition, the structure of the data used in the above-described method may be recorded in a computer-readable recording medium through various means. The computer-readable recording medium includes a storage medium such as a magnetic storage medium (eg, ROM, RAM, USB, floppy disk, hard disk, etc.) and an optically readable medium (eg, CD-ROM, DVD, etc.) do.

The above description of the present invention is for illustration, and those of ordinary skill in the art to which the present invention pertains can understand that it can be easily modified into other specific forms without changing the technical spirit or essential features of the present invention. will be. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. For example, each component described as a single type may be implemented in a dispersed form, and likewise components described as distributed may be implemented in a combined form.

The scope of the present invention is indicated by the following claims, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts should be construed as being included in the scope of the present invention.

100: information processing device
110: communication unit 120: processor
130: storage

Claims (21)

An information processing method comprising:
receiving first information expressed in a one-dimensional format and including a categorical variable;
determining one or more criteria used for rearranging the first information based on a characteristic of the categorical variable included in the first information;
converting the first information into a two-dimensional format based on the one or more criteria to obtain second information; and
Determining a characteristic of the first information by applying a Convolutional Neural Network (CNN) to the second information converted into the two-dimensional format.
The method of claim 1,
The step of determining the characteristic of the first information by applying the CNN
determining an attribute of the first information based on a local characteristic between locations of the information included in the second information.
3. The method of claim 2,
the categorical variable includes a nominal variable,
The first information includes a bitstream in which information on variables indicated by the nominal variable is sequentially provided.
4. The method of claim 3,
The categorical variable further includes an ordinal variable,
wherein the one or more criteria comprises an order by which the ordinal variable is based.
The method of claim 1,
The first information includes purchase information, and the categorical variable includes category information about a product for which purchase occurs,
wherein the one or more criteria comprises the category.
The method of claim 1,
The first information includes purchase information, and the categorical variable includes information on a commercial area or industry in which a purchase occurs,
The one or more criteria include the commercial district or the industry, the method.
The method of claim 1,
The second information includes the first information,
The step of determining the characteristic of the first information by applying the CNN
filtering the second information to obtain main information; and
determining a characteristic of the first information based on the primary information.
The method of claim 1,
classifying the first information into a first matrix, a second matrix and a third matrix according to the one or more criteria;
obtaining an image in which the first matrix, the second matrix, and the third matrix are represented by R, G, and B, respectively; and
determining the characteristic of the first information by applying the CNN to the image.
5. The method of claim 4,
The first information further includes a numeric variable,
obtaining, from the first information, 1-1 information consisting of the categorical variable and 1-2 information consisting of the numerical variable;
obtaining 2-1 information by converting the 1-1 information into a two-dimensional format based on the characteristics of the categorical variable;
obtaining 2-2 information by converting the 1-2 information into a two-dimensional format based on the characteristics of the numeric variable;
Based on the inter-location local characteristics of the information included in the 2-1 information and the inter-location local characteristics of the information included in the 2-2 information, the 2-1 information and the 2-2 information are applying each of the CNNs; and
determining the characteristic of the first information using a result of applying the CNN to the 2-1 information and the 2-2 information.
10. The method of claim 9,
obtaining, from the 1-1 information, 1-3 th information composed of the nominal variable and 1-4 th information composed of the ordinal variable;
converting the 1-3 th information into a two-dimensional format to obtain 2-3 th information based on the characteristic of the nominal variable;
converting the 1-4th information into a two-dimensional format to obtain 2-4th information based on the characteristics of the ordinal variable;
Based on the inter-location local characteristics of the information included in the 2-3 information and the inter-location local characteristics of the information included in the 2-4 information, the 2-3 information and the 2-4 information applying each of the CNNs; and
determining the characteristic of the first information using a result of applying the CNN to the 2-3th information and the 2-4th information;
An information processing device comprising:
a communication unit for receiving first information expressed in a one-dimensional format and including a categorical variable; and
One or more criteria used for rearranging the first information are determined based on the characteristics of the categorical variable included in the first information, and the first information is converted into a two-dimensional format based on the one or more criteria. A processor that acquires second information and determines a characteristic of the first information by applying a Convolutional Neural Network (CNN) to the second information converted into the two-dimensional format.
12. The method of claim 11,
the processor
and determine an attribute of the first information based on a local characteristic between locations of the information included in the second information.
13. The method of claim 12,
the categorical variable includes a nominal variable,
The first information includes a bitstream in which information on variables indicated by the nominal variable is sequentially provided.
14. The method of claim 13,
The categorical variable further includes an ordinal variable,
wherein the one or more criteria comprises an order in which the ordinal variable is based.
12. The method of claim 11,
The first information includes purchase information, and the categorical variable includes category information about a product for which purchase occurs,
wherein the one or more criteria comprises the category.
12. The method of claim 11,
The first information includes purchase information, and the categorical variable includes information on a commercial area or industry in which a purchase occurs,
The one or more criteria include the commercial district or the industry, the device.
12. The method of claim 11,
The second information includes the first information,
the processor
The apparatus of claim 1, wherein the second information is filtered to obtain main information, and a characteristic of the first information is determined based on the main information.
12. The method of claim 11,
the processor
classifying the first information into a first matrix, a second matrix and a third matrix according to the one or more criteria;
obtaining an image in which the first matrix, the second matrix and the third matrix are represented by R, G, and B, respectively;
Applies the CNN to the image to determine the characteristic of the first information.
15. The method of claim 14,
The first information further includes a numeric variable,
the processor
Obtaining 1-1 information consisting of the categorical variable and 1-2 information consisting of the numerical variable from the first information,
converting the 1-1 information into a two-dimensional format based on the characteristics of the categorical variable to obtain 2-1 information,
Converting the 1-2 information into a two-dimensional format based on the characteristics of the numeric variable to obtain 2-2 information,
Based on the inter-location local characteristics of the information included in the 2-1 information and the inter-location local characteristics of the information included in the 2-2 information, the 2-1 information and the 2-2 information are Each of the CNNs is applied,
An apparatus for determining a characteristic of the first information by using a result of applying the CNN to the 2-1 information and the 2-2 information.
20. The method of claim 19,
the processor
obtaining 1-3 th information including the nominal variable and 1-4 th information including the ordinal variable from the 1-1 information;
transforming the 1-3 information into a two-dimensional format based on the characteristic of the nominal variable to obtain the 2-3 information;
converting the 1-4th information into a two-dimensional format based on the characteristics of the ordinal variable to obtain 2-4th information,
Based on the inter-location local characteristics of the information included in the 2-3 information and the inter-location local characteristics of the information included in the 2-4 information, the 2-3 information and the 2-4 information Each of the CNNs is applied,
and determining a characteristic of the first information by using a result of applying the CNN to the 2-3th information and the 2-4th information.
A computer-readable recording medium in which a program for executing the method of any one of claims 1 to 10 in a computer is recorded.

Images