KR102119518B1 - Method and system for recommending product based style space created using artificial intelligence - Google Patents

Abstract

It provides a product recommendation method and system based on a style space created using artificial intelligence. The product recommendation method according to embodiments of the present invention uses MDAE (Multimodal Deep Autoencoder) to learn the correlation of products by category included in the learning image, and based on the correlation learned in the MDAE (Multimodal Deep Autoencoder). You can recommend products.

Description

Product recommendation method and system based on style space created using artificial intelligence {METHOD AND SYSTEM FOR RECOMMENDING PRODUCT BASED STYLE SPACE CREATED USING ARTIFICIAL INTELLIGENCE}

The following description relates to a product recommendation method and system based on a style space created using artificial intelligence.

The fashion coordinator system is a system that helps a user to select another garment suitable for the wearer to select one, and is a system for recommending a product or service. As described above, the system for recommending a product or service is largely divided into a collaborative filtering method using similarity among users and a content-based filtering method using characteristic values of products and users. Are distinguished. For example, Korean Patent Registration No. 10-0882716 discloses a technique for automatically recommending product information about a product using a product code, product name, or model name of the product recognized by the information recommendation agent installed on the user terminal.

Product recommendation method for learning the correlation of products by category included in the learning image using MDAE (Multimodal Deep Autoencoder), and recommending a product based on the correlation learned in MDAE (Multimodal Deep Autoencoder), the product A computer device performing a recommendation method, a computer program coupled to a computer, and a computer program stored in a computer-readable recording medium to execute the product recommendation method on a computer, and a recording medium therefor.

Extracting products for each category included in the learning image set; Learning a multimodal deep autoencoder (MDAE)-based product recommendation model to project the extracted category-specific products at a point on a style space based on a correlation between product categories by category included in the same learning image; Projecting a plurality of recommendation target products to a plurality of first points on the style space through the learned product recommendation model; Projecting the product to be searched to the second point on the style space through the learned product recommendation model; And selecting at least one recommendation target product among the plurality of recommendation target products based on the distance between each of the first points and the second point in the style space. to provide.

In combination with a computer, a computer program stored in a computer-readable recording medium is provided to execute the product recommendation method on a computer.

Provided is a computer-readable recording medium, characterized in that a program for executing the product recommendation method is recorded on a computer.

A computer device, comprising: at least one processor implemented to execute instructions readable by a computer, and by the at least one processor, extracting products by category included in a set of learning images, and included in the same learning image Based on the correlation between products by category, a multimodal deep autoencoder (MDAE)-based product recommendation model is trained to project the extracted category-specific products to a point in a style space, and the plurality of recommendation target products are Projected to a plurality of first points in the style space through the learned product recommendation model, projected the product to be searched to the second point in the style space through the learned product recommendation model, and the product in the style space. Provided is a computer device characterized in that at least one of the plurality of recommendation target products is selected based on the distance between each of the one point and the second point.

MDAE (Multimodal Deep Autoencoder) can be used to learn the correlation of products for each category included in the learning image, and products can be recommended based on the correlation learned in MDAE (Multimodal Deep Autoencoder).

By using the features of the reconstructed products reconstructed through the MDAE decoder as the correct answer data, by learning the product recommendation model without using the incorrect answer data, the negative sampling process, which has the largest amount of computation in the existing distance model, can be omitted. Can be.

1 is a diagram showing an example of a network environment according to an embodiment of the present invention.
2 is a block diagram showing an example of a computer device according to an embodiment of the present invention.
3 is a flowchart illustrating an example of a product recommendation method according to an embodiment of the present invention.
4 is a view showing an example of MDAE according to an embodiment of the present invention.
5 and 6 are views showing an example of a distance model.
7 is a diagram illustrating an example of a style space generation form according to an embodiment of the present invention.
8 is a diagram illustrating an example of learning MDAE in an embodiment of the present invention.
9 is a view showing an example of a style space according to an embodiment of the present invention.

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

The product recommendation method according to the embodiments of the present invention may be implemented through a computer device such as an electronic device or a server to be described later. At this time, a computer program according to an embodiment of the present invention may be installed and driven on the computer device, and the computer device may perform a product recommendation method according to embodiments of the present invention under the control of the driven computer program. . The above-described computer program may be stored in a computer-readable recording medium to execute a product recommendation method on a computer in combination with a computer device.

1 is a diagram showing an example of a network environment according to an embodiment of the present invention. The network environment of FIG. 1 shows an example including a plurality of electronic devices 110, 120, 130, 140, a plurality of servers 150, 160, and a network 170. 1 is not limited to the number of electronic devices or the number of servers as an example for describing the invention. In addition, the network environment of FIG. 1 is merely an example of one of the environments applicable to the embodiments, and the environment applicable to the embodiments is not limited to the network environment of FIG. 1.

The plurality of electronic devices 110, 120, 130, and 140 may be a fixed terminal or a mobile terminal implemented as a computer device. For example, a plurality of electronic devices (110, 120, 130, 140), smart phones (smart phone), mobile phones, navigation, computers, notebooks, digital broadcasting terminals, PDAs (Personal Digital Assistants), PMP (Portable Multimedia Player) , Tablet PC and so on. For example, although the shape of a smartphone is shown as an example of the electronic device 1 110 in FIG. 1, in the embodiments of the present invention, the electronic device 1 110 substantially uses the wireless or wired communication method to connect the network 170. It may mean one of various physical computer devices capable of communicating with other electronic devices 120, 130, 140 and/or servers 150, 160.

The communication method is not limited, and a communication method using a communication network (for example, a mobile communication network, a wired Internet, a wireless Internet, and a broadcasting network) that the network 170 may include may include short-range wireless communication between devices. For example, the network 170 includes a personal area network (PAN), a local area network (LAN), a campus area network (CAN), a metropolitan area network (MAN), a wide area network (WAN), and a broadband network (BBN). , Any one or more of the networks such as the Internet. Further, the network 170 may include any one or more of a network topology including a bus network, a star network, a ring network, a mesh network, a star-bus network, a tree, or a hierarchical network. It is not limited.

Each of the servers 150 and 160 communicates with a plurality of electronic devices 110, 120, 130, and 140 through a network 170 to provide commands, codes, files, contents, services, or the like, or a plurality of computers. It can be implemented with devices. For example, the server 150 is a service (for example, a social network service, a messaging service, a search service, a mail service) with a plurality of electronic devices 110, 120, 130, and 140 accessed through the network 170. It may be a system that provides a content providing service.

2 is a block diagram showing an example of a computer device according to an embodiment of the present invention. Each of the plurality of electronic devices 110, 120, 130, 140 described above, or each of the servers 150, 160 may be implemented by the computer device 200 shown through FIG. 2, according to an embodiment. The product recommendation method may be performed by a product recommendation system implemented by the computer device 200.

2, the computer device 200 may include a memory 210, a processor 220, a communication interface 230, and an input/output interface 240. The memory 210 is a computer-readable recording medium, and may include a non-permanent mass storage device such as random access memory (RAM), read only memory (ROM), and a disk drive. Here, a non-destructive large-capacity recording device such as a ROM and a disk drive may be included in the computer device 200 as a separate permanent storage device separate from the memory 210. In addition, an operating system and at least one program code may be stored in the memory 210. These software components may be loaded into the memory 210 from a computer-readable recording medium separate from the memory 210. Such a separate computer-readable recording medium may include a computer-readable recording medium such as a floppy drive, disk, tape, DVD/CD-ROM drive, and memory card. In other embodiments, software components may be loaded into memory 210 through communication interface 230 rather than a computer-readable recording medium. For example, software components can be loaded into the memory 210 of the computer device 200 based on a computer program installed by files received over the network 170.

The processor 220 may be configured to process instructions of a computer program by performing basic arithmetic, logic, and input/output operations. Instructions may be provided to processor 220 by memory 210 or communication interface 230. For example, the processor 220 may be configured to execute a received command according to program code stored in a recording device such as the memory 210.

The communication interface 230 may provide a function for the computer device 200 to communicate with other devices (eg, the storage devices described above) through the network 170. For example, requests, commands, data, files, etc. generated by the processor 220 of the computer device 200 according to program codes stored in a recording device such as the memory 210 are controlled by the communication interface 230. 170) to other devices. Conversely, signals, commands, data, files, etc. from other devices may be received through the network 170 to the computer device 200 through the communication interface 230 of the computer device 200. Signals, commands, data, etc. received through the communication interface 230 may be transferred to the processor 220 or the memory 210, and files and the like may be further stored by the computer device 200 (described above) Permanent storage device).

The input/output interface 240 may be a means for interfacing with the input/output device 250. For example, the input device may include a device such as a microphone, keyboard or mouse, and the output device may include a device such as a display or speaker. As another example, the input/output interface 240 may be a means for interfacing with a device in which functions for input and output are integrated into one, such as a touch screen. The input/output device 250 may be configured as a computer device 200 and a single device.

Further, in other embodiments, the computer device 200 may include fewer or more components than those in FIG. 2. However, there is no need to clearly show most prior art components. For example, the computer device 200 may be implemented to include at least some of the input/output devices 250 described above, or may further include other components such as a transceiver, a database, and the like.

3 is a flowchart illustrating an example of a product recommendation method according to an embodiment of the present invention. The product recommendation method according to the present embodiment may be performed by the computer device 200 described above. For example, the processor 220 of the computer device 200 may be implemented to execute control instructions according to code of an operating system included in the memory 210 or code of at least one program. Here, the processor 220 is the computer device 200 to perform the steps (310 to 370) of the method of Figure 3, the computer device 200 according to the control command provided by the code stored in the computer device 200 Can be controlled.

In step 310, the computer device 200 may extract products for each category included in the learning image set. For example, the computer device 200 may receive a set of learning images including a plurality of learning images, and extract products for each category from each of the plurality of learning images. In this case, the categories may be variously defined as, for example, tops, bottoms, hats, shoes, and the like. The number of predefined categories may also be variously set according to categories of products to be provided. For example, only two categories of tops and bottoms may be used, or a larger number of categories may be used, such as tops, bottoms, hats, accessories, shoes, and the like.

In step 320, the computer device 200 is multimodal deep autoencoder (MDAE) to project the extracted products for each category based on a correlation between products for each category included in the same learning image at a point in the style space. You can train the model based on product recommendation. For example, the learning images are images that include products that match each other. For example, a product image coordinated by a merchandiser (MD) such as a fashion pictorial, magazine, shopping mall, or a seller, or a coordinated evaluation score by general users. The above-mentioned coordination product images and the like can be collected. In other words, the learning image itself may include a correct answer image in which products by category included in the same learning image correlate with each other. The MDAE-based product recommendation model may be implemented to learn correlations between products included in the learning image, and in particular, products that match each other may be trained to be projected at relatively close distance points in the style space. .

As a more specific example, the above-described style space may be classified into a plurality of spaces, and MDAE may be a model that classifies a corresponding product into one of a plurality of spaces based on an image characteristic of an input product. In this case, points in the classified space may be based on the relationship between the spaces. In this case, the computer device 200 projects the products for each category extracted in step 320 to points in the style space, but points to the same space among a plurality of spaces in which products for each category included in the same learning image are classified. You can train the MDAE's encoder to project. Points in the same space will be relatively close, and points in different spaces are likely to be relatively long. Therefore, the MDAE-based product recommendation model according to the present embodiment is configured by arranging products by category included in the same learning image, which are assumed to include products that match each other, in the same space among a plurality of spaces classified on the style space. , It is possible to learn the correlation between products by category included in the same learning image.

In addition, in step 320, the computer device 200 extracts the image feature through each of the subsets excluding the empty set for the set having the products for each category included in the same learning image as an element, and inputs data for the product recommendation model Can generate For example, suppose that the extracted products for each category are products corresponding to tops, products corresponding to bottoms, and products corresponding to shoes. In this case, when defining a set of products with categorized products as elements {top, bottom, footwear}, the subsets excluding the empty set are {top}, {bottom}, {shoes}, {top, bottom}, { 7 sets of top shoes}, {bottom shoes} and {top, bottom, shoes}. In this case, the computer device 200 has image characteristics for the image of the product corresponding to {top}, image characteristics for the image of the product corresponding to {bottom}, and image characteristics for the image of the product corresponding to {shoes} , Image characteristics for images of products corresponding to each of {top, bottom}, image characteristics for images of products corresponding to {footwear}, images of products corresponding to each {bottom} Data of seven image features, such as image features and image features for images of products corresponding to each of {top, bottom, and footwear} may be generated as input data. Depending on the embodiment, only some of the subsets may be utilized.

In addition, the computer device 200 may use the correct answer data for features of products reconstructed through the encoder and decoder of MDAE in step 320, and train the product recommendation model without using additional incorrect data. . As will be explained in more detail later, the features reconstructed through the MDAE encoder and decoder are separate and incorrect data (learning that includes products that do not fit together) because image features for products by category extracted from the same learning image are shared and combined. Even if data is not used, the product recommendation model can be trained to select products that match each other with high accuracy.

In operation 330, the computer device 200 may project a plurality of recommended target products to a plurality of first points in the style space through the learned product recommendation model. For example, the computer device 200 may generate input data for each of a plurality of recommendation target products through a plurality of images including a plurality of recommendation target products, and the product recommendation model learned from the generated input data As a result, a plurality of recommended target products can be projected at a plurality of first points in the style space as input as.

In step 340, the computer device 200 may project the search target product to the second point on the style space through the learned product recommendation model. For example, the computer device 200 may generate input data through an image including the product to be searched, and as a result of inputting the generated input data into the trained product recommendation model, as a result, the second in the style space. You can project the product to be searched for points.

In operation 350, the computer device 200 may select at least one recommendation target product among a plurality of recommendation target products based on the distance between each of the first points and the second point in the style space. As already described, the distance between points in the style space can correspond to the degree of matching between products corresponding to the points, and the computer device 200 selects the recommended target product that goes well with the search target product based on the distance between the points. You can choose.

For example, in step 350, the computer device 200 determines N recommendation targets corresponding to N (N is a natural number) first points in the order of distance from the second point among the plurality of recommendation target products. You can choose a product. In other words, the recommended target product corresponding to the first point closest to the second point on which the searched product is projected may be selected as the recommended target product that best matches the searched product, and the second point and the first product N recommended products may be selected in the order in which the distances of the points are close. At this time, at least one recommendation target product among the selected recommendation target products may be provided to a searcher.

Meanwhile, the recommendation target of the product may be the same category as the search target product or a different category according to the search condition. Or, the searcher may be a specific category. For example, the computer device 200 may select a category of a search target product from among a plurality of recommendation target products based on a distance. As another example, the computer device 200 may select a recommended target product of a category selected by a searcher among a plurality of recommended target products based on a distance.

차원 벡터는 아래 수학식 1과 같이 표현될 수 있다. 4 is a view showing an example of MDAE according to an embodiment of the present invention. Given a set V of images, each image v _i may be an element of set V. At this time, the virtual features of the product of the category Cj of the image v _i

The dimensional vector can be expressed as Equation 1 below.

이미지 v _i 의 j 번째 카테고리의 상품의 가상 특징을 나타내는 각각의 엘리먼트

를 갖는 특성 매트릭스를 나타낼 수 있다. In addition, x ^Cj is

Each element representing the virtual characteristics of the product of the jth category of the image v _i

It can represent a characteristic matrix having a.

Table 1 shows an example of the MDAE algorithm when two categories exist.

차원 벡터를, X ^b 는 이미지 v _i 의 두 가지 카테고리 중 제2 카테고리의 상품의 가상 특징들을 나타내는

차원 벡터를 각각 의미할 수 있다.In Table 1, X ^t represents virtual characteristics of the product of the first category among the two categories of the image v _i

The dimensional vector, X ^b represents the virtual characteristics of the product of the second category of the two categories of the image v _i

Each may mean a dimensional vector.

In FIG. 4, image features 410 for three categories of categories C 1, C 2, and C 3 enter the input of the MDAE 420 and generate a style space according to a shared expression through an encoder network. Here, an example of reconstructing the image feature 410 input through the decoder network again and outputting the reconstructed input 430 is illustrated. The input of the MDAE 420 is formed of 7 subsets (2 ³ -1=7) except for the empty set that can be included in the set of features of categories C 1, C 2, and C 3 as shown in FIG. 4. Can be. If there are 2 categories, the subsets, excluding the empty set of features, will be 3 (2 ² -1=3), and if there are 4 categories, 15 subsets, excluding the empty set of features, (2 ⁴ -1=15).

와

간의 차이를 평가하기 위한 비용 함수 J(W, b)는 아래 수학식 2와 같이 표현될 수 있다.Also, the reconstructed input 430 in FIG. 4

Wow

The cost function J(W, b) for evaluating the difference between can be expressed as Equation 2 below.

5 and 6 are views showing an example of a distance model.

FIG. 5 extracts image characteristics for each product from an image as input data 510, obtains an image that does not match the product, and includes an image set (hereinafter, “correct answer data”) including images of products that match each other and products that do not match each other. An example of learning a style space 530 by generating an image set including images (hereinafter referred to as'incorrect data') as a model input form 520 and inputting it into a product recommendation model is shown. In this case, the MDAE-based product recommendation model may be trained such that products of the correct answer data are projected at points of relatively close distance in the style space, and products of the incorrect data are projected at points of relatively long distance in the style space. .

FIG. 6 shows an example of extracting images of products from the learning image 610 for the product recommendation model and projecting them into the image vector space 620. At this time, the product recommendation model may project each of these products to the style space 630, and as described above, the products of the correct answer data are projected at points of relatively close distance in the style space, and the incorrect data of Products can be learned to project at relatively distant points in the style space.

Meanwhile, as illustrated in FIG. 6, in order to generate information to indicate that products that do not match each other are far apart, a large amount of calculation may be consumed to generate and utilize incorrect data, which is an image set between products that do not match each other, The amount of calculation consumed may increase as the number of categories increases. Therefore, hereinafter, an embodiment in which distances between products can be properly calculated in a style space without generating and utilizing such incorrect data will be described.

7 is a diagram illustrating an example of a style space generation form according to an embodiment of the present invention.

7 shows an example of extracting products by category from an image as input data 710 and generating a model input form 720 by extracting image features for each product of each category. At this time, as described above, the image feature may be extracted through each of the subsets excluding the empty set for the set having the products for each category included in the same learning image as an element to generate input data for the product recommendation model. The model input form 720 of FIG. 7 shows some of the subsets. The input data of the generated model input form 720 may be input 730 as an MDAE-based product recommendation model. At this time, the MDAE-based product recommendation model can project products in a style space based on input data. This model input form 720 allows recommendations for all products of various categories by simply linking image features of products of a plurality of categories, not just image features of two products, to use as input data of MDAE. It can be possible. The style space may be classified into a plurality of spaces, and products extracted from the same learning image may be projected into the style space such that products are located in the same space among the plurality of spaces. In other words, products extracted from the same learning image can be projected at relatively close points. 7 shows that three inputs 740 corresponding to the three products extracted from the input data 710 are located in the same space among the plurality of spaces that classify the style space.

Meanwhile, the correct answer data 750 may be data reconstructed while the input data passes through the MDAE encoder and decoder. MDAE's encoder that creates the style space and projects the products can learn products extracted from the same learning image as points of the same space without additional incorrect data as it is learned based on the correct answer data 750. Therefore, by omitting the process of generating and using incorrect data having the largest amount of calculation in the distance model, learning and recommendation for multiple categories can be possible even in a realistic time.

As described above, MDAE is an application of a denoising autoencoder technology that removes noise from test data containing noise by intentionally adding noise to the training data during training. By adding noise (blind processing), the entire data can be inferred from only one or a whole subset of data. In order to learn that the clothes shown as the correct answer data 750 in FIG. 7 match each other, in the distance model described with reference to FIGS. 5 and 6, information on disagreement must be input, whereas MDAE-based products according to the present embodiment The recommended model is a style by using the data reconstructed using an autoencoder (eg, the reconstructed input 430 of FIG. 4) and the features of the clothes in the shared expression as learning by using the correct answer data 750, style Space can be created.

8 is a diagram illustrating an example of learning MDAE in an embodiment of the present invention. MDAE includes the original data shown in FIG. 8 (a subset containing all elements of the subset of the set of features, 810) and similar data processed blind (a subset containing some elements of the subset of the set of features, As 820 and 830 are learned to be the same data, the value of the shared expression in which the feature is learned can be similarly learned. In this case, when learning about MDAE is completed, correct answer data 750 may be generated from one category of images (blind processed similar data 820 and 830), and the shared representation of the blind processed similar data is correct answer Similar to the shared representation of data 750. In other words, since the shared expressions of each of the subset categories of the correct answer data 750 are all similar, the distances in the style space are close to each other. As a result, the shared expressions of each subset are distributed in the style space according to their own characteristics, but in the case of similar styles (styles from the same correct answer data 750), they are located in spaces close to each other.

As described above, the MDAE-based product recommendation model according to the present embodiment uses the incorrect data sampling in the distance model to forcibly study the distance from the distance when the styles are different, and the characteristics of the existing image without additional incorrect data sampling. By automatically projecting images in different spaces, images having similar characteristics (characteristics from the same correct answer data) can be placed at close distances from each other in style space.

9 is a view showing an example of a style space according to an embodiment of the present invention. In FIG. 9, an image vector space 920 for each of the products extracted from the learning image 910 is generated, and a style space 930 is generated based on MDAE for input data having image characteristics as an input to style. An example of projecting products in space 930 is shown. As already described, MDAE may be learned so that products extracted from the same learning image are projected to points on the same space in the style space 930, and accordingly, in FIG. 9, images extracted from the same learning image are style space 930. Indicates projected at points close to each other. In addition, in FIG. 9, as products extracted from different learning images are learned to project at points on different spaces in the style space 930, products extracted from different learning images are relative to the style space 930. Indicates that they are projected at points distant from each other. In addition, MDAE encodes the image vector space 920 and decodes it again to generate a reconstructed image vector space 940. By utilizing the reconstructed image features, products can be appropriately stored in the style space without additional incorrect data generation and utilization. MDAE's encoder can be trained to project to a point.

As described above, according to embodiments of the present invention, the correlation of products by category included in the learning image is learned using MDAE (Multimodal Deep Autoencoder), and the product is based on the correlation learned in MDAE (Multimodal Deep Autoencoder). Can recommend In addition, by using the features of the reconstructed products reconstructed through the MDAE decoder as correct answer data, by learning the product recommendation model without using the incorrect answer data, a negative sampling process having the largest computation in the existing distance model is performed. Can be omitted.

The system or device described above may be implemented as a hardware component, or a combination of hardware components and software components. For example, the devices and components described in the embodiments include, for example, processors, controllers, arithmetic logic units (ALUs), digital signal processors (micro signal processors), microcomputers, field programmable gate arrays (FPGAs). , A programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions, may be implemented using one or more general purpose computers or special purpose computers. The processing device may perform an operating system (OS) and one or more software applications running on the operating system. In addition, the processing device may access, store, manipulate, process, and generate data in response to the execution of the software. For convenience of understanding, a processing device may be described as one being used, but a person having ordinary skill in the art, the processing device may include a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that may include. For example, the processing device may include a plurality of processors or a processor and a controller. In addition, other processing configurations, such as parallel processors, are possible.

The software may include a computer program, code, instruction, or a combination of one or more of these, and configure the processing device to operate as desired, or process independently or collectively You can command the device. Software and/or data may be interpreted by a processing device, or to provide instructions or data to a processing device, of any type of machine, component, physical device, virtual equipment, computer storage medium or device. Can be embodied in The software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, or the like alone or in combination. The medium may be a computer that continuously stores executable programs or may be temporarily stored for execution or download. In addition, the medium may be various recording means or storage means in the form of a combination of single or several hardware, and is not limited to a medium directly connected to a computer system, but may be distributed on a network. Examples of the medium include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical recording media such as CD-ROMs and DVDs, and magneto-optical media such as floptical disks, And program instructions including ROM, RAM, flash memory, and the like. In addition, examples of other media may include an application store for distributing applications or a recording medium or storage medium managed by a site, server, or the like that supplies or distributes various software. Examples of program instructions include high-level language code that can be executed by a computer using an interpreter, etc., as well as machine language codes produced by a compiler.

As described above, although the embodiments have been described by a limited embodiment and drawings, those skilled in the art can make various modifications and variations from the above description. For example, the described techniques are performed in a different order than the described method, and/or the components of the described system, structure, device, circuit, etc. are combined or combined in a different form from the described method, or other components Alternatively, even if replaced or substituted by equivalents, appropriate results can be achieved.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims (15)

A method for recommending a product of a computer device including at least one processor, the method comprising:
Extracting products for each category included in the learning image set by the at least one processor;
MDAE (Multimodal Deep Autoencoder) based on the at least one processor to project the extracted category-specific products to points in a style space based on correlations between category-specific products included in the same learning image. Learning a product recommendation model of the product;
Projecting, by the at least one processor, a plurality of recommendation target products to a plurality of first points on the style space through the learned product recommendation model;
Projecting, by the at least one processor, a product to be searched through the learned product recommendation model to a second point on the style space; And
Selecting at least one recommendation target product among the plurality of recommendation target products based on the distance between each of the first points and the second point in the style space by the at least one processor;
Including,
The style space is classified into a plurality of spaces,
The step of learning the product recommendation model,
It is to project the extracted products for each category to points on a style space, but to train the encoder of the MDAE such that products for each category included in the same learning image are projected at points of the same space among the classified plurality of spaces. Featured product recommendation method. delete According to claim 1,
The step of learning,
A product recommendation method characterized by extracting image features through each of the sub-sets excluding the empty set for a set having products by category included in the same learning image as elements, and generating input data for the product recommendation model. According to claim 1,
The step of learning the product recommendation model,
A product recommendation method characterized by using the correct answer data for features of products reconstructed through the encoder and decoder of the MDAE and learning the product recommendation model without using additional incorrect data. According to claim 1,
The step of selecting,
A product recommendation method comprising selecting N recommendation target products corresponding to N (N is a natural number) first points in an order of distance from the second point among the plurality of recommendation target products. According to claim 1,
The step of selecting,
A method of recommending a product, wherein a product of a category different from the category of the product to be searched is selected from among the plurality of products to be recommended based on the distance. According to claim 1,
A product recommendation method characterized in that the products for each category included in the same learning image include three or more products for each category. A computer program stored in a computer readable recording medium in combination with a computer to execute the method of claim 1 or any one of claims 3 to 7 on a computer. A computer-readable recording medium, characterized in that a program for executing the method of any one of claims 1 or 3 to 7 is recorded on a computer. In the computer device,
At least one processor implemented to execute computer readable instructions
Including,
By the at least one processor,
Extract products by category included in the learning image set,
A multimodal deep autoencoder (MDAE)-based product recommendation model is trained to project the extracted category-specific products to points on a style space based on correlations between category-specific products included in the same learning image,
Projecting a plurality of recommended target products to a plurality of first points on the style space through the learned product recommendation model,
The product to be searched is projected to the second point on the style space through the learned product recommendation model,
On the style space, based on the distance between each of the first points and the second point, select at least one recommended target product among the plurality of recommended target products,
The style space is classified into a plurality of spaces,
By the at least one processor,
Learning the encoder of the MDAE such that the extracted products for each category are projected to points on a style space, and products for each category included in the same learning image are projected at points in the same space among the plurality of classified spaces.
Computer device characterized in that. delete The method of claim 10,
By the at least one processor,
A computer device characterized by extracting image features through each of the subsets excluding the empty set for a set having elements by category included in the same learning image as elements, and generating input data for the product recommendation model. The method of claim 10,
By the at least one processor,
A computer device characterized by using the correct answer data for features of products reconstructed through the encoder and decoder of the MDAE, and training the product recommendation model without using additional incorrect data. The method of claim 10,
By the at least one processor,
A computer device characterized by selecting N recommended target products corresponding to N (where N is a natural number) first points in a sequence close to the second point among the plurality of recommended target products. The method of claim 10,
By the at least one processor,
A computer device, characterized in that, among the plurality of recommended target products, a category of the search target product and a recommended target product of a different category are selected based on the distance.

Images