JP2018181326A - Personalized product recommendation utilizing deep learning - Google Patents

Abstract

[Problem] To provide personalized product recommendation technology utilizing deep learning.
SOLUTION: A personalized merchandise recommendation method includes merchandise information to be recommended for a specific service among a plurality of services provided by the personalized merchandise recommendation system, and a user using at least one of a plurality of services. Collecting user information including behavior history information, using commodity information and the user information, using deep learning to learn purchase predictions for recommended products, and purchase prediction learning results for recommended products Providing product recommendation results through specific services based on
[Selected figure] Figure 4

Description

  The following description relates to a technique for recommending a product suitable for the user.

  Generally, the product recommendation in the Internet market selects and recommends the most suitable products for the individual based on the personal purchase history. At this time, as a method mainly used, a user whose purchase history is similar to a user who is a target of product recommendation is selected, and a similar user has already purchased a product but a recommendation target user has not yet purchased a product Such as Collaborative Filtering (Collaborative Filtering) recommending a product, and Association Rule Mining (Associative Rule Mining) recommending a product with a possibility of simultaneous purchase or a possibility of simultaneous click by using products purchased by multiple users. Algorithm is used.

  For example, in Patent Document 1 (registration date: February 02, 2009), product information on a product using a product code, a product name, or a model of a product recognized by an information recommendation agent installed in a user terminal A technique for automatically recommending

Korean Registered Patent No. 10-0882716

  A method and system for recommending a product suitable for a user by utilizing deep learning.

  A method and system for providing a product recommendation result to a product on one service by using user information including user behavior history on another service in a cross domain form in a portal environment providing multiple services. provide.

  A personalized product recommendation method implemented by a personalized product recommendation system implemented by a computer, comprising: product information to be recommended for a specific service among a plurality of services provided by the personalized product recommendation system; Collecting user information including user behavior history information in at least one service among the plurality of services; using the product information and the user information; utilizing a deep learning to make a recommendation target product And providing the product recommendation result through the specific service based on the learning of the purchase prediction for the recommended product and the learning result of the purchase prediction for the recommended product.

  According to one aspect, in the learning, the product information and the user information are based on a feedforward neural network (FFNN) learning model in which the product information and the user information are configured as input vectors. The purchase prediction for the recommended product may be learned by learning together.

  According to another aspect, the step of learning may include the step of dividing the vector set corresponding to the user action history information into a predetermined section and constructing an input vector by an average value of each section.

  According to another aspect, the step of learning may include the step of forming an input vector by including the number of times of history of each type and the number-of-items information in the user action history information.

  According to another aspect, the product information includes at least one of product meta information and a product image, and the learning includes converting the product meta information into an embedding vector as an input vector. The method may comprise constructing or constructing a feature vector including a visual pattern of the product image as an input vector.

  According to another aspect, the providing step is performed based on a crossover form between the plurality of services, using the user behavior history information in a service different from the specific service. You may provide a product recommendation result for the service of.

  According to another aspect, the step of learning calculates the predicted purchase score for the recommendation target product based on the FFNN learning model in which the product information and the user information are configured as an input vector, and the provision is performed A part of products to be recommended in the specific service may be recommended based on the purchase prediction score.

  According to still another aspect, the providing may recommend a part of products corresponding to the query based on the predicted purchase score when a query for a search is input.

  A computer program recorded on a computer-readable recording medium to execute a personalized product recommendation method in combination with a computer system, wherein the personalized product recommendation method includes a plurality of provided by the personalized product recommendation system. Collecting user information including product information to be recommended for a specific service among services and user behavior history information on at least one of the plurality of services; the product information and the user Using information and learning purchase prediction for the recommendation target product using deep learning, and providing a product recommendation result through the specific service based on the learning result of the purchase prediction for the recommendation target product , Provide computer programs.

  A computer implemented personalized product recommendation system, comprising at least one processor implemented to execute computer readable instructions, the at least one processor being provided by the personalized product recommendation system An information management unit for collecting user information including product information to be recommended for a specific service among a plurality of services and user behavior history information for at least one of the plurality of services, the product information And the user information to learn the purchase prediction for the recommendation target product using deep learning, and the product recommendation result through the specific service based on the learning result of the purchase prediction for the recommendation target product. Provided is a personalized product recommendation system including a product recommendation unit to be provided.

  According to the embodiment of the present invention, deep learning can be used to recommend products suitable for the individual user.

  According to an embodiment of the present invention, based on a cross domain configuration in a portal environment providing a plurality of services, a product by using user information including user behavior history in another service for a product of a service It can provide recommendation results.

FIG. 1 illustrates an example of a network environment in an embodiment of the present invention. It is a block diagram for demonstrating the internal structure of an electronic device and a server in one Embodiment of this invention. FIG. 5 illustrates an example of components that may be included in a processor of a server, in an embodiment of the present invention. FIG. 5 is a flow chart illustrating an example of a method that a server may perform in an embodiment of the present invention. It is an illustration figure for demonstrating the FFNN learning model which predicts goods purchase in one embodiment of the present invention. FIG. 6 is a diagram illustrating an example of a process of collecting user information according to an embodiment of the present invention. It is a figure showing an example of a process in which user information is constituted by an input vector of a FFNN learning model in one embodiment of the present invention. It is a figure showing an example of a process in which user information is constituted by an input vector of a FFNN learning model in one embodiment of the present invention. It is the figure which showed an example of the process in which goods information is comprised with the input vector of FFNN learning model in one Embodiment of this invention. It is the figure which showed an example of the process in which goods information is comprised with the input vector of FFNN learning model in one Embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings.

  Embodiments of the present invention relate to a technology for recommending a product suitable for a user.

  Embodiments including matters specifically disclosed in the present specification may utilize a deep neural network to construct a personalized merchandise recommendation model, thereby allowing for the quality of service and utilization rate. Achieve considerable advantages in terms of improvement, efficiency, convenience, cost savings etc.

  FIG. 1 is a diagram illustrating an example of a network environment in an embodiment of the present invention. The network environment of FIG. 1 illustrates an example including a plurality of electronic devices 110, 120, 130, 140, a plurality of servers 150, 160, and a network 170. Such FIG. 1 is merely an example for describing the invention, and the number of electronic devices and the number of servers are not limited as in FIG. 1.

  The plurality of electronic devices 110, 120, 130, and 140 may be fixed terminals or mobile terminals implemented by computer devices. Examples of the plurality of electronic devices 110, 120, 130, and 140 include a smartphone, a mobile phone, a tablet, a navigation, a PC (personal computer), a laptop computer, a terminal for digital broadcasting, a PDA (Personal Digital Assistant), and a PMP (Portable). Multimedia Player) etc. As an example, the first electronic device 110 may communicate with other electronic devices 120, 130, 140 and / or servers 150, 160 via the network 170 using a wireless or wired communication scheme.

  The communication method is not limited, and not only a communication method utilizing a communication network (for example, a mobile communication network, a wired Internet, a wireless Internet, a broadcast network) which can be included in the network 170, but also a short distance between devices Wireless communication may be included. For example, the network 170 may be 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), a broadband network (BBN), the Internet, etc. Any one or more of the networks may be included. Further, network 170 may include any one or more of a network topology, including bus networks, star networks, ring networks, mesh networks, star bus networks, trees or hierarchical networks, etc. There is no limitation to these.

  As an example, the server 160 may provide the first electronic device 110 connected via the network 170 with a file for installing an application. In this case, the first electronic device 110 may install the application using the file provided from the server 160. In addition, the server 150 is connected to the server 150 according to the control of an operating system (OS) included in the first electronic device 110 and at least one program (for example, a browser or the above installed application). And you may receive the provision of content. For example, when the first electronic device 110 transmits a service request message to the server 150 via the network 170 according to control of the application, the server 150 may transmit a code corresponding to the service request message to the first electronic device 110, The first electronic device 110 may provide content to the user by configuring and displaying a screen based on a code according to control of the application.

  FIG. 2 is a block diagram for explaining an internal configuration of the electronic device and the server in the embodiment of the present invention. 2, the internal configuration of the first electronic device 110 will be described as an example for one electronic device, and the internal configuration of the server 150 will be described as an example for one server. The other electronic devices 120, 130, 140 and the server 160 may have the same or similar internal configuration.

  The first electronic device 110 and the server 150 may include memories 211, 221, processors 212, 222, communication modules 213, 223, and input / output interfaces 214, 224, respectively. The memories 211 and 221 are computer readable recording media, and include random access memory (RAM), read only memory (ROM), and permanent mass storage device such as a disk drive. Good. In addition, the memory 211, 221 may store an operating system and at least one program code (for example, a code for an application installed and driven in the first electronic device 110). Such software components may be loaded from a computer readable storage medium separate from the memories 211, 221. Such other computer readable recording media may include computer readable recording media such as floppy drives, disks, tapes, DVD / CD-ROM drives, memory cards and the like. In another embodiment, software components may be loaded into memory 211, 221 through communication module 213, 223 which is not a computer readable storage medium. For example, at least one program may be installed by a file provided by a file distribution system (for example, the server 160 described above) for distributing the installation file of the developer or the application by the file provided via the network 170 (for example, described above Application may be loaded into the memory 211, 221.

  Processors 212, 222 may be configured to process computer program instructions by performing basic arithmetic, logic, and input / output operations. The instructions may be provided to the processors 212, 222 by the memories 211, 221 or the communication modules 213, 223. For example, the processors 212, 222 may be configured to execute instructions received in accordance with program code stored in a recording device, such as the memories 211, 221.

  The communication modules 213 and 223 may provide a function for the first electronic device 110 and the server 150 to communicate with each other via the network 170, or another electronic device (for example, the second electronic device 120). Or, it may provide functionality to communicate with other servers (as an example, server 160). As an example, a request (for example, a search request) generated by the processor 212 of the first electronic device 110 according to the program code stored in the recording device such as the memory 211 is a server via the network 170 according to the control of the communication module 213 It may be transmitted to 150. On the contrary, control signals, instructions, contents, files, etc. provided under the control of the processor 222 of the server 150 are transmitted through the communication module 223 and the network 170 to the first electronic device through the communication module 213 of the first electronic device 110. 110 may be received. For example, control signals and instructions of the server 150 received through the communication module 213 may be transmitted to the processor 212 and the memory 211, and contents, files and the like may be stored in a storage medium that the first electronic device 110 may further include. May be done.

  Input / output interface 214 may be a means for interfacing with input / output device 215. For example, the input device may include a device such as a keyboard or a mouse, and the output device may include a device such as a display for displaying a communication session of the application. As another example, the input / output interface 214 may be a means for interfacing with a device such as a touch screen in which functions for input and output are integrated. As a more specific example, the processor 212 of the first electronic device 110 is configured using data provided by the server 150 or the second electronic device 120 when processing the instructions of the computer program loaded in the memory 211. Service screens and content may be displayed on the display through the input / output interface 214. Similarly, the input / output interface 224 may output information configured using data provided by the server 150 as the processor 222 of the server 150 processes the instructions of the computer program loaded into the memory 221.

  Also, in other embodiments, the first electronic device 110 and the server 150 may include more components than the components of FIG. 2. However, most prior art components need not be shown explicitly in the figures. For example, the first electronic device 110 may be realized to include at least a part of the input / output device 215 described above, or may be a transceiver, a GPS (Global Positioning System) module, a camera, various sensors, a database, etc. May further include other components such as As a more specific example, when the first electronic device 110 is a smartphone, an acceleration sensor or a gyro sensor generally included in the smartphone, a camera, various physical buttons, a button using a touch panel, an input / output port, It will be appreciated that various components such as a vibrator for vibration may be realized to be further included in the first electronic device 110.

  In the following, specific embodiments of a personalized merchandise recommendation method and system utilizing deep neural learning (DNN) will be described.

  FIG. 3 is a diagram illustrating an example of components that may be included in a processor of a server according to an embodiment of the present invention, and FIG. 4 is a server executable according to an embodiment of the present invention 3 is a flow chart illustrating an example of a method.

  The server 150 targets a plurality of electronic devices 110, 120, 130, 140, which are clients, such as search services, shopping, communities (eg, groups, blogs, etc.), emails, news, directions, etc. It plays the role of a platform that integrates and provides various services. A large number of products exist in such a service platform, and new products are frequently registered, but in the search channel etc., a very small number of products, mainly popular products, are displayed. is there.

  In the present invention, the server 150 uses deep learning to recommend products that are suitable for individual users so that more products can be connected with more users. In particular, the server 150 makes a product recommendation by using user information including user behavior history in another service for a product of a specific service based on a cross domain configuration between a plurality of services provided in a portal environment. The results can be provided. In other words, the server 150 may recommend different domains or different types of items from the items that the user normally consumed based on the cross over recommendation logic.

  As shown in FIG. 3, the processor 222 of the server 150 may include an information management unit 310, a learning processing unit 320, and a product recommendation unit 330 as components. Such processor 222 and components of processor 222 may control server 150 to perform steps 410-430 included in the method of FIG. At this time, the processor 222 and the components of the processor 222 may be implemented to execute an instruction according to the code of the operating system and the code of at least one program included in the memory 221. Also, the components of the processor 222 may be representations of different functions performed by the processor 222 according to the operating system and control instructions provided by the at least one program. For example, the information management unit 310 may be used as a functional expression that controls the processor 222 to maintain and manage the product information and the user information according to the control command described above.

  In step 410, the information management unit 310 includes product information to be recommended for a specific service among the plurality of services that can be provided by the server 150, and a user behavior history for at least one of the plurality of services. User information may be collected and maintained. In order to implement crossover recommendation logic that recommends items of a type / domain different from items that the user has normally consumed, the information management unit 310 performs user behavior in services other than domains that require product recommendation. Historical information may be collected and managed.

  In step 420, the learning processing unit 320 may learn the purchase prediction for the recommendation target product by utilizing deep learning in which the product information and the user information are configured as input vectors. As an example, the learning processing unit 320 may predict the purchase possibility of a product based on user information and product information based on a forward propagation neural network (FFNN) learning model. A collaborative filtering model that recommends items recommended by the user group similar to the user, and a content-based filtering model that recommends other items whose content is similar to the item that the user has selected in the past are combined In order to realize a hybrid (Hybrid) recommendation model, the learning processing unit 320 simultaneously learns, on the same learning model, user information including a user behavior history in a service different from the product information to be recommended in the specific service. You may The learning processing unit 320 described above may learn the purchase prediction for each product by using the given product information and the user information (user action history). At this time, the learning processing unit 320 can also perform short-term purchase prediction and long-term purchase prediction based on learning data indicating the user behavior history.

  In step 430, the product recommendation unit 330 may provide the product recommendation result personalized to the user through a specific service based on the purchase prediction learning result for the recommendation target product. The product recommendation unit 330 targets a user who uses a specific service, provides a product recommendation result through the corresponding service, but includes user information including user behavior history in a service different from the product information on the corresponding service. Personalized products may be recommended to the user based on the result of learning simultaneously utilizing the same deep learning model. In other words, in providing the product recommendation result through the service corresponding to the specific domain, the product recommendation unit 330 learns the user information including the user log in the other domain together with the product information on the corresponding domain. Based on the product recommendation results may be provided. In some cases, for example, when providing a search result corresponding to a user query in the shopping service, the user preference is analyzed by integrating the user history in the shopping service as well as the user history in other services, The recommendation result corresponding to may be included in the search result in the shopping service. The product recommendation model for the cross domain is to provide product recommendation results by utilizing different types of user qualification data, for example, to provide shopping product recommendation results based on a camp reservation history, or to be based on a fashion purchase history. Sound source recommendation results may be provided.

  The present invention utilizes deep learning for individualized product recommendation, but as an example, heterogenous information integration learning related to the product, semantic learning by embedding, optimal parameter automatic learning, etc. The included FFNN learning model may be applied.

  FIG. 5 is an exemplary view for explaining an FFNN learning model that predicts whether or not to purchase a product in the embodiment of the present invention.

  The FFNN learning model 500 according to the present invention is provided with user information 510 including an action history indicated by any user u in a plurality of services overall, and product information 520 to be recommended for a specific service. It relates to the problem of predicting whether the user u purchases the product i. The user information 510 and the item information 520 may be represented by embedded vectors corresponding to the input layer of the FFNN learning model 500.

  For personalized product recommendation, a process of collecting user information 510 from a portal environment providing various services is required. For example, as shown in FIG. 6, profile information 601 including ID, gender, age, etc., which is basic information of the user, may be collected, and as user behavior history for each service, for example, user behavior history in shopping service Information 602 includes user product search history (shopping query history), user product click history (including product meta information such as product name, category, price, image, etc.), user product purchase history (product name, category, price, You may collect product meta information such as images etc.) as user action history information 603 in all services except shopping service, user content search history (content query history), user content click history (content meta information) User content consumption history (including Including the Tentsumeta information) may be collected and the like. Thereby, the information management unit 310 may collect, from the platform provided by the server 150, the user information 510 including the user behavior history of all services used by the user.

  7 and 8 are diagrams showing an example of a process of configuring user information as an input vector of the FFNN learning model in the embodiment of the present invention.

  Referring to FIG. 7, FFNN learning model by concatenating profile information 601 collected as user information 510, user action history information 602 in shopping service, and user action history information 603 in all services other than shopping service. It may be configured as 500 input vectors. The user behavior history information 602 in the shopping service may be utilized to analyze the user's shopping preferences, and the user behavior history information 603 in other services in general is utilized to analyze the user's usual interests Good. By collecting user behavior history not only for shopping services but also for other services in general, the user's interest information can be widely utilized, and even if the user does not directly perform shopping-related behavior, the user can perform other services. Personalized product recommendations can be made by using the usual interests shown in.

  The user behavior history information described above may be configured as an input vector by converting the meta information of the related product into a high-dimensional vector by embedding. For example, in the case of the user product click history, as shown in FIG. 8, the product click analysis results of the user may be connected in order to construct an input vector. When the order information is considered in the user history information, the historical order of interest may be applied instead of the recent history order. Effectively grasp the general interests of the user by analyzing in advance the history that the user has shown interest within a certain period and utilizing it, rather than simply considering the user's recent historical order become able to. The FFNN learning model 500 may construct an input vector to indicate features of historical information that can be ordered or collated. As shown in FIG. 8, after dividing the vector set corresponding to the user action history information, that is, the embedded vector of the history object into a certain section, an average value (average) of the certain section is output to construct an input vector. Thus, when an embedded vector is introduced to construct an input vector, more information can be included than when the entire average is taken. At this time, it is possible to consider active factors such as product click activity based on continuous values obtained for the number of related actions / the number of products. In other words, an input vector may be configured by including the number of times of history of each type and the number of articles of the user with respect to the user action history information, whereby active or inactive for each type of history information You can use it to analyze user interests by determining

  FIGS. 9 and 10 are diagrams showing an example of the process of configuring product information as an input vector of the FFNN learning model in the embodiment of the present invention.

  The product information 520 may include information displayed on the service page, and for example, as shown in FIG. 9, basic text of the product such as product name, category, price, sales office name, product ID, etc. Merchandise meta information 901 which is information, merchandise visual information 902 such as a merchandise image and the like may be collected.

  Referring to FIG. 10, the product meta information 901 collected as the product information 520 and the product visual information 902 may be connected and configured as an input vector of the FFNN learning model 500 together with the user information 510. Product meta information 901 which is text information such as product name, category, price, sales office name, product ID may be embedded and utilized, and in the case of product visual information 902, converted by image learning Feature vectors may be used. For example, the product image may be converted into an image feature vector including visual patterns by a convolutional neural network (CNN), and compressed and represented using an auto-encoder. In the case of product categories where visual information is important, such as fashion clothing and fashion items, image feature vectors may be used to analyze user interests.

  As described above, in the FFNN learning model 500, the user information 510 and the product information 520 may be configured as an input vector of the input layer (input layer), and as shown in FIG. 5, the user information 510 and the product information 520 At the same time, it may be learned and a predicted purchase score for each product may be derived by a hidden layer activation function (e.g., ReLU). Therefore, the learning processing unit 320 can calculate the purchase prediction score for the recommendation target product by using the FFNN learning model 500 in which the product information and the user information are configured as input vectors. As a result, the product recommendation unit 330 corresponds to a product that has a predicted purchase score equal to or higher than a certain level among products to be recommended for a specific service, or a certain number of products corresponding to the highest ranked purchase predicted score. It may be recommended to the user through the service of At this time, when the user inputs a query for a search in a specific service, the product recommendation unit 330 uses at least some of the products corresponding to the query as a search result based on the predicted purchase score. It may be included and recommended to the user.

  As described above, according to the embodiment of the present invention, it is possible to use the user behavior history information collected for all services, not for a specific service, even if there is no user behavior history for a particular service (cold start problem ( Also in the case of cold-start problems), it is possible to recommend a personalized product for a specific service by learning more abundant information. In providing personalized product recommendation results to users, it is not popular because it reflects user preferences analyzed for the entire service, not providing limited recommendation results focusing on popular products Even products or new products can be included, and a wide range of recommendation results can be provided according to the user's preferences. In addition, according to an embodiment of the present invention, visual aspect can be obtained by learning together visual information such as a product image as well as text information such as a product name and a category name to analyze user interests. However, it is possible to recommend personalized products in which user preferences are reflected. Furthermore, according to the embodiment of the present invention, the input vector of the FFNN learning model is effectively combined to improve the learning performance by dividing the embedded vector set input to the input layer for the user history into fixed sections and configuring them. It can be done.

  The above-described apparatus may be realized by hardware components, software components, and / or a combination of hardware components and software components. For example, the devices and components described in the embodiments include a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA), and a program. It may be implemented using one or more general purpose computers or special purpose computers, such as programmable logic units (PLUs), microprocessors, or various devices capable of executing and responding to instructions. . The processing device may execute an operating system (OS) and one or more software applications running on the OS. The processing device may also respond to software execution, access data, and store, manipulate, process, and generate data. Although one processor may be described as being used for ease of understanding, one skilled in the art will appreciate that the processor may include more than one processing element and / or more than one type of processing element. You will understand. For example, the processing device may include multiple processors or one processor and one controller. Other processing arrangements are also possible, such as parallel processors.

  The software may include a computer program, code, instructions, or a combination of one or more of these, configuring the processing unit to operate as desired, and instructing the processing unit independently or collectively. You may The software and / or data may be embodied in any type of machine, component, physical device, virtual device, computer storage medium or device to be interpreted based on the processing device or to provide instructions or data to the processing device. May be The software may be distributed on computer systems connected by a network, and may be stored or executed in a distributed manner. The software and data may be stored on one or more computer readable recording media.

  The method according to the embodiments may be realized in the form of program instructions executable by various computer means and recorded on a computer readable medium. At this time, the medium may store the computer executable program permanently or may temporarily store it for execution or download. Also, the medium may be various recording means or storage means in the form of combination of one or more hardwares, but it is not limited to the medium directly connected to a certain computer system, but on a network It may be distributed and present. Examples of media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs, DVDs, magneto-optical media such as floppy disks, And program instructions such as ROM, RAM, flash memory, etc. may be stored. In addition, as an example of another medium, an application store that distributes an application or a site that supplies or distributes various other software, a recording medium managed by a server, or a storage medium may be used.

  Although the embodiments have been described based on the limited embodiments and the drawings as described above, various modifications and variations can be made by those skilled in the art from the above description. For example, components described in the systems, structures, devices, circuits, and the like may be performed in a different order from that of the methods described and / or different from those described in the methods described. Appropriate results can be achieved even if combined or combined, or replaced or substituted by other components or equivalents.

  Therefore, even different embodiments, which are equivalent to the scope of the claims, belong to the appended claims.

222: Processor 310: Information management unit 320: Learning processing unit 330: Product recommendation unit

Claims (17)

A personalized merchandise recommendation method implemented by a personalized merchandise recommendation system implemented by a computer, comprising:
Product information to be recommended for a specific service among a plurality of services provided by the personalized product recommendation system, and user information including user action history information on at least one of the plurality of services is collected Stage to
The product recommendation result through the specific service based on learning the purchase prediction for the recommendation target product using deep learning using the product information and the user information, and the learning result of the purchase prediction for the recommendation target product Providing personalized product recommendation methods. The learning phase is
Both the product information and the user information are learned together based on a forward propagation neural network (FFNN) learning model in which the product information and the user information are configured as input vectors, and a purchase forecast for the recommendation target product is learned. The personalized merchandise recommendation method according to claim 1, characterized in that The learning phase is
The method according to claim 1, further comprising the step of: after dividing the vector set corresponding to the user action history information into predetermined sections, forming an input vector by an average value of each section. The learning phase is
The personalized merchandise recommendation method according to claim 1, further comprising the step of forming an input vector by including the history count of each type and the number-of-items information with respect to the user behavior history information. The product information includes at least one of product meta information and a product image,
The learning phase is
The method according to claim 1, comprising: converting the product meta information into an embedded vector and configuring as an input vector, or configuring a feature vector including a visual pattern of the product image as an input vector. . The providing step may
The method according to claim 1, wherein the product recommendation result of the specific service is provided by using user behavior history information in a service different from the specific service according to a crossover form between the plurality of services. Personalized product recommendation method described. The learning phase is
Based on an FFNN learning model in which the product information and the user information are configured as input vectors, a predicted purchase score for the product to be recommended is calculated,
The providing step may
The personalized merchandise recommendation method according to claim 1, further comprising: recommending a part of merchandise to be recommended in the specific service based on the purchase prediction score. The providing step may
The personalized merchandise recommendation method according to claim 7, characterized in that when a query for search is input, a part of products corresponding to the query is recommended based on the purchase prediction score. A computer program for causing a computer system to execute a personalized merchandise recommendation method, comprising:
The personalized product recommendation method is
Product information to be recommended for a specific service among a plurality of services provided by the personalized product recommendation system, and user information including user action history information on at least one of the plurality of services is collected Stage,
The product recommendation result through the specific service based on learning the purchase prediction for the recommendation target product using deep learning using the product information and the user information, and the learning result of the purchase prediction for the recommendation target product Providing a computer program. A personalized product recommendation system implemented by a computer, comprising:
At least one processor implemented to execute computer readable instructions;
The at least one processor is
Product information to be recommended for a specific service among a plurality of services provided by the personalized product recommendation system, and user information including user action history information on at least one of the plurality of services is collected Information Management Department,
A learning processing unit that uses deep learning to learn purchase prediction for a recommendation target product using the product information and the user information, and a product through the specific service based on learning results of purchase prediction for the recommendation target product. A personalized product recommendation system comprising a product recommendation unit that provides a recommendation result. The learning processing unit
The product information and the user information are both learned based on an FFNN learning model in which the product information and the user information are configured as input vectors, and a purchase forecast for the recommendation target product is learned. The personalized product recommendation system according to Item 10. The learning processing unit
The personalized merchandise recommendation system according to claim 10, wherein the vector set corresponding to the user action history information is divided into predetermined sections, and then an input vector is formed by an average value of each section. The learning processing unit
11. The personalized merchandise recommendation system according to claim 10, wherein an input vector is configured by including the number of times of history of each type and the number-of-items information with respect to the user action history information. The product information includes at least one of product meta information and a product image,
The learning processing unit
The personalized product according to claim 10, wherein the product meta information is converted into an embedded vector and configured as an input vector, or a feature vector including a visual pattern of the product image is configured as an input vector. Recommendation system. The product recommendation unit
The method according to claim 10, wherein the product recommendation result of the specific service is provided by using user behavior history information in a service different from the specific service according to a crossover form between the plurality of services. Personalized product recommendation system described. The learning processing unit
Based on an FFNN learning model in which the product information and the user information are configured as input vectors, a predicted purchase score for the product to be recommended is calculated,
The product recommendation unit
The personalized merchandise recommendation system according to claim 10, wherein a part of products to be recommended in the specific service is recommended based on the purchase prediction score. The product recommendation unit
The personalized merchandise recommendation system according to claim 16, wherein when a query for search is input, a part of products corresponding to the query is recommended based on the purchase prediction score.

Images