KR20240027478A - Method, computer device, and computer program to recommend products to buy together - Google Patents

Abstract

A method, computer device, and computer program for recommending products to be purchased together are disclosed. The product recommendation method may include recommending a product to be purchased together with a reference product using a language model learned from joint purchase data indicating a history of purchasing two or more products together within a certain time, and the language model may be used for each product. It may be a model in which relationships between products are learned using linguistic information made up of natural language related to products.

Description

Method for recommending products to purchase together, computer devices, and computer programs {METHOD, COMPUTER DEVICE, AND COMPUTER PROGRAM TO RECOMMEND PRODUCTS TO BUY TOGETHER}

The explanation below is about product recommendation technology.

Generally, when a user selects a product on the Internet, products similar to the selected product may be recommended.

Methods mainly used for product recommendation can be broadly divided into collaborative filtering methods that utilize similarities between users, and content-based filtering methods that utilize characteristic values of products and users. You can.

For example, Korean Patent Publication No. 10-2021-0131198 (published on November 2, 2021) discloses a technology for recommending similar products based on images extracted through an artificial neural network.

Based on actual purchase data, products in categories that are likely to be purchased along with the standard product can be recommended.

Considering the linguistic characteristics of the standard product, products with characteristics that are frequently purchased together can be recommended.

By using a category dictionary for purchase data along with a language model learned from purchase data, products suitable for overall purchase behavior can be recommended.

In a product recommendation method executed on a computer device, the computer device includes at least one processor configured to execute computer-readable instructions included in a memory, and the product recommendation method includes, by the at least one processor, A step of recommending a product to be purchased together with a reference product using a language model learned from joint purchase data indicating a history of purchasing two or more products together within a period of time, wherein the language model is a language composed of natural language in relation to each product. We provide a product recommendation method characterized by a model in which relationships between products are learned using enemy information.

According to one aspect, the language model is learned for each higher-level category, and the recommending step may include selecting a recommended product through a language model learned for the higher-level category to which the reference product belongs.

According to another aspect, the language model may be trained with purchase data for different periods of time according to the upper category.

According to another aspect, the recommending step may include recommending a product that has a linguistic relationship with the reference product among products whose categories are different from the reference product using the language model.

According to another aspect, the recommendation step uses a language model learned for the upper category to which the reference product belongs to determine if the lower category among products in the higher category is different from the reference product and establishes a linguistic relationship with the reference product. It may include the step of recommending existing products.

According to another aspect, the recommending step includes reordering recommended products selected through the language model using a category dictionary in which categories of products purchased together are aggregated based on the joint purchase data. It can be included.

According to another aspect, the category dictionary may be composed of the results of aggregating the categories of the products purchased together for the entire category set.

According to another aspect, the category dictionary is created for each higher-level category and may be composed of an aggregate result of sub-category units that subdivide the higher-level category.

According to another aspect, in the reordering step, at least one of the order of the recommended products and the ratio by category can be rearranged by checking the together purchase category for the category to which the reference product belongs through the category dictionary.

According to another aspect, the recommending step may include rearranging recommended products selected through the language model based on the product-related history of an individual user or a user group to which the user belongs.

According to another aspect, the category dictionary may be personalized by aggregating categories of products purchased together based on the joint purchase data by an individual user or a user group to which the user belongs.

According to another aspect, the recommending step may include rearranging recommended products selected through the language model based on keywords related to the reference product.

According to another aspect, the recommending step may include determining the exposure time of the recommended product selected through the language model in consideration of the product purchase behavior of the individual user or the user group to which the user belongs.

According to another aspect, the determining step may expose the recommended product at the point when the user's behavior reaches the product purchasing behavior.

A computer program stored in a computer-readable recording medium is provided to execute the product recommendation method on a computer device.

A computer device comprising: at least one processor configured to execute computer-readable instructions included in a memory, wherein the at least one processor learns from joint purchase data indicating a history of purchasing two or more products together within a certain period of time; The language model is used to process the process of recommending products to be purchased together with the reference product, and the language model is a model in which the relationship between products is learned using linguistic information made up of natural language in relation to each product. Provide computer equipment.

According to embodiments of the present invention, recommendation quality can be improved by recommending products in a category that have a high probability of being purchased together with a reference product based on actual purchase data.

According to embodiments of the present invention, reliability and accuracy of recommendation can be increased by recommending products with characteristics that are frequently purchased together by considering the linguistic characteristics of the reference product.

According to embodiments of the present invention, a language model learned from purchase data is used along with a category dictionary for purchase data to recommend products suitable for overall purchase behavior, thereby providing a product recommendation model that is robust to the cold start problem. can do.

1 is a diagram illustrating an example of a network environment according to an embodiment of the present invention.
Figure 2 is a block diagram showing an example of a computer device according to an embodiment of the present invention.
Figure 3 is a flowchart showing an example of a method that can be performed by a computer device according to an embodiment of the present invention.
Figure 4 shows an example of purchase data in one embodiment of the present invention.
Figures 5 and 6 show an example of a language model learning process in one embodiment of the present invention.
Figure 7 shows an example of pre-configuring purchase categories in one embodiment of the present invention.
Figures 8 and 9 show an example of a process for recommending products to be purchased together in an embodiment of the present invention.
Figure 10 shows another example of a process for recommending products to be purchased together in one embodiment of the present invention.
Figure 11 shows another example of a process for recommending products to be purchased together in one embodiment of the present invention.
Figure 12 shows another example of a process for recommending products to be purchased together in one embodiment of the present invention.
Figure 13 shows an example of recommending products to be purchased together in one embodiment of the present 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 technology for recommending products.

Embodiments including those specifically disclosed in this specification can recommend products in a category with a high probability of being purchased together through a language model that learns the linguistic relationship between products purchased together.

The product recommendation system according to embodiments of the present invention may be implemented by at least one computer device, and the product recommendation method according to embodiments of the present invention may be performed through at least one computer device included in the product recommendation system. It can be. At this time, the computer program according to an embodiment of the present invention may be installed and driven in the computer device, and the computer device may perform the product recommendation method according to the embodiments of the present invention under the control of the driven computer program. . The above-described computer program can be combined with a computer device and stored in a computer-readable recording medium to execute the product recommendation method on the computer.

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

The plurality of electronic devices 110, 120, 130, and 140 may be fixed terminals or mobile terminals implemented as computer devices. Examples of the plurality of electronic devices 110, 120, 130, and 140 include smart phones, mobile phones, navigation devices, computers, laptops, digital broadcasting terminals, Personal Digital Assistants (PDAs), and Portable Multimedia Players (PMPs). ), tablet PC, etc. For example, in FIG. 1, the shape of a smartphone is shown as an example of the electronic device 110. However, in embodiments of the present invention, the electronic device 110 actually communicates with other devices through the network 170 using a wireless or wired communication method. It may refer to one of various physical computer devices capable of communicating with electronic devices 120, 130, 140 and/or servers 150, 160.

The communication method is not limited, and may include not only a communication method utilizing a communication network that the network 170 may include (for example, a mobile communication network, wired Internet, wireless Internet, and a broadcast network), but also short-range wireless communication between devices. For example, the network 170 may include 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). , may include one or more arbitrary networks such as the Internet. Additionally, the network 170 may include any one or more of network topologies including a bus network, star network, ring network, mesh network, star-bus network, tree or hierarchical network, etc. Not limited.

Each of the servers 150 and 160 is a computer device or a plurality of computers that communicate with a plurality of electronic devices 110, 120, 130, 140 and a network 170 to provide commands, codes, files, content, services, etc. It can be implemented with devices. For example, the server 150 may be a system that provides a service (for example, a product recommendation service, etc.) to a plurality of electronic devices 110, 120, 130, and 140 connected through the network 170.

Figure 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, and 140 described above or each of the servers 150 and 160 may be implemented by the computer device 200 shown in FIG. 2.

As shown in FIG. 2, this 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, non-perishable large-capacity recording devices such as ROM and disk drives may be included in the computer device 200 as a separate permanent storage device that is distinct from the memory 210. Additionally, 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 separate computer-readable recording media may include computer-readable recording media such as floppy drives, disks, tapes, DVD/CD-ROM drives, and memory cards. In another embodiment, software components may be loaded into the memory 210 through the communication interface 230 rather than a computer-readable recording medium. For example, software components may be loaded into memory 210 of computer device 200 based on computer programs installed by files received over network 170.

The processor 220 may be configured to process instructions of a computer program by performing basic arithmetic, logic, and input/output operations. Commands may be provided to the processor 220 by the memory 210 or the communication interface 230. For example, processor 220 may be configured to execute received instructions according to program code stored in a recording device such as 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, a request, command, data, file, etc. generated by the processor 220 of the computer device 200 according to a program code stored in a recording device such as memory 210 is transmitted to the network ( 170) and can be transmitted to other devices. Conversely, signals, commands, data, files, etc. from other devices may be received by the computer device 200 through the communication interface 230 of the computer device 200 via the network 170. Signals, commands, data, etc. received through the communication interface 230 may be transmitted to the processor 220 or memory 210, and files, etc. may be stored in a storage medium (as described above) that the computer device 200 may further include. It can be stored as a permanent storage device).

The input/output interface 240 may be a means for interfacing with the input/output device 250. For example, input devices may include devices such as a microphone, keyboard, or mouse, and output devices may include devices such as displays and speakers. As another example, the input/output interface 240 may be a means for interfacing with a device that integrates input and output functions, such as a touch screen. The input/output device 250 may be configured as a single device with the computer device 200.

Additionally, in other embodiments, computer device 200 may include fewer or more components than those of 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, etc.

Hereinafter, specific embodiments of a method and device for recommending products to be purchased together will be described.

To recommend related products in the shopping domain, recommendation models based on collaborative filtering, recommendation models based on MPV (Meta Product2Vector), and recommendation models based on Siamese networks and triplet networks are being used.

The recommendation performance of the above product recommendation models depends on how large or small the user's purchase history is. Since recommendations are made primarily for products with a certain amount of purchase history, it is easy for a phenomenon of recommendation concentration, where purchase history is concentrated on some products, to occur. In other words, there is a problem that cold items that are sold for the first time or have a lack of purchase history do not receive recommendation opportunities compared to warm items that have a lot of history.

These embodiments provide a new methodology capable of recommending various products even when there is no or insufficient purchase history, and provide a comprehensive understanding of various linguistic information such as product name, category, brand, price, material, and tag to identify products purchased together. Using a language model that learns the linguistic relationships between products, it is possible to provide a product recommendation model that is robust to the cold start problem.

The computer device 200 according to this embodiment can provide a product recommendation service to clients through a dedicated application installed on the client or through access to a web/mobile site related to the computer device 200. The computer device 200 may be configured with a computer-implemented product recommendation system. For example, a product recommendation system may be implemented in the form of a program that operates independently, or may be implemented in the form of an in-app of a specific application so that it can operate on the specific application.

The processor 220 of the computer device 200 may be implemented as a component for performing the following product recommendation method. Depending on the embodiment, components of the processor 220 may be selectively included in or excluded from the processor 220. Additionally, depending on the embodiment, components of the processor 220 may be separated or merged to express the functions of the processor 220.

The processor 220 and the components of the processor 220 can control the computer device 200 to perform the steps included in the product recommendation method below. For example, the processor 220 and its components may be implemented to execute instructions according to the code of an operating system included in the memory 210 and the code of at least one program.

Here, the components of the processor 220 may be expressions of different functions performed by the processor 220 according to instructions provided by program codes stored in the computer device 200.

The processor 220 may read necessary instructions from the memory 210 where instructions related to controlling the computer device 200 are loaded. In this case, the read command may include an command for controlling the processor 220 to execute steps that will be described later.

Steps included in the product recommendation method to be described later may be performed in an order different from the order shown, and some of the steps may be omitted or additional processes may be included.

Figure 3 is a flowchart showing an example of a method that can be performed by a computer device according to an embodiment of the present invention.

Referring to FIG. 3, in step S310, the processor 220 can generate a language model that learns the relationship between products using the linguistic information of each product for products purchased together as the user's purchase history. there is. Linguistic information is side information made up of natural language related to a product and may include, for example, product name, category, brand, price, material, tag, etc. For example, the processor 220 may represent products in the shopping domain as vectors (embeddings) based on a language model to create a model that better understands the semantic relationship between products purchased together. The processor 220 can prepare a language model that learns the linguistic relationship between products purchased together by comprehensively understanding various linguistic information such as product name, category, brand, price, material, and tag. The processor 220 may prepare a language model that learns purchase data within the category for each higher-level category based on the higher-level category that classifies the product. The upper category refers to the root category and is a product classification standard, for example, living/health, food, digital/home appliances, childbirth/childcare, furniture/interior, cosmetics/beauty, sports/leisure, fashion, etc. may include. In this embodiment, joint purchase data may refer to a history of purchasing products in the same upper category within a certain period of time (eg, one day, 24 hours, etc.). The processor 220 may learn a language model by refining purchase data for each upper category over a recent period of time. Together purchase data for an equal period of time (e.g., the last 1 year) for all top categories can be used as learning data, and depending on the embodiment, a purification period of the together purchase data to be used as learning data according to the top category. This can be applied differently. For example, in the case of digital/home appliances or childbirth/childcare, the recent year's joint purchase data is used to learn the language model, and in the case of sports/leisure and fashion, the recent three months' joint purchase data is used to learn the language model. You can.

In step S320, the processor 220 may generate a category dictionary by aggregating categories of products purchased together using joint purchase data. The processor 220 may together create a dictionary in units of subcategories based on purchase data. A subcategory refers to a leaf category that subdivides the upper category. It is a standard for subcategorizing products. For example, the subcategories of fashion include T-shirts, pants, skirts, outerwear, theme clothing, shoes, bags, jewelry, etc. It can be classified, and sub-categories of furniture/interior can be classified into sofa, bed/mattress, dressing room/hanger, living room cabinet/table, desk/bookcase, chair, dining table/kitchen furniture, shelf/storage furniture, etc. The processor 220 may collect categories of products purchased together for the entire leaf category set and create a joint purchase category dictionary. For example, based on the 'T-shirt' category, a category dictionary can be created by linking the 'Pants' category, 'Outerwear' category, and 'Skirt' category as categories of products purchased together with products in the 'T-shirt' category. Depending on the embodiment, a category dictionary for jointly purchased products may be created by refining joint purchase data for a recent certain period of time for each upper category. Since the volume of together purchase data may be different for each parent category, the purification period of the together purchase data used to create the category dictionary may be applied differently.

In step S330, the processor 220 may use a language model and a category dictionary for a given standard product to recommend a product to be purchased together with the standard product based on the linguistic information and category information of the standard product. The processor 220 may recommend products in a category that have a high probability of being purchased together with the standard product, using products clicked or searched by the user, products registered in the shopping cart or as a favorite product, etc., as standard products. The processor 220 may select recommended products focusing on products with a short vector distance indicating a linguistic relationship with the reference product among all products in the upper category through a language model learned for the upper category to which the reference product belongs. At this time, the processor 220 may reorder recommended products selected as a language model using a category dictionary. The processor 220 may check the together purchase categories for the subcategory to which the reference product belongs in the category dictionary and rearrange the recommended products selected by the language model according to at least one of the order and ratio of the together purchase categories.

The process of creating a product recommendation model that is robust to the cold start problem is described in detail as follows.

Figure 4 shows an example of purchase data in one embodiment of the present invention.

Referring to FIG. 4, the processor 220 collects together purchase data 401, 402, and 403 for products purchased together among the purchase history within the category in order to learn a language model in units of higher categories. You can.

At this time, the joint purchase data 401, 402, and 403 may refer to a history of products in the same upper category being purchased together by the same user over a certain period of time (eg, one day). The processor 220 can collect various linguistic information consisting of natural language such as product name, category, brand, price, material, tag, etc. of each product purchased together as purchase data 401, 402, and 403.

When collecting purchase data 401, 402, and 403 together, the processor 220 may exclude products of the same subcategory from collection. Likewise, if multiple identical products are purchased, they can be excluded from the purchase data (401, 402, 403).

In other words, the processor 220 may use the joint purchase data 401, 402, and 403, which indicates a history of purchasing two or more types of products with different lower categories among products in the upper category, for language model training.

The processor 220 may perform preprocessing, including normalization, when the purchase data 401, 402, and 403 are given together. Together, noise data such as stop words such as special characters, articles, prepositions, particles, and conjunctions can be removed from purchase data (401, 402, 403) and then used for language model learning.

Figures 5 and 6 show an example of a language model learning process in one embodiment of the present invention.

In this embodiment, it is possible to prepare a language model (for example, BERT (Bidirectional Encoder Representations from Transformers), etc.) that understands the semantic relationship between products purchased together by learning purchase data together through triplet learning. there is.

A triplet network-based recommendation model, which is one of the methodologies for solving the cold start problem, can use a method of learning embeddings encoded with linguistic information of products using triplet loss.

Triplet learning is a distance-based learning methodology called metric learning. There is anchor information, positive information, and negative information, and the correct answer is learned to be closer to the standard and the incorrect answer is learned to be farther away from the standard. It is done.

Referring to FIG. 5 , the processor 220 may configure the correct answer data pair 510 using the purchase data 501 together. When there is a list of products (Product A, Product B, Product C, Product D) of different subcategories that a specific user purchased together in a specific parent category on a specific date, this is used as label information and the answer data pair (510 ) can be made. Among products in the upper category for which the together purchase data 501 is collected, other products that are not included in the together purchase data 501 can be used as incorrect answer data. For example, when product A is considered standard information, product B, product C, and product D can be correct information, and the remaining products can be incorrect information.

Meanwhile, the processor 220 may create a list of products purchased by a specific user from a specific store on a specific date as the correct answer data pair 510. Among the products of a specific store for which the purchase data 501 is collected, other products that are not included in the purchase data 501 can be used as incorrect answer data. Additionally, learning data can be constructed in the same way using purchase records of products in a specific higher category among purchase records at a specific store.

As shown in FIG. 6, in the process of learning the purchase data 501 together, the processor 220 learns product B, which is the correct answer information, to be closer to product A with respect to product A, which is the reference information (i.e., a joint purchase relationship). ), product X, which is incorrect information, can be learned to move away from product A.

The processor 220 not only recommends product B, product C, and product D to a user viewing product A through a language model learned with purchase data 501 through triplet learning, but also recommends product A and product D. You can recommend products including other similar products. For example, a linguistic relationship in which product A and product Q can be purchased together even if there is no purchase history for product Q (for example, a linguistic relationship in which product A and product Q can be purchased together even if the similarity of the letters themselves is low, such as "T-shirt" and "shorts") is high), product Q can be provided as a recommended product for product A.

In the case of language models for categories where differences in seasonality exist even for products within the same category among higher-level categories, additional learning can be conducted including data related to purchase point and seasonality.

In the case of a language model, because it embeds various linguistic information of a product and understands the semantic relationship between products purchased together, generalization is possible in understanding the product, and it can also recommend products with no or insufficient purchase history.

Figure 7 shows an example of pre-configuring purchase categories in one embodiment of the present invention.

The processor 220 may generate a joint purchase category dictionary by aggregating categories of products purchased together using joint purchase data for a set of categories (e.g., an entire set of subcategories). For example, as shown in Figure 7, when the product categories are Category A and Category W, the categories of products purchased together with products of Category A are aggregated, and the categories of products purchased together with products of Category W are calculated. By counting, you can create a dictionary of purchase categories (700) together. At this time, the purchase category dictionary 700 may include the product purchase rate for each category. A purchase category dictionary (700) can be created by tallying up the purchase rate of products in Category A and the purchase rates of products in Category B, Category C, and Category D, which are categories of products purchased together with products in Category A.

In the above, it is explained that a purchase category dictionary is created together without distinguishing the hierarchy of categories, but it is not limited to this, and depending on the embodiment, it is a category dictionary with a hierarchical structure, and a category dictionary of lower category units is created for each upper category. It is also possible to make it.

The together purchase category dictionary 700 can be automatically created periodically through data aggregation using together purchase data, and the update cycle or timing of the category dictionary 700 can be determined by considering seasonality, etc.

When creating the category dictionary 700, the processor 220 may also utilize time data of purchase data to create a category dictionary that takes into account the direction according to the purchase order. It is valid to recommend an air purifier filter to a user who purchased an air purifier, but it may not be desirable to recommend an air purifier to a user who purchased an air purifier filter. Taking this into consideration, it is possible to create a purchase category dictionary 700 that takes into account the product purchase order.

Figure 8 shows an example of a process for recommending products to be purchased together in one embodiment of the present invention.

Referring to FIG. 8, in step S801, the processor 220 may extract a recommended product with a high probability of being purchased together with a reference product through a language model learned with purchase data. The processor 220 can use a language model learned for the upper category to which the reference product belongs to select recommended products focusing on products with a short vector distance indicating a linguistic relationship with the reference product among all products in the higher category. . At this time, the processor 220 may select a product whose lower category is different from the standard product among products in the upper category as a recommended product.

In step S802, the processor 220 may rearrange and display the recommended products selected in step S801 using a purchase category dictionary together with the reference product. Products belonging to the same category have high linguistic similarity, such as keywords used, so when only products with short vector distances are selected through a language model, products in the same subcategory as the standard product sometimes become the main products. To improve these results, you can exclude products in the same subcategory as the reference product and recommend products in a different subcategory from the reference product as products to be purchased together with the reference product. You can rearrange the products you want to purchase.

Referring to FIG. 9, the processor 220 may extract a recommended product 910 to be purchased together with a reference product belonging to category W through a language model, and at this time, the recommended product 910 extracted through the language model is first divided into vector distances. You can sort them in descending order (highest language model score).

The processor 220 may check the category dictionary based on the subcategory of the standard product. In the together purchase category dictionary, the categories of products purchased together with products of Category W are Category Alternatively, the recommended products 910 extracted through the language model can be rearranged according to the ratio.

For example, assume that for a specific t-shirt z, knit a, jumper b, and pants c are extracted through a language model in the order of similar linguistic relationships. If a t-shirt and pants are counted as purchased together in the category dictionary, the products to be purchased with t-shirt z are not recommended in the order of knit, jumper b, and pants c. Instead, the category dictionary information is reflected in the order of pants c, knit, and jumper b. It can be recommended by rearranging.

As another example, if the result of counting the categories purchased together with the <Milk> category is 90 soy milk, 70 lactic acid bacteria, and 40 juice, the product to be purchased along with the standard product <Easily digestible sterilized milk 190ml> is used as a language model. After selection, soy milk products, lactic acid bacteria products, and juice products among the selected products can be rearranged and exposed in a ratio of 9:7:4.

Not only can the recommended products (910) extracted through the language model be rearranged according to the product purchase rate for each category, but also the weight for the combined purchase categories can be determined according to the product purchase rate, and the corresponding product in the recommended product (910) can be determined. It is also possible to reorder to reflect the weight according to the purchase rate of products in the category to which it belongs.

In this embodiment, the limited category recommendation problem, which can be considered a limitation of the language model, can be solved by using a category dictionary.

Figure 10 shows another example of a process for recommending products to be purchased together in one embodiment of the present invention.

In this embodiment, the reordering of recommended products using a category dictionary can be personalized.

Referring to FIG. 10 , in step S1001, the processor 220 may extract a recommended product with a high probability of being purchased together with the reference product through a language model learned with joint purchase data of the upper category to which the reference product belongs.

In step S1002, the processor 220 may personalize the rearrangement of recommended products based on personal history related to products in the upper category to which the reference product belongs when rearranging the recommended products selected in step S1001. For example, the processor 220 uses purchase data of an individual user or a user group to which the user belongs (e.g., a group of users with similar profiles, interests, shopping patterns, search patterns, etc.) in the upper category to which the reference product belongs. Thus, a dictionary of purchase categories can be constructed in a form that takes into account the tastes of individuals or groups. As another example, the processor 220 creates a category dictionary focusing on products clicked or searched by the user during a recent certain period (for example, 30 days) or products registered as a shopping cart or favorite product within the upper category to which the standard product belongs. It can be configured.

For example, when recommending products to purchase together for User A and User B who are looking at the product <Milk Sterilized Low Fat 2% 200ml>, User A is recommended the product <Juice Fruit Vegetable Salad Red 200ml/Healthy according to each user's taste. Contains dietary fiber and lycopene> for user B, and <health protein supplement whey protein powder protein stick chocolate> to user B. Additionally, when a user who was searching for <Han River Date> and <Lunch Box> is looking at the <Mat> product, products in the <Picnic Basket> category that were frequently purchased by a group of users with similar search/shopping patterns as the user may be recommended. You can.

The processor 220 may display products in a category that matches the user's taste or products in a category that corresponds to recent shopping intention among the recommended products selected by the language model, sorted at the top.

Figure 11 shows another example of a process for recommending products to be purchased together in one embodiment of the present invention.

In this embodiment, the explainability of the recommendation results can be improved by recommending products based on keywords.

Referring to FIG. 11 , in step S1101, the processor 220 may extract a recommended product with a high probability of being purchased together with the reference product through a language model learned with joint purchase data of the upper category to which the reference product belongs.

In step S1102, the processor 220 may rearrange the recommended products selected in step S1101 based on keywords related to the standard product. For example, for users who are looking at the product <Spicy dumpling dumplings 385Gx4 bags>, products such as <Namhae Spicy Oyster Sauce 350G> and <Cheongyangcho Spicy Mulnaengmyeon for 2 954.4G> were recommended using the keyword ‘spicy’. You can use the keyword ‘Wang Gyoza’ to recommend products such as <Dumpling Kimchi Wang Gyoja 420Gx4 bags>, product <Dumpling and Shrimp Wang Gyoja 315Gx6 pieces>, etc.

In order to rearrange recommended products based on keywords, you can map the main keywords related to each product in advance and use them for recommendations by mapping keywords between products. For example, the keywords 'spicy' and 'dumpling' were mapped to the products <Spicy dumpling dumplings 385Gx4 bags> and <Dumpling Kimchi Wang dumplings 420Gx4 bags>, and the keywords 'shrimp', ‘Seafood’ and ‘dumplings’ were mapped, and the keywords ‘low fat’ and ‘milk’ were mapped to the product <Milk Sterilized Low Fat 2% 200ml>, and the product <Juice Fruit and Vegetable Salad Red 200ml/Contains dietary fiber and lycopene for health> You can map the keywords 'fruit juice' and 'diet' to the product <Health Protein Supplement Whey Protein Powder Protein Stick Chocolate> and the keywords 'protein powder' and 'chocolate'. In this way, key keywords for each product can be mapped in advance, and depending on the embodiment, it is also possible to create and utilize a model that predicts key keywords from the linguistic information of the product.

Figure 12 shows another example of a process for recommending products to be purchased together in one embodiment of the present invention.

In this embodiment, recommended products can be exposed at a time when the user's individual product purchase behavior is taken into consideration.

Referring to FIG. 12, in step S1201, the processor 220 may extract a recommended product with a high probability of being purchased together with the reference product through a language model learned with joint purchase data of the higher category to which the reference product belongs.

In step S1202, the processor 220 may rearrange the recommended products selected in step S1201 using the purchase category dictionary.

In step S1203, the processor 220 may determine the exposure time of the recommended products rearranged in step S1202 by considering the user's individual product purchase behavior. The timing of product recommendation can be said to be important because each user's product purchasing behavior varies and the point in time when existing shopping intentions are completed and new shopping intentions can be formed is different. It is possible to create a model that learns product purchase behavior, such as how many times a product is checked on average before a purchase is made or what actions are taken immediately before purchase. Through this, users who show similar product purchase behavior can be grouped together to create different products for each user group. Recommended products can be displayed at the right time. If product B, product C, and product D are selected as recommended products that are likely to be purchased together with product A, product B, product C, and product D can be recommended to each user at different times. For example, to User A, who shows purchasing behavior after registering a favorite product, Product B, Product C, and Product D can be recommended at the time Product A is registered as a favorite product, and 5 similar products of the same category can be added to the shopping cart. Product B, Product C, and Product D can be recommended to User B, who registers more than 5 products and compares them before purchasing. When 5 or more other products in the same category as Product A are registered.

Personal purchase behavior can be used to determine the exposure time of recommended products to be purchased together, and depending on the embodiment, it is also possible to use personal purchase behavior to determine the category of recommended products. For example, when recommending a product to User B, who registers more than 5 similar products of the same category in the shopping cart and shows the behavior of comparing and then purchasing, if there are less than 5 similar products registered in the shopping cart, the standard product and Other products in the same subcategory are recommended first, and later, when five or more similar products are registered in the shopping cart, a product in a subcategory different from the standard product that is likely to be purchased together with the standard product can be recommended.

Figure 13 shows an example of recommending products to be purchased together in one embodiment of the present invention.

The processor 220 selects products that have a similar linguistic relationship to the reference product among products with different subcategories from the reference product through a language model learned from the purchase data together, and then uses a category dictionary composed of an aggregation of the purchase data to create a language model. Selected products can be reordered to recommend products in categories that are likely to be purchased together with the standard product.

For example, as shown in FIG. 13, the processor 220 may provide a recommended product 1320 in a category with a high probability of being purchased together with the reference product 1310 whose subcategory is <treatment>.

The processor 220 has a linguistic relationship with the reference product 1310 among products of subcategories (e.g., shampoo, hair dye, hair loss care, hair pack, rinse, toothpaste, etc.) different from the reference product 1310 through the language model. A similar product can be selected as a recommended product (1320).

The processor 220 may use the purchase category dictionary to rearrange and display the recommended products 1320 in the order of subcategories of products frequently purchased together with the reference product 1310.

Before the user's behavior reaches the past product purchase behavior shown by the individual user or the user group to which the user belongs, a product (not shown) of the same subcategory as the reference product 1310 may be recommended, and thereafter, the user's behavior in the past may be recommended. At the point when product purchase behavior is reached, products (1320) of other subcategories that are likely to be purchased along with the standard product (1310) can be recommended through an active interface using pop-ups.

The processor 220 may provide a recommended product 1320 at a personalized time by considering product purchase behavior with different patterns for each user.

In this way, according to embodiments of the present invention, recommendation quality can be improved by recommending products in a category with a high probability of being purchased together with the reference product based on actual purchase data. And, according to embodiments of the present invention, the reliability and accuracy of recommendation can be increased by recommending products with characteristics that are frequently purchased together by considering the linguistic characteristics of the reference product. In addition, according to embodiments of the present invention, it is possible to provide a product recommendation model that is robust to the cold start problem by recommending products suitable for overall purchase behavior by using a category dictionary for purchase data along with a language model learned from purchase data. there is.

The device described above may be implemented with 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 programmable logic unit (PLU). It may be implemented using one or more general-purpose or special-purpose computers, such as a logic unit, microprocessor, or any other device 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 operating system. Additionally, a processing device may access, store, manipulate, process, and generate data in response to the execution of software. For ease of understanding, a single processing device may be described as being used; however, those skilled in the art will understand that a processing device includes multiple processing elements and/or multiple types of processing elements. It can be seen that it may include. For example, a processing device may include a plurality of processors or one processor and one controller. Additionally, other processing configurations, such as parallel processors, are possible.

Software may include a computer program, code, instructions, or a combination of one or more of these, which may configure a processing unit to operate as desired, or may be processed independently or collectively. You can command the device. The software and/or data may be embodied in any type of machine, component, physical device, computer storage medium or device for the purpose of being interpreted by or providing instructions or data to the processing device. there is. 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. At this time, the medium may continuously store a computer-executable program, or temporarily store it for execution or download. In addition, the medium may be a variety of recording or storage means in the form of a single or several pieces of hardware combined. It is not limited to a medium directly connected to a computer system and may be distributed over a network. Examples of media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, And there may be something configured to store program instructions, including ROM, RAM, flash memory, etc. Additionally, examples of other media include recording or storage media managed by app stores that distribute applications, sites or servers that supply or distribute various other software, etc.

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

Therefore, other implementations, other embodiments, and equivalents of the claims also fall within the scope of the claims described below.

Claims (20)

In a product recommendation method executed on a computer device,
The computer device includes at least one processor configured to execute computer-readable instructions contained in a memory,
The method of recommending the above product is:
Recommending, by the at least one processor, a product to be purchased together with a reference product using a language model learned with joint purchase data indicating a history of purchasing two or more products together within a certain time period.
Including,
The language model is a model in which the relationships between products are learned using linguistic information made up of natural language in relation to each product.
A product recommendation method characterized by . According to paragraph 1,
The language model is learned for each upper category,
The recommended steps above are:
Step of selecting a recommended product through a language model learned for the upper category to which the reference product belongs
A product recommendation method including. According to paragraph 2,
The language model is learned with purchase data from different periods of time according to the parent category.
A product recommendation method characterized by . According to paragraph 1,
The recommended steps above are:
Recommending a product that has a linguistic relationship with the standard product among products that have a different category from the standard product using the language model
A product recommendation method including. According to paragraph 2,
The recommended steps above are:
Using a language model learned for the upper category to which the standard product belongs, recommending products that have a different subcategory from the standard product and have a linguistic relationship with the standard product among products in the higher category
A product recommendation method including. According to paragraph 1,
The recommended steps above are:
A step of reordering recommended products selected through the language model using a category dictionary in which categories of products purchased together are aggregated based on the joint purchase data.
A product recommendation method including. According to clause 6,
The category dictionary is composed of the results of aggregating the categories of products purchased together for the entire category set.
A product recommendation method characterized by . According to clause 6,
The category dictionary is created for each high-level category and consists of the aggregate results of sub-category units that subdivide the high-level category.
A product recommendation method characterized by . According to clause 6,
The rearrangement step is,
Confirming together purchase categories for the category to which the reference product belongs through the category dictionary and rearranging at least one of the order of the recommended products and the ratio by category
A product recommendation method characterized by . According to paragraph 1,
The recommended steps above are:
A step of rearranging recommended products selected through the language model based on the product-related history of the individual user or the user group to which the user belongs.
A product recommendation method including. According to clause 6,
The category dictionary aggregates and personalizes the categories of products purchased together based on the joint purchase data by the individual user or the user group to which the user belongs.
A product recommendation method characterized by . According to paragraph 1,
The recommended steps above are:
A step of rearranging recommended products selected through the language model based on keywords related to the reference product
A product recommendation method including. According to paragraph 1,
The recommended steps above are:
A step of determining the exposure time of the recommended product selected through the language model in consideration of the product purchase behavior of the individual user or the user group to which the user belongs.
A product recommendation method including. According to clause 13,
The determining step is,
Exposing the recommended product at the point when the user's behavior reaches the purchasing behavior of the product
A product recommendation method characterized by . A computer program stored in a computer-readable recording medium for executing the product recommendation method of any one of claims 1 to 14 on a computer device. In computer devices,
At least one processor configured to execute computer readable instructions contained in memory
Including,
The at least one processor,
The process of recommending products to be purchased together with a reference product using a language model learned with joint purchase data that represents the history of purchasing two or more products together within a certain period of time
Process it,
The language model is a model in which the relationships between products are learned using linguistic information made up of natural language in relation to each product.
A computer device characterized by a. According to clause 16,
The at least one processor,
Reordering the recommended products selected through the language model using a category dictionary in which categories of products purchased together are aggregated based on the joint purchase data.
A computer device characterized by a. According to clause 17,
The at least one processor,
Confirming together purchase categories for the category to which the reference product belongs through the category dictionary and rearranging at least one of the order of the recommended products and the ratio by category
A computer device characterized by a. According to clause 16,
The at least one processor,
Reordering recommended products selected through the language model based on the product-related history of the individual user or the user group to which the user belongs.
A computer device characterized by a. According to clause 16,
The at least one processor,
Determining the exposure time of recommended products selected through the language model by considering the product purchase behavior of the individual user or the user group to which the user belongs
A computer device characterized by a.

Images