Method for providing electronic commerce service to user terminal, electronic device for performing method, and non-transitory computer-readable recording medium on which computer program for performing method is stored

Abstract

A method for providing an electronic commerce service to a user terminal, according to one embodiment, may be provided, the method comprising the steps of: determining a metric by collecting user behavior data and product-related data; storing information related to the metric in a meta store and a feature store; storing, on the basis of the information stored in the meta store and the feature store, an artificial intelligence model corresponding to a result modeled through profiling; and providing the electronic commerce service to the user terminal on the basis of output data for the artificial intelligence model. According to another embodiment, an electronic device for performing the method may be provided. According to another embodiment, a non-transitory computer-readable recording medium on which a computer program for performing the method is stored may be provided.

Description

A method for providing e-commerce services to a user terminal, an electronic device for performing the method, and a non-transient computer-readable recording medium containing a computer program for performing the method

[0001] The present disclosure relates to an invention for providing e-commerce services to a user terminal, and more specifically to a technology for modeling an artificial intelligence model to provide e-commerce services optimized for the user.

[0002] In e-commerce and marketing services, there is a growing demand to provide personalized experiences, such as customized product recommendations and promotions, by analyzing customers' purchase history, interests, and web or app usage patterns. Traditionally, basic classification and targeting have been performed using simple rule-based algorithms, such as rules like "recommend similar products to users who purchased a specific item within the last week." While rule-based algorithms offer the advantages of simple implementation and intuitiveness, the development of e-commerce services and the increasing diversity of users and services have led to increasingly complex delivery methods. Furthermore, as the volume of data to be handled has grown massively, problems have arisen regarding conflicts or the accumulation of overly limited rules. Additionally, limitations have emerged in that it is difficult to capture hidden correlations or subtle patterns using only rule-based algorithms.

[0003] Against this backdrop, there is a growing need for an integrated profiling solution that leverages the automatic learning and inference capabilities of machine learning (ML) algorithms to ensure accuracy and flexibility in e-commerce service provision, reduce redundant rule settings and conflicts, and enable rapid decision-making while reflecting the diverse needs of e-commerce providers and users.

[0004] The technical problems that this embodiment aims to solve are not limited to those described above, and other technical problems can be inferred from the following embodiments.

[0005] According to one embodiment, a method for providing an e-commerce service to a user terminal may be provided, comprising: a step of determining a metric by collecting user behavior data and product-related data; a step of storing information related to the metric in a meta store and a feature store; a step of storing an artificial intelligence model corresponding to a result modeled through profiling based on the information stored in the meta store and the feature store; and a step of providing an e-commerce service to the user terminal based on output data for the artificial intelligence model.

[0006] According to another embodiment, an electronic device for providing e-commerce services to a user terminal may be provided, comprising: a memory; a transceiver; and at least one processor, wherein the at least one processor controls the memory and the transceiver to enable the electronic device to: collect user behavior data and product-related data to determine a metric, store information related to the metric in a meta store and a feature store, store an artificial intelligence model corresponding to a result modeled through profiling based on the information stored in the meta store and the feature store, and provide e-commerce services to the user terminal based on output data for the artificial intelligence model.

[0007] According to another embodiment, a non-transient computer-readable recording medium may be provided that contains a computer program for performing a method of providing an e-commerce service to a user terminal, wherein the method comprises: a step of collecting user behavior data and product-related data to generate a metric; a step of storing information related to the metric in a meta store and a feature store; a step of storing an artificial intelligence model corresponding to a result modeled through profiling based on the information stored in the meta store and the feature store; and a step of providing an e-commerce service to a user terminal based on output data for the artificial intelligence model.

[0008] According to the proposed embodiment, one or more of the following effects can be expected.

[0009] By utilizing the artificial intelligence model of the present disclosure, the problems associated with profiling performed using conventional rule-based algorithms—namely, the burden of managing complex rules, conflicts, and limitations in the level of personalization—can be significantly improved. Furthermore, by automatically learning the hidden correlations between user behavior patterns and product characteristics through the method utilizing the artificial intelligence model of the present disclosure, recommendations or marketing optimized for individual users can be performed, thereby simultaneously ensuring service quality and scalability.

[0010] The effects of the invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by a person skilled in the art from the description in the claims.

[0011] FIG. 1 illustrates an overall method performed by an electronic device according to one embodiment.

[0012] FIG. 2 illustrates a flowchart of a method according to one embodiment.

[0013] FIG. 3 is a flowchart illustrating a method for modeling an artificial intelligence model according to one embodiment.

[0014] FIG. 4 is a flowchart illustrating a method of applying an artificial intelligence model determined according to one embodiment to an e-commerce service.

[0015] FIG. 5 is a flowchart illustrating a method for providing an e-commerce service to a test subject user through experimental operation using an artificial intelligence model according to one embodiment.

[0016] FIG. 6 is a diagram illustrating how an artificial intelligence model determined based on the results of an electronic device performing modeling according to one embodiment is utilized.

[0017] FIG. 7 illustrates a block diagram of an electronic device according to one embodiment.

[0018] According to one embodiment, a method for providing an e-commerce service to a user terminal may be provided, comprising: a step of determining a metric by collecting user behavior data and product-related data; a step of storing information related to the metric in a meta store and a feature store; a step of storing an artificial intelligence model corresponding to a result modeled through profiling based on the information stored in the meta store and the feature store; and a step of providing an e-commerce service to the user terminal based on output data for the artificial intelligence model.

[0019] The step of storing in a metastore and a feature store of a method according to one embodiment may include: a step of generating a feature to be input into an artificial intelligence model based on a metric and storing it in a feature store; and a step of storing metadata related to the feature in a metastore.

[0020] The step of storing an artificial intelligence model of a method according to one embodiment may further include: a step of performing profiling for entities distinguished by at least one of users and products based on information stored in a meta store and a feature store; a step of determining an artificial intelligence model by performing modeling based on the profiling results; and a step of outputting and storing a file containing information about the determined artificial intelligence model.

[0021] The step of determining an artificial intelligence model of a method according to one embodiment may include: determining input data and output data for an artificial intelligence model based on at least one metric included in information stored in a meta store and a feature store; and determining an artificial intelligence model by performing modeling using the determined input data and output data.

[0022] The step of providing an e-commerce service according to one embodiment of the method may further include: a step of performing an experimental serving of an artificial intelligence model; and a step of performing a formal distribution based on whether the result of the experimental serving satisfies predetermined conditions.

[0023] The step of performing an experimental operation of a method according to one embodiment may further include: determining at least one experimental user among a plurality of users who satisfies predetermined conditions; and providing an e-commerce service to at least one experimental user based on the output data of an artificial intelligence model.

[0024] The step of performing formal distribution of a method according to one embodiment may include: collecting activity information regarding an e-commerce service of at least one experimental subject user; and performing formal distribution when the activity information corresponds to the input data and output data of an artificial intelligence model.

[0025] The step of performing formal distribution of the method according to one embodiment may further include: a step of re-modeling the artificial intelligence model based on activity information when the activity information does not correspond to the input data and output data of the artificial intelligence model; and a step of performing experimental operation of the re-modeled artificial intelligence model.

[0026] The step of providing an e-commerce service based on output data of an artificial intelligence model according to one embodiment of the method may include: tagging at least one experimental subject user; calling an experimental API when the tagged at least one experimental subject user accesses the e-commerce service; and providing the e-commerce service based on output data of the artificial intelligence model to the tagged at least one experimental subject user in response to the call of the experimental API.

[0027] According to another embodiment, an electronic device for providing e-commerce services to a user terminal may be provided, comprising: a memory; a transceiver; and at least one processor, wherein the at least one processor controls the memory and the transceiver to enable the electronic device to: collect user behavior data and product-related data to determine a metric, store information related to the metric in a meta store and a feature store, store an artificial intelligence model corresponding to a result modeled through profiling based on the information stored in the meta store and the feature store, and provide e-commerce services to the user terminal based on output data for the artificial intelligence model.

[0028] According to one embodiment, at least one processor may be configured to generate a feature for input to an artificial intelligence model based on a metric and store it in a feature store, and to store metadata related to the feature in a meta store.

[0029] According to one embodiment, at least one processor may be further configured to perform profiling for entities distinguished by at least one of users and products based on information stored in a metastore and a feature store, perform modeling based on the profiling results to determine an artificial intelligence model, and output and store a file containing information about the determined artificial intelligence model.

[0030] According to one embodiment, at least one processor may be configured to determine input data and output data for an artificial intelligence model based on at least one metric included in information stored in a metastore and a feature store, and to determine an artificial intelligence model by performing modeling using the determined input data and output data.

[0031] According to one embodiment, at least one processor may be further configured to perform experimental operation of an artificial intelligence model and to perform formal deployment based on whether the result of the experimental operation satisfies predetermined conditions.

[0032] According to one embodiment, at least one processor may be further configured to determine at least one experimental user among a plurality of users who satisfies predetermined conditions, and to provide an e-commerce service to at least one experimental user based on the output data of an artificial intelligence model.

[0033] According to one embodiment, at least one processor may be configured to collect activity information regarding an e-commerce service of at least one experimental subject user, and to perform formal distribution when the activity information corresponds to the input data and output data of an artificial intelligence model.

[0034] According to one embodiment, at least one processor may be further configured to re-model an artificial intelligence model based on activity information and perform experimental operation of the re-modeled artificial intelligence model when activity information does not correspond to the input data and output data of the artificial intelligence model.

[0035] According to one embodiment, at least one processor may be configured to tag at least one experimental user, and when the tagged at least one experimental user accesses an e-commerce service, call an experimental API and provide the e-commerce service to the tagged at least one experimental user based on the output data of an artificial intelligence model in response to the call to the experimental API.

[0036] According to another embodiment, a non-transient computer-readable recording medium may be provided that contains a computer program for performing a method of providing an e-commerce service to a user terminal, wherein the method comprises: a step of collecting user behavior data and product-related data to generate a metric; a step of storing information related to the metric in a meta store and a feature store; a step of storing an artificial intelligence model corresponding to a result modeled through profiling based on the information stored in the meta store and the feature store; and a step of providing an e-commerce service to a user terminal based on output data for the artificial intelligence model.

[0037] Specific details of other embodiments are included in the detailed description and drawings.

[0038] The terms used in the embodiments have been selected to be as widely used as possible, taking into account their functions in the present disclosure; however, these may vary depending on the intent of those skilled in the art, case law, the emergence of new technologies, etc. Additionally, in specific cases, terms have been arbitrarily selected by the applicant, and in such cases, their meanings will be described in detail in the relevant explanatory section. Therefore, terms used in the present disclosure should be defined not merely by their names, but based on their meanings and the overall content of the present disclosure.

[0039] When a part of a specification is described as "including" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components.

[0040] The expression "at least one of a, b, and c" described throughout the specification may include 'a alone', 'b alone', 'c alone', 'a and b', 'a and c', 'b and c', or 'a, b, and c all'.

[0041] The "terminal" mentioned below may be implemented as a computer or portable terminal capable of connecting to a server or other terminal via a network. Here, the computer includes, for example, a notebook, desktop, or laptop equipped with a web browser, and the portable terminal may include, for example, a wireless communication device that ensures portability and mobility, and may include all types of handheld-based wireless communication devices such as IMT (International Mobile Telecommunication), CDMA (Code Division Multiple Access), W-CDMA (W-Code Division Multiple Access), LTE (Long Term Evolution), communication-based terminals, smartphones, tablet PCs, etc.

[0042] Embodiments of the present disclosure are described below with reference to the attached drawings so that those skilled in the art can easily implement them. However, the present disclosure may be embodied in various different types and is not limited to the embodiments described herein.

[0043] Embodiments of the present disclosure will be described in detail below with reference to the drawings.

[0044] FIG. 1 illustrates an overall method performed by an electronic device (100) according to one embodiment. Referring to FIG. 1, a project owner (110) may correspond to an operator or user performing a rule-based algorithm (hereinafter referred to as the "first manager"). A data scientist and a machine learning scientist (120) may correspond to a manager performing a machine learning (ML)-based approach (hereinafter referred to as the "second manager"). The electronic device (100) may implement various embodiments of the present disclosure using a profiling platform (130), a profiling modeling engine (140), and a learning platform (150), and the operation of each component may be performed through at least one processor. In the following, communication between the first manager (110) and the second manager (120) and the electronic device (100) may be interpreted as being performed through their respective manager terminals. According to one embodiment, the administrator terminal may include various electronic devices such as mobile devices, servers, PCs, and laptops.

[0045] According to one embodiment, the first manager (110) may provide information for defining user segmentation or logic using simple conditional statements (such as if-then-else) in the process of implementing a machine learning-based algorithm using the electronic device (100) of the present disclosure. According to one embodiment, the first manager (110) may define requirements for an e-commerce service or set user segmentation, nudge advertising types, etc. using a simple rule-based algorithm. For example, a rule such as "users who have purchased a specific brand within the last 7 days are classified into Segment A" may be registered in the profiling platform (130) to generate simple conditional classification or aggregation metrics.

[0046] According to one embodiment, the second manager (120) is a data scientist (DS) or machine learning scientist who intends to classify users or products more complexly or develop prediction models through machine learning algorithms, performs machine learning model definition, hyperparameter setting, learning schedule setting, etc., and can produce various metrics (e.g., indicators such as "conversion probability", "user propensity") generated based on data input and output through e-commerce services. The second manager (120) can model an artificial intelligence model through a profiling platform (130) and a profiling modeling engine (140). Specifically, the second manager (120) can provide information for modeling an artificial intelligence model from vast amounts of input and output data (e.g., click logs, purchase history, review text, etc.) obtained through e-commerce services, and can perform the role of evaluating and updating the performance of the artificial intelligence model using offline statistical indicators (accuracy, recall, AUC) or experimental results (A / B test). The second manager (120) can transmit input metrics, output metrics, etc. through the profiling platform (130) and instruct the profiling modeling engine (140) to create a machine learning model.

[0047] According to one embodiment, the electronic device (100) may use input metrics and output metrics, as well as goal metrics, for modeling. The goal metric may include information for evaluating how accurately the artificial intelligence model has inferred output metrics based on input metrics. According to one embodiment, the goal metric may vary depending on the modeling type of the artificial intelligence model. For example, in an artificial intelligence model for binary classification, AUC, Precision, Recall, F1-score, Logloss, etc. may be included in the goal metric, and in a recommendation problem, NDCG, MAP, Precision@K, etc. may be included in the goal metric.

[0048] According to one embodiment, the electronic device (100) may utilize a target metric that includes information on actual behavioral indicators suitable for business purposes. For example, the number of nudge clicks exposed to a user may be extracted from user behavior data and product-related data and set as the target metric. Since the target metric can be used as a criterion for evaluating how accurately the output value generated by the artificial intelligence model actually induced user behavior, the electronic device (100) can perform modeling by utilizing this target metric as information for modeling, thereby judging the performance of the artificial intelligence model that infers output metrics based on input metrics based on the target metric. According to this embodiment

[0049] According to one embodiment, user behavior data may include information regarding various activities displayed by a user within a platform where an e-commerce service is provided, such as viewing products, entering search terms, clicking, adding to a cart, purchase history, and writing reviews. According to one embodiment, product-related data may include elements such as product name, description, category, price, stock, rating, sales volume, and images, as information about the product itself. Product-related data does not need to be interpreted as being limited to data related to physical inventory products and may include data related to services provided by the e-commerce service (e.g., food delivery services, streaming services, purchasing agency services, etc. associated with the business entity providing the e-commerce service). The types of such data do not need to be interpreted as being limited to the data described above, and may be interpreted as including various types of data that the electronic device (100) can refer to to model an artificial intelligence model for providing a personalized e-commerce service.

[0050] According to one embodiment, the input and output data utilized in the modeling process may correspond to the input and output metrics utilized in a rule-based algorithm. For example, the electronic device (100) can determine input metrics for each user and output metrics determined based thereon based on collected user behavior data and product-related data, and can perform modeling through a learning process based on these input and output metrics. That is, the electronic device (100) can utilize various indicators that were calculated according to conditions directly defined by humans in conventional systems by converting them into input values ​​and correct answers for learning an artificial intelligence model. For example, the electronic device (100) can collect user behavior data and analyze it to calculate input metrics such as quantitative figures related to user behavior data for individual users (e.g., number of clicks in the last week, frequency of viewing of products in a specific category, number of times items were added to the cart, elapsed time since the last purchase, or average review rating, etc.). The electronic device (100) can process the input metrics to use features that reflect each user's purchasing propensity, interest, and behavioral patterns.

[0051] Additionally, output metrics serve as target indicators for the user's expected behavior, such as whether a specific product is purchased (yes / no), click probability, conversion rate, or churn probability score, and these may also be included in the information acquired and stored by the electronic device (100) as a result of execution according to a rule-based algorithm. Since output metrics are derived based on the results of user behavior that actually occurred in past data, the electronic device (100) can use these output metrics as correct answers during supervised learning of the artificial intelligence model.

[0052] In this way, the electronic device (100) can perform modeling by generating training data by configuring input metrics and output metrics as pairs and training an artificial intelligence model based on this. The model trained in this way can perform inference based on the same input metrics for new users or situations, and through this, it can be applied to various e-commerce services such as generating recommendation results, user segmentation, and personalized configurations and content to be displayed on a user interface.

[0053] According to one embodiment, the profiling platform (130) can centrally collect and manage user and product-related data, activity information, logs, and metrics, as well as features based thereon. According to one embodiment, the profiling platform (130) effectively processes user behavior data and product-related data in an environment providing e-commerce services, and enables integrated operation that satisfies both rule-based algorithm and machine learning-based requirements based thereon.

[0054] According to one embodiment, the profiling platform (130) can explore features from metrics determined based on user behavior data and product-related data collected during the operation of an e-commerce service. For example, data required for modeling can be determined by analyzing how often a user has placed an order (order-total-count) or how long a specific product page has been viewed recently (SDP-recent-view-time). For example, when the first manager (110), who is the project owner, intends to determine a new metric called "transition-to-wow-possibility," the profiling platform (130) can determine this metric based on existing collected data.

[0055] According to one embodiment, the profiling platform (130) can, in the process of collecting user behavior data and product-related data, retrieve data in real time from an Integrated Workload Layer (IWL) where user behavior information is stored or from an external database and convert it into a structured form. The collected data is stored in a feature store and can subsequently be used in training or inference tasks for an artificial intelligence model.

[0056] According to one embodiment, the profiling platform (130) can manage a data pipeline to process raw data through an ETL (Extract, Transform, Load) process and convert it into features suitable for modeling an artificial intelligence model. For example, the profiling platform (130) can perform tasks such as normalizing the user's recent view time by time zone or aggregating the number of orders by specific user segment.

[0057] According to one embodiment, the profiling platform (130) can manage input metrics and output metrics of an artificial intelligence model. The profiling platform (130) can organize input metrics and output metrics (e.g., the number of orders and the number of clicks on a specific nudge) set by the first manager (110) and record them as metadata in a metastore. If the second manager (120) requests specific features or output metrics for a new artificial intelligence model, the profiling platform (130) can structure them and store them in a feature store.

[0058] According to one embodiment, the profiling platform (130) can integrate rule-based and machine learning-based operations. For example, when a first manager (110) creates a rule to filter "users who have placed three or more orders in the last 7 days," the profiling platform (130) calculates the result of processing the data. At the same time, when a second manager (120) requests a model to predict "conversion probability of users with long recent view times," the profiling platform (130) prepares the relevant data and provides it to the model training pipeline. A structure in which these two approaches coexist and operate complementarily can significantly increase the efficiency and flexibility of data management.

[0059] According to one embodiment, the profiling modeling engine (140) performs actual modeling work or performs modeling work based on data and configuration information (such as rules and definitions of machine learning models) transmitted from the profiling platform (130). For example, the profiling modeling engine (140) performs training, deployment, testing, and performance analysis of an artificial intelligence model in an electronic device (100), and performs a process of receiving data from the profiling platform (130), creating an artificial intelligence model, and finally evaluating it through experimental operation.

[0060] According to one embodiment, based on input metrics and output metrics provided by the profiling platform (130), the profiling modeling engine (140) can perform learning using the learning platform (150). For example, when using input metrics including the number of orders and the number of nudge clicks to create a "transition-to-wow-possibility" model to infer the conversion rate (output metric) of users to paid users, the profiling modeling engine (140) can train an artificial intelligence model based on the input metrics and output metrics, and evaluate the results using metrics such as precision, recall, and AUC (Area Under the Curve). For example, if the artificial intelligence model inferred that 10 users would convert to paid users and 8 actually converted, the precision is 80%, and if there were 12 actual paid users and the artificial intelligence model correctly predicted 9, the recall is 75%, and the AUC, which is the result of the artificial intelligence model dividing those who will become paid users and those who will not, can be determined to be 0.9 (maximum 1).

[0061] According to one embodiment, the profiling modeling engine (140) can export the completed artificial intelligence model in the form of a file and perform experimental serving based thereon. Since the artificial intelligence model file contains weights and metadata, the artificial intelligence model can be utilized based on the file. For example, if the "transition-to-wow-possibility" model is deployed to an experiment that displays a nudge to the top 5% of users, the engine can quickly deliver it to the execution environment.

[0062] According to one embodiment, the profiling modeling engine (140) can perform A / B testing for experimental operation and, for this purpose, analyze the effect based on the results applied only to specific users (i.e., experimental target users). For example, the electronic device (100) can apply a new artificial intelligence model to only 10% of the total traffic and maintain a conventional algorithm for the remainder, and then track changes in conversion rates or click rates. User behavior data generated in this process can be synchronized with the IWL and used for subsequent analysis.

[0063] According to one embodiment, the profiling modeling engine (140) can analyze the performance and business performance of the modeled artificial intelligence model through a reporting function. The profiling modeling engine (140) can collect accuracy, recall, and AUC as technical performance indicators of the artificial intelligence model, as well as business indicators such as GMV (Gross Merchandise Value), Churn Rate, and CLTV (Customer Lifetime Value). For example, it can generate a report stating that GMV increased by 5% after applying the "transition-to-wow-possibility" model and deliver it to the first manager (110) and / or the second manager (120).

[0064] According to one embodiment, the profiling modeling engine (140) of the electronic device (100) can perform training of an artificial intelligence model using a learning platform (150) based on data and metrics prepared in the profiling platform (130) through organic linkage with the profiling platform (130), or schedule batch predictions through Spark EMR in its own pipeline. For example, when a second manager (120) requests a new artificial intelligence model, the profiling platform (130) determines and structures the metrics, and the profiling modeling engine (140) can perform training and inference using them.

[0065] According to one embodiment, the profiling modeling engine (140) may perform preprocessing and feature transformation for configuring data in a trainable form for modeling an artificial intelligence model based on input metrics and output metrics. Since the data input from rule-based algorithms and machine learning-based algorithms may exist in various forms (e.g., categorical, numerical, text, etc.), the profiling modeling engine (140) includes a process of structuring it into features suitable for model training. For example, simple classification conditions defined by rule-based algorithms (e.g., whether a purchase was made within the last 7 days) may be simplified into Boolean values ​​or preprocessed to be expanded into high-dimensional features used in machine learning models (e.g., average time spent on product pages, click frequency in the last 30 days, etc.). The results of this series of feature processing are automatically performed in the pre-model training stage and contribute to improving the consistency and quality of the training pipeline.

[0066] According to one embodiment, the profiling modeling engine (140) may include a scheduler and an orchestration function for automatically managing the training, inference, and deployment schedules of an artificial intelligence model defined based on preprocessed metrics. For example, if the second person in charge (120) sets the artificial intelligence model to be trained offline at 2:00 AM every day, the profiling modeling engine (140) can automatically execute the relevant pipeline according to the set time to update the model with a retrained model using the latest data. According to one embodiment, the scheduler function can perform full-cycle orchestration, including not only simple model training but also performing A / B tests on the trained model, collecting evaluation metrics, and automatically generating reports. Accordingly, since regular verification and iterative improvement of the performance of the artificial intelligence model are possible, the electronic device (100) can automate the operation of the artificial intelligence model without manual intervention by the first manager (110) and / or the second manager (120). This structure enables rule-based algorithms and machine learning-based algorithms to coexist on the same platform without conflict, and has the effect of enhancing the stability and efficiency of the entire profiling and personalization service.

[0067] According to one embodiment, the learning platform (150) may include an external or internal system for training a machine learning model with large-scale data or performing batch inference. Various environments may be included, such as Spark, Hadoop, EMR (AWS), and internal platforms of servers providing other e-commerce services.

[0068] Specific operations in the present disclosure based on the features of such electronic device (100) will be described later through the following embodiments.

[0069] FIG. 2 illustrates a flowchart of a method according to one embodiment.

[0070] In step S210, the electronic device (100) can generate metrics by collecting user behavior data and product-related data. According to one embodiment, the electronic device (100) may include user behavior data that can be obtained through an e-commerce service, such as the user's search history, click events, page viewing records, whether items were added to a shopping cart, actual purchase history, whether a review was written, and wishlist registration activities. Such user behavior data is mainly collected in the form of logs, and each behavior may be stored in a structured form along with the time of occurrence, target product, category, etc. According to one embodiment, the electronic device (100) may include product-related data that can be obtained through an e-commerce service, such as price, discount rate, stock quantity, category information, seller type, average rating, number of reviews, sentiment keywords, and daily sales volume.

[0071] According to one embodiment, the electronic device (100) can generate metrics based on collected user behavior data and product-related data. For example, the electronic device (100) can generate metrics such as order-total-count, page clicks in the last 7 days, click-through rate (CTR) for a specific category, cart-abandonment-rate, review sentiment score, wishlist-to-purchase-rate, average purchase interval, and customer lifetime value (LTV).

[0072] In step S220, the electronic device (100) can store information related to the metric determined in step S210 in a meta store and a feature store. Based on the metric determined in step S210, the electronic device (100) can process it into a feature to be input into an artificial intelligence model or use it as a target indicator to measure business performance. Input metrics are quantitative indicators that express specific behaviors or states of a user or product numerically, and may refer to information that serves as the basis for feature creation. For example, items such as "number of orders in the last 30 days," "number of clicks in a specific category," and "whether a review has been written" correspond to input metrics, and the electronic device (100) can determine these from user behavior data and / or product-related data. Input features are the result of converting the input metrics into a form that can be directly learned by a machine learning model, and can be used as input values ​​for the actual artificial intelligence model. For example, the electronic device (100) can generate an input feature in the form of a "normalized order count value" or a "log scale order frequency" by undergoing processes such as missing value correction, range normalization, and log transformation based on an input metric called "order count". At this time, the input feature is stored in a feature store and can be called during the model training and inference phases.

[0073] According to one embodiment, the metastore stores and manages metadata including definition and relationship information for various features defined or generated through the profiling platform (130). According to one embodiment, for input metrics, information such as the metric identifier, name, description, data source (e.g., log table), calculation method, aggregation period, creation date and time, and creator may be stored in the metastore. For input features, metadata including which metric the feature was derived from, which preprocessing rule was applied, data type, null handling method, version information of the feature, and associated model identifier may be stored in the metastore. This information enables reliable data tracking in subsequent processes such as feature reuse, artificial intelligence model retraining, and regression analysis.

[0074] According to one embodiment, the feature store and meta store may be managed in memory within the electronic device (100) or stored on a server cloud-based or on-premises data infrastructure. For the feature store, a data lake or data store such as AWS S3, Google Cloud Storage, BigQuery, Snowflake, etc. may be utilized as an offline storage for large batch data, and a Key-Value based database such as Redis, Cassandra, DynamoDB, etc. may be used as an online storage for real-time inference. According to one embodiment, the meta store may be configured on a relational database such as MySQL, PostgreSQL, etc. or a metadata catalog system such as Hive Metastore, AWS Glue Data Catalog, MLflow.

[0075] In step S230, the electronic device (100) may store an artificial intelligence model corresponding to the result of modeling through profiling based on information stored in a meta store and a feature store according to one embodiment.

[0076] According to one embodiment, an electronic device (100) can construct a training dataset by loading definitions and version information of metrics and features required by an artificial intelligence model from a metastore and loading actual input values ​​from a feature store in order to create an artificial intelligence model. Subsequently, the profiling modeling engine (140) of the electronic device (100) can perform modeling of an artificial intelligence model (e.g., classification, regression, recommendation, sentiment analysis, etc.) based on this.

[0077] According to one embodiment, modeling can be performed to infer results such as user interest, potential for conversion to a paid membership, potential for purchase, and churn risk rate. Such inference work may set various goals according to each business purpose and application scenario, and the electronic device (100) may use output metrics that correspond to the inference goals during the modeling process. According to one embodiment, the output metrics may be information input by the second manager (120). Specifically, the output metrics correspond to labels in supervised learning that determine what result value to predict for features input during modeling, and may be set considering the purpose of inference, business needs, and the possibility of data collection. For example, when modeling a model that predicts a user's potential for purchase, the second manager (120) may designate events such as "whether a purchase was made within the last 7 days" or "whether a purchase occurred after a coupon was displayed" as output metrics. The electronic device (100) may derive output metrics from collected user behavior data, product-related data, or transaction records. According to one embodiment, the second manager (120) may set a single or multiple output metrics according to the purpose of the experiment, or design the modeling by distinguishing types such as continuous, binary, or multi-class.

[0078] For example, the electronic device (100) utilizes user behavior data, such as how long a user stayed on a specific product detail page and how often they browsed the relevant category, as input metrics for modeling a model to infer a user's interest in a product, and corresponding output metrics may be set such as "whether a specific banner was clicked," "whether a wishlist was added," "whether a CTA button on the page was clicked," and "conversion of the user account to a paid membership account." These output metrics are indicators that reflect behavior based on the user's actual interest and can be used as correct labels in the modeling process.

[0079] As another example, for modeling a model to infer a user's purchase probability, an electronic device (100) may use past purchase frequency, recent click count, shopping cart history, etc. as input metrics and define and use whether a purchase was completed within a certain period (e.g., whether a purchase was made within the last 7 days) as output metrics in a binary classification form.

[0080] As another example, the electronic device (100) may use input features such as a user's recent activity reduction pattern, long-term inactivity, and shopping cart drop rate for modeling a model for predicting churn risk, and may use output metrics such as inactivity of service, discontinuation of purchases, and withdrawal processing within a certain period.

[0081] As another example, the electronic device (100) may use input metrics such as view history related to the promotion, similar campaign responsiveness, and price sensitivity for modeling a model that infers the possibility of conversion to a paid membership, and output metrics such as "whether a nudge click was made," "whether a discount coupon was used," or "whether a product was purchased within the campaign."

[0082] According to one embodiment, the electronic device (100) may store the artificial intelligence model determined through step S230 for future reuse, evaluation, or distribution. The information stored may include not only model parameters and weights for the artificial intelligence model, but also additional metadata such as the unique identifier of the artificial intelligence model, creation date and time, list and version of input features, output metric information, data period used for training, performance indicators (AUC, Precision, Recall, etc.), associated experiment ID, and distribution status (Experimental, Mature, etc.). The electronic device (100) may manage the information regarding the artificial intelligence model so that it can be utilized later by storing it in a file format. When the artificial intelligence model is stored in this way, it is possible to implement prediction results under the same input feature conditions, and comparative analysis with past versions of the artificial intelligence model may be easy. Furthermore, if the artificial intelligence model satisfies a predetermined condition based on the response data of the experimental subject collected during the experimental operation process, the stored artificial intelligence model can be easily distributed to a formal operating environment (Mature Serving).

[0083] According to one embodiment, the electronic device (100) may be used to provide recommendations or analysis results to actual users by pre-verifying the performance of the model in an experimental operating environment and then formally distributing it in an operating environment. The electronic device (100) may be configured to perform offline learning on large-scale user behavior data and product-related data by linking the output model file to a distributed processing system.

[0084] FIG. 3 is a flowchart illustrating a method for modeling an artificial intelligence model according to one embodiment. Specifically, steps S332 to S338 of FIG. 3 may correspond to subdivided steps related to step S230 of FIG. 2. Steps S310, S320, and S340 of FIG. 3 may have features corresponding to steps S210, S220, and S240 of FIG. 2, so a detailed description is omitted.

[0085] In step S332, according to one embodiment, the electronic device (100) may extract characteristics for each entity based on information related to metrics stored in the meta store and feature store in step S320 (e.g., metadata such as the actual value of a feature based on the metric, the method or data source in which the feature was created, the update cycle, the scope of application, the version history, etc.) and perform profiling based thereon. According to one embodiment, the electronic device (100) may set a user and / or a product as the entity to be profiled. According to one embodiment, the electronic device (100) may dynamically determine the entity to be profiled among a user, a product, and a combination thereof according to the system settings. For example, if the electronic device (100) determines a user as the target for profiling, it may determine a user profile representing the characteristics of that user by analyzing features such as the user's recent purchase history, search keywords, click patterns, preferred categories, and review sentiment reactions. As another example, when the electronic device (100) determines a product to be profiled, it may determine a product profile indicating the popularity, reliability, and sales potential of the product based on information such as the average rating, number of reviews, conversion rate, and cart rate of the product. As yet another example, the electronic device (100) may analyze user and product interaction data together to generate a composite profile including individual response characteristics, prediction scores, and recommendation priorities for the combination of user and product.

[0086] In step S334, the electronic device (100) may determine an artificial intelligence model by performing modeling based on the profiling results performed in step S332 according to one embodiment. According to one embodiment, after performing profiling on an entity, the electronic device (100) may perform modeling to determine an appropriate artificial intelligence model based on the profiling results. This modeling process is configured differently depending on the characteristics of the entity, the purpose of use, the distribution of data, and the characteristics of the metric, and one or more of various types of artificial intelligence models such as classification, regression, recommendation, clustering, and sentiment analysis may be selected. According to one embodiment, the electronic device (100) performs modeling by utilizing the profiling results as input values ​​for the artificial intelligence model, and the artificial intelligence model determined through this modeling process can be utilized within an e-commerce service for user-customized recommendations, targeted marketing, risk prediction, etc.

[0087] In step S336, the electronic device (100) can output a file containing information about an artificial intelligence model.

[0088] According to one embodiment, the electronic device (100) can utilize metrics generated through profiling (e.g., number of user clicks, review score, conversion rate, etc.) as input features of the artificial intelligence model to perform learning and tuning of the artificial intelligence model, and can save the artificial intelligence model determined according to the results of such performance in the form of a model file so that it can be used in an actual service environment.

[0089] According to one embodiment, the model file may be stored in a format commonly used in machine learning frameworks, for example (e.g., .pkl, .joblib, .pb, .onnx, .pt, .h5, .csv, etc.), and the model file may contain information such as model structure, weights, hyperparameters, and input feature definitions. According to one embodiment, the model file may be transmitted to a central repository so that it can be referenced in a pipeline and operational environment for offline training and inference, as well as in a plurality of distinct systems including a first manager (110) and / or a second manager (120). Model metadata, such as version information, creation date and time, and coverage of the model file, may also be recorded. The model file may be used to perform experimental operations by streaming in a Kafka manner using a data pipeline.

[0090] In step S338, the electronic device (100) can redesign the pipeline by integrating the output files according to one embodiment. According to one embodiment, the electronic device (100) re-architectures the existing data processing pipeline so that the output model files can be operated integrally, rather than merely being used individually. That is, the electronic device (100) can expand or reconfigure the pipeline structure so that a plurality of artificial intelligence models can be combined and utilized.

[0091] For example, the electronic device (100) can be utilized as an artificial intelligence model capable of outputting an integrated inference result by modeling a model that predicts user churn risk, a model that calculates product conversion rates, and a model that performs user-customized recommendations, respectively, and connecting them. Specifically, a pipeline can be configured to first calculate the churn risk based on a user profile, then predict the conversion rate for product groups of interest, and then provide a screen optimized for the user based on the results. This integrated pipeline can be executed on a distributed processing-based cluster environment (e.g., a Spark cluster) and is designed to efficiently perform offline inference operations on large-scale data. To integrate multiple model files, the inference result generated by each model can be stored as an intermediate result and then linked to a subsequent model or post-processing algorithm. For example, the electronic device (100) can perform redesigns to optimize the flow of inference by filtering only users with a high churn risk from the data, which are the intermediate inference results of each model, and inputting them into a personalized recommendation model, or by filtering only product groups with a high probability of conversion and applying an emphasis model.

[0092] According to one embodiment, when multiple models refer to the same feature, the electronic device (100) can prevent duplicate processing during the feature creation and loading process and optimize resource usage.

[0093] In step S340, the electronic device (100) may provide e-commerce services to a user terminal based on an artificial intelligence model corresponding to the result of redesigning the pipeline in step S338 according to one embodiment. According to one embodiment, the inference result of the integrated model may be utilized for marketing campaigns, recommendation systems, personalized services, etc.

[0094] FIG. 4 is a flowchart illustrating a method of applying an artificial intelligence model determined according to one embodiment to an e-commerce service.

[0095] In step S400, the electronic device (100) may perform experimental operations based on output data for an artificial intelligence model according to one embodiment. According to one embodiment, the electronic device (100) may perform experimental operations to verify performance under limited environmental conditions before the artificial intelligence model, upon completion of training, is officially deployed to an operating environment. During the execution of the experimental operations, the electronic device (100) may utilize the inference results of the modeled artificial intelligence model to provide services to predetermined experimental target users and collect response data. Through this, the electronic device (100) can evaluate the accuracy and effectiveness of the entire modeled artificial intelligence model in advance before official deployment.

[0096] In step S402, the electronic device (100) may determine at least one experimental user among a plurality of users who satisfies predetermined conditions according to one embodiment. According to one embodiment, the electronic device (100) may determine predetermined conditions to determine the experimental user, such as a user who has exhibited a specific behavior within a certain period, a user who has an interest in a specific product group, or a group of new subscribers. According to one embodiment, the electronic device (100) may select at least one experimental user who satisfies the predetermined conditions and set them as experimental operation targets, thereby increasing the significance of the inference of the artificial intelligence model based on input metrics and output metrics, and collecting feedback from the targeted user accordingly.

[0097] In step S404, the electronic device (100) may collect activity information regarding an e-commerce service from at least one experimental subject user. According to one embodiment, the electronic device (100) observes what actions the experimental subject user determined in step S402 actually took within the e-commerce service and collects behavior information regarding this. For example, the user behavior information to be collected may include activities such as clicking, searching, adding to cart, purchasing, and writing reviews, as well as response time, bounce rate, and page dwell time. This behavior information may be considered during the performance evaluation and re-modeling process of the artificial intelligence model.

[0098] In step S406, the electronic device (100) can determine whether the result of the experimental operation satisfies a predetermined condition. According to one embodiment, the electronic device (100) can compare the result of the experimental operation with the predetermined condition based on behavioral information of at least one experimental subject user collected in step S404. The electronic device (100) can store all behavioral information that occurred during the process of the experimental subject users using the e-commerce service and calculate various performance indicators based on the results of analysis through comparison with the inference results of the artificial intelligence model on which the experimental operation was performed. According to one embodiment, the performance indicators calculated may include various indicators such as click-through rate (CTR), conversion rate, session dwell time, number of page views, user response rate, bounce rate reduction effect, model precision, recall rate, AUC, etc.

[0099] According to one embodiment, the electronic device (100) may predetermine criteria for evaluating an artificial intelligence model based on a service goal set at the start of experimental operation or a request received from a first manager. For example, conditions such as whether the click-through rate is above a certain level or whether the conversion rate has increased by more than a certain percentage compared to the existing rate may be included. The evaluation criteria may be determined as quantitative figures, and if necessary, the reliability of the experimental results may be supplemented through statistical significance testing, etc.

[0100] According to one embodiment, if the result of the experimental operation is determined to satisfy predetermined conditions, the electronic device (100) may perform formal distribution of the artificial intelligence model in step S408 according to one embodiment. Through this, the artificial intelligence model is applied to all users utilizing the e-commerce service, and functions such as recommendation, classification, and prediction can be fully provided.

[0101] According to one embodiment, if it is determined that the results of the experimental operation do not satisfy predetermined conditions, in step S410, the electronic device (100) may re-model the artificial intelligence model based on collected behavioral information of the experimental subject user instead of formally distributing the existing artificial intelligence model as is according to one embodiment. Through the re-modeling process, the electronic device (100) may modify and add input features, modify the architecture of the artificial intelligence model, or replace the algorithm itself based on the results of comparing the collected behavioral information of the experimental subject user with the predetermined conditions. According to one embodiment, the electronic device (100) may receive and utilize information (input metrics, output metrics, etc.) different from that used in the previous modeling process from the first manager (110) and / or the second manager (120) for re-modeling.

[0102] In step S412, the electronic device (100) may perform experimental operations again based on the artificial intelligence model remodeled through step S410 according to one embodiment. According to one embodiment, the electronic device (100) may perform steps S400 through S410 again until it is determined that formal deployment can be performed based on the remodeled artificial intelligence model.

[0103] FIG. 5 is a flowchart illustrating a method for providing an e-commerce service to a test subject user through experimental operation using an artificial intelligence model according to one embodiment.

[0104] In step S502, the electronic device (100) may determine at least one user satisfying predetermined conditions as an experimental user to perform experimental operation of an artificial intelligence model according to one embodiment, and may tag the experimental user to distinguish it from other users. The tagging may be managed as metadata that explicitly indicates that the user is included in the experimental group. For example, the electronic device (100) may determine a user as an experimental user by assigning a tag of "experimental subject" to a user who has viewed a product of a specific category three or more times within the last 30 days, and subsequently provide an e-commerce service based on output data from the artificial intelligence model to the experimental user during experimental operation.

[0105] In step S504, the electronic device (100) may perform experimental operations based on output data for an artificial intelligence model according to one embodiment. According to one embodiment, the electronic device (100) may experimentally provide functions such as recommendation, classification, and personalization only to experimental subjects prior to formal distribution, and may indirectly evaluate the quality or effectiveness of the artificial intelligence model in an actual service environment by collecting behavioral information of the experimental subjects corresponding thereto. According to one embodiment, the output data of the artificial intelligence model may include a recommended product list, a priority score, etc.

[0106] In step S506, the electronic device (100) may determine whether a user accessing an e-commerce service is a tagged experimental subject user according to one embodiment. According to one embodiment, the electronic device (100) may determine whether to provide a personalized configuration of the application's user interface based on the output data of an artificial intelligence model, based on whether the user of the terminal that accesses the e-commerce service and is required to provide a user interface is tagged as an experimental subject user when performing experimental operation.

[0107] According to one embodiment, if it is determined that a user accessing an e-commerce service is a tagged experimental user, in step S508, the electronic device (100) may call an experimental API to provide an experimental operating environment according to one embodiment, and may obtain information related to content to be provided to the experimental user through the user interface of the user terminal. The information regarding the obtained content may correspond to personalized information for the experimental user accessing the e-commerce service based on the output data of an artificial intelligence model. According to one embodiment, the input of the experimental API may include a user identifier, feature value, experimental tag, session information, etc., and the output of the experimental API may include a personalized recommendation result, prediction score, UI change element, text or banner content, etc. The call to the experimental API may be implemented by caching real-time inference (online inference) or offline inference results.

[0108] In step S510, the electronic device (100) may provide personalized content obtained based on the output data of an artificial intelligence model according to one embodiment to each of at least one experimental user through the user interface of a user terminal. Specifically, the content provided to the experimental user may correspond to information personalized based on an artificial intelligence model that is differentiated from other general users. For example, the arrangement of recommended products provided when accessing the home screen of an e-commerce service through the experimental user's user terminal, the priority of search results on the screen provided when a specific keyword is searched, keywords to be highlighted on a review page, and the types of promotion banners to be displayed on individual pages receiving the e-commerce service may be displayed on the experimental user's user terminal as personalized components based on the data output by the artificial intelligence model according to input features calculated by considering input metrics related to the experimental user. To this end, the electronic device (100) may perform rendering based on the output data of the artificial intelligence model on a front-end (web / app user interface) page so that personalized information is displayed on the screen of the experimental user's user terminal.

[0109] According to one embodiment, if it is determined that a user accessing an e-commerce service is not a tagged experimental subject user, in step S512, the electronic device (100) may provide the user with an e-commerce service that applies a conventional rule-based algorithm, rather than personalized content based on an artificial intelligence model, according to one embodiment. For example, on a home screen or product list screen provided to a general user who is not an experimental subject user, "products sold the most in the last 7 days based on all users," "products sorted in order of highest discount rate," or "popular products by category fixed by the administrator" may be displayed according to a rule-based algorithm.

[0110] In the case corresponding to performing experimental operation using a re-modeled artificial intelligence model according to another embodiment, the electronic device (100) may provide an e-commerce service based on the output data of the artificial intelligence model prior to re-modeling to a user terminal of a user that is not the subject of the currently ongoing experimental operation.

[0111] According to one embodiment, the electronic device (100) can be used to analyze experimental operation results based on behavioral information collected according to the result of providing different user interface components to experimental subjects and general users in a differentiated manner according to the judgment result in step S506. That is, the electronic device (100) can quantitatively and qualitatively analyze how effective the method of providing e-commerce services based on personalized information based on an artificial intelligence model is compared to the existing method by establishing a control group to clearly compare the difference between experimental subjects and general users to whom the artificial intelligence model is applied, and by collecting and analyzing behavioral data of the two groups. Through this, the electronic device (100) can determine whether to formally distribute the artificial intelligence model in the future, whether re-modeling is necessary, or whether to establish a target user segmentation strategy.

[0112] FIG. 6 is a diagram illustrating how an artificial intelligence model determined based on the results of an electronic device performing modeling according to one embodiment is utilized. The input metric, output metric, number of users, output data of the artificial intelligence model, etc., illustrated in FIG. 6 are schematically depicted for convenience of explanation and should not be interpreted as being limited thereto.

[0113] Referring to FIG. 6, for modeling purposes, input metrics (e.g., fresh order count / amount, eats order count / amount, play order count / amount, etc.) for a certain number of users (X1, X2, X3, X4) can be collected for each user. For example, in the case of the input metric for fresh order count, the electronic device (100) can obtain information on the number of fresh product orders for 7, 14, 21, 30, 60, 60, 90, and / or 360 days (fresh order count 7d / 14d / 21d / 30d / 60d / 90d / 360d in FIG. 6) for each user and use it as an input metric.

[0114] According to one embodiment, as an output metric that can be referenced for modeling, the electronic device (100) may use information related to the behavior of each user based on experimental results using an existing algorithm. Referring to FIG. 6, each user (X1, X2, X3, and X4) corresponds to a user whose account status has been converted to a paid membership account, and the electronic device (100) may obtain information (Y1, Y2, Y3, and Y4) regarding which product or service each user used to convert to a paid membership account. For example, the electronic device (100) may determine that the accounts of user X1 and user X2 were converted to paid membership accounts by clicking and purchasing a fresh product based on the output metric value represented by Y1 and Y2 being fresh, that the account of user X3 was converted to a paid membership account by using a food delivery service based on the output metric value represented by Y3 being eats, and that the account of user X4 was converted to a paid membership account by using a streaming service based on the output metric value represented by Y4 being play. The electronic device (100) can perform modeling by considering the relationship between the input metric and the output metric for each of these users.

[0115] According to one embodiment, the electronic device (100) can determine an artificial intelligence model based on the results of performing modeling and can distribute the model file of such artificial intelligence model. Since the artificial intelligence model distributed according to one embodiment is learned based on input metrics and output metrics during the modeling process, it can output data through modeling even if user behavior data and product-related data other than the information about users and products considered during the learning process are input. Referring to FIG. 6, unlike the number of users (X1, X2, X3, and X4) considered during the modeling process, the number of users generally using e-commerce services may correspond to a very large number, such as 30 million, and the electronic device (100) can determine user behavior data and product-related data collected according to the behavior of each user as input metrics and obtain output data of the artificial intelligence model based thereon. The output data may be output for each user, and the output metrics may include a score per product and / or service to determine whether the user's account could be converted to a paid membership account through which product and / or service, as considered during the modeling process. Referring to FIG. 6, the output metric (Y1) for user X1 may include a score (0.5) for food delivery services, a score (0.7) for streaming services, and a score (0.8) for purchasing fresh products. According to one embodiment, the electronic device (100) can determine which products and services have the highest probability of converting to a paid membership account by analyzing a plurality of scores calculated per user according to a predetermined standard.For example, the electronic device (100) may determine the product and service having the highest score among the scores included in the output metric output by the artificial intelligence model based on the input metric for each user as information related to the product and service having the highest probability of conversion to a paid membership account for each user. Alternatively, the electronic device (100) may determine the product and service having a score higher than the threshold score set by the first manager (110) to determine information related to the product and service having the highest probability of conversion to a paid membership account for each user.

[0116] FIG. 7 illustrates a block diagram of an electronic device (700) according to one embodiment. The electronic device (700) of FIG. 7 may correspond to the electronic device (100) of FIG. 1. The memory (710), transceiver (720), and at least one processor (730) of the electronic device (700) of FIG. 7 may be configured to perform the operations performed by the electronic device (100) in the embodiments described above.

[0117] Referring to FIG. 7, an electronic device (700) is configured as at least part of a system providing an e-commerce service according to an embodiment of the present disclosure and can interact with a terminal (750). The electronic device (700) may include a memory (710), a transceiver (720), and at least one processor (730). Only the components of the electronic device (700) illustrated in FIG. 7 that are related to an embodiment of the present disclosure are illustrated. Therefore, it will be understood by those skilled in the art related to the present embodiment that other general-purpose components may be included in addition to the components illustrated in FIG. 7. In the embodiment, the transceiver (720) may be included in a communication device. Also, in the embodiment, at least one processor (730) may be included in a controller.

[0118] According to one embodiment, the memory (710) may store data acquired by the electronic device (700) and / or data generated by processing. According to one embodiment, the memory (710) may store data from a meta store and / or feature store from which the electronic device (700) utilizes information necessary for modeling.

[0119] According to one embodiment, the transceiver (720) is a communication module for the electronic device (700) to transmit and receive data with an external electronic device, and the transceiver (720) can perform wired / wireless communication. The external electronic device may be a terminal (750) or a server. In addition, the communication technology used by the transceiver (720) may include GSM (Global System for Mobile communication), CDMA (Code Division Multi Access), LTE (Long Term Evolution), 5G, WLAN (Wireless LAN), Wi-Fi (Wireless-Fidelity), Bluetooth, RFID (Radio Frequency Identification), Infrared Data Association (IrDA), ZigBee, NFC (Near Field Communication), etc.

[0120] According to one embodiment, the terminal (750) may include at least one terminal outside of various electronic devices (700), including a terminal of a consumer using an e-commerce service, a terminal of a first manager (110), and a terminal of a second manager (120).

[0121] According to one embodiment, if a metastore and / or feature store that utilizes information necessary for modeling is stored externally, the transceiver (720) may be configured to receive data from the metastore and / or feature store.

[0122] According to one embodiment, at least one processor (730) can control at least one of the transceiver (720) and the transceiver (720) and process data and signals for the overall operation of the electronic device (100) described above. At least one processor (730) may be composed of at least one hardware unit. It may be operated by one or more software modules generated by executing program code stored in memory (710). At least one processor (730) may include memory (710), and at least one processor (730) can control the overall operation of the electronic device (700) and process data and signals by executing program code stored in memory (710). According to one embodiment, at least one processor (730) is a central processing unit of the electronic device (700) and can process data received from the transceiver (720) and store necessary information in memory (710). At least one processor (730) may include a high-performance processing unit such as an ARM-based CPU, GPU, or ASIC (Application-Specific Integrated Circuit). The processor supports a multi-core architecture for parallel processing of data and can perform low-latency operations when real-time processing is required. At least one processor (730) may also execute machine learning (ML) algorithms or artificial intelligence models to predict user behavior or make optimized decisions regarding the display format of the shopping cart page. At least one processor (730) may perform modeling based on data received through the transceiver (720) and / or data stored in memory (710), and may render a user interface to provide personalized information using an artificial intelligence model corresponding to the result of the modeling.

[0123] Although not illustrated in FIG. 7, the electronic device (700) may further include an input / output interface (not illustrated) for the user to directly input or provide necessary information to the user. According to one embodiment, the input / output interface provides physical and logical interfaces for interaction between the electronic device (700) and the user. The input / output interface may include various hardware components such as a display device (e.g., LCD, OLED, etc.), a touchscreen, a microphone, a speaker, a physical button, a vibration motor, a camera, etc. The input / output interface supports sophisticated input recognition technologies (e.g., capacitive touch detection, voice recognition, gesture recognition) to accurately recognize user commands and transmit them to at least one processor (730). In terms of output, the input / output interface may visually display data such as text, images, and graphs through a high-resolution display and may enhance the user experience by supporting HDR (High Dynamic Range) or a high refresh rate. The input / output interface may include functions such as adaptive brightness adjustment and a low-power mode to optimize the energy efficiency of the electronic device (700). In addition, the interface provides a layout based on UX (User Experience) design principles to reduce visual fatigue for users, and supports dynamic rendering features to effectively display graphs and detailed information.

[0124] The electronic device (700) according to the embodiments described above may include a processor, memory for storing and executing program data, permanent storage such as a disk drive, a communication port for communicating with an external device, and user interface devices such as a touch panel, a key, a button, etc. Methods implemented as software modules or algorithms may be stored on a computer-readable recording medium as computer-readable code or program instructions executable on the processor. Here, computer-readable recording media include magnetic storage media (e.g., ROM (read-only memory), RAM (random-access memory), floppy disk, hard disk, etc.) and optical reading media (e.g., CD-ROM, DVD (Digital Versatile Disc)). Computer-readable recording media may be distributed across networked computer systems, allowing computer-readable code to be stored and executed in a distributed manner. The medium may be readable by a computer, stored in memory, and executed by a processor.

[0125] Embodiments of the present disclosure may be represented by functional block configurations and various processing steps. These functional blocks may be implemented by various numbers of hardware and / or software configurations that execute specific functions. For example, the embodiments may employ integrated circuit configurations such as memory, processing, logic, look-up tables, etc., which can execute various functions by the control of one or more microprocessors or other control devices. Similar to how components may be implemented as software programming or software elements, the embodiments may be implemented in programming or scripting languages ​​such as C, C++, Java, assembler, etc., including various algorithms implemented as combinations of data structures, processes, routines, or other programming configurations. Functional aspects may be implemented as algorithms executed on one or more processors. Additionally, the embodiments may employ prior art for electronic configuration, signal processing, and / or data processing. Terms such as "mechanism," "element," "means," and "configuration" may be used broadly and are not limited to mechanical and physical configurations. The above terms may include the meaning of a series of software processes (routines) in conjunction with processors, etc.

[0126] The above-described embodiments may be implemented as artificial intelligence (AI) through the processor and memory of the electronic device (100). The processor may be composed of one or more processors, and the one or more processors may be general-purpose processors such as CPUs, APs, and DSPs (digital signal processors), graphics-dedicated processors such as GPUs and VPUs (vision processing units), or artificial intelligence-dedicated processors such as NPUs. The one or more processors may be controlled to process input data according to predefined operation rules or artificial intelligence models stored in memory. Alternatively, if the one or more processors are artificial intelligence-dedicated processors, the artificial intelligence-dedicated processors may be designed with a hardware structure specialized for processing a specific artificial intelligence model.

[0127] The predefined rules of operation or artificial intelligence models are characterized by being created through learning. Here, being created through learning means that a predefined rules of operation or artificial intelligence models configured to perform a desired characteristic (or objective) are created by a basic artificial intelligence model being trained using a number of learning data by a learning algorithm. Such learning may be performed on the electronic device itself where the artificial intelligence according to the present disclosure is executed, or it may be performed through a separate server and / or system. Examples of learning algorithms include supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning, but are not limited to the examples described above.

[0128] An artificial intelligence model may be composed of multiple neural network layers. Each of the multiple neural network layers has multiple weight values ​​and can perform neural network operations through operations between the results of previous layers and the multiple weights. The multiple weights possessed by the multiple neural network layers can be optimized based on the learning results of the artificial intelligence model. For example, the multiple weights can be updated so that the loss value or cost value obtained from the artificial intelligence model during the learning process is reduced or minimized. The artificial neural network may include, but is not limited to, deep neural networks (DNN), convolutional neural networks (CNN), recurrent neural networks (RNN), restricted Boltzmann machines (RBM), deep belief networks (DBN), bidirectional recurrent deep neural networks (BRDNN), or deep Q-networks.

[0129] The aforementioned embodiments are merely examples, and other embodiments may be implemented within the scope of the claims set forth below.

Claims

1. A method for providing e-commerce services to a user terminal, A step of determining metrics by collecting user behavior data and product-related data; A step of storing information related to the above metrics in a meta store and a feature store; A step of storing an artificial intelligence model corresponding to the result of modeling through profiling based on the information stored in the metastore and the feature store; and A method comprising the step of providing the e-commerce service to the user terminal based on output data of the artificial intelligence model.

2. In claim 1, the step of storing in the meta store and the feature store is, Based on the above metric, a step of generating a feature for input to the artificial intelligence model and storing it in the feature store; and A method comprising the step of storing metadata related to the above feature in the meta store.

3. In paragraph 1, the step of storing the artificial intelligence model is, A step of performing profiling for entities distinguished by at least one of users and products based on information stored in the meta store and the feature store; A step of determining the artificial intelligence model by performing modeling according to the above profiling results; and A method further comprising the step of outputting and saving a file containing information about the artificial intelligence model determined above.

4. In paragraph 3, the step of determining the artificial intelligence model is, A step of determining input data and output data for the artificial intelligence model based on at least one metric included in the information stored in the metastore and the feature store; and A method comprising the step of determining the artificial intelligence model by performing modeling using the determined input data and output data.

5. In paragraph 1, the step of providing the e-commerce service is, A step of performing experimental serving of the above artificial intelligence model; and A method further comprising the step of performing a formal distribution based on whether the results of the above experimental operation satisfy predetermined conditions.

6. In paragraph 5, the step of performing the above experimental operation is, A step of determining at least one experimental subject user satisfying predetermined conditions among a plurality of users; and A method further comprising the step of providing an e-commerce service to at least one experimental subject user based on the output data of the artificial intelligence model.

7. In Paragraph 6, the step of performing the formal distribution above is, A step of collecting activity information regarding the e-commerce service of at least one experimental subject user; and A method comprising the step of performing the formal distribution when the above activity information corresponds to the input data and output data of the above artificial intelligence model.

8. In Paragraph 7, the step of performing the formal distribution above is, If the above activity information does not correspond to the input data and output data of the above artificial intelligence model, a step of re-modeling the above artificial intelligence model based on the above activity information; and A method further comprising the step of performing an experimental operation of the re-modeled artificial intelligence model.

9. In paragraph 6, the step of providing e-commerce services based on the output data of the artificial intelligence model is, A step of tagging at least one experimental subject user; When at least one of the above-mentioned tagged experimental subjects accesses the above-mentioned e-commerce service, a step of calling the experimental API; and A method comprising the step of providing an e-commerce service to at least one tagged experimental subject user based on the output data of the artificial intelligence model in response to a call to the experimental API.

10. In an electronic device that provides e-commerce services to a user terminal, Memory; transceiver; and It includes at least one processor, and the at least one processor controls the memory and the transceiver to enable the electronic device: Collect user behavior data and product-related data to determine metrics, and Store information related to the above metrics in the meta store and feature store; Based on the information stored in the above metastore and the above feature store, an artificial intelligence model corresponding to the result modeled through profiling is stored, and An electronic device configured to provide the e-commerce service to the user terminal based on output data for the artificial intelligence model.

11. In paragraph 10, the above at least one processor is, Based on the above metric, a feature for inputting into the artificial intelligence model is generated and stored in the feature store, and An electronic device configured to store metadata related to the above feature in the meta store.

12. In paragraph 10, the above at least one processor is, Based on the information stored in the meta store and the feature store, profiling is performed for each entity distinguished by at least one of the user and the product, and Based on the above profiling results, modeling is performed to determine the above artificial intelligence model, and An electronic device further configured to output and save a file containing information about the artificial intelligence model determined above.

13. In paragraph 12, the above at least one processor is, Based on at least one metric included in the information stored in the metastore and the feature store, input data and output data for the artificial intelligence model are determined, and An electronic device configured to determine the artificial intelligence model by performing modeling using the above-determined input data and output data.

14. In paragraph 10, the above at least one processor is, Perform experimental operation of the above artificial intelligence model, and An electronic device further configured to perform formal distribution based on whether the results of the above experimental operation satisfy predetermined conditions.

15. In paragraph 14, the above at least one processor is, Among multiple users, at least one experimental subject user satisfying predetermined conditions is determined, and An electronic device further configured to provide e-commerce services to at least one experimental subject user based on the output data of the artificial intelligence model.

16. In paragraph 15, the above at least one processor is, Collecting activity information regarding the e-commerce service of at least one experimental subject user, and An electronic device configured to perform the formal distribution when the above activity information corresponds to the input data and output data of the above artificial intelligence model.

17. In paragraph 16, the above at least one processor, If the above activity information does not correspond to the input and output data of the above artificial intelligence model, the above artificial intelligence model is re-modeled based on the above activity information, and An electronic device further configured to perform experimental operation of the above-mentioned re-modeled artificial intelligence model.

18. In paragraph 15, the above at least one processor, Tagging at least one experimental subject user, and When at least one of the above-tagged experimental subjects accesses the above-mentioned e-commerce service, the experimental API is called, and An electronic device configured to provide an e-commerce service to at least one tagged experimental subject user based on the output data of the artificial intelligence model in response to a call to the above experimental API.

19. A non-transient computer-readable recording medium containing a computer program for performing a method of providing an e-commerce service to a user terminal, wherein the method comprises: A step of generating metrics by collecting user behavior data and product-related data; A step of storing information related to the above metrics in a meta store and a feature store; A step of storing an artificial intelligence model corresponding to the result of modeling through profiling based on the information stored in the metastore and the feature store; and A non-transient computer-readable recording medium comprising the step of providing the e-commerce service to the user terminal based on output data for the artificial intelligence model.

Images