KR20200029647A - Generalization method for curated e-Commerce system by user personalization - Google Patents

Abstract

The present invention relates to a method for generalizing personalized information, which provides customized prediction information for individuals to a plurality of external sites by means of personalized information at an online e-commerce site, instead of independently operating a conventional site. According to an embodiment of the present invention, a system for recommending required products to a purchaser in an online e-commerce system has been generalized. However, the system is only operated within a specific service site, and there is no universal implementation method for providing, to a third party, customized prediction information as in an online shopping mall operated externally. The reason is that it is difficult to implement a common method since each online shopping mall has different types of products and users have various unique tendencies. The present invention constructs a tree structure by a category classification method, which is a method for generally managing products, to solve the problem, and defines and learns characteristics information for each category, thereby being capable of recommending appropriate products according to the characteristics of the purchaser, and manages this category structure, thereby being capable of dynamically dealing with external various requirements. That is, the purpose of the present invention is to provide a system for providing an appropriate recommendation system to online shopping malls that sell various types of items.

Description

Generalization method for curated e-Commerce system by user personalization}

The present invention provides a curated e-commerce system that targets multiple product groups and generalizes to a number of service providers with respect to a recommendation system used to provide optimal customized information according to user's personalized information in online e-commerce. It's about how to implement.

Artificial intelligence technology has been studied for a very long time with the aim of enabling computers to communicate between humans and computers, as well as intellectual behaviors that make humans similar recognition and judgment.

Early AI required a lot of computation, so it suffered a slowdown due to the processing speed problem and the limitation of the accuracy of the result. Recently, the problem of processing speed has improved dramatically with the development of hardware, and the accuracy has been dramatically improved with the advent of deep learning. The improvement has resulted in a lot of attention and attention. As a result of these interests and efforts, new learning techniques such as reinforcement learning and Generative Artificial Network (GAN) are emerging.

Neural network is a field of artificial intelligence technology that predicts or classifies results by inputting characteristic information. Deep Learning, proposed by Professor Jeffrey Hinton of Canada, has opened a new chapter in artificial intelligence by overcoming the limitations of existing neural networks.

In the field of deep learning, various techniques such as Deep Belief Network, Convolutional Neural Network, and Recurrent Neural Network have been developed and used according to each feature. This deep learning is excellent in extracting features from a given large amount of data and classifying them or making predictions by learning.

The present invention implements a matching system that predicts preferences based on user information and determines and presents items associated with a recommendation system that recommends suitable purchase items based on the user's information. In addition, it is possible to implement a prediction system such as prediction of future consumption patterns for a specific item based on the consumer's purchasing patterns and social phenomena or issues, and it is possible to provide multiple sites other than a specific site for learning and prediction. It is characterized by suggesting a generalization implementation methodology.

 The recommendation system includes a collaborative filtering method that recommends an item suitable for a visitor using unique data of a site visitor such as a search history and a purchase history, and a content analysis method that recommends items related to the visitor by analyzing information of the content itself. The method of implementing the recommendation system was mainly used by statistics, but recently, it is preferred by artificial intelligence incorporating natural language processing techniques.

The Natural Language Process is able to be used in various meaningful systems based on this by providing information in a form that the computer can recognize by analyzing and interpreting the natural language. Natural language processing analyzes and extracts natural language sentences into morphemes, converts them into vector forms with meaning, and processes them to be recognized and usable by the machine. The vector information is classified or learned by artificial intelligence techniques and used for prediction.

In online e-commerce, there is an online shopping mall (online shopping mall) where the seller posts a product and the buyer searches for the product and then purchases it. There are online purchasing malls (electronic purchasing, electronic procurement, reverse auction) where transactions are made by submitting.

The present invention targets multiple product groups for a recommendation system that provides optimal personalized information based on personalized information of a user in online e-commerce and recommends items such as appropriate products or services based on the information. As a method for realizing a generalized curated e-commerce system for service providers in Korea, learn personal information based on personalized characteristic information about product groups to extract and recommend personal basic information and history information for personalization. The challenge is to build a recommendation model for recommending the appropriate product.

In order to solve the above-described problems, the curated e-commerce system according to an embodiment of the present invention includes a learning processing unit that learns a purchasing tendency through information collected from a user, purchase information history information, environmental information, and the like, and the learned processing unit. It consists of a prediction processing unit that performs user's purchase preference or sales forecast through, and a system operation unit that manages collected data, sets necessary parameters for learning and prediction, and verifies and tunes learning accuracy.

Also, it is composed in conjunction with the Front system unit that collects information from external users through online sites or provides personal prediction information based on user's characteristic information and external environment information to provide product recommendation or sales forecast information for individuals. Lose. The information collected by the front system unit is stored and stored in a learning object information database inside the system for use in learning, and such learning data is received and stored in real time according to user actions or received in batch form at regular intervals. Can be.

The learning target information collected by the above may be reflected in learning as soon as it is collected or reflected in learning with a certain period, and such a processing procedure may set and control parameters in the system operation unit. The learning processing unit performs learning based on a separate setting for items to be predicted by the artificial intelligence technique, and stores the results of the learning in neuron information. The neuron information is a set of weight values generated by the AI learning model, and is used for prediction by the same AI prediction model.

In order to collect the learning object information, the front system unit collects information according to the user's behavior and transmits it to the curate system or provides an application programming interface (API) module provided by the system to receive the predicted result value from the system. It is mounted and started. Through the above API, the user's behavior information and user's characteristic information are transmitted to the database inside the curate system, and the personalized information provided by the system serves as a relay to the front system.

The learning processing unit includes a data pre-processing function that reconstructs the collected learning object information into data in a form required by the learning processing module, and is implemented by a natural language processing (NLP) technique. Based on the data generated by the data pre-processing function, the AI learning processing module learns.

The system operation unit can set or manage the structure and properties of the object to be predicted, and set the appropriate parameters for efficient learning in the learning processing module and improve the prediction performance through evaluating the learning results.

The system operation unit sets the structure of the managed product category (product group) in the form of a tree to define a curated e-commerce generalization according to an embodiment of the present invention, and defines features for learning and prediction processing for each category. Includes the ability to set values. In addition, at the bottom of each category, products matching the feature defined in the corresponding category can be mapped and used to specify the prediction target.

In an online e-commerce system according to an embodiment of the present invention, a system for recommending a product required to a buyer has become common. However, all of these systems are implemented only within a specific service site, and there is no universal implementation method to provide to third parties, such as an online shopping mall operated externally. The reason is that it is difficult to implement in common because the types of products handled by each online shopping mall are different and the users also have various characteristic tendencies. According to the present invention, in order to solve the above problems, a tree structure is constructed according to a category classification method, which is a method of generally managing products, and by defining and learning characteristic information for each category, an appropriate product according to a buyer's characteristics can be recommended. In addition, by managing this category structure, it is possible to actively respond to various external needs. According to the present invention, by providing an appropriate recommendation system for a plurality of online shopping malls selling various types of items, it is possible to provide a personalized service using an artificial intelligence recommendation system in an online shopping mall at a very low cost.

1 is a configuration diagram between systems of the present invention. It consists of a curated e-commerce operation system (hereinafter referred to as a 'curate e-commerce system' or 'curate system') and a number of online shopping mall sites connected to provide customized information to users and purchase users connected to each site.
2 is an internal structure and data flow diagram of a curated e-commerce system.
FIG. 3 shows a structure in which a target product group (category) for providing customized information for generalization of curated e-commerce, which is a characteristic of the present invention, is managed in a tree form, and is referred to as a 'Curate Tree Map'. .
4, 5, and 6 are examples of features set in the curate tree map, and select and set items that may affect preference determination in each product category.
7 is an example of a process for learning a padded product by using the user's purchase information as an example of a learning process method.
FIG. 8 is a flow diagram of a user-system interaction process for the learning process of FIG. 7.
9 is an example of a process for learning a padding product using personal log information of a user as an example of the learning processing method.
10 is an example of a process for learning a padding product using environmental information as an example of a learning process method.
11 is an example of processing for predicting a padding product using personal characteristics of a user as an example of a prediction processing method.
FIG. 12 is a flow chart of a user-system interaction process for the prediction process of FIG. 11.
13 is a processing example of predicting padding products using environmental information as an example of a prediction processing method.

Prior to the description, when a certain part of the specification refers to a "equipment" or "includes" a certain component, it may not include other components but include other components, unless specifically stated otherwise. It means that there is. In addition, terms such as "... unit" and "... module" in the specification mean a unit that processes at least one function or operation, which is hardware, software, or hardware. And software.

In addition, in this specification, the term "an embodiment" means to serve as an example, example, or illustration, but the subject matter of the invention is not limited by such examples. Also, “comprising”, “having”, “having” and other similar terms are used, but when used in the claims, as an open transition word that does not exclude any additional or other components. Comprising "is used generically in a similar manner.

The various techniques described herein can be implemented with hardware or software, or combinations of all of them where appropriate. As used herein, terms such as “Unit”, “Module”, and “System” likewise refer to computer-related entities, ie, combinations of hardware, hardware and software, software, or It can be treated as equivalent to the software at the time of execution. For example, the program module may be composed of one component and equivalent or a combination of two or more components. In addition, in the present invention, both application programs and hardware executed in a server or terminal may be configured in a module unit, or may be recorded in one physical memory or distributed between two or more memories and recording media.

In the curated e-commerce generalization system and method according to an embodiment of the present invention, providing personalized personalized information such as product recommendation in an e-commerce site such as an online shopping mall can be applied to multiple online shopping malls. do.

To this end, in an embodiment of the present invention, a product group named `` Curate tree map '' (hereinafter referred to as a `` category '') is managed in a tree structure, and a 'Feature indicating a characteristic of the product in each category. 'Is defined to store the learning result information and to generate a result using this information in future prediction. This 'Feature' consists of items for determining the preference of the product group, and the system operation manager manages creation / modification / deletion.

In each category of the curated tree map, preferences for features of a corresponding product group can be learned and managed by storing these learning results in neuron information. This neuron information is then used in prediction and serves to improve accuracy by receiving behavior information such as the user's purchase result and re-learning. The neuron information is used as a meaning of storing learning information of a neural network, but also includes information generated by a machine learning technique other than the neural network.

1 is a configuration diagram between systems of the present invention, and is a connection diagram between the curate system 101 and an external system. The shopping mall 1 (103), the shopping mall 2 (105), and the shopping mall 3 (107) are Internet online shopping malls connected by separate APIs through the Internet 102, and the characteristics of users 104, 106, and 108 connected to each shopping mall. Information or information about an activity such as purchase or surfing is collected and transmitted to the curated system 101. Curate system stores and manages the received information and processes the data if necessary to use it for learning.

2 is an internal configuration diagram of a curated e-commerce system. The system is largely performed by the system operator 202 that controls the system through the service operator terminal 207, and the collected learning target information 210, and stores the result in the neuron information 211. It consists of a learning processing unit 203, and a prediction processing unit 204 that predicts an individual's preference or sales amount based on the learned neuron information 211. In addition, there is a front system unit 205 of a site that provides an e-commerce service by securing a separate customer, such as an independent online shopping mall, outside the system.

Inside the learning processing unit 203 is composed of a learning processing module 208 and an evaluation processing module 209 for learning, and the evaluation processing module evaluates the accuracy of the learning processing results. In addition to the collected learning object information 210, the learning processing module 208 may receive user behavior information 216, such as a purchase, search, or inquiry collected by the front system unit 205 to perform learning. The learned results are stored or updated in the neuron information 211.

When a request for a custom prediction is received from the information collection module 215 of the front system unit 205 using the neuron information 211 which is the result information learned from the learning processing unit, prediction information corresponding to the corresponding user characteristic information is derived. Thus, the prediction result 218 is transmitted to the expression unit 214 (www or CRM) of the front system unit 205 and provided to the user.

3 is a structure diagram named 'Curate tree map' for learning and predicting various product groups (categories) as a key methodology for generalizing personalized prediction information by personalization, which is a characteristic of the present invention, and providing it to a plurality of service providers. to be. The structure is largely composed of a category level 301 and a product level 302, and each node 303 at the category level has a plurality of sub-nodes 306. In addition, each node is composed of neuron information 304 containing learning information about the node and Feature 305 for determining characteristics of the corresponding category. In addition, a product 308 suitable for the corresponding node may be defined at the bottom of each node, and the corresponding product may be applied to learning and prediction results according to characteristics of the corresponding category node.

The feature may be defined as items suitable for each category as shown in FIGS. 4, 5, and 6, and such a task is performed in the system operating unit 202 of FIG. 2.

4, 5, and 6 are examples of defining the attribute of the node at the category level. FIG. 4 is a definition of padding in the clothing field, FIG. 5 is a defined example of a laptop, and FIG. 6 is a defined example of a beverage category, and a similar method can define features of other product groups (categories).

7 illustrates a case in which the user's purchase information is received and processed as an example of learning processing. When receiving the personal purchase information 701, the information is set to Y (output variable, independent variable) of the learning data, and the personal property information 702 of the corresponding user is set to X (input variable, dependent variable) to which the corresponding product belongs. Learning is performed on the neuron information 706 in the category 705 and the result is updated on the neuron information 706. At this time, the learning structure is performed according to the pre-defined Feature (707, 708) of the corresponding category. The other categories are also updated by learning the neuron information of the corresponding category as described above.

8 is a flowchart illustrating the processing of FIG. 7. When the user 801 purchases (802) the product from the front system of the online shopping mall site, the corresponding purchase information is collected and stored (803), and this processing may be performed in real time. The collected information is processed into personal characteristic information 806 and personal purchase information 807 for learning through the information pre-processing 805 of the information processing system 804, and then the learning request request from the learning processing system 809 ( 808). The learning processing system 809 receiving the learning request improves the accuracy by learning 810 the received information using the neuron information of the corresponding category and then updating the result to the neuron information (811).

The above process flow chart describes an example of receiving the user's purchase information. In the case of FIGS. 9 and 20, the process flow chart description is omitted, and it is revealed that processing is possible in the same manner as in FIG.

9 is a processing configuration diagram for learning based on the user's log information (information that the user has searched for or searched for a specific product), and the detailed processing flow diagram is processed in the same manner as in FIG. 8, but in this case, only the received data is different. Will be applied.

FIG. 10 is a processing configuration diagram for learning based on environmental information (information obtained by analyzing a newspaper article or magazine information and extracting a purchase trend of a corresponding product), and the detailed processing flow chart is processed in the same manner as in FIG. 8. According to the personal purchasing information, the students will learn together, but in some cases, only this learning can be conducted through a separate procedure.

FIG. 11 describes a series of processes for predicting and providing personalized personal information, and uses the personal property information 1101 of the user received from the front system to predict neuron information 1104 in the target category 1103. ) To perform prediction. The result of the prediction is derived (1106) by the structure of the previously defined feature. The prediction result 1106 is expressed in the form of a score for each item, and in the present invention, a value between 0.0 and 1.0 is displayed. For example, among the items, for example, 'color', the probability of favoring White is the highest with 0.4 (40%) and the lowest with Black being 0.1 (10%).

12 is a flowchart of the processing of FIG. 11, and receives personal personality information 1203 of the user 1201 from the front system unit, processes the information for prediction in the information preprocessing system 1204, and personal personality information After generating (1205), the corresponding information is requested by the prediction processing system (1208) to make a prediction request (1207), and the requested prediction processing system (1208) receives the prediction result (1212) according to the feature of the target category to predict. ).

13 is a processing configuration diagram for predicting based on environmental information (information obtained by analyzing a newspaper article or a magazine and extracting a purchase trend of a corresponding product), and the detailed processing flow chart is processed in the same manner as in FIG. 13. At the time of the prediction process, the final prediction result can be derived by combining the personal prediction result with the user's personal characteristic information 1101 and this result. At this time, the weight can be reflected by giving importance according to the importance of the result 1106 by personal characteristic information and the result 1306 by environmental information.

-Curated e-commerce: E-commerce that provides personalized prediction information by analyzing personalized information and various data.
-Neuron information: Model values generated as a result of learning by artificial intelligence, which are used for prediction and collectively refer to various learning / prediction models.
-API: As an application programming interface, it refers to a program that transmits and receives information according to the rules set for information exchange between systems.
-Feature: It is the structure that defines the items representing the characteristics of the product group (category).
-Front system: An external e-commerce or web service site.
-www: generic term for e-commerce sites or other websites.
-CRM: A system that collects and manages customer information through Customer Relation Management.
-Natural language processing technique: Also called NLP (Natural Language Processing), it is a processing technique that interprets sentences written in natural language into a structure that can be recognized by a computer.
-Personal characteristic information: Information that indicates personal characteristics such as the user's age, gender, occupation, residence, and income level.
-Personal purchase information: Information purchased by users online or offline. It includes information such as product type, price, color, or feature, and is used for learning. Such purchase information may be shy in real time when an action occurs or may be collected from the sales history data of the store.
-Behavior information, log information, and history information: Refers to the user's behavior such as purchase, search, and inquiry.
-Environmental information: Information about the social atmosphere or phenomena, such as economic conditions, political situations, and trendy trends, is mainly analyzed through articles, magazines, or online postings.
-Customized information: Refers to all information provided as a prediction product, such as a recommendation product and matching between a supplier and a buyer.
-Input variable: Also called dependent variable, it is indicated as 'X' and refers to the variable that is the condition for deriving the result.
-Output variable: Also called independent variable, it is indicated as 'Y' and refers to the variable corresponding to the result.

Claims (7)

A system operation unit that sets parameters to be used for learning and prediction by a service operator and manages learning and prediction processing;
A learning target information database unit that collects and stores personal characteristics of users, user behavior information, and social environment information;
A learning processing unit that performs mechanical learning using the information of the learning object information database and stores learning information in a neuron information database;
A prediction processing unit that performs personalized prediction using the neuron information and personal characteristic information;
An e-commerce personalized information service system, characterized by consisting of a number of external front system units that collect information on behaviors such as user's purchase or search / retrieve from a user terminal or provide customized prediction information to a user. According to claim 1,
The learning processing unit defines a target product group to be predicted as a step-by-step category of a tree structure;
Defining a feature capable of representing a feature of a product group in each category in the step-by-step category;
An e-commerce personalized information service system comprising learning by each category using the user's personal characteristic information and behavior information and storing the result in neuron information of the category. According to claim 1,
The prediction processing unit receives the personal characteristic information of the user from the front system unit;
Searching for neuron information of a category to be predicted according to the request;
Deriving a predicted value matching the received personal characteristic information using the always searched neuron information;
And providing the derived predicted value information to a user of the front system unit. According to claim 2,
The learning processing unit collects information when behavior information such as a user's purchase or product search and inquiry is generated in the front system unit;
Learning by using the behavior information and the user's characteristic information;
E-commerce personalized information system comprising the step of storing the learning results in the corresponding neuron information. According to claim 2,
The learning processing unit analyzes a sentence according to a natural language processing technique (NLP) and analyzes a purchase trend or a pattern as a function of analyzing a purchase trend for each individual characteristic information and social environment information;
Learning the analysis result in a category unit;
E-commerce personalized information system comprising the step of storing the learning results in the corresponding neuron information. According to claim 2,
The learning processing unit collects information about the user's behavior in the front system unit, re-learns it, and updates the results to neuron information. The method of claim 3,
The prediction processing unit receives the personal characteristic information of the user from the front system unit;
Reading a value for environmental information from the learning object information database unit;
Searching for neuron information of a category to be predicted according to the request;
Deriving a predictive value matching the received personal characteristic information and environmental information using the always searched neuron information;
And providing the derived prediction value information to a user of the Front system.

Images