KR100756658B1 - Automatic Construction Method Of Electronic Catalogs - Google Patents

Abstract

The present invention relates to a method for automatically constructing an electronic catalog, wherein a learning data generation step of generating learning data by performing a normalization and standardization process by extending a representation by extracting a key word with reference to a lexical dictionary for prebuilt electronic catalog data Wow; A catalog classifier learning step of delivering the training data to a catalog classifier to train the catalog classifier; Vocabulary expansion and standardization step of classification-requested e-catalog data to be classified to classify the e-catalog data to be categorized by extending the expression through key word extraction, performing normalization and standardization, and delivering it to the catalog classifier. Wow; The catalog classifier includes a catalog classification step of performing classification learning using the learning data and recommending a suitable classification for the classification requested electronic catalog data.

E-catalogue, automatic categorization, catalog properties, vocabulary extensions

Description

Automatic Construction Method Of Electronic Catalogs

1 is a view schematically showing a conventional electronic catalog construction procedure;

2 is a schematic process diagram illustrating a method for automatically building an electronic catalog according to the present invention;

3 is a view schematically showing an operating process of the catalog automatic classifier according to the present invention;

4 is a diagram illustrating a processing flowchart of an automated tool when a new request item comes off-line or on-line in the automatic electronic catalog construction method according to the present invention;

5 is a flowchart at the time of item approval of the newly requested data in the electronic catalog automatic construction method according to the present invention.

The present invention relates to a method for automatically building an electronic catalog, and more particularly, to an electronic catalog for automatically classifying and refining product information (commodity catalog) distributed in electronic commerce according to a classification system of a market place. The present invention relates to an electronic catalog automatic construction method.

Currently, e-commerce businesses such as B2B marketplaces handle large quantities of goods. At this time, the recognition and judgment of the goods for each participating entity is mainly made by the classification to which the goods belong, which requires the details of the electronic goods of the supplier and the buyer. The electronic catalog is an electronic catalog, and the supplier and the buyer check the electronic catalog and perform a transaction on the required product. And since this is a transaction request and processing made electronically, it can be said to be the core foundation of the electronic commerce system in that it enables automated processing as well as electronic processing of orders and order documents.

If you build an e-catalogue system and operate a B2C or B2B marketplace, you will need to register additional products in the e-catalogue of the marketplace through your customer or product manager.

In building an electronic catalog, a lot of manpower and money should be invested in the existing classification of industry-specific standards and the classification of individual items according to the standards. In addition, accurate classification of individual articles requires domain experts with product knowledge.

When additional product registration is required, if the number of requests for additional product registration is small, the input time of the marketplace operator, the purchasing manager, or the catalog manager does not take much time, but the number of products to be registered in the electronic catalog is large. If the data or the classification of the product to be registered are diversified and diversified, the time required for the operation of the operator or the person in charge is increased. Accordingly, the marketplace can increase the manpower or the cost in order to register and order the product in a timely manner. You pay. 1 is a diagram illustrating an existing product registration process.

In the existing general electronic catalog system, product registration is performed in the same procedure as the electronic catalog product registration process of FIG. 1. This is described as follows.

That is, a product registration request from a buyer is made in three ways. New input of individual items using the existing web interface, new input of interface (I / F) of batch items made in real time with established third-party computer networks, and new input of large quantities of bulk items due to new launch of the site or large-scale business to be.

Requested new data will be saved immediately after cleansing work along each path, for web input and I / F input, and separate from Excel data for bulk bulk items. A C / S program creates a private, passive new request database.

After the catalog manager checks the classification (up to the segment of UNSPSC), the catalog manager finds the catalog manager in charge of the leaf node of the actual taxonomy and assigns it directly. The catalog manager of the item decides on which UNSPSC leaf node the requested item will be created, and then individually checks all items for the same product as the existing product and registers the product in the electronic catalog.

In detail, in order to register a product in an electronic catalog, the following process is performed.

In other words, the product registration request data that the buyer requests to register with the marketplace is atypical, and the data cleansing operation, which is a data shaping operation that is maintained to register such atypical new request data in the marketplace e-catalog system. This is necessary.

In addition, when a product requested by a buyer is registered in an e-catalog, there is a lot of work to be handled by one catalog manager when there are a large number of products handled by a marketplace where the business is active in the market, or when the number of product registration requests comes in large quantities or multiple products. As a result, the large B2B and B2C marketplaces, which are actively trading, share their work with as few as five to as many as twenty. Therefore, when a large number of new product registration requests come in, it is necessary for the catalog manager to assign a request to the catalog manager corresponding to the classification and product group.

In addition, when each catalog owner is determined for a new product registration request, the catalog manager compares the new product registration requests with his existing product registration information, and if there is already established product information, Eliminate duplicates of items by mapping them to the item without building it.

In addition, if the request for registering a new product does not overlap with the existing data, the catalog manager registers the new item after classifying it to the final leaf class of the classification system, or after completing the new request approval process. The requested buyer will be notified.

The bottleneck zone, which requires the most manpower and time input when registering the requested product, includes the data cleansing part, the cataloging part assignment according to the product classification and characteristics, and the appropriate UNSPSC for product registration. The classification system decision part.

As described above, as the marketplace based on the e-catalog is operated, the business expands, and accordingly, the increase, mass, and variety of products to be registered in the e-catalog require a lot of time and manpower to register the e-catalog. . In the conventional product registration process, when a business is activated, there is a problem in that the recruitment of the catalog staff of the market place is necessary. In addition, if a buyer's request for new product registration is delayed, the lead time of catalog product registration is increased, resulting in an increase in time for subsequent processing such as sourcing, ordering, and delivery, which may increase customer dissatisfaction.

Therefore, reducing catalog entry lead times is one big issue for each marketplace to resolve customer complaints, respond quickly to customer requests, and create a successful online B2B marketplace. Eliminating delays by automating three major bottleneck zones that delay catalog product registration lead times among existing product registration processes that process items requested for product registration requests. In other words, there is a need for automation of data cleansing during catalog product registration, automation of catalog staff assignment, and automation of classification and item recommendation.

Accordingly, the present invention has been made to solve the above-mentioned problems of the prior art, and data cleansing and catalog assigning and classification / item recommendation are made to fit the product information (commodity catalog) distributed in electronic commerce in accordance with the marketplace classification system. It is an object of the present invention to provide a method for automatically building an electronic catalog, which significantly reduces lead time of catalog product registration.

In order to achieve the above object, the automatic electronic catalog construction method according to the present invention performs the normalization and standardization process by extending the expression by extracting key words with reference to the lexical dictionary for the pre-built electronic catalog data to perform the learning data. Generating training data to generate; A catalog classifier learning step of delivering the training data to a catalog classifier to train the catalog classifier; Vocabulary expansion and standardization step of classification-requested e-catalog data to be classified to classify the e-catalog data to be categorized by extending the expression through key word extraction, performing normalization and standardization, and delivering it to the catalog classifier. Wow; The catalog classifier includes a catalog classification step of performing classification learning using the learning data and recommending a suitable classification for the classification requested electronic catalog data.

Hereinafter, an electronic catalog automatic construction method according to a preferred embodiment of the present invention with reference to the accompanying drawings will be described in detail.

The present invention aims to solve the three bottleneck spots that have an absolute influence on the catalog lead time in the electronic catalog construction by using an automation tool. In the existing e-catalog building process, the major delays were data cleansing, the assignment of the catalog manager responsible for the actual items of the general manager, the classification to find the classification in which the items are located, and the item recommendation. It was a manual process that had to be checked or searched visually.

FIG. 2 is a diagram illustrating a sequential flow of detailed business processes at the time of automatic electronic catalog construction according to a preferred embodiment of the present invention.

First, when a buyer uses the market place and a required product is not registered in the electronic catalog in the market place, a purchase order PO cannot be issued, so a new product registration request step is made.

These new requests can be made online using the marketplace system you are using, and in the case of B2B marketplaces, when a buyer company or a new buyer company is launched, The off-line or on-line catalogs of buyer companies used in off-line or internal systems will be registered in bulk in the marketplace electronic catalog.

When the buyer requests new product registration, the hidden bulk process is converted into a normalization process. Buyers will no longer inform item owners of item bulk data.

The bulk data that buyers have is entered into existing DB as if it came through existing web and I / F through a prescribed process.

All of these new raw data coming in consistently go through the Auto Cleansing process of the Automated Tool and the Auto Catalog Assignment Process of the Automated Tool. After registration of new requested data through automatic classifier and automatic recommendation of items, through request complementary function process, product registration is completed in electronic catalog.

Here, in case of a new item request that cannot be confirmed and determined by the sales catalog manager or the automation tool because the quality of the data that the buyer requested the product is significantly low, the request supplement function is requested to reinforce the data contents to the buyer company. It is desirable to provide. The catalog manager makes a request for supplementary request. The buyer company checks the details and then reinforces the data in the market place and registers the product in the electronic catalog.

In the present invention, data cleansing automation is performed as follows.

There are two forms of data that a buyer requests for new product registration. First, the first form is an on-line request using a marketplace system. In the other case, when a large amount of data is required for a new product registration request, the existing buyer company usually In many cases, the Excel file or the electronic catalog file of the EDI system used internally is delivered to the marketplace in free format. The new product registration request data received from the buyer is raw raw data which is often in a pre-format format. Therefore, some form of data available in the marketplace and data cleansing is required to process.

Automated data cleansing tools use rule-based Auto Data Cleansing Systems to define cleansing rules in a pre-formatted format. In addition, the buyer's request data is cleaned by the data cleansing rules set in the market place. Once cleaned, the data will be converted into a format that can be used in the marketplace's electronic catalog system.

In addition, the catalog manager assignment automation in the present invention is performed as follows.

After the data has been cleaned by the buyer, the cleaned data is stored in the DB as a new request database in the marketplace's electronic catalog system. Each newly requested data is assigned to an individual catalog manager who manages the items in each taxonomy (automated catalog manager job assignment in the automated tool).

Previously, since the general manager confirmed each request and decided the person who would handle each new product registration request, it took a lot of time for the general manager to check. However, in the present invention, a person in each category is automatically assigned by operating an learning system that introduces an automated tool and learns a history of newly registered data on newly requested data. In this case, the assignee can be automatically assigned to each category by the result of the automatic classifier, or the assignment can be made by adjusting the amount of work for each person in charge, thereby eliminating the waste of time required for job allocation of the general manager. And if you want to change the assignments automatically made by the automation tool, you can adjust the assignment of the contact person directly from the administration screen.

In addition, the classification and item recommendation automation in the present invention is performed as follows.

Newly requested data is provided through the automated sorter system of the automated tool, and the auto-recommendation function is provided so that the catalog manager in charge of each classification can easily handle the newly requested data.

Previously, the catalog manager decided on which category to generate the requested item based on the user's experience or search confirmation on all the data requested for registration of new products. Based on the UNSPSC classification system and items recommended by the company, new product registrations can be judged faster and more accurately.

The catalog automatic classifier employed in the automatic electronic catalog construction system according to the present invention configured as described above will be described.

In the present invention, the classification code of the catalog is recommended using the developed automatic classifier. The classifier uses a classification model that extends the Naive Bayesian Classifier.

The catalog classification uses the catalog information specified by the item name, specification, and manufacturer, and can use the registered synonym information, based on the Naive Bayesian Theory, and recommends classification codes by learning. have.

The operation process of this catalog automatic classifier is shown in FIG.

As shown in FIG. 3, the expression is expanded through key word extraction with reference to a lexical dictionary (a standard vocabulary list including a simple noun dictionary to an ontology extension) with respect to previously constructed data (prebuilt electronic catalog data). Normalization and standardization processing is performed to generate learning data (electronic catalog data designed for learning) and deliver it to the catalog classifier (S1, S3, S4). Accordingly, the catalog classifier learns using the received training data.

 On the other hand, with respect to the request data (e-catalogue data to be classified), the lexical dictionary (a standard lexical list including a simple noun dictionary to an ontology extension) is referred to to extend the expression through key word extraction to perform normalization and standardization processing. To the catalog classifier (S2, S5).

The catalog classifier performs classification learning using learning data, and recommends a suitable classification (classification catalog) with respect to the input request data (S6). Here, the catalog classifier recommends, for example, about three classification catalogs for one request data, and the recommended classification catalog is subjected to verification, mapping, and redundancy check by a catalog classification officer. The classification catalog is selected as an electronic catalog for the corresponding request data and stored in the integrated catalog DB (S7).

Catalog learning and classification according to the present invention applies different classification weights according to catalog attributes (e.g., product name, specification, manufacturer, model number, etc.), which adopts the extended Naive Bayesian classification and learning algorithm. To achieve.

The basic technology of the automated classifier of the automated tool uses various learning theories, and Navie Bayesian classification and learning theory are used as the core part of learning and classification.

Duplicate determination on the catalog is performed based on the following criteria.

We recommend using the Mass Catalog Search Engine (MCI) to list existing items that correctly contain the words contained in the new item request. The system basically recommends items with words that resemble the most requested items based on product name, specification, and model number. The attributes to be searched can be recommended by combining four attributes such as product name, specification, manufacturer and model number. Even if special characters are used and spaces are incorrect, the item with the most similar attribute value is recommended. Duplicate catalog discrimination criteria can be changed from time to time, and the first three items are recommended to provide an interface that allows a person to directly confirm the duplicate.

Next, applying the treatment of the present invention to a new requested item is as follows.

4 is a diagram illustrating a processing flow diagram of an automated tool when a new requested item comes in off-line or on-line.

When raw data in a raw form (for example, data in EXEL file format for a large amount of data) is input in a bulk form (S10) off-line from a buyer (S10), Automatically clean the input value according to the defined rule (S12) and automatically convert the input data (S14), recommend the converted data, and convert the document into data selected from the recommended data to obtain the standardized data (S16 to S20). ).

On the other hand, raw data (for example, a large amount of data) in a raw form is input in bulk (BULK) form through I / F from a buyer on-line (S10) or raw (raw) form through the web. When data (for example, a large amount of data) is input (S24), the input data is automatically cleaned according to a rule defined in the system (S26) and converted into a standardized data form (S20).

The formalized data is registered and stored in the database, and at the same time, the buyer is notified that the data is registered (S28, S30). The data stored in the database is automatically classified by the general manager using the automatic classifier, selects the data recommended for classification by the automatic classifier, and assigns a catalog manager to process each data (S32 to S36). Accordingly, the system notifies the assigned person in charge of the registration registration (S38).

5 is a flowchart at the time of item approval of newly requested data.

According to the new request search by the catalog manager, the system searches for a new request and asks the person in charge whether to reclassify the searched new request (S40, S42).

When the catalog manager reclassifies, the catalog sorter performs the automatic sorting and receives a classification recommendation by the catalog sorter (S44, 46). At this time, the catalog manager decides whether or not to classify. If not, the catalog manager requests re-allocation to the general manager (S48 and S50).

On the other hand, if the catalog manager continues to classify the existing items are automatically recommended through the automatic classification process (S52) and the search process (S54) (S56), and the items automatically classified and automatically classified in the S44 and S46 is recommended Check whether the existing items are automatically recommended (S58).

If it is confirmed in step S58 that the existing item is duplicated, the corresponding existing item is mapped and the item is designated and stored in the electronic catalog database (S62, S64, S66). On the other hand, if it is confirmed in step S58 that the existing item is not duplicated, the new item is designated and stored in the electronic catalog database (S64, S66).

In order to compare before and after applying the automation tool according to the present invention described above, the time required for measuring 100 new items as a new launch was measured for the work time, and the results are shown in Table 1.

TABLE 1

Before applying the automation tool according to the present invention Application of automation tool according to the present invention    Job history 1) General manager (1 person): .Catalog manager designation for new launch RAW data distribution: 3 hours. Classification system designation and catalog manager final allocation: 3 hours 2) Catalog manager (5 people). Classification system designation, etc .: 2 hours each (total 10 hours) 1) General manager (1 person). Loading time on automated tool system: 45 minutes 2) Automated tool system operation (cleaning, classification, etc.): 15 minutes 3) Catalog manager (1 person) Final confirmation: 3 hours Total work time 16 hours 4 hours

Therefore, according to the present invention, the lead time of catalog product registration by fully automating data cleansing, catalog assignee, and classification / item recommendation to match the product information (product catalog) distributed in electronic commerce according to the marketplace classification system. Can be significantly reduced.

In addition, this invention is not limited to the above-mentioned specific embodiment, It can variously deform and modify within the range which does not deviate from the summary of this invention, and can implement.

As described above, according to the present invention, even if the e-commerce is activated and the number of products to be registered in the e-catalog is increased, the number of mass products and the variety of products are increased, the time and manpower required to register the e-catalog can be significantly reduced. This can reduce catalog entry lead times, resolve customer complaints, and respond quickly to customer requests, helping to create a successful online B2B marketplace.

Claims (4)

delete A learning data generation step of generating learning data by performing a normalization and standardization process by extending an expression through key word extraction with reference to a lexical dictionary with respect to the previously constructed electronic catalog data; A catalog classifier learning step of delivering the training data to a catalog classifier to train the catalog classifier; Vocabulary expansion and standardization step of classification-requested e-catalog data to be classified to classify the e-catalog data to be categorized by extending the expression through key word extraction, performing normalization and standardization, and delivering it to the catalog classifier. Wow; A catalog classification step of performing, by the catalog classifier, classification learning using the learning data and recommending a classification catalog as a suitable classification for the classification-requested electronic catalog data; And performing a duplicate check by referring to an existing catalog which is automatically recommended through automatic classification and search for the classified catalog recommended in the catalog classification step. The method of claim 2, And in the catalog classification step, the catalog classifier recommends a plurality of classification catalogs for one classification requested electronic catalog data. The method of claim 3, The lexical dictionary includes a standard lexical list including a simple noun dictionary and an ontology extension.

Images