KR100289065B1 - Method for processing natural query language for searching commodity information of electronic commerce - Google Patents

Abstract

PURPOSE: A method for processing a natural query language is provided to convert into an SQL(structured query language) which is a query language of an RDBMS(relational database management system) by extracting a commodity term in a natural query language when a user requests a search according to a desired commodity using a natural language. CONSTITUTION: If a buyer requests a search for a commodity by a natural query language for buying the commodity through Internet virtual stores(S1), a query language is extracted by an extractor(S3) and stored in a storing place array through an HTTPa(40) which is a web server in an electronic commerce system(S2). A token of the stored natural query language is separated by a separator by making various conditions as a space, a pause, a period, and a symbol such as a delimiters(S4). If a term which indicates a commodity price is existed in the separated token, the term is stored in a storage location array and removes the corresponding token(S5). After a postposition is removed in each token(S6), an unused token is removed(S7). If an ESQL(embedded SQL) is created according to an extracted term and an electronic commerce term table is searched, characteristics(CIC(Company Identification Code), PCC(product Classification Code), BN(Brand name), CFC(Chief Function Code)) of each term may be sensed from a control field(S8). A term, which is not possessed in five categories(company identification code, product classification code, brand name, chief function code, and price), is excluded in a processing object, and an ESQL is created according to the four terms except the price and an electronic commerce synonym table is searched, thus a synonym/similar language/foreign language may be obtained(S9). The synonym/similar language/foreign language is processed as one term group and is not considered in a normal creation according to a query language(S10).

Description

Natural Language Query Processing Method for E-Commerce Product Information Retrieval

According to the present invention, when a purchaser who wants to purchase a product in an electronic commerce request a search for product information desired by a natural language query, it is extracted and provided from a relational database (hereinafter referred to as RDB). It is about a method.

In the past, since the search service of this field is performed on the Internet World Web, when the search service for the natural language query is performed, the search speed is reduced due to the complexity of parsing the natural language query. Is not able to perform a search service using a natural language query, and inputs the product information to be searched separately in the 'Hyper Text Markup Language' (form HTML) form tag. Method has been used.

Therefore, the user interface is not natural, causing inconvenience in use, and it is difficult for the general public to easily use it.

In order to solve the above problems, the present invention, when a user requests a search for a desired product using a natural language, a relational database management system by extracting a product information search term from a natural language query language so that it can be provided quickly and accurately. An object of the present invention is to provide a method of converting a structured query language (hereinafter referred to as SQL) that is a query word of a management system (hereinafter referred to as RDBMS).

1 is an overall configuration diagram of an electronic commerce information service system according to the present invention;

Figure 2 is a flow chart of a natural language query processing method for e-commerce product information search to which the present invention is applied.

<Explanation of symbols for main parts of drawing>

10: Buyer web browser 20: Seller web browser

30: Merchant System 40: Web Server

50: high performance web-database gateway application server

51: Electronic Commerce Information Search System

52: Electronic Commerce Information Directory Management System

53: e-commerce information indexing system

60: relational database management system

70: e-commerce information database 80: search term extractor

90: query regular expression generator 91: term database

92: Synonym Database 100: Search Result Generator

101: commodity database

In order to achieve the above object, the present invention extracts a search target term, generates a regular expression for a query of each extracted term, and uses the generated regular expression as a structural query (SQL) that is a query word of a relational database management system (RDBMS). It is characterized by generating a search result by converting to.

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

1 is an overall configuration diagram of an e-commerce information service system according to the present invention, a buyer web browser 10, a seller web browser 20, a merchant system 30, a web server 40, an e-commerce information search system ( 51), a high performance web-database gateway application server 50 consisting of three parts of the e-commerce information directory management system 52 and the e-commerce information indexing system 53, the relational database management system 60, and the e-commerce information database ( 70).

1 is a purchaser who wants to purchase a product through an Internet virtual store, and requests information of a product he / she wants to purchase using a natural language query using a web browser 10, an e-commerce information service system. Through the fast common gateway interface (hereinafter referred to as FastCGI) of the hypertext transfer protocol daemon (hereinafter referred to as HTTPd) 40, which is a web server within the web server, the e-commerce information retrieval system 51 When delivered, converted to SQL, and delivered to the relational database management system (RDBMS) 60, the product information is extracted from the e-commerce information database 70, and then passed back to the e-commerce information search system 51 to search results Is converted into an HTML document and transmitted to the buyer's web browser 10 through the HTTPd 40.

A seller who wants to sell a product by building an internet virtual store enters a home page document location (hereinafter referred to as a URL) and seller information using his web browser 20, and then uses HTTPd (40). After passing through the FastCGI to the e-commerce information directory management system 52 and passing the seller authentication process, SQL is generated, passed to the RDBMS 60, and the seller's directory is registered in the e-commerce information database 70 and again. The registration information is transmitted to the electronic commerce information directory management system 60, and the registration result is dynamically converted into an HTML document and transmitted to the web browser 20 of the seller through the HTTPd 40.

The virtual store registered in the e-commerce information database 70 goes to the merchant's merchant system 30 through the HTTPd 40 periodically by the indexing web agent in the e-commerce information indexing system 53. When the HTML document with information is found and collected again in the e-commerce information indexing system 53 through the HTTPd 40, the product information is extracted by the product information indexer, and the SQL for this is generated and transmitted to the RDBMS 60. The product information is registered in the information database 70.

Table 1, Table 2, and Table 3 show the configuration of the database used in the natural language query processing method for e-commerce product information search.

The database used in the process of generating SQL for product information retrieval from the user's natural language query language includes the e-commerce information service term table of Table 1, the e-commerce information service synonym table of Table 2, and the e-commerce information service product information of Table 3 It consists of a table.

The term table and the synonym table are constructed by the cooperation of the e-commerce information indexing system 53 and the person who is the e-commerce information service system administrator, and the product information table is automatically constructed by the e-commerce information indexing system 53. do.

The term table consists of terms, control fields, and fields indicating registration dates. The synonyms table consists of representative terms, synonyms, and registration dates. The product information table includes the manufacturer's name, product classification name, main specifications, product name, and model number. , Price, document location, and date of registration.

E-commerce information service term table Field (column) name Field (Column) Specifications Terminology (TERM) variable Character Control Field (CFIELD) Variable Character Registration date (RDATE) DATE

E-Commerce Information Service Synonyms Table Field (column) name Field (Column) Specifications Representative term (RTERM) Variable Character Synonyms (SYNONYM) Variable Character Registration date (RDATE) DATE

E-commerce information service product information table Field (column) name Field (Column) Specifications Manufacturer Name (CIC) Variable Character Product Classification Name (PCC) Variable Character Brand name (BN) Variable Character Main Specifications (CFC) Variable Character Model Number (MN) Variable Character PRICE Integer Document location (URL) Variable Character Registration date (RDATE) DATE

2 is a flowchart of a natural language query processing method for e-commerce product information retrieval to which the present invention is applied, and a buyer who wants to purchase a product through an Internet virtual store stores the information of a product he / she wants to purchase. When the user requests a search using the natural language query using the browser 10 (S1), the query is extracted by the user natural language query extractor through the HTTPd 40 which is a web server in the e-commerce information service system (S2). ) Is stored in the storage array.

The stored natural language query words are separated by a query token separator using various conditions such as spaces, commas, periods, and symbols as delimiters (S4).

Investigate whether there is a term representing commodity prices in the separated tokens, if any, store them in a storage array and remove the corresponding tokens (S5), and after the investigation is removed from each token (S6), stopwords are removed (S7). ).

In the present invention, the term indicating the commodity price to be processed is, for example, a section specification type of 'between 500,000 won and 1 million won', a section tracking type of '1 million won', and a 'smallest' 3 It is eggplant.

When the above processing is completed, the search term is extracted from the first user's natural language query (80).

The product information on the query subject to be processed in the present invention includes a company name (Company Identification Code, hereinafter referred to as CIC), a product classification code (hereinafter referred to as PCC), a brand name (Brand Name, hereinafter referred to as BN). , Chief function code (CFC) and price (PRICE).

The buyer requests the search using the language used in everyday life by combining the five product information.

For each extracted term, an embedded structural query (hereinafter referred to as ESQL) is generated and the E-commerce information service term table 91 is searched for the characteristics of each term (CIC, PCC, BN, CFC) from the control field. ) Can be grasped (S8).

Terms not belonging to the above five categories are excluded from processing, and ESQL is generated for four terms excluding price, and the e-commerce information service synonym table 92 is searched to obtain synonyms, similar words, and foreign words for each. It may be (S9).

A synonym / similar word / foreign word searched for a term is treated as one term set for that term and is not considered when generating a regular expression for the query word (S10).

The natural language query regular expression generation is performed by selecting a logical operator from the query and determining a type of a term representing a product price.

At this time, the process of selecting a logical operator from the query, the first step of separating the natural language query language into logical thin tokens, the second step of extracting the search term by removing the stopword token, the characteristics of the extracted search target term And a query step is generated by a logical operator using the characteristics of the search target term.

The query for e-commerce product information search is made from the natural language query through each of the above steps. In the following description, a process for generating the query for e-commerce product information search from the natural language query is described below.

First, when the user asked, "I want to buy Samsung's refrigerator for 1 million won" in natural language, the user's natural language, "Samsung", "refrigerator", "for 1 million won", " The first stage, which is divided into tokens of "accident" and "I want", and the logical stages, which are logically separated, are the first stage, in which the terminology tokens are separated through a search dictionary, and the logically separated tokens, through a search dictionary. The second step is performed by obtaining a search target term of "Samsung", "fridge" and "1 million won" by removing the token, and using the term definition dictionary table for e-commerce product information search, the search term is found. Samsung "-Manufacturer's name (ICC)," Refrigerator "-Product classification name (PCC)," 1 million won "-Price is redefined.

When the characteristics of the search term are obtained as described above, a structural query word (SQL), which is a query word of a relational database management system for extracting information stored in the database, should be generated. Perform the fourth step by using.

In this case, a more complicated natural language query is described as an example of the fourth step of generating a query by a logical operator using characteristics of a search term. For example, "I want to buy a TV and a refrigerator of 1 million won for Samsung and Daewoo." After the first, second and third stages, the following results are produced.

"Samsung", "Daewoo"-Manufacturer (CIC)

"Refrigerator", "TV"-Product Name (PCC)

"1 million won"-PRICE

The query regular expression generated in the present invention is shown in Equation 1 below.

((CIC)) * ((PCC)) * ((BIN)) * ((CFC)) * (PRICE)). … Formula (1)

'*' In Formula 1 represents a logical operator AND, and '(())' represents zero or more as possible numbers.

For example, when a buyer requests a search with the query 'I want to buy a TV and a refrigerator 1 million won for Samsung and Daewoo,' the regular expression generated from Equation 1 is as follows.

(Samsung) * (TV) * (1 million won)

(Samsung) * (Refrigerator) * (Million KRW)

(Daewoo) * (TV) * (1 million won)

(Daewoo) * (Fridge) * (1 million won)

In this case, the four regular expressions that apply to SQL are derived as follows.

First, the search terms that have the same characteristics are OR-related, and the other characteristics are AND-related.

Ie OR relationship: (Samsung or Daewoo),

(TV or refrigerator)

AND relationship; (Samsung or Daewoo) and (TV or refrigerator) and (1 million won)

Second, the number of SQL metabolic regular expressions is calculated by multiplying the occurrence frequency of the OR relationship by the number of AND relationships, i.e., in the above example, two CICs, two PCCs, and one Price occur. 1 = 4

The following is an example of generating the most typical SQL using the regular expression.

select * from Product Details

where: CIC = 'Samsung' OR: CIC = 'Daewoo'

AND: PCC = 'TV' OR: PCC = 'Fridge'

AND: Price between 1000000 and 1999999;

Here, "one million won" represents an embodiment that can be searched by converting between 1000000 won and 1999999 won.

In the present invention, it does not parse correctly in processing a user's natural language query, but look at the query regular expression (Equation 1), but the subject of interest is five characteristics, providing the user with as much information as possible to select The focus is on widening. In other words, the complex natural language query shown below shows the same result as the query word.

"I want to buy Samsung TV and Daewoo refrigerator for 1 million won each."

(I want to buy a million won TV and refrigerator for Samsung and Daewoo.)

Comparing the two query terms, the difference is that "Samsung refrigerator 1 million won" and "Daewoo TV 1 million won" is additionally shown as a result, but the user is not required but presented as a data for comparison.

Another reason for doing this is that when the correct parsing is performed, the system response time is remarkably reduced, which is not suitable for the search service environment on the Internet web used by the unspecified majority.

Next, the judgment on which type the term representing the commodity price belongs to includes, for example, a section specification type (for example, between 500,000 won and 1 million won), a section fall type (for example, 1 million won), and a minimum maximum type (for example, For example, the most expensive and the least expensive) is used. During the first, second and third steps of selecting a logical operator from the query word, the term for price refers to an electronic dictionary, that is, an electronic commerce information service. Terminology table 'and' e-commerce information service synonym table '.

For example, 'won,' 10,000 won ',' wondae, 'wonwondae', 'about 10,000 won', etc. are in the above section, 'from circle', 'from 10,000 won', 'from circle', 'between', ' 10,000 won ”and the like are the above-mentioned section type, and 'expensive', 'cheap', 'cheap' and the like can be identified as the search target term indicating the price by using the registered items classified as the minimum and maximum types. You can find out which type they belong to.

When the processing procedure is completed, the query regular expression is generated from the term to be searched (90).

The following is to generate the SQL that is a query of the RDBMS from the generated query regular expression (S11).

At this time, the generated SQL is static SQL. In consideration of the number of cases that can be generated by the query regular expression, the host variable is included in one SQL statement configured to retrieve all related information in one execution. Is created by substituting

Of course, it is possible to generate and respond to dynamic SQL corresponding to the query regular expression generated at that time, but generating and searching dynamic SQL in the search service environment on the Internet World Wide Web used by the unspecified majority not only takes a long processing time but also an RDBMS. Since the query optimization function of is not used effectively, it is excluded in the present invention.

The e-commerce information service product table 101 is searched using the generated structured query word to obtain related information (S12).

The product information on the query subject to be processed in the present invention is CIC, PCC, BN, CFC and PRICE, but the product information provided to the buyer as a search result is superimposed on these five pieces of information to sell the model number and the corresponding product. The URL in the store and the home page URL of the virtual store extracted from the URL are additionally provided.

The search results for the eight pieces of information are dynamically generated in the form of an HTML document (S13) and transmitted to the buyer's web browser 10 through the HTTPd 40 (S14).

As described above, the present invention provides a convenient user interface because the user requests the desired product information in the natural language query used by the user everyday, and provides a high-quality search service through accurate product information extraction and fast processing time for the natural language query. It is effective to provide.

Claims (4)

In the natural language query processing method for e-commerce product information search, A first process of extracting a search term; Generating a query regular expression for each extracted term; And a third step of converting the generated regular expression into a structural query (SQL), which is a query of a relational database management system (RDBMS), to generate a search result. The method of claim 1, wherein the first process is A first step of requesting a purchaser to purchase a product through an internet virtual shop by using a web browser to search information of the product using a natural language query; A second step of requesting a search using a natural language query and extracting the query word through a user natural language query extractor through a web server in an e-commerce information service system and storing the query in a storage arrangement; A third step of separating the token from the stored natural language query word using various conditions such as spaces, commas, periods, and symbols as cutters; If there is a commodity price term in the separated token, the e-commerce product information comprising the fourth step of storing the token in the storage array and removing the corresponding token, removing the investigation from each token and removing the stopword token. Natural language query processing method for search. The method of claim 1, wherein the second process Separating a natural language query word into a logical enable token, removing a stop word token, extracting a search target term, identifying characteristics of the extracted search target termong, and generating a query term using a logical operator using the same; And a second step of identifying the form of the term representing the commodity price by dividing the natural language query into a section specification type, section tracking type, and minimum size type. 4. . The method of claim 1, wherein the third process is A first step of converting the generated regular expression into SQL which is a query word of an RDBMS; E-Commerce Information Service Search the product table to find the manufacturer name (CIC), product classification name (PCC), product name (BN), main specifications (CFC), model number (MN), price (PRICE), and document location (URL). A second step of obtaining product information; And extracting a homepage URL from the URL, attaching it to a search result, and dynamically generating an HTML document and providing the same to a user through a web server.

Images