KR20030039334A - search engine and method using the semantic relationship information for document or keyword - Google Patents

Abstract

PURPOSE: A device and a method for searchiing data using semantic relation data of a search term are provided to extract the plural semantic relation data for a document by automating a process classifying and indexing an input document through an intelligent inference engine. CONSTITUTION: The new morpheme of the input document is extracted by processing a morpheme analysis and stopwords after receiving the input document of the news(S15). A combination vector comprising the isomorphic morpheme and a corresponding discrimination code is extracted through the comparison between the dictionary database I and the news morpheme by using the inference engine I, and the core morpheme combination data is extracted by applying an inference rule to the combination vector(S30-S40). The semantic connection code data according to the core morpheme combination data is automatically extracted from the dictionary database II by using the inference engine II(S50-S60). The core morpheme combination data combined with the connection code data is stored in the connection dictionary database III and is attached to the input document, and the input document is stored in a database(S65-S75).

Description

Search engine and method using the semantic relationship information for document or keyword}

The present invention relates to an information retrieval apparatus and method that enables semantic information retrieval instead of word based by using semantic connection code information automatically generated for a specific document through the intelligent inference engine as a condition for DB retrieval. More specifically, the search terms entered by the user are not used as the discriminating condition of the DB search itself. Instead, the key morphemes of the original document are extracted using the intelligent inference engine, and the semantic linking code information is automatically generated. By using this information as a new discrimination condition of search, it relates to an extended information retrieval apparatus and method for presenting a search result for documents inferred to be semantically related, although they do not contain the same morpheme as the search term. .

Currently used information retrieval methods can be classified into four types.

First, a word matching method that extracts a document containing the same word collectively using one word as a search criterion. In this case, if there is no identical word in the document, the word matching method is not presented as a search result and the meaning is the same. If only words exist, the search results are high.

For example, petroleum prices are usually abbreviated as 'oil prices', which leads to documents that have no semantic association with the search term, such as '-for one reason--', 'family meetings', or 'securities'. Search error rate is high.

Second, subdividing a word into lower words, assigning a code to each word, and applying the distance vector concept to these words to extract other documents of the closest type. Since only the segmentation is performed, the search cannot be performed for words at the same level or higher level as the search word.

Third, it is a method of finding one optimal search category using a plurality of words, which in turn extracts only the documents that fall into one category, which reduces the variety of search and results according to the weighting method for each word. There is a disadvantage in that the difference is large.

Fourth, as a method of discovering a plurality of optimal search categories using a plurality of words, the difference in results is large according to the weighting method for each word.

In addition, all four methods use a word-matching method that uses a one-to-one comparison between search terms and comparison documents as a basic condition of a DB search, so that only documents that contain the search term are limited to search targets. The problem of retrieval error rate is also large.

In the existing method, since the classification code indexing process for the original document depends on the user's direct input, the scope of classification is narrow and many unclassified documents appear. For example, a media company that generates more than 500 news information a day follows the reporter's method of inputting indexing of each news information. As a result, the major categories (political news, economic news, Only social news, etc.) is possible, but it is practically impossible to directly index linkage codes of subclassification, subclassification, and semantics.

The present invention has been made to solve the problems posed by the classification and indexing of information and the search process of word matching method. First, the classification and indexing process of the original document is automated by using an intelligent reasoning engine. It is possible to extract a plurality of semantic information related to the information, and also aims to accurately index the connection code for it.

By using the link code index information in the user search phase, although the documents are not the same as the search terms in the document, the documents inferred to be semantically relevant are presented as valid search results, so the error rate of the search according to the existing word matching method The aim is to solve the problem.

1 is an exemplary diagram for a step-by-step procedure for automatically generating connection code information of the semantic center of a specific document according to the present invention

2 is a block diagram of the automatic generation device of the connection code information presented in the present invention

3 is a block diagram of the inference engine I proposed in the present invention

4 is a block diagram of the inference engine II presented in the present invention

5 is an exemplary diagram illustrating a step-by-step procedure for extracting a semantic related document instead of a word based word for an input search word or a selected document according to the present invention.

6 is a block diagram of an apparatus for extracting a semantic-oriented related document for an input search word or a selected document according to the present invention;

7 is a block diagram of the inference engine III proposed in the present invention

Figure 8 is a simple illustration of the dictionary DB I in accordance with the present invention

Figure 9 is a simplified illustration of the dictionary DB II according to the present invention

10 is a simple illustration of the connection dictionary DB III according to the present invention

In order to achieve the above object, the present invention broadly divides the core morphological information of the original document by applying the intelligent inference engine I, and automatically applies the connection code information to the core morphological information by applying the intelligent inference engine II. Indexing and extracting the connection code information corresponding to the user input search word using the intelligent reasoning engine III and extracting the document corresponding to the code information from the database and transmitting it to the user.

Hereinafter, exemplary embodiments of the present invention will be described with reference to the accompanying drawings.

Before that, a few terms defined and used in the present invention will be described.

(1) input document: text-based information as initial document information generated directly by the author, either internally or externally.

(2) news morpheme: A morpheme set of news information generated after morphological analysis and stopword processing of news information, which is a kind of original document.

(3) database morpheme: All types of morphemes included in the dictionary database created by the administrator.

(4) Isomorphic morpheme: A morpheme included in a dictionary morpheme among the news morphemes, the pattern being the same.

(5) core morpheme: A morpheme automatically extracted by the inference engine I of the present invention as a single or a plurality of morphemes which are the core subjects of the original document among the same morphemes.

(6) input keyword: A word or combination of words that a user directly enters to retrieve a particular document from a database.

1 is an exemplary diagram illustrating a stepwise procedure of automatically generating semantic connection code information for an original document according to the present invention. The news information receiving step (S10), news stemming extraction step (S15), and the news stemming Determining whether there is a dictionary DB I (S20), in the case of a news stemmer not in the dictionary DB I step (S25), in the case of news stemmers in the dictionary DB I, the same stemming in the dictionary DB I and the corresponding Extracting a classification code (S30), generating a combination vector of the same morpheme and the classification code (S35), extracting a core morphological combination from the combination vector using an inference engine I (S40), and Determining whether or not the discriminating code of the core morpheme combination is 1 (S45), if it is not 1, searching for the same morpheme for a line of code that matches the discriminating code in the dictionary DB II (S50), and the dictionary DB. Ⅱ Extracting the item morpheme and the industry code corresponding to the same morpheme (S55), extracting the connection code information corresponding to the S55 or S40 information from the dictionary DB II using the inference engine II (S60), the S40 Combining the results of and S60 and storing them in the connected dictionary DB III (S65), adding the results of the S40 to the news information of S10 (S70), saves the results of S70 in the database and ends (S75 ~ S80). It consists of

In particular, in the present embodiment will be described focusing on the process of processing the news information that occurs in real time (in the case of news is produced more than 500 per day because the average is produced by one every three minutes). In this case, the information producer is defined as a journalist's reporter, the receiver is a media DBMS or an external DBMS, and the user is a user of Internet news information through a wired or wireless communication network.

In addition, the conventional general retrieval apparatus is composed of only three steps, such as S10-S75-S80 when the steps according to the reception and storage of data are based on FIG. 1.

In other words, the original document is simply stored in the database in chronological order without any processing, and when a search word is entered, the corresponding document is searched in the database by word matching method based on the search word.

The news information receiving step S10 of FIG. 1 corresponds to a general data transmission / reception system of a communication system, and the news morpheme extraction step S15 corresponds to a role of a morpheme analyzer which is a basic element of an information retrieval apparatus. It is not a new suggestion, so detailed description thereof will be omitted. The morpheme analyzer, however, serves to separate a given phrase or sentence into morphemes, which are the smallest units of meaningful words.

In the case of S20, it is determined whether the news morpheme of S15 exists in the dictionary DB I of the type shown in FIG. The dictionary DB I summarizes the morphemes appearing in the news by assigning a classification code for each type predefined by the administrator. The morphemes located on the left side of the column in terms of mathematical inclusion relations correspond to higher levels. Conversely, the one on the right becomes the lower level.

As a simple example, if each step is applied to the news information "Samyangsa, flour price reduction" as follows.

(S10) Samyang, wheat prices cut

(S15) Samyangsa, flour, price, cuts, wheat, flour

(S20) Samyangsa; flour ; price ; Cuts

In this case, S20 extracts only the morphemes that exist in the dictionary DB I, that is, the same morphemes, from the morphemes of S15. If the classification code for the same morpheme of S20 is given by using the dictionary DB I of FIG.

(S30) (Samyang, 1), (Flour, 3), (Price, 5), (Reduction, 6)

In the dictionary DB I of FIG. 8, the classification code means a classification code for each type of morpheme, which is assigned to 1 for an individual company (item name), 2 for an industry type, and 3 for an industry type factor affecting the industry type. As the administrator defines the classification criteria in advance, it is necessary to classify each morpheme according to the criteria.

Then, according to S35, the combination vector of the classification code and the morpheme is obtained as follows.

(S35) (1356; Samyangsa, flour, price, cuts)

At this time, in the method of preparing the combination vector, all the distinguishing codes of S30 may be sequentially listed without any spaces and separators, and after that, the separators may be inserted, and then all the morphemes of S30 may be separated and separated sequentially.

In S40, the core morpheme is extracted based on the combination of the classification codes from the combination vector using the inference engine I. The results are as follows.

(S40) (1, Samyangsa)

Inference engine I is a kind of computing device in which a stepwise inference rule is coded into a computer program to obtain a type of combination vector consisting of the core morpheme and the distinguishing code of the original document from the morpheme combination vector of S35. In this case, as an example, the representative classification code is first obtained by using the combination of the classification code of S35 as '13 ** = 1 ',' 2 **** = 2 ', '31 ** = 3', and then the General operation rules for obtaining the morpheme corresponding to the location information identical to the representative classification code as the core morpheme should be defined, and a detailed description thereof will be made with reference to FIG. 3. In conclusion, inference about the combination vector of S35 according to this rule can lead to the conclusion that '1356 = 1', and finally, the core morphological combination of the original document becomes (1, Samyangsa) as S40.

As another example, if the news, "Flour, North-North Economic Cooperation," is analyzed as above.

(S10) Wheat Flour, Inter-Korean Economic Issues

(S15) flour, inter-Korean economic cooperation, important issues, wheat, flour, inter-Korean cooperation

(S20) wheat flour; Inter-Korean economic cooperation; issue

(S30) (Flour, 3), (North and North Korea Cooperation, 4), (Issue, 6)

(S35) (346; Flour, Inter-Korean Cooperation, Issue)

(S40) (3, flour)

S45 is a step of determining the type of discrimination code shown in the core morpheme combination of S40. This is important to prevent retrieval errors when automatically extracting semantic connection code information for the core morphology of S40 from the dictionary DB II of FIG. The two examples will be described in detail.

1) In case of classification code = 1

In the case of "Samyangsa, flour price reduction", the core morpheme combination is (1, Samyangsa) and the division code = 1 in S45, so in step S60 'Samyangsa' in dictionary DB II using the inference engine II of FIG. You need to extract the connection code information for. At this time, as shown in the structure of FIG. 9, the pre-DB II shows all types of company information that has a business relationship with the company, such as the type of business, raw materials and products, competition and subsidiaries, and foreign affiliates. It is a database structure predefined by the administrator. The reasoning engine II of FIG. 4 is a series of arithmetic devices embedded with arithmetic rules for automatically extracting semantic connection code information for core morphemes of S40 using the dictionary DB II. Eventually, in this case, the following connection code information can be generated in S60.

(S60) (1, Samyang Corporation; food, feed, chemical, fiber; raw sugar, corn, TPA, EG; sugar, etc.)

That is, when the classification code of the morpheme combination obtained in S40 is 1, the classification code and the item name morpheme matching the morpheme combination of S40 are searched in the dictionary DB II, and then all morphemes belonging to the lower level of the morpheme combination are extracted. Just do it.

As can be seen from the results of S60, in the present invention, first, it is possible to automatically extract all the morphemes that are semantically related to the morphemes by extracting the core morphology of 'Samyangsa' from the original document. Second, when the user enters the search term 'Samyangsa', the present invention uses the result of S60 as a condition for determining DB search, so there is no morpheme called Samyangsa as shown in S60 as well as the same morpheme as Samyangsa. News information can also be searched, so the variety of search is expanded. Third, the search speed and accuracy are improved because only the morphological combination information existing at a specific location of S60 can be compared even in the DB search process for the search word.

However, if the discrimination code in S40 is not 1, inference is made through a more complicated procedure.

2) If the classification code is not 1

In the case of the "wheat flour, inter-Korean economic issues," the optimal morpheme is (3, wheat flour) and the distinguishing code is not 1 in S40, and thus the column of code lines corresponding to the distinguishing code in the dictionary DB II according to S50. Search for identical morphemes that match key morphemes. In addition, according to S55, the item morpheme and the industry code including the same morpheme must be extracted from the dictionary DB II, and the result is as follows.

(S55) (Samyangsa; 1), (Cheil Jedang; 1), (Korean Sugar; 1)

'1' in the (Samyangsa; 1) means, for example, one of a plurality of industry codes of Samyangsa, that is, an industry code delimiter indicating the food industry, which is the first industry.

After that, using the result of S55 as the inference standard of inference engine II, extracting the relevant connection code information from dictionary DB II (S60), the result is as follows.

(S60) (3, wheat flour; Samyangsa, 1; raw sugar, corn; sugar, cooking oil, wheat flour: Jeil sugar, 1; raw sugar, corn, starch; sugar, cooking oil, flour: Daehan sugar, 1; raw sugar, corn, starch; Sugar, cooking oil, flour)

Here, ',', ';' and ':' correspond to the division operator.

In this case, as in the case of 1) above, first, the morphologically relevant morphemes can be automatically indexed using the core morphology of 'flour', and second, the user searches for the word 'flour'. Even if input, the present invention uses the result of S60 as the condition for determining DB search, so it is possible to search not only the same morphemes as flour but also news information without morphemes such as Samyangsa, food, raw sugar, corn, etc. Third, DB search process Even in S60, only the information that exists in the specific location of S60 needs to be compared and searched, thereby improving the search speed and accuracy.

Comparing the results of 1) and 2) there are differences as follows.

In the case of 1), the classification code in S40 is 1, so all lower-level morphemes of the core morphemes are automatically extracted as concatenated codes in the dictionary DB II, but in the case of 2), the classification code in S40 corresponds to 3, namely commodity or raw morphemes Therefore, in S55, all item morphemes and industry morphemes of the upper level including the morphemes must be extracted, and other morphemes of the lower level directly linked to the industry morphemes must be extracted as the connection code information.

The reason for differentiating the discrimination criteria for each classification code is, for example, extracting the relevant item morphemes from the dictionary DB II for the morpheme 'flour' as described above, and then simply connecting all sub-morphemes of the item morphemes as in the case of 1). This is because the search error rate increases because 'chemical industry' and subordinate morphemes, which are not substantially related to wheat flour, are extracted as linked information.

In the present embodiment, the connection code information extracted in S60 for the above 1) and 2) is shown in the form of the combination code and the morpheme as described above, but if there is a separate code number for each morpheme, the classification code and the code number It can also be represented by the combination of.

In step S65, the connection code information obtained in S60 is stored in the connection dictionary DB III in the form shown in FIG. 10 in a sequential or overwriting format, and S70 adds the result of S40 to a specific position of the original document. Step. In this case, the reason for adding the result of S40 rather than the result of S60 to S10 is that after the user inputs a specific search word, the corresponding connection code information is extracted from the connection dictionary DB III and added to the connection code information and the original document. This is because the database search speed and accuracy can be improved by only comparing the information with S40.

In the following, (1) shows the data structure of the original document, and (2) shows the data structure subjected to the processing of S70, where the additional position of S40 can be arbitrarily arranged.

(1) Data structure of the original document

Major Code Date and time information Headline main text

(2) Modified original document processed in S70

Major Code Date and time information Information of S40 Headline main text

And S75 ~ S80 is a step of ending after storing the result of S70 in the database.

Figure 2 is an embodiment of a configuration diagram for a device for automatically generating a connection-oriented information of the semantic center presented in the present invention, the data receiver 100, the morpheme analysis unit 150, the inference engine I (200), the morpheme DB (10) ), Dictionary DB I (20), Reasoning Engine II (300), Dictionary DB II (30), Administration Tool (40), Morphological Classification (400), Linked Dictionary DB III (50), and Database (500). It is composed.

The data receiving unit 100 is a general data receiving apparatus capable of receiving an original document generated externally or internally, and the morpheme analysis unit 150 extracts a news morpheme from the original document. Although processors are required, they are already generalized techniques, and thus detailed explanations are omitted. And the reasoning engine I (200) will be described in detail later in the description process of FIG. 3, the same morphemes and corresponding morphemes by comparing the news morphemes extracted from the morpheme analysis unit 150 with the dictionary morphology of the dictionary DB I (20) After the classification code is extracted, the combination vector of the same morpheme and the classification code is generated, and the core morpheme combination for the original document is extracted according to the inference rule.

The dictionary DB I 20 will be described in detail later with reference to FIG. 8, but all the morphemes appearing in the news information are classified according to a predefined criterion, and a classification code is assigned to each of them. It follows the general database structure that can add / delete or modify stemming.

Inference engine II (300) is a device that automatically extracts the semantic-oriented connection code information from the dictionary DB II (30) according to the type of classification code of the core morpheme combination. In this case, when the classification code is '1', the same item morpheme as the core morpheme may be extracted from the dictionary DB II, and all morphemes of the lower level may be extracted as the connection code information. Steps S50 to S60 in 1 should be followed. The dictionary DB II (30) will be described in detail in FIG. 9, but is a data structure system in which all types of morphemes related to business, such as products, raw materials, competitors, and affiliated industries, are sorted by type. Although the management tool 40 does not present the actual screen example in the present invention, it is a device capable of modifying and updating the contents of the dictionary DB I and II, and the contents of the stemming DB 10 and the dictionary DB I 20 and It should contain the contents of II (30).

As described in the present invention, the reason for generating and giving semantic connection code information in the reception step of the original document is to improve the retrieval speed of the document. For example, the generation interval of the original document is about 1 minute on average, but the user access characteristics range from 1 second to 10 to 100 people when there is a lot of traffic. If you provide a large amount of information providing time, at the same time the load on the server is also great. However, in some cases, the reverse process may be possible.

The morpheme classification unit 400 combines the core morpheme information and the connection code information and stores them in the connection dictionary DB III (50), and simultaneously adds the core morpheme combination information to a specific position of the original document. At this time, the connection dictionary DB III (50) will be described in Figure 10, but is a kind of database structure for sequentially storing the results derived from the inference engine I (200) and II (300).

The database 500 is a general storage device capable of storing and managing data.

3 shows a detailed configuration diagram of the inference engine I 200 of FIG. 2, which is a data receiver 210, a morpheme comparison operator 220, a morpheme DB 10, a dictionary DB I (20), and a management tool. (40), the classification code extraction unit 230, the combination vector generation unit 240, the core morphological extraction unit 250, and the inference rule (60).

The data receiver 210 is a device for receiving each news morpheme extracted from the morpheme analysis unit 150 of FIG. 2, and the morpheme comparison operation unit 220 compares the news morpheme with the dictionary morpheme of the dictionary DBI 20. To determine the existence of the same morpheme. If the division code extracting unit 230 finds the same morpheme in the comparison operation unit 220, extracts the corresponding division code from the dictionary DB I 20, and stores the morpheme DB 10 in the case of a news morpheme without the same morpheme. It plays a role. Finally, the morpheme DB 10 is a storage device for temporarily storing news morphemes not in the dictionary DB I, and the morphemes stored in the morpheme DB are distributed to the dictionary DB I and II by the administrator using the management tool 40, respectively. Create a circulation cycle to manage the new morphemes.

The combination vector generator 240 generates a pair of combination vectors for the morphemes and the classification codes extracted from the apparatuses 220 and 230.

The core morpheme extracting unit 250 applies the inference rule 60 defined in the present invention to the morpheme combination vector, and the combination information consisting of a single or a plurality of core morphemes representing the original document and a distinguishing code according to the original document. It serves to extract. At this time, the reasoning rule is the reason for inferring the combination form of the division code among the above information. In general, in case of Korean, the subject noun existing at the beginning of a sentence becomes a core morpheme. For example, we look at inference rules that extract key morphemes from the headlines of news information. In the case of headline information, in most cases, it is generally composed of five or less morphemes when the stopwords such as postpositional words are removed. And if the classification code (1,2,3,4,5) is given by dividing all types of morphemes into five in dictionary DB I, all headline information can be represented by combining five numbers from 1 to 5. In this case, the number of all cases can be repeated combination, so 5 * 5 * 5 * 5 * 5 = 3125. In other words, the syntax characteristics of the 3125 cases can be grasped in advance to find regularity such as 12324 = 1, 13145 = 11, 35433 = 3, and then coded into a computer program as follows.

i = 1 to 5if code (i) = 1if code (i + 1) = 2if code (i + 2) = 3 ....... ..elsethen core_code = 1 ......................... endendendend

4 shows a detailed configuration diagram of the inference engine II 300 of FIG. 2, which is a data receiver 310, a division code comparison operation unit 320, a morpheme extraction unit 330, a dictionary DB II 30, It consists of the management tool 40 and the connection code extraction unit I 340.

The data receiver 310 is a device that receives the result of the core morpheme extractor 250 of FIG. 3, and the division code comparison operator 320 is a type of the discriminating code of the core morpheme combination information received by the data receiver 310. It is a device to compare. Such a distinction code may exist in 1, 2, 3, ... N branches as in the structure of the dictionary DB I (20) shown in FIG. 8, and the concatenation code type of the core morphological combination information is '1'. It transmits the result to the code extraction unit I 340, and otherwise transmits the result to the morpheme extraction unit 330.

The morpheme extraction unit 330 is a device that is executed only when the result of the division code comparison operation unit 320 is not '1'. In this case, a column of code lines matching the division code in the dictionary DB II 30 is used. It extracts the same morpheme from the dictionary morphemes in). That is, if the distinguishing code is '1' instead of '1', the same morpheme is extracted based on the column of the code line of the dictionary DB II based on 3, and the same on the column of the N if the code line is 'N'. It is responsible for extracting morphemes.

In addition, the connection code extracting unit I 340 is divided into two cases. First, when the classification code is 1 in the division code comparison operation unit 320, a dictionary morpheme matching the core morpheme at that time is extracted from the dictionary DB II. Later, it extracts all the stems of the lower level as connection code information. On the other hand, if the discriminating code is not 1, the morpheme information of the lower level including the 'item name' and the 'business code' including the same morpheme extracted by the morpheme extracting unit 330 is extracted from the dictionary DB II. It serves to extract.

For example, if the data receiver 310 receives a key morpheme (3, cooking oil), the classification code comparison operation unit 320, since the classification code is '3', the morpheme extraction unit 330 is a dictionary DB II (30) Extract the same stem from the column corresponding to line 3 of the code in. As a result, three morphemes of 'edible oil' are searched, and accordingly, the connection code extracting unit I 340 extracts item morphemes and industry codes including the morphemes, and the results are as follows.

(Samyang Corporation; 1), (Cheil Jedang; 1), (Oriental flow rate; 2)

Then, using the result, the following connection code information can be obtained from the connection code extraction unit I 340.

(1) (Samyangsa; 1)-(1, Samyangsa; food; raw sugar, corn; sugar, cooking oil, flour, ..)

(2) (Cheil Jedang; 1)-(1, Cheil Jedang; Food; Corn, Oil, Edible Oil, Flour, ..)

(3) (east oil flow rate; 2)-(1, east oil flow rate; food; corn, soybean; cooking oil, sesame oil, ..)

FIG. 5 is an exemplary diagram illustrating a step-by-step procedure for searching for a relevant document of meaning-centeredness instead of a word-centered word for a user-entered search term, specifically, a user's search term input or selection step (S1010), the input A step of receiving a search word (S1020), a step of determining whether a search word exists in the dictionary DBⅠ (S1030), a step of extracting the classification code of the search word from the dictionary DBⅠ (S1040), the classification code and morpheme of the news selected in the S1010 Database Extracting at step S1050, extracting the latest connection code information matching the division code and the search term (or morpheme) from the connection dictionary DB III using the inference engine III (S1060), matching the connection code information Searching for a document in the database (S1070), and displays the search results in a predetermined screen frame for each division code and then ends (S1080 ~ S1). 090).

In this case, S1010 is a step in which a user inputs a specific search word or selects a headline of a specific news through a service providing screen, and S1020 is a step of receiving user input contents through a web server. In operation S1030, it is determined whether the search word input by the user exists in the dictionary DB I (20), and if not present, feeds back the initial input step (S1010), and if present, transmits the result to the next step. The step S1040 is a step of comparing and extracting the classification code of the search word input by the user in the dictionary DB I (20).

In step S1010, when the user selects a headline of a specific news, the corresponding news is extracted from the database 500 through the web server, and then the information of S40 added to the original news through S70 of FIG. 1 in the news information. This step extracts the identification code and key morphemes.

The step S1060 is a step of extracting connection code information matching the division code and the search word (or key morpheme) from the information stored in the connection dictionary DB III 50 through the inference engine III in S65 of FIG. 1. The reason why the most recent information is extracted from the contents of the connected dictionary DB III is because the dictionary DB I and II are continuously updated through the management tool 40, and thus the connected dictionary DB III is updated at the same time.

Further, the S1070 is a step of searching document information corresponding to the connection code information extracted in the previous step in the database. In this case, the criterion of the database search is the connection code information of the S1060, even though the user inputs or selects only a specific keyword in S1010. In addition to documents containing an input search word, documents that are determined to be semantically related to documents not included can be searched.

In addition, the steps S1080 and S1090 display and end the result searched in the previous step in the optimum screen frame preset for each division code. To pursue optimization.

FIG. 6 is a block diagram of an apparatus for automatically searching a semantic-oriented related document for the input search term described in FIG. 5. The user system 600, the communication network 650, the web server 700, and the search term analyzer 800 are described. ), Dictionary DB I (20), Management Tool (40), Reasoning Engine III (900), Linked Dictionary DB III (50), and Database (500).

The user system 600 corresponds to any device capable of wired / wireless computer communication such as a PC, a portable terminal, etc. and capable of activating a service screen of an information provider, and the communication network 650 means a wired / wireless computer communication network including the Internet. . The web server 700 is a general operating system capable of receiving and receiving user systems and information, further storing basic information on users, and receiving and receiving internal databases and information.

The search word analyzing unit 800 is an apparatus for extracting the same morpheme (or core morpheme) and a corresponding classification code from the dictionary DB I 20 and the database 500 with respect to the input or selected content by the user. And the reasoning engine III (900) will be described in detail later in Figure 7, but using the information on the morpheme and the identification code extracts the connection code information matching in the connection dictionary DB III, and corresponds to the connection code information in the database It is a device to write SQL statements to search documents.

The management tool 40, the connection dictionary DB III 50, and the database 500 are devices that perform the same functions as previously described, and thus detailed description thereof will be omitted.

7 is a block diagram of the inference engine III proposed in the present invention, a data receiver 910, a connection code extraction unit II (920), a connection dictionary DB III (50), SQL generation unit 930, data transmission unit ( 940).

The data receiver 910 is a device that receives the results of the search word analyzer 800 of FIG. 6, and the connection code extractor II 920 is an input search word (or key morpheme) received from the data receiver 910. The device compares the classification code information with the information of the connection dictionary DB III (50) and extracts the corresponding connection code information. Here, the connection code extraction unit I 340 of FIG. 4 and the connection code extraction unit II 920 of FIG. 7 are compared as follows.

The connection code extracting unit I 340 is a device that first generates the semantic center connection code information in the dictionary DB II 30 by using the core morpheme and division code information generated by the core morph extracting unit 250 of FIG. 3. The connection code extracting unit II 920 is a morpheme extracted from the search word analyzing unit 800 of FIG. 6 based on the information already generated by the connection code extracting unit I 340 and stored in the connection dictionary DB III 50. And extracting semantic connection code information corresponding to the classification code information.

The connection dictionary DB III (50) is the same as the previous description, and detailed description thereof will be omitted in FIG. 10. The SQL generation unit 930 automatically generates a database query statement, that is, an SQL statement necessary for retrieving all document information corresponding to the connection code information extracted from the connection code extracting unit II (920) in the database (500). to be.

In addition, the data transmission unit 940 is a device for transmitting the SQL statement written by the SQL generation unit 930 to the Database (500).

Figure 8 is a simple example of the dictionary DB I (20) according to the present invention, the first classification of all types of news stemming presented in the news information according to the classification criteria defined by the administrator, the database to which the corresponding classification code is assigned to each type As a device of structure, the type of classification code may be defined variously as 1, 2, 3, ..., N according to the classification criteria. In particular, in the present invention, the classification criteria of the morphemes are set for the stock market, and the morphemes corresponding to the individual items and their synonyms are set to '1', the business morphemes to '2', the business morphemes to 3, and the like. The larger the value is, the higher level the set can be. In addition, the configuration method of dictionary DB I may be based on a row basis as shown in this example. Alternatively, the dictionary DB I may be based on a column basis, but in order to satisfy the consistency of results and the one-to-one correspondence of information, the dictionary DB The morpheme and information in II should be identical. In addition, the dictionary DB I must be equipped with a management tool as an accessory device in that the management tool 40 for easily and simply adding and changing a newly generated morpheme can be provided at the same time to improve the accuracy of the search result. Although the present invention does not provide a detailed illustration of such a management tool 40, it should be able to include the contents of the pre-DB I (20) and II (30), and the stem stem DB (10) in consideration of the management efficiency. do.

9 is a simple illustration of the dictionary DB II (30) according to the present invention, in the present invention, based on the individual items of the stock market, the classification criteria that pre-define all the morphemes related to the item from the business point of view (belonging to Industry, raw materials, products, affiliated companies, etc.) and assigned the corresponding code information. In this case, the code information has been described in connection code extraction unit I (340) of Figure 4 and connection code extraction unit II (930) of Figure 7 for the accuracy of information retrieval, in the present invention 2 Eggplant types were included. Here, as described in the dictionary DB I (20), the classification code is assigned to '1' in common because it corresponds to an individual company, and in the case of the industry code, all the types of industries included in the company are simply numbered sequentially.

In order to shorten the processing time of the morpheme extracting unit 330 in the inference engine II 300, the code line should be included in the first line of the DB II. This is because the retrieval time is reduced by utilizing such code line information when extracting the association information from the dictionary DB II using the classification code of the core morpheme combination extracted in S40 of FIG. 1. To this end, the code type included in the code line must be prepared in the same manner as the classification criteria defined in the dictionary DB I 230. Similarly, in order to satisfy the consistency of the results and the one-to-one correspondence of the information, the morphological information in the prior DB I and II should be identical. That is, the morphemes in the dictionary DB I must also exist in the dictionary DB II and vice versa.

10 is a simplified illustration of the connection dictionary DB III (50) according to the present invention, which is a kind of database structure to sequentially store the results of S40 and S60 of FIG. Here, the same information as that of the newly generated information of S40 among the past information of the connection dictionary DB III may be overwritten.

The present invention constituted by the above apparatus and method can expect the following effects.

First, in the existing retrieval apparatus, the administrator directly performed the classification and indexing process for the input document. Therefore, the limitation and error of the classification were large. However, in the present invention, the semantic-oriented related information extraction for the document is performed using an intelligent inference engine. Link code indexing can be done quickly and accurately, and the search speed in response to user requests can be improved.

Second, although the existing search apparatus is a word matching method for searching only documents including a search word input by a user, the search error rate is high. However, in the present invention, by using semantic connection code index information automatically provided through an inference engine, Documents inferred with semantically relevant but not semantically related documents are presented as valid search results, thereby increasing the variety of search results and at the same time solving the problem of search error rate.

Claims (7)

A first step of extracting news morphemes of the original document by receiving the original document of the news generated externally or internally, and performing morphological analysis and stopword processing; The combination vector consisting of the same morpheme and the corresponding classification code is extracted by comparing the dictionary DB I with the news morpheme using the inference engine Ⅰ, and the inference rules are applied to the combination vector to combine one or more core morphemes. A second step of extracting information; A third step of automatically extracting semantic connection code information according to the core morphological combination information from the dictionary DB II using the inference engine II; And The semantic meaning of the search term is characterized by combining the core morphological combination information and the connection code information and storing it in the linked dictionary DB III, adding the core morphological combination information to the original document, and storing and terminating it in the database. Information retrieval method using relevance information. Data receiving unit for receiving the original document of the news generated externally or internally; A morpheme analysis unit extracting news morphemes after performing morphological analysis and stopword processing on the original document; An inference engine I for comparing the news morphemes with the dictionary morphemes of the dictionary DB I, extracting the same morphemes and classification codes, generating a combination vector thereof, and then extracting a single or a plurality of key morpheme combinations for the combination vectors; Inference engine II for automatically generating semantic connection code information in dictionary DB II for the core morpheme combination; A morpheme classification unit which combines the core morpheme combination information and the connection code information and stores them in the connection dictionary DB III, and adds the core morpheme combination information to the original document and stores the information in the database; And And a database for storing the original document and key morpheme combination information. The inference engine I of claim 2, Data receiving unit for receiving the news morpheme extracted from the morpheme analysis unit; A morpheme comparison operation unit for extracting the same morpheme by comparing the news morpheme with the dictionary morpheme of the dictionary DB I; A division code extraction unit for extracting a division code of the same morpheme from dictionary DB I; A combination vector generation unit combining the same morpheme and a classification code; And An information retrieval apparatus using semantic relevance information of a search word further comprising a core morphological extracting unit extracting single or plural pieces of key morphological information by applying an inference rule to the combination vector. The inference engine II of claim 2, A data receiver receiving the core morpheme combination information; A division code comparison operation unit for determining a classification code type of the core morpheme information; A morpheme extraction unit for extracting the same morphemes for the core morphemes from the dictionary DB II according to the type of the classification code; And An information retrieval apparatus using semantic relevance information of the search word, further comprising a connection code extracting unit I for automatically extracting semantic connection code information of the same morpheme from the dictionary DB II. A first step in which a user inputs a search word or selects specific document information on a service screen provided through a wired or wireless communication network; When the document information is selected in the above step, the classification code and key morpheme information are extracted from the original document stored in the database, and when the user inputs a search word, the same morpheme and the classification code are extracted from the dictionary DB I for the corresponding search word. Second step; A third step of extracting the latest connection code information corresponding to the classification code and the morpheme from the connection dictionary DB III using the inference engine III; A fourth step of extracting document information corresponding to the connection code information from a database; And And a fifth step of displaying the database search result on a predetermined user screen frame for each division code and ending the same. A user system capable of inputting a specific search word or selecting specific document information on a service providing screen; Wired and wireless computer networks; Web server; A search word analysis unit for extracting the corresponding classification code from the dictionary DB I for the user input search word and extracting the classification code and key morpheme information of the corresponding document from the database for the user selected document information; An inference engine III for extracting connection code information corresponding to the classification code and morpheme information from a connection dictionary DB III; And An information retrieval apparatus using semantic correlation information of a search term, further comprising a database storing an original document and key morpheme combination information. The method of claim 6, wherein the inference engine III, A connection code extraction unit Ⅱ for automatically extracting semantic connection code information by comparing the classification code and morpheme information of the search word analysis unit with the information of the connection dictionary DB III; An SQL generator for automatically generating an SQL statement with respect to the connection code information; And An information retrieval apparatus using semantic relevance information of a search word, further comprising a data transmission unit.