KR100818742B1 - Document retrieval method using relevance of index word's location information in document - Google Patents

Abstract

The present invention relates to a document retrieval method using the relevance of the index word in the document position information, generating a document information for the document stored in the database in the search engine, generating a word information, word-document information Generating a query word input to the search engine as the index word in the generated word-document information, and when the query word is two or more words using the location information of the word in the text, each query word is clustered in the document. In the document retrieval method using the location information in the document of the existing index word to efficiently search for the document containing the query in the case of forming a high relevance, up to three position information of the word in the text The number of locations where words exist and the location information of the words are limited to meaningful ones. Generating a location information structure to be generated, and obtaining a relationship by using the location information structure in the form of (word identification x word frequency in a document) + weight + cluster degree according to an index field in a document, and in the step And selecting a document having a high relevance value as a document to be searched by a user input search word based on the calculated relevance. A document search method.

According to the present invention, by limiting the position information of the words in the text to only three meaningful, by generating a position information structure consisting of the number of positions where the words exist, the position information of the words to calculate the relevance according to the document As the existing document search method, the disk is inevitably generated in the process of forming a separate index file according to the position of the index word in the variable document and using the position and the key value. It is possible to prevent the exponential deterioration of the search speed due to the IO, thereby enabling the rapid search speed and accurate output of the search results without reducing the recall.

Description

Search method using the relevance of index word position information in the document {Search Methode Using Word Position Data}

The present invention relates to a document retrieval method using the relevance of the index word in the document position information, generating a document information for the document stored in the database in the search engine, generating a word information, word-document information Generating a query word input to the search engine as the index word in the generated word-document information, and if the query word is two or more words using the location information of the word in the text, each query word is clustered in the document. In the document retrieval method using the location information in the document of the existing index word to efficiently search for the document containing the query in the case of forming a high relevance, up to three position information of the word in the text The number of locations where words exist and the location information of the words are limited to meaningful ones. Generating a location information structure comprising the steps of: calculating a relationship by using the location information structure (the word identity × word frequency in a document) + weight according to the index field in the document + the cluster degree; Selecting a document having a high relevance value as a document to be searched by a user input search word based on the relevance calculated in the step; It relates to a document retrieval method using the relevance to the location information in the document of the index word characterized in that it further comprises.

In general, documents that are search targets of a search engine that input one or more words (search terms, index terms) in an Internet portal site or a computer system to search for a document having related contents are presented in natural language form as shown in FIG. It is expressed as a series of sentences, and the sentence is ultimately expressed as a sequence of words. The search engine is a system that expresses (indexes) this document around words and searches for documents using words. It considers three objects: documents, words and word-documents as main elements.

Among the elements of consideration of the search engine, "document" is a minimum unit representing related information. In a search engine, a part belonging to document information such as a large and small classification name, a created date, a title, and an author, etc. It can have attributes such as text that represents information. Within the search engine, documents are managed as 4 byte integers (Document ID: Document ID (DocID)) rather than names handled outside the search engine, such as the document name. Managed. The files related to this are associative tables of external IDs of documents and internally managed IDs. External document interworking key tables (Drid: Document Relation ID) and document related information of the types shown in FIGS. 4, which stores the key value, the DocID as the key value, and stores document storage information (Dmst: Document Master Information) in the form shown in FIGS. 3 and 3, and field information of the document in a stream form. Document header information (Dinf: Document Information), document source information (Dorg: Original Document) of the form shown in FIG. 5 for storing the original body of the document in the form of a stream, and ASCII text body of the document in the form of a stream (index There is document text information (Dtxt: Text Document) of the form shown in FIG.

TDridBK TLONG ExtBK expansion Bucket  number TSINT SlotCnt Slot count TDridSL aDridSL [ MAX _ DRID _ SL ] External Document Linkage Key Record

TDridSL  ( External Document Linkage Key Record ) char Drid [MAX_DRID_LEN] DRID (Document Relation ID): Document Association ID TLONG DocID Document ID, 2 ³¹ = 2,147,483,648

TDmst char InfFileNo Dinf file number TLONG Infoff Dinf start position TSINT InfLen Dinf length char OrgFileNo Dorg file number TLONG OrgOff Dorg start location TLONG OrgLen Dorg length char TxtFileNo Dtxt file number TLONG TxtOff Dtxt starting location TLONG TxtLen Dtxt length

On the other hand, the second element of the search engine consideration factor "word" is an entity created by the search engine during the indexing process, the number of documents in which the word appears in the indexed document set, the accumulated index frequency, the accumulated index field It has attributes such as information. Words are also managed as 4 byte integers (WordIDs) inside the search engine, and Word: WordID is managed as a separate table for interworking with external search engines such as query words. Related files include Word and internal WordID association table. External Word Interworking Table (Wrid: Word Relation ID) and IDL (Inverted Document List) as shown in FIGS. 7 and 4 and 5. ), There is an index word information (Wmst: Word Master Information) table having the form shown in Figs.

TWridBK TLONG ExtBK Expansion bucket number TSINT SlotCnt Number of slots recorded TWridSL aWridSL (External Word Association Record) RID_SL]

TWridSL Char Wrid [MAX_WRID_LEN] word TLONG WordID Word ID, 2³¹ = 2,147,483,648

TWmst TLONG IndexCount Cumulative Index Frequency TDWORD IndexField Accumulated Index Fields TLONG IndexDoc Cumulative Index Documents TLONG TmpOff Tmpstart position of "word-document" TLONG ExtOff Ext start location of "word-document" char ExtFileNo Extfile number for "word-document"

Next, "word-document", which is the last element of the search engine's consideration factor, is generated by the search engine during the indexing process, is a property of a word in the document, and is indexed in the document. And the location of words in the text and the first block (in 256 bytes). This information is managed as a list with WordID as a key value, and how the information is managed and used can determine the performance of the search engine. Related files include an Idl.tmp file having the form shown in FIG. 9 as a primary buffering file for removing a search speed degradation due to fragmentation of an Inverted Document List Bucket (IdlBK). In other words, there is an Idl.ext file that stores an Inverted Document List Bucket (IdlBK) structure in which fragmentation is removed and has the structure of FIG. 10 and Tables 7-9.

TIdHD char ExtFileNo Ext file times TLONG ExtOff Ext starting position TSINT SlotCnt Number of slots recorded in ExtBK TLONG SrtDocID Initial DocID TLONG EndDocID Last DocID

TIdlBK TIdlSL aIdlSL [MAX_IDL_SL]

TIdlSL TLONG DocID Document ID with word indexed TDWORD IndexField The field where the word is indexed TWORD IndexCount Frequency at which words are indexed TWORD Commentblock The block from which the word was first indexed TIdlPos IdlPos Geolocation structure

The index structure in which the above information is processed is stored and processed in the memory of the search engine in the structure of the existing general search engine as shown in FIG. 11, and in this case, the generated and processed files are as shown in FIG. 12. Is treated as a structure. In the general search method of such a conventional search engine, in order to increase the accuracy by applying the position information of the word-document information to the relevance calculation algorithm, and to maintain the reproducibility and the retrieval speed, there is a slight Put constraints The premise for improving the accuracy of the search with such a restriction is that "when the query has two or more words, if each of these queries forms a cluster within the document, the document is more relevant to the user than a document that does not form a cluster. " For example, if the search query is two words of A and B, the position of the query A and the query B close to each other to form a cluster, such as "... A..B ........" The document has a relevance to the contents of the search query entered by the user compared to the document where the locations of Query A and Query B are relatively far, such as "... A ........... B .." Highly likely document. Under this premise, the necessary information is the position information of the index word in the document represented by the position information structure at the bottom of Table 9. However, in the existing method used in the search engine, the position information could not be fixed because it is variable. As shown in Fig. 13, the location information is recorded in a separate index file, and the location and number are kept separately as key values and used for calculation. This method has the advantage of being able to use all the location information appearing in the document, but it causes a lot of disk IO and ultimately decreases the search speed exponentially, making it difficult to apply to large search systems that need to satisfy simultaneous users.

The present invention has been made to solve the above problems, the step of generating document information for the document stored in the database in the search engine, generating the word information, generating the word-document information, the generation In the case of the word-document information, when the query word input to the search engine is an index word and the query word is two words or more by using the location information of the word in the text, each of the query words forms a cluster in the document. In the document search method using the location information in the document of the existing index word to determine that the document containing the query efficiently, the word is limited by limiting the position information of the word in the text to only three meaningful Create a location information structure consisting of the number of existing locations, the location information of the word (B) obtaining a relevance using the position information structure using (word identity x word frequency in a document) + weight according to an index field in a document + cluster degree; and relevance based on the relevance calculated in the step. Providing a document retrieval method using the relevance to the location information in the document of the index word, characterized in that it further comprises the step of selecting a document having a high value as a document to be searched by the user's input search word, As in the existing document search method, exponential search speed due to the sudden disk IO inevitably occurred in the process of forming a separate index file according to the position of the index word in the variable document and using the position and the key value. Speed and accurate search results without sacrificing recall The main purpose is to enable to output.

의 공식에 의하여 단어 식별성(IDF:Inverted Document Frequency)을 구하는 단계와, 제목 필드(Title Field) > 예비 필드(Reserved Field) > 저자명 필드(Author Field)의 순과 같이 문서내 색인 필드에 따른 가중치를 구하는 단계와, 상기 위치 정보 구조체를 이용하여 각 각의 색인 단어 사이의 군집도를 수치적으로 구하는 군집도 연산(Within 연산)에 의하여 군집도를 구하는 단계와, (상기 단어 식별성×문서내 단어 빈도수)+ 상기 문서내 색인 필드에 따른 가중치+ 상기 군집도의 식으로 관련성을 구하는 단계와, 상기 단계에서 계산된 관련성에 근거하여 관련성이 높은 값을 갖는 문서를 사용자가 입력한 검색어로 검색하고자 하는 문서로 선택하는 단계를 더 포함하여 구성되는 것을 특징으로 한다.In order to achieve the above object, a document retrieval method using the relevance of the index word in the document to the location information of the present invention is a document as a minimum unit expressing related information with respect to a document stored in a database in a search engine. Indexing and generating document information, which is an object having attributes relating to a part of document information such as a large and small classification name belonging to it, a date created, a title, an author, and a text representing information, the document of the indexed document information Generating word information that is an object having attributes related to the number of documents in which the word appears in the set, the accumulated index frequency, the accumulated index field information, and the like, and the word is associated with the indexed document information and the word information. The attributes of the attribute in the document, and the number of indexed fields and frequencies in the document, Generating word-document information having attributes relating to the position and the first block (256-byte unit) information; a word in the text using the query word input to the search engine as the index word in the generated word-document information. If the query is more than two words by using the location information of, it is determined that the query is highly related when each query is formed in the document, and the location information in the document of the existing index word is searched to efficiently search for the document including the query. A document retrieval method comprising the steps of: generating a location information structure consisting of the number of locations where a word exists and the location information of the word by limiting the location information of the word in the text to only three meanings; The index frequency of the most indexed words in the entire index document using the geolocation structure N, which is the number of index documents * 3), and n, which is the index frequency of the word

Obtaining the word identity (IDF: Inverted Document Frequency) according to the formula, and weighting the index field in the document in the order of Title Field> Reserved Field> Author Field. Obtaining a cluster by a clustering operation (within operation) that numerically obtains a cluster between each index word using the location information structure; and (the word discrimination x word frequency in a document). ) + A weighting value according to the index field in the document + a document for which a user inputs a search word for a document having a high relevance value based on the relevance calculated in the step; Characterized in that further comprises the step of selecting.

The clustering operation (within operation) for obtaining the clustering degree may include: 1) initializing the clustering ratio to 0 using a distance at which each word of the position structure information clusters, and 2) position of the word 1 In a double recursive loop whose number and position of word 2 are variables, calculate the variable Within = absolute value (position of word 1 [number of position of word 1] -position of word 2 [number of position of word 2]) + 1 And cumulatively adding (30 / Within) to the cluster degree, and 3) multiplying the cluster degree calculated in step 2) by a weight.

The clustering operation (within operation) for calculating the clustering degree may include: 1) initializing the clustering ratio = 0 using a frequency at which each word of the position structure information clusters, and 2) position of the word 1 In a double circular loop whose variables are the number of words and the number of positions of word 2, the absolute value (position of word 1 [number of position of word 1] -position of word 2 [number of position of word 2]) <15 Cumulatively adding 1, and 3) multiplying the cluster degree calculated in step 2) by a weight.

According to the present invention, by limiting the position information of the words in the text to only three meaningful means to generate a position information structure consisting of the number of positions where the words exist, the position information of the words to calculate the relevance accordingly As the document search process is performed, as in the existing document search method, a sudden index that is inevitably generated in the process of forming a separate index file according to the variable index word position in the document and using the position and key value is calculated. It is possible to prevent the exponential deterioration of the search speed due to the disk IO, thereby enabling the fast search speed and accurate output of the search results without reducing the refresh rate.

Hereinafter, with reference to the accompanying drawings, a document retrieval method using the relevance of the position information in the document of the index word according to an embodiment of the present invention will be described in detail. First, in the drawings, the same components or parts are to be noted that as indicated by the same reference numerals as possible. In describing the present invention, detailed descriptions of related well-known functions or configurations are omitted in order not to obscure the subject matter of the present invention.

The present invention broadly refers to 1) generating document information, 2) generating word information, 3) generating word-document information, and 4) up to three location information of words in the text. Generating a location information structure consisting of the number of locations where the word exists, the location information structure of the word, 5) obtaining the IDF (Inverted Document Frequency), 6) in the index field in the document. Obtaining a weighting according to the method; 7) obtaining a clustering degree by a clustering operation (within operation), which numerically obtains a clustering degree between each index word using the location information structure, and 8) (the word identification × word frequency in the document) + weighting according to the index field in the document + calculating the relevance in the formula of the cluster and 9) calculating a highly relevant value based on the relevance calculated in the step And selecting a document to be searched for by a search word input by the user. The method of generating document information, word information and word-document information from steps 1) to 3) of this step to generate related files and data tables is described in the section "Technology to which the Invention belongs and Prior Art in the Field". In the same manner as the conventional search method described above, since it is a known method widely known and used in the technical field to which the present invention belongs, a detailed description thereof will be omitted.

에서 N(가장 많이 색인된 단어의 빈도수)값 설정시, 전체 문서수 * 3으로 한다는 것을 고려할 때, 한 문서 내에서 색인된 단어의 위치 정보를 3개 까지만 의미 있는 것으로 제한하고 , 그에 맞는 최적의 위치 정보 구조체를 비트 필드(Bit Field)를 이용해서 만들어 검색엔진에 응용하는 것이 바람직하다. 그 위치 정보 구조체는 기존의 위치 정보 구조체와 달리, 아래 표 10 및 도 14에 도시한 것과 같은 구조를 가지는 것이 바람직하다.Next, a step of generating a positional information structure will be described. In general, as in the method of generating the location information structure used by the existing search engine described above, the location information of the index word in the document is treated as variable and the location information is recorded in a separate index file without being fixed and the location and number In the case of calculating and storing the key as a key value, as described above, there is a problem that it is difficult to apply to a large search system that needs to satisfy simultaneous users by causing a large number of disk IOs and thus reducing the search speed exponentially. In order to solve this problem, an efficient method that can be represented by fixed-length information is needed to maintain the search speed, the average size of web documents and other text documents is less than 4K Byte, and IDF (Inverted Document Frequency) value. The formula used to calculate

In setting the value of N (the frequency of the most indexed words) in, considering that the total number of documents * 3, limit the location information of the indexed words within a document to only three meaningful ones. It is desirable to create a location information structure using a bit field and apply it to a search engine. Unlike the existing location information structure, the location information structure preferably has a structure as shown in Table 10 and FIG. 14 below.

TIdlPos (Location information structure) TDWORD PosCnt: 2 Location history count TDWORD Pos0: 10 First position TDWORD Pos1: 10 Second position TDWORD Pos2: 10 Third position

Here, the bit field of the location information structure is composed of TIdlPos = PosCnt + 0 [Position] 2, and the number of location records occupies 2Bit because only up to three pieces of location information of the indexed word are stored. On the other hand, if the bit for recording the location information of the index word in the document is 10Bit, the entire location information structure is composed of 2Bit for recording the number of three position records and three 10Bit for recording the location information for the index word in the document. The size of the entire location information structure is 2 + 3 × 10 = 32 Bit = 4 bytes. On the other hand, the number that can be represented by 10Bit representing the position information of one index word is 2 ¹⁰ = 1024, and if the average length of the index word is 3 Korean characters (6 bytes), the position information structure of the above structure is used. Location information can be recorded without error in text document of maximum 6K Byte.

Next, a step of obtaining word inverted document frequency (IDF) will be described. The step of obtaining the IDF (Inverted Document Frequency) is to obtain the index frequency (the total number of index documents * 3) of the word with the most index in the entire index document set using the location information structure. When the index frequency of the word n is calculated by the following formula (1).

The word discrimination value converges in a range of values from 1 to 13, and the higher the identity, the larger the value.

Next, a step of obtaining weights according to index fields in the document will be described. The step of obtaining the weight acts as an important factor in calculating the relevance when the query word is a single word, because the weighting factor is different for each user (searcher). Therefore, it is possible to subjectively combine various orders in the order according to a user (searcher). As an example of such weighting, the most common value of weight is title field> reserved field> author field, and field and weight values according to the characteristics of the document set being indexed. Can be adjusted.

Next, the step of obtaining a cluster degree is demonstrated. The step of calculating the cluster degree is a step of numerically calculating the cluster degree between each index word. Various methods can be used to obtain such clusters. In one embodiment, the clustering operation (WithIn operation) for calculating the clustering degree focuses on the distance at which each word of the positional structure information clusters, and the distance at which each word of the positional structure information clusters. 1) initializing the cluster degree = 0, and 2) in a double circular loop where the number of positions of word 1 and the number of positions of word 2 are variables of the number of cycles, the variable Within = absolute value (word 1 of It is preferable to calculate the position-position 2 + 1), add (30 / Within) to the cluster, and 3) multiply the cluster by the weight in the cluster. If the above step is expressed by a program, it can be expressed as follows. In this case, the word 1 position and the word 2 position are variables each having the number of positions of word 1 and the number of positions of word 2 as array variables (that is, in the program, for example, "word 1 position [word 1 position number" and In the recursive loop according to the number of positions of word 1 and the number of positions of word 2, the word 1 position and the word 2 position are sequentially applied.

Relevance = 0

Number of positions of FOR word1

Number of positions of FOR word2

Within = absolute value (word 1 position [number of position of word 1] -word 2 position [number of position of word 2]) + 1

Relevance + = (30 / Within)

END FOR

END FOR

Relevance * = 10 (weighted)

On the other hand, according to the second embodiment of calculating the clustering degree, the clustering operation (within operation) for calculating the clustering degree uses a frequency in which each word is clustered by focusing on the frequency of clustering of each word of the position structure information element. 1) initializing the cluster degree = 0; and 2) in a double circular loop having the number of positions of word 1 and the number of positions of word 2 as variables, the absolute value (position of word 1-position of word 2) < In the case of 15, it may be configured to accumulate and add 1 to the cluster and 3) multiply the weight by the cluster calculated in the step 2). If the above step is expressed by a program, it can be expressed as follows. In this case, the word 1 position and the word 2 position are each a variable having the number of positions of word 1 and the number of positions of word 2 as array variables (that is, in the program, for example, "word 1 position [word 1 position number]]. "Coding", the word 1 position and the word 2 position are sequentially applied in a cyclic loop according to the number of positions of word 1 and the number of positions of word 2.

Relevance = 0

Number of positions of FOR word1

Number of positions of FOR word2

IF absolute value (word 1 position [number of position of word 1]-word 2 position [number of position of word 2]) <15 THEN

Relevance + = 1

END FOR

END FOR

Relevance * = 10 (weighted)

A first embodiment of such a clustering calculation will be briefly described taking a document having a word arrangement as shown in FIG. 15 as an example. In the document having the word arrangement as shown in FIG. 15, since the word A appears four times, the number of words A is four, and the word B appears three times, so the number of words B is three. However, in the document retrieval method using the relevance of the index word in the document position information of the present invention, since only the first three positions of each word are fixed to be valid and stored in the position information structure, the number of positions of the word B is While it is still 3, the number of positions of the word A is treated as 3 rather than 4. Therefore, in the position information structure TIdlSL of the word A having the structure as shown in Table 10, 3 is the position number PosCnt of the word A, and 100, 350,400 is the position of the word A in the position of the word A. Each of these is stored in Pos0, Pos1, and Pos2. On the other hand, if the process is applied to the word B, the position information structure TIdlSL of the word B having the structure shown in Table 10 of the word B is 3 in the position number PosCnt of the word B, and the position of the word B ), 107, 362, and 420, which are positions in the sentence of word B, are stored in Pos0, Pos1, and Pos2, respectively.

By using the position information structures of words A and B stored in this way, a cluster degree is calculated according to the first embodiment, which focuses on the distance that each word of the position structure information clusters.

First, the number of cycles of the double cyclic loop of the first embodiment is determined by the inner loop 3 cycle and the outer loop 3 cycle by the number 3 of the word A and the number 3 of the word B. Then, when performing the first outer loop for word A, the inner loop is cycled for each of positions B, 107, 362, and 420 for 100, the first position of word A, and the absolute value of the difference. Calculate and add the sum of 30 divided by the absolute value to the cluster, ie (30/7) + (30/262) + (30/320) is calculated as cluster. Next, when we perform the second outer loop for word A, we loop through the inner loop for each of the positions B in words 107, 362, and 420 for 350, the second position of word A, and return the absolute value of the difference. Calculate and add the value obtained by dividing 30 by its absolute value to the cluster. In this case, (30/243) + (30/12) + (30/70) accumulates and adds to the previous clustering value. Finally, if we perform the third outer loop for word A, we loop through the inner loop for each of the positions of word B, 107, 362, and 420, for 400, the third position of word A, and return the absolute value of the difference. Calculate the cluster by adding the values obtained by dividing 30 by the absolute value to the cluster. In this case, (30/293) + (30/38) + (30/20) accumulates and adds to the cluster calculated in the previous step. Thereafter, a value obtained by multiplying the calculated cluster degree by a weight (10 in the first embodiment) is obtained.

On the other hand, according to the second embodiment which focuses on the frequency of grouping each word of the position structure information body using the position information structures of the words A and B generated for the document shown in FIG. The clustering is calculated as follows.

First, the number of cycles of the double cyclic loop of the first embodiment is determined by the inner loop 3 cycle and the outer loop 3 cycle by the number 3 of the word A and the number 3 of the word B. Then, when performing the first outer loop for word A, the inner loop is cycled for each of positions B, 107, 362, and 420 for 100, the first position of word A, and the absolute value of the difference. If the value is smaller than 15 (that is, if the distance between the words A and B is smaller than 15), the word A and the word B are determined to be clustered with each other and increase the clustering rate by 1. In this case, since the position of the word B with the distance less than 15 with respect to the first position 100 of the word A is only when the position of the word B is 107, the cluster degree increases by one through the first inner loop. Next, when performing a second outer circular loop on word A, for a second location 350 of word A, the location of word B with a distance less than 15 is only when the position of word B is 362, so the cluster is second inside. It is increased by 1 through the loop and equals 2. Finally, when performing the third outer loop for word A, the position of word B with a distance less than 15 for the third position 400 of word A does not exist, so the clustering increases during the last outer loop. I never do that. The cluster degree 20 is obtained by multiplying the cluster degree by a weight (10 in the first embodiment).

Next, the steps for finding relevance will be described. The relevance may include obtaining word inverted document frequency (IDF), obtaining weights according to index fields in the document, and calculating word identity, weight, and clusterity, respectively. Comprehensive step is to numerically obtain the degree that the document is suitable for the user's request in relation to the search query. The relation may be calculated by defining Equation 2 below. If the relational expression is based on the importance of human thinking, the item extracted to satisfy the relation mathematically is used for the "calculated relation", so "the relationship between the calculated relation and the actual relation" It can be generally considered that "if the calculated relevance is high, the actual relevance is also high." The reason for the term “general” is that even when the same searched document is viewed, each user (searcher) often considers the importance and relevance of the search word differently.

Next, a document having a high relevance value is selected and displayed as a document to be searched by a user input search word based on the relevance calculated in the above step. A document retrieval method using relevance is constructed.

When the document retrieval method using the relevance of the index word of the present invention described above with respect to the location information in the document, the time required for the numerical calculation in the search engine is increased for the relevance calculation. However, the time required for numerical calculation in such a search engine is the time required for the calculation to be performed in the microprocessor, which is a negligible time compared to the disk IO time which is physically constrained. The time required for the numerical calculation in the search engine is very small compared to the unnecessary disk IO time reduced by the present invention, and can significantly improve the search speed of the search engine.

In the foregoing description, optimal embodiments have been disclosed in the drawings and the specification. Although specific terms have been used herein, they are used only for the purpose of describing the present invention and are not used to limit the scope of the present invention as defined in the meaning or claims. Therefore, those skilled in the art will understand that various modifications and equivalent other embodiments are possible. Therefore, the true technical protection scope of the present invention will be defined by the technical spirit of the appended claims.

1 is a schematic diagram of a document to be searched for in a document retrieval method using the relevance of the index word to the document position information according to an embodiment of the present invention.

2 is a block diagram of an external document linking key table DRid of a document retrieval method using an association of an index word with document position information according to an exemplary embodiment of the present invention.

3 is a block diagram of document storage information (Dmst: Document Master Information) of a document retrieval method using the relevance of the index word in the document position information according to an embodiment of the present invention

4 is a diagram illustrating a document header information (Dinf: Document Information) of a document retrieval method using an association of an index word with document position information according to an exemplary embodiment of the present invention.

5 is a configuration diagram of document original information (Dorg: Original Document) of a document retrieval method using a relation of index words with respect to location information in a document according to an exemplary embodiment of the present invention.

6 is a block diagram of document text information (Dtxt: Text Document) of a document retrieval method using the relevance of the index word in the document position information according to an embodiment of the present invention

7 is a block diagram of an external word association table (Wrid: Word Relation ID) of a document retrieval method using a relation of an index word to document position information according to an embodiment of the present invention.

8 is a block diagram of index word information (Wmst: Word Master Information) of a document retrieval method using the relevance of the index word in the document position information according to an embodiment of the present invention

9 is a block diagram of an Idl.tmp file of a document retrieval method using a relevance of index words in document position information according to an embodiment of the present invention

10 is a block diagram of an Idl.txt file of a document retrieval method using the relevance of the index word in the document position information according to an embodiment of the present invention

11 is a memory structure diagram of a document retrieval method using an association of an index word with document position information according to an embodiment of the present invention;

12 is a file configuration diagram of a document retrieval method using a relevance of index words in document position information according to an embodiment of the present invention;

13 is a block diagram of a location structure information element of a document retrieval method using the relevance of index words to document position information according to an embodiment of the present invention;

14 is a block diagram of a location structure information element of a document retrieval method using the relevance of index words to document position information according to an embodiment of the present invention;

15 is an exemplary view of a document to be searched for in a document searching method using a relevance of index words to document position information according to an embodiment of the present invention.

Claims (3)

The minimum unit expressing relevant information about documents stored in the database in the search engine. The text representing parts and information belonging to the document information such as the name of the document, large and small classification names, created date, title, and author. Indexing and generating document information, which is an object having an attribute relating to an object, etc .; an object having attributes related to the number of documents in which the word appears in the document set of the indexed document information, the accumulated index frequency, and the accumulated index field information. Generating the word information, and associating the indexed document information with the word information as attributes of the property of the word in the document, the indexed fields and frequency in the document, the position of the word in the text, Generating a word-document information having an attribute relating to the first block (256-byte unit) information; the generated word- In the case of the query information input to the search engine in the index information using the location information of the words in the text using the location information of the word in the text, if the query is formed by grouping each query in the document is determined to be highly relevant In a document retrieval method using information on a location of a document of an existing index word that efficiently retrieves a document including a query word, By limiting the positional information of each word in the text to only three meanings, a positional information structure consisting of the word position number which is the number of positions where each word exists and the word position which is the position information of each word is generated. Making; Using the location information structure, N, the index frequency of the word with the highest index among the entire index documents (the total number of index documents * 3), and n, the index frequency of the word to be searched, is used.

Obtaining word identity (IDF: Inverted Document Frequency) by a formula of; Obtaining weights according to index fields in the document in the order of Title Field> Reserved Field> Author Field; Obtaining a cluster by a clustering operation (within operation) that numerically obtains a cluster between each index word using the location information structure; Obtaining relevance in terms of (word discrimination x word frequency in a document) + weight according to the index field in the document + the cluster degree; Selecting a document having a high relevance value as a document to be searched by a user input search word based on the relevance calculated in the step; Document search method using the relevance to the location information in the document of the index word, characterized in that it further comprises. The method according to claim 1, The clustering operation (within operation) for obtaining the clustering degree is performed by using a distance in which each word of the position structure information clusters. 1) Initializing the cluster degree = 0 2) In a double circular loop where the number of positions of word 1 and the number of positions of word 2 are the number of cycles, the variable Within = absolute value (position of word 1 [number of position of word 1] -position of word 2 [word 2) Counting the number of positions]) + 1, and accumulating and adding (30 / Within) to the cluster; 3) multiplying the cluster degree calculated in step 2) by a weight; Document search method using the relevance to the location information in the document of the index word, characterized in that it further comprises. The method according to claim 1, The clustering operation (within operation) for obtaining the clustering degree is performed by using a frequency at which each word of the position structure information clusters. 1) Initializing the cluster degree = 0 2) In a double circular loop where the number of positions of word 1 and the number of positions of word 2 are the number of cycles, the absolute value (position of word 1 [number of position of word 1] -position of word 2 [number of position of word 1) ]) <15 cumulatively adding 1 to the cluster; 3) multiplying the cluster degree calculated in step 2) by a weight; Document search method using the relevance to the location information in the document of the index word, characterized in that it further comprises.

Images