KR20180101955A - Document scoring method and document searching system - Google Patents

Abstract

The present invention relates to a method for scoring a document and a system thereof. According to the present invention, the method comprises the following steps: extracting a query language by analyzing a query statement inputted from a user and extracting a similar term and similarity with respect to the extracted query; inquiring pre-stored user data and calculating a weight for the query language, the similar term, and document classification; reconstructing the query statement using the extracted query language, similar term, and similarity; performing document search and scoring using the reconstructed query statement; and outputting a document search result for the query statement. Therefore, the efficiency of the system may increase.

Description

[0001] DOCUMENT SCORING METHOD AND DOCUMENT SEARCHING SYSTEM [0002]

The present invention relates to a document scoring method and a document retrieval system.

Various methods such as word embedding for expressing words included in natural language in a natural language in a document retrieval system have been studied and CBOW to skip-gram methods are being actively developed in conjunction with recent deep learning.

This method has been devised as a preprocessing method for deep running, but it has a performance that can yield meaningful results just like the extraction of synonyms. However, the computational complexity of applying the word embedding itself directly to a system requiring a real-time response such as a retrieval system is large. Therefore, proper preprocessing is required for introduction into real-time systems.

In addition, although a search system can be constructed using natural language and classification for a document to which a classification is given, a user who does not understand and understand the classification system is difficult to use the classification properly in a search system, do.

Therefore, it is necessary to increase the efficiency of the search system by allowing the classification to be used internally even if the user uses only natural language.

In this way, when a query word (query statement) is input from a user, the search system is not enough to provide the user with a search result for the inputted query statement, but in this case, it is not enough to reflect the user's intention in the search result. Therefore, a proper scheme for reflecting the user's intention to the search result is required.

Korean Patent Publication No. 10-2014-0109729 (published on September 16, 2014)

It is an object of the present invention to provide a document scoring method and a document retrieval system in which, when a user makes a retrieval request using only natural language, the efficiency of the retrieval system is increased using the classification internally.

It is another object of the present invention to provide a document scoring method and a document retrieval system in which a user's intention is reflected in document retrieval.

The technical problems of the present invention are not limited to the above-mentioned technical problems, and other technical problems which are not mentioned can be understood by those skilled in the art from the following description.

According to an aspect of the present invention, there is provided a document scoring method for extracting a query term by analyzing a query input from a user, extracting similar terms and similarities with respect to the extracted query terms, Retrieving stored user data, calculating a weight for the query term, the similar term, and document classification; reconstructing a query statement using the extracted query term, the similar term, and the similarity; Performing document search and scoring using the document search result, and outputting a document search result for the inquiry statement.

The user data is characterized in that adoption or non-adoption information input from the user to the previous document search result for the query term is stored.

The step of calculating the weights may further comprise the steps of: identifying the adopted and non-adopted documents selected by the user from the user data; calculating weighted and non-adopted documents according to the frequency of the adopted and non- Is calculated.

The step of calculating the weight is characterized by calculating a weight using the query term, the similar term, and a weight function unique to each document classification.

The step of calculating the weighting value is characterized by calculating the similarity degree of the similar term as a weight value when there is no pre-stored user data.

Wherein the step of calculating the weights includes the step of arbitrarily determining that some documents having a high document score are included in the document for the query term if the pre-stored user data does not exist, the query term and the similar term, And the weight is calculated according to the frequency of the adopted document.

The reconstructing of the query statement is characterized by reconstructing the query statement by combining the query term and the similar term in a logical OR relationship.

The step of performing the document searching and scoring may include searching an index term corresponding to a query term included in the reconstructed query sentence from an index repository storing index data according to the index structure of the indexed document and the index structure, Extracting an index score given in advance corresponding to an index word, and assigning a document score to the document using the extracted index score and a weight calculated for the query term and the similar term, .

The step of calculating the document score may include adding a value obtained by multiplying the weighted value calculated for the query term and the similar term by the index score to the document score of the document.

In addition, the document scoring method according to an embodiment of the present invention may include the steps of: receiving information on adoption or non-adoption from the user on the document search result; identifying information on adoption or non- Storing the user data in the user data together with the information, and adjusting the weight for the query term, the similar term, and the document classification using the stored user data.

The step of adjusting the weights may include adjusting the parameter values of the weight function for the query term, the similar term, and the document classification.

The step of extracting the similar terms and similarities is characterized by extracting similar terms in which the degree of similarity to the query term among the terms included in the document is equal to or greater than a predetermined threshold value.

The step of outputting the document search result may include sorting and outputting documents in the document search result according to a document search condition for the query.

The outputting of the document search result is characterized in that each document is sorted according to the size of the document score given to the documents in the document search result.

According to another aspect of the present invention, there is provided a document retrieval system for analyzing a query input from a user to extract a query term, similar terms and similarities to the query term, An operation module for calculating the weight of the extracted query word, the similar term, and the document classification, and reconstructing the query sentence using the extracted query term, the similar term, and the similarity, And a search module for performing document search and scoring using the search module.

And the search module performs scoring on the search documents based on the weights.

According to the present invention, when the user makes a search request using only natural language, the efficiency of the search system is increased by using the classification internally. In addition, according to the present invention, there is an advantage that a user's intention can be reflected in a document search result.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram of a document search system according to an embodiment of the present invention; FIG.
FIG. 2 is a detailed block diagram of the operation module of FIG. 1. Referring to FIG.
3 is a graph illustrating a weight function graph according to an embodiment of the present invention.
4 is a detailed block diagram of the search module of FIG.
5 and 6 are diagrams illustrating an operation flow for a document scoring method according to an embodiment of the present invention.
Figure 7 is a diagram of a computing system in which a method according to one embodiment of the invention is implemented.

Hereinafter, some embodiments of the present invention will be described in detail with reference to exemplary drawings. It should be noted that, in adding reference numerals to the constituent elements of the drawings, the same constituent elements are denoted by the same reference symbols as possible even if they are shown in different drawings. In the following description of the embodiments of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the difference that the embodiments of the present invention are not conclusive.

In describing the components of the embodiment of the present invention, terms such as first, second, A, B, (a), and (b) may be used. These terms are intended to distinguish the constituent elements from other constituent elements, and the terms do not limit the nature, order or order of the constituent elements. Also, unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the relevant art and are to be interpreted in an ideal or overly formal sense unless explicitly defined in the present application Do not.

The document search system according to an embodiment of the present invention performs document search and scoring using a query input from a user and provides a document search result to a user. The document search system includes a user's intention (e.g., adoption or non-adoption) ) To the document search and scoring.

The document searching system according to the embodiment of the present invention may be any system that searches one or more documents according to a query and provides the documents to a user. For example, the document search system of the present invention may be a patent search system. Of course, it is not limited thereto.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram of a document search system according to an embodiment of the present invention; FIG.

Referring to FIG. 1, the document search system 100 may include a calculation module 110, a search module 130, and a search DB 150.

The operation module 110 extracts a query term and similar term by analyzing the inputted query sentence when a query is input from a user, calculates a weight for the extracted query term, similar term, classification of the document, can do. In addition, the operation module 110 performs a function of providing a document retrieval result retrieved by the retrieval module 130 to a user. An embodiment of the detailed configuration of the computing module 110 will be described in more detail with reference to FIG.

The search module 130 performs document search and scoring using a query input by the user or a query sentence reconstructed by the operation module 110. [ An embodiment of the detailed configuration of the search module 130 will be described in more detail with reference to FIG.

The search DB 150 may store data and algorithms required to operate the document search system 100 according to an embodiment of the present invention.

The search DB 150 may include a document repository 151, an index repository 153, a term vector aggregation repository 155, a similar term repository 157, and a user data repository 159.

The document repository 151 stores one or more document data. Each document data stored in the document repository 151 may be assigned a classification value and stored together with the corresponding document data.

An embodiment of the classification value corresponding to each document data can be expressed as follows.

The index store 153 may store the index data stored according to the index structure and the index structure.

 The index store 153 may store the index score corresponding to each index together. At this time, the index score may be given to the index structure in the form of a field weight, and may be given to the index data itself. In addition, the index score may be given a different weight for each position (field) from which the corresponding index is extracted.

As an example, in the patent search system, the indexer score of the index A (title) is 1.0, the indexer score of the index B (abstract) and the indexer score of the index C (claim) are 0.8 and the indexer score of the index C Respectively.

Here, the indexer score can be given as a value of tf-idf.

The term vector store 155 may store terms (word, phrase) and their vectors.

Word embedding of terms can be performed to reduce dimension reduction and hence computational complexity, and also to extract similar terms through inter-term computation (e.g., cosine similarity). Here, the vectorization of terms can be performed on the premise that the words used in a similar context have a distributional hypothesis that shares a similar semantic.

Vectorization of terms can be performed using frequency-based methods such as LSA, PCA, LDA, etc., CBOW for predicting words when given contexts, and Skip-grams for predicting contexts when words are given .

When the similarity term and similarity extraction event are generated by the operation module 110, the term vector store may extract similar terms and similarities corresponding to the query terms and provide them to the operation module 110.

Extraction of similar terms is performed by extracting texts according to documents from the document repository 151 and performing a preprocessing process of removing tables, formulas, and formulas, and then performing morphological analysis, morphological classification, sentence separation, dependency analysis, Contexts and predictions can be constructed and learned in a skip-gram manner using recognized phrases and / or words.

The similar term storage 157 may store similar terms extracted from corresponding terms of the query, query terms, and similarity information between similar terms among the terms included in the document. At this time, the similar term storage 157 may store a similar term in which the similarity degree of the extracted similar term is greater than or equal to a predetermined threshold value. Here, the predetermined threshold is applied to all terms in the same manner, but it is preferable to set the threshold value to a low value such that a sufficient number of similar words are secured, but to use a value higher than the threshold value in extracting the similar term. However, the present invention is not limited thereto.

The following embodiment shows an example in which similar terms and similarities for a query term are constructed by the JSON method (including an arrangement format in which the similarity is sorted in descending order).

The following embodiment shows an example in which similar terms and similarities for a query term are constructed in the term table of the RDB scheme or the similar_term table format.

<term table>

<similar_term table>

The user data store 159 may store adoption or non-adoption information input from the user to the documents retrieved using the query term included in the query of the user.

At this time, the adoption or non-adoption information input from the user may be stored together with the identification number and classification information of the document.

Accordingly, the operation module 110 may inquire information stored in the user data storage 159 and reflect the calculated information in terms of the query terms and / or the classification weight.

2 is a diagram illustrating a configuration of a document scoring apparatus according to an embodiment of the present invention.

FIG. 2 is a detailed block diagram of the operation module of FIG. 1. Referring to FIG.

2, a calculation module 110 according to an embodiment of the present invention includes a query analyzing unit 111, a weight calculating unit 113, a query generating unit 115, and a query determining unit 117 can do.

First, the query analyzing unit 111 analyzes a query sentence input from a user and extracts a query. In addition, the query term analyzing unit 111 extracts similar terms and similarities for the query term from the term vector aggregation storage 155.

For example, if the query is 'hybrid car', the query analyzing unit 111 determines 'vehicle' corresponding to the query term 'car' included in the query. 'motor', 'automobile', and so on.

The query analyzing unit 111 stores information on the extracted similar terms and similarities in the similar term storage 157.

The weight calculation unit 113 calculates a weight for the query term and / or similar term extracted by the query term analysis unit 111. [ In addition, the weight calculation unit 113 calculates a weight for each document classification.

At this time, the weight calculation unit 113 can calculate a weight for each query term (including similar terms) and for each document classification using the weight function f (x). Here, the weight function f (x) is referred to the following formula (1).

In Equation (1), f (x) is the weight, x is the frequency of adopted and non-adopted documents, L is the maximum weight, and k is the slope. Here, the frequency x may be a value obtained by subtracting the number of non-adopted documents from the number of adopted documents.

For example, if there are 4 adopted documents from the user and 0 non-adopted documents for the query term 'car' in the query input from the user, the weight of the query term 'car' may be f (4-0) . The weight of the similar term 'vehicle' can be f (1-2) when there is one adopted document from the user and two non-adopted documents for the similar term 'vehicle' of the query term 'car'. On the other hand, if two adopted documents from the user and two non-adopted documents are related to 'Class A' in the classification of the document, the weight of 'Class A' may be f (2-2). In addition, when there are 0 adopted documents from the user and 5 non-adopted documents with respect to 'Class B' in the classification of the document, the weight of 'Class B' may be f (-5).

Also, the slope k can be adjusted so that the similarity of the adopted document is greater than the similarity of the non-adopted document.

If the maximum similarity of the non-adopted documents is larger than the minimum similarity of the adopted documents, the operation module 110 may increase the inclination of the weight function for the classification assigned to the non-adopted documents, It is possible to reduce the inclination of the weight function. In this case, if the error situation is not solved by the logistic gradient for classification, the operation module 110 may adjust the inclination of the weight function for the query term.

The graph of the weight function f (x) thus calculated can be represented as shown in FIG.

Since weights are determined by the frequency of the adopted and non-adopted documents, weights may not be calculated in the initial searches for which the adopted and non-adopted documents are absent.

In this case, the weight calculation unit 113 can arbitrarily set the weight initial value for the query term and / or the similar term.

For example, the weight calculation unit 113 may set the weight initial value to 1, which is a fixed value, for the query term in the initial search. In addition, the weight calculation unit 113 can set the weight for the similar term at the time of the first search to the degree of similarity between the query term and the similar term. Assuming that the similarity of the similar term 'motor' to the query term 'car' is 0.381901, the initial weight value of the similar term 'motor' can be set to 0.381901.

As another example, the weight calculation unit 113 may calculate the weight initial value for the query term in the state where the document about the query term in the initial search is assumed to be the adopted document.

In this case, it can be arbitrarily decided that some documents having a high document score among the documents for the query term have been adopted.

The weighted initial values of query terms and similar terms may be updated with weights calculated by the frequency of adopted and non-adopted documents, if adoption or non-adoption of the retrieved documents is determined.

On the other hand, the weight calculation unit 113 can calculate a weight for the sub-classification on the classification hierarchy for each document classification when calculating the weight for the document classification. At this time, the weight assigned to each document classification can be applied to grant a document score to each document in the search module.

The query statement generation unit 115 reconstructs the query statement using the query term extracted by the query term analysis unit 111, similar terms, and similarity. At this time, the query generation unit 115 may combine one or more similar terms extracted corresponding to the query terms with the query terms in a logical-OR relationship.

For example, the query term is 'hybrid car', and the similar term extracted from the query term analysis unit 111 corresponding to the query term 'car' is 'vehicle'. 'motor' and 'automobile', the query term generator generates the query terms 'hybrid' and 'car', and the similar term 'vehicle'. You can combine 'motor' and 'automobile' to reconstruct the query like 'hybrid & (car + vehicle + motor + automobile)'.

Of course, the query generating unit 115 may retrieve similar terms and similarities for each query combination if the query has three or more query terms, and reconstruct the query for each combination.

When providing the document search result to the user, the query determining unit 117 may inquire whether the search result provided to the user is adopted or not. When the adoption or non-adoption information of the document search result is input from the user, the appointment decision unit 117 determines whether the adoption or non-acceptance of each of all or part of the documents included in the document search result based on the information input from the user And stores the result in the user data store 159. [

At this time, the weight calculation unit 113 calculates the weight (or the parameter value of the weight function) calculated for each query word and each document classification based on the frequencies of the adopted and non-adopted documents determined by the query determining unit 117 Can be adjusted.

4 is a detailed block diagram of the search module of FIG.

Referring to FIG. 4, the search module 130 according to an embodiment of the present invention may include a document search unit 131 and a document score calculation unit 135.

First, when a query is input from a user, the document search unit 131 may search for a document corresponding to the query through a search engine.

Also, when the query statement is reconstructed by the query statement generation unit 115 of the operation module 110, the document search unit 131 can search through the search engine for a document corresponding to the reconstructed query statement.

Here, the search engine can search documents in the index repository 153 using query terms (and similar terms). Since the information of the indexed document is stored in the index store 153 as an index structure (e.g., B-Tree), the search engine searches the index structure of the index store 153 for an index word corresponding to the query term can do.

In addition, the index repository 153 may store indexed documents together with indexed documents for the documents. Accordingly, the search engine can extract the index score information for the searched index term when searching for an index term for a query term and a similar term.

The document score calculating unit 135 searches the document searching unit 131 for a document corresponding to the query statement or the reconstructed query statement, and at the same time calculates the document score for the retrieved document.

Here, the document score calculator 135 calculates document scores for the corresponding documents using the weighted values for the respective query terms and the similar terms, and the index word scores for the index words searched for using the respective query terms and similar terms . At this time, the document score calculating unit 135 may calculate the value obtained by multiplying the weight of each query word and each similar term by the score of the indexer as the document score of the corresponding document.

The document score calculating unit 135 may calculate the document score of the document by using the weight assigned to each document classification by the weight calculating unit 113. [

The document search system 100 according to an exemplary embodiment of the present invention may be implemented as an independent hardware device, and may include at least one processor, such as a microprocessor or a general-purpose computer system, As shown in FIG.

The operation flow of the apparatus according to the present invention will be described in more detail as follows.

5 and 6 illustrate operational flows for a method according to an embodiment of the present invention.

5 is a flowchart illustrating an operation of a document scoring method according to an embodiment of the present invention.

Referring to FIG. 5, the operation module 110 of the retrieval system analyzes a query sentence by extracting a query term from the user (S130) (S140). Here, the operation module 110 may extract similar terms and similarities to the query terms based on the information stored in the similar term vector storage. In addition, the operation module 110 may store the extracted similar terms and similarities in the similar term storage 157.

If it is the first search for the query input in step S110, the search module 130 may perform a document search based on the query input in step S110 (step S121). At this time, when information such as adopted or non-adopted information is input from the user to the document retrieved in the process of 'S121', the operation module 110 transmits the input information as user data to the user data storage 159, (S125). The user data stored in the process 'S125' can be inquired by the calculation module 110 in the process 'S150' to be described later.

The operation module 110 then inquires the user data stored in the user data storage 159 (S150), and calculates a weight for each query word (including a similar term) and / or a weight for each classification of documents (S160).

Here, the user data store 159 may store adoption or non-adoption information input from the user for the previous search record.

Accordingly, the computing module 110 can identify adopted or non-adopted documents stored in the user data store 159 and can calculate weights according to the frequency of each query and / or adopted and non-adopted documents for each class have.

In the process of 'S160', the weights can be calculated as a weight function f (x) unique to each query word and / or each classification.

The operation module 110 of the retrieval system reconstructs the query using the extracted query term in step S130 and similar terms and similarities extracted in step S140 in operation S170.

At this time, the operation module 110 may combine similar terms with a query word in a logical-OR relationship.

When the query module is reconstructed by the operation module 110, the search module 130 searches the document using the reconstructed query statement and scans the search document (S180).

In addition, the search module 130 may simultaneously perform scoring of the retrieved documents while searching the documents.

Here, the search module 130 may assign a document score to the corresponding document using the calculated weight for each query word and each similar term, and the index word score for the index word searched using each query word and each similar term. At this time, the search module 130 may assign a value obtained by multiplying the weight of each query word and each similar term by the score of the indexer as the document score of the corresponding document.

When the document search and the scoring are completed by the search module 130, the calculation module 110 can output the document search result (S190). For example, the document search result may be provided to the user in an ordered state according to the size of the document score assigned to each document.

As another example, the computing module 110 may provide the user with a document between the adopted and non-adopted literature scores in the document search results. As another example, the computing module 110 may provide the user with a document whose document score is less than the adopted document score or the non-adopted document score in the document search result according to the search condition.

FIG. 6 is a flowchart illustrating an operation of adjusting a weight function applied to the process of 'S160' of FIG.

Referring to FIG. 6, when the document retrieval and scoring for the query input from the user is completed, the computation module 110 may inquire whether the search result provided to the user is adopted or not. Accordingly, the user can select the adoption or non-adoption of all or part of the documents retrieved through the search result screen. Of course, the user may choose to adopt or not to accept all or some of the documents searched, even if there is no separate query.

If the adoption or non-adoption information is input from the user to the search result provided to the user (S210), the calculation module 110 determines whether to adopt or not adopt each document based on the information input from the user (S220) And the corresponding information may be stored in the user data storage 159 (S230). Of course, there may exist documents in which the adoption or non-adoption from the user is not inputted among the search results provided to the user.

If adoption or non-adoption of all or some of the retrieved documents is determined, the computation module 110 may adjust the value of the function that computes weights according to the frequency of the adopted and non-adopted documents (S240). In this case, the weight of each query word and each classification can be updated.

Therefore, the calculation module 110 can reconstruct the query statement by applying the updated query weight for each query term and each classification at the next search.

Figure 7 is a diagram of a computing system in which a method according to one embodiment of the invention is implemented.

7, a computing system 1000 includes at least one processor 1100, a memory 1300, a user interface input device 1400, a user interface output device 1500, (1600), and a network interface (1700).

The processor 1100 may be a central processing unit (CPU) or a memory device 1300 and / or a semiconductor device that performs processing for instructions stored in the storage 1600. Memory 1300 and storage 1600 may include different types of volatile or non-volatile storage media. For example, the memory 1300 may include a ROM (Read Only Memory) and a RAM (Random Access Memory).

Thus, the steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by processor 1100, or in a combination of the two. The software module may reside in a storage medium (i.e., memory 1300 and / or storage 1600) such as a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a removable disk, You may. An exemplary storage medium is coupled to the processor 1100, which can read information from, and write information to, the storage medium. Alternatively, the storage medium may be integral to the processor 1100. [ The processor and the storage medium may reside within an application specific integrated circuit (ASIC). The ASIC may reside within the user terminal. Alternatively, the processor and the storage medium may reside as discrete components in a user terminal.

The description above is merely illustrative of the technical idea of the present invention. It will be apparent to those skilled in the art that various modifications and variations may be made without departing from the essential characteristics of the present invention.

Therefore, the embodiments disclosed in the present invention are intended to illustrate rather than limit the scope of the present invention, and the scope of the technical idea of the present invention is not limited by these embodiments. The scope of protection of the present invention should be construed according to the following claims, and all technical ideas within the scope of equivalents should be construed as falling within the scope of the present invention.

100: Document search system 110: Operation module
111: Query term analysis unit 113: Weight calculation unit
115: query statement generation unit 117: query decision unit
130: Search module 131:
135: Document score calculation unit 150: Search DB
151: Document repository 153: Index repository
155: Term vector set storage 157: Similar term storage
159: User Data Store

Claims (17)

Extracting a query term by analyzing a query input from a user, and extracting a similar term and similarity with respect to the extracted query term;
Retrieving the stored user data, calculating a weight for the query term, the similar term, and the document classification;
Reconstructing a query sentence using the extracted query term, the similar term, and the similarity;
Performing document search and scoring using the reconstructed query statement; And
And outputting a document search result for the query statement. The method according to claim 1,
Wherein the user data comprises:
Wherein adoption or non-adoption information input from a user to a previous document search result for the query term is stored. The method of claim 2,
The step of calculating the weight includes:
And a weight calculation unit for calculating a weight based on the query term and the similar term and the frequency of the adopted and non-adopted documents for each document classification from the user data by identifying the adopted and non-adopted documents selected by the user, Scoring method. The method according to claim 1,
The step of calculating the weight includes:
And calculating a weight to be calculated using the query term, the similar term, and a weight function unique to each document classification. The method according to claim 1,
The step of calculating the weight includes:
And if the pre-stored user data does not exist, the degree of similarity of the similar term is calculated as a weight value. The method according to claim 1,
The step of calculating the weight includes:
And if the pre-stored user data does not exist, a decision is made that the document about the query term is adopted, and a weight is calculated according to the frequency of the adopted document for each document classification. The method according to claim 1,
The step of reconstructing the query statement comprises:
And the query term is reconstructed by combining the query term and the similar term in a logical OR relationship. The method according to claim 1,
Wherein performing the document search and scoring comprises:
Retrieving an index word corresponding to a query term included in the reconstructed query sentence from an index repository storing index data according to the index structure of the indexed document and the index structure;
Extracting an indexer score given in advance corresponding to the searched indexer; And
A weight calculated for the query word and the similar term, and a document score for the document using the extracted index word score. The method of claim 8,
The step of calculating the document score includes:
And a value obtained by multiplying the weighted value calculated for the query term and the similar term by the score of the indexer is given as a document score of the document. The method according to claim 1,
Receiving information on adoption or non-adoption from the user on the document search result;
Storing the information on the adopted or non-adopted information in the user data together with the identification information of the corresponding document;
Further comprising adjusting a weight for the query term, the similar term, and the document classification using the stored user data. The method of claim 10,
The step of adjusting the weight includes:
Adjusting the parameter value of the weight function for the query term, the similar term, and the document classification. The method according to claim 1,
Wherein the step of extracting the similar terms and similarities comprises:
And a similar term in which the degree of similarity with the query term is equal to or greater than a predetermined threshold value is extracted from each term included in the document. Extracts a query term, similar terms and similarities with respect to the query term by analyzing the query sent from the user, calculates a weight for the extracted query term, the similar term, and the document classification by querying the previously stored user data, An operation module for reconstructing the query sentence using the extracted query term, the similar term, and the similarity; And
And a retrieval module for performing document retrieval and scoring using the reconstructed query statement,
Wherein the search module comprises:
And scoring the search documents based on the weights. 14. The method of claim 13,
Wherein the user data comprises:
Wherein adoption or non-adoption information input from a user to the previous document search result for the query term is stored. 15. The method of claim 14,
The operation module includes:
And a weight calculation unit for calculating a weight based on the query term and the similar term and the frequency of the adopted and non-adopted documents for each document classification from the user data by identifying the adopted and non-adopted documents selected by the user, Search system. 14. The method of claim 13,
The operation module includes:
And if the pre-stored user data does not exist, calculates the similarity degree of the similar term as a weight value. 14. The method of claim 13,
The operation module includes:
If the pre-stored user data does not exist, it is arbitrarily determined that both the query term and the similar term have been adopted, and the query term, the similar term, and the weight of the adopted document for each document classification are calculated Wherein the document retrieval system comprises:

Images