JP5179564B2 - Query segment position determination device - Google Patents

Description

  The present invention relates to an operation technique for a search query in a search engine or the like.

  A site called a search engine is provided on the Internet as a tool for searching for desired information from a vast amount of information.

  The search engine is provided with a search DB (Data Base) in which a search keyword and a page including the search keyword are associated with each other in advance mechanically or manually, and a search query input by a user from a terminal device is provided. Based on the search DB, the title or snippet of the hit page is displayed on the terminal device as a search result with a URL (Uniform Resource Locator) link. When the user wants to see the actual page after viewing the title, snippet, etc., the screen transitions to the linked page by selecting the title, etc.

  By the way, as a search query, in addition to designating a single word or the like, a space “□” or the like is often designated as a separator, such as “Manga □ browsing”. In this case, a page including both “manga” and “browsing” is searched (AND search). The search engine performs an AND operation on the search result based on “manga” and the search result based on “browsing” and returns the result to the user as a final search result. Separation by a separator is called segmentation, and the position of the separator on the search query is called a query segment position.

  However, the user does not always put a separator at an appropriate position, and the search result changes depending on the query segment position and the desired search result cannot be obtained even if the character string is the same except for the separator. There is. For example, if “Manga standing reading” is entered instead of “Manga standing reading” in the above example, the search engine searches for “Manga standing” and “reading” separately, and performs an AND operation on each search result. This is different from the final search result of “Manga □ Browsing”. As a more realistic example, when trying to search for a celebrity song title, etc., when the song title can be divided into two or more words, a separator is inserted between the search query and the search query. The search results will differ depending on the search query. In this case, since the title of the music is often registered in the search DB without being divided, a desired search result can be obtained when the search query is made continuously without being divided, but in the case of dividing the general term In many cases, the desired search results cannot be found due to noise.

JP 2009-301140 A

  As described above, simply performing a search based on a search query input by the user causes a decrease in search accuracy, and a countermeasure for improving the search accuracy has been demanded.

  On the other hand, Patent Document 1 discloses a technique for performing text segmentation on text data accumulated in a text database, but is not intended for a search query.

  The present invention has been proposed in view of the above-described conventional problems, and its object is to improve search accuracy by correcting the query segment position of a search query input by a user to an appropriate position. There is.

  In order to solve the above-described problems, in the present invention, as described in claim 1, a search log acquisition unit that acquires a search log, and a separator deletion that deletes a separator from a search query of the acquired search log And a query dividing means for dividing the search query from which the separator is deleted into a plurality of character strings, a first search means for searching a search database based on the divided character strings, and a search result. Query segment position determination comprising: analysis means for calculating as a score the frequency at which a single character string is connected and co-occurring with adjacent character strings; and query segment position determination means for determining a query segment position based on the calculated score The gist of the device.

  In addition, as described in claim 2, in the query segment position determination device according to claim 1, the query dividing unit can divide the search query into a plurality of character strings by morphological analysis. .

  Moreover, as described in claim 3, in the query segment position determination device according to claim 1, the query dividing unit selects a plurality of search queries by selecting a pattern having the largest number of divisions from the search log. It can be divided into character strings.

  In addition, as described in claim 4, in the query segment position determination device according to claim 1, the query dividing unit generates a plurality of search queries by generating a plurality of character string combination models from the search log. It can be divided into character strings.

  In addition, as described in claim 5, a search log acquisition unit that acquires a search log, a separator deletion unit that deletes a separator from a search query of the acquired search log, and a plurality of search queries from which the separator has been deleted Query dividing means for dividing the character string, first search means for searching the search database based on the divided character string, and one divided character string based on the search result are connected to the adjacent character string. Analyzing means for calculating the frequency of occurrence as a score, query segment position determining means for determining a query segment position based on the calculated score, and learning means for learning the pattern recognition unit based on the determined query segment position And the search query input from the user is judged by the pattern recognition unit to determine the suitability of the query segment position. It can be constructed and query correction means corrects the segment position as a search device and a second search means for searching the search database based on the modified search query.

  In addition, as described in claim 6, a receiving unit that receives a search query from a user, a separator deleting unit that deletes a separator from the received search query, and a search query from which the separator has been deleted is divided into a plurality of character strings. Query dividing means, first search means for searching a search database based on the divided character strings, and frequency of one divided character string concatenated with adjacent character strings based on the search result Analyzing means for calculating as a score, query segment position determining means for determining a query segment position based on the calculated score, query correcting means for correcting a search query based on the determined query segment position determining means, and correction And a second search means for searching the search database based on the search query. It can be.

  In addition, as described in claim 7, a search log acquisition step in which the control unit of the search device acquires a search log, and a separator deletion step in which the control unit deletes a separator from the search query of the acquired search log A query dividing step in which the control unit divides the search query from which the separator has been deleted into a plurality of character strings; and a first search step in which the control unit searches a search database based on the divided character strings; The control unit calculates, as a score, the frequency at which one divided character string is concatenated and co-occurs with an adjacent character string based on the search result; and the control unit is based on the calculated score. A query segment position determining step for determining a query segment position, and a learning step in which the control unit learns the pattern recognition unit based on the determined query segment position; The control unit determines the suitability of the query segment position by the pattern recognition unit for the search query input from the user, and corrects the query segment position to an appropriate query segment position, and the control unit performs the corrected search. It can comprise as a search control method provided with the 2nd search process of searching the search database based on a query.

  In addition, as described in claim 8, the control unit of the search device receives a search query from a user, the separator delete step of deleting the separator from the received search query, and the control A query dividing step of dividing the search query from which the separator is removed into a plurality of character strings, a first search step of searching the search database based on the divided character strings, and the control unit An analysis step of calculating, as a score, the frequency with which one divided character string is concatenated and co-occurring with an adjacent character string based on the search result, and the control unit determines a query segment position based on the calculated score A query segment position determining step for determining the query, and the control unit correcting the search query based on the determined query segment position determining means When the control unit can be configured as a search control method and a second search step of searching the search database based on the modified search query.

  According to a ninth aspect of the present invention, a computer constituting the search device includes a search log acquisition unit that acquires a search log, a separator deletion unit that deletes a separator from a search query of the acquired search log, and the separator is deleted. Query dividing means for dividing the search query into a plurality of character strings, first search means for searching the search database based on the divided character strings, and characters adjacent to one divided character string based on the search results Analyzing means for calculating the frequency of concatenated co-occurrence with a column as a score, query segment position determining means for determining a query segment position based on the calculated score, and learning of the pattern recognition unit based on the determined query segment position The learning means determines whether the query segment position is appropriate by the pattern recognition unit based on the search query input by the user. Sectional and can be configured as an appropriate query correction means corrects the query segment location, the search control program to function as the second searching means for searching the search database based on the modified search query.

  According to a tenth aspect of the present invention, a computer constituting the search device includes: a reception unit that receives a search query from a user; a separator deletion unit that deletes a separator from the received search query; and a search query from which the separator has been deleted. Query dividing means for dividing into a plurality of character strings, first search means for searching a search database based on the divided character strings, and one divided character string connected to an adjacent character string based on the search result Analyzing means for calculating the frequency of occurrence as a score, query segment position determining means for determining a query segment position based on the calculated score, query correcting means for correcting a search query based on the determined query segment position determining means, Second search means for searching the search database based on the corrected search query It can be configured as a search control program to function.

  In the present invention, it is possible to improve the search accuracy by correcting the query segment position of the search query input by the user to an appropriate position.

It is a figure which shows the structural example of the system concerning the 1st Embodiment of this invention. It is a figure which shows the example of a data structure of a search log. It is a figure which shows the data structure example of search DB. It is a figure which shows the hardware structural example of a search device. It is a flowchart which shows the process example of 1st Embodiment. It is a figure (the 1) which shows the mode of a data transition. It is a figure which shows the example of learning data. It is a figure (the 2) which shows the mode of a data transition. It is a figure which shows the structural example of the system concerning the 2nd Embodiment of this invention. It is a flowchart which shows the process example of 2nd Embodiment.

  Hereinafter, preferred embodiments of the present invention will be described.

<First Embodiment>
FIG. 1 is a diagram showing a configuration example of a system according to the first embodiment of the present invention.

In FIG. 1, a network 1 such as the Internet includes a PC (Personal Computer), a mobile phone, a PDA (Personal Digital) operated by a user.
A plurality of user terminals 2 such as Assistants) are connected. The user terminal 2 includes a general browser (Web browser) 21. The browser 21 is in accordance with HTTP (Hyper Text Transfer Protocol), which is a standard protocol of the Internet, etc.
It has a function of requesting / acquiring / displaying page data written in a language such as Text Markup Language) and transmitting form data.

  On the other hand, the network 1 is connected to a search device 3 that performs a Web search for access from the browser 21 of the user terminal 2 operated by the user and returns the search result to the browser 21 of the user terminal 2.

  The search device 3 includes a query segment position learning unit 301, a pattern recognition unit 309, a search query receiving unit 310, a search query correcting unit 311, a search unit 312 and a search result response unit 313 as functional units. The query segment position learning unit 301 includes a search log acquisition unit 302, a separator deletion unit 303, a query division unit 304, a search unit 305, a search result analysis unit 306, a query segment position determination unit 307, a learning data generation / learning request unit 308, It has.

These functional units include a CPU (Central Processing Unit) and a ROM (Read Only) of the computer constituting the search device 3.
The present invention is realized by a computer program executed on hardware resources such as a memory (RAM) and a random access memory (RAM). These functional units do not need to be arranged on a single computer, and may be distributed as necessary.

  A search log 314 and a search DB 315 are provided as databases used by the search device 3. These databases systematically hold predetermined data on a storage medium such as an HDD (Hard Disk Drive) in a computer constituting the search device 3. Note that the search log 314 and the search DB 315 need not be arranged in the search device 3 and may be arranged on another device.

  FIG. 2 is a diagram showing an example of the data structure of the search log 314. The search log 314 includes raw search log data and search log total data. The search log raw data includes items such as “search date and time” and “search query”. “Search date and time” is the date and time when the search was executed. “Search query” is a search expression used for the search. The search log total data includes items such as “search query” and “number of searches (frequency)”. “Search query” is a search expression used for the search. “Number of searches (frequency)” is the number or frequency of searches for the same search query.

  FIG. 3 is a diagram illustrating an example of the data structure of the search DB 315. The search DB 315 includes items such as “search keyword”, “page URL”, “title”, and “summary”. “Search keyword” is a character string used for the search. “Page URL” is an address of a page including a search keyword. “Title” is the title of the page. “Summary” is a character string of a part or all of the contents of the page.

  Returning to FIG. 1, the query segment position learning unit 301 of the search device 3 operates as a batch process, searches the search DB 315 based on the search log 314, determines an appropriate query segment position, and supports an SVM (Support Vector Machine). ) And the like to have the pattern recognition unit 309 learn.

  The search log acquisition unit 302 has a function of acquiring a search query from raw search log data of the search log 314.

  The separator deletion unit 303 has a function of deleting a separator such as a space from the search query acquired by the search log acquisition unit 302.

  The query dividing unit 304 has a function of dividing the search query from which the separator has been deleted by the separator deleting unit 303 into a plurality of character strings within a meaningful range.

  The search unit 305 has a function of searching the search DB 315 based on a plurality of character strings divided by the query division unit 304.

  The search result analysis unit 306 analyzes the search result obtained by the search of the search unit 305, and each of the plurality of character strings divided by the query dividing unit 304 is connected to the adjacent character string on the page. It has a function of calculating a score representing the frequency of existence (joint co-occurrence).

  The query segment position determination unit 307 has a function of determining an appropriate query segment position based on the score calculated by the search result analysis unit 306.

  The learning data generation / learning request unit 308 generates learning data indicating the characteristics of the query segment position determined by the query segment position determination unit 307 and requests the pattern recognition unit 309 to perform learning based on the learning data. It has a function to do.

  The pattern recognition unit 309 has a well-known mechanism such as SVM, and performs learning based on the feature vector and the teacher signal. At the time of pattern recognition, the input feature vector is assigned to any class (in this example, a query as a search query). It has a function of outputting a recognition result indicating whether the segment position belongs or not.

  The search query receiving unit 310 has a function of receiving a search query when a search request is issued from the browser 21 of the user terminal 2.

  The search query correcting unit 311 determines whether the query segment position of the search query received by the search query receiving unit 310 is appropriate or not by shifting the query segment position in order between all character strings by the pattern recognition unit 309. Has a function to correct.

  The search unit 312 has a function of searching the search DB 315 based on a search query (a search query corrected as necessary) via the search query correction unit 311.

  The search result response unit 313 has a function of returning the search result of the search unit 312 to the browser 21 of the requesting user terminal 2.

  FIG. 4 is a diagram illustrating a hardware configuration example of the search device 3.

  In FIG. 4, the search device 3 is connected to the CPU 32, ROM 33, RAM 34, NVRAM (Non-Volatile Random Access Memory) 35, I / F (Interface) 36, and I / F 36 connected to the system bus 31. A keyboard, mouse, monitor, CD / DVD (Compact Disk / Digital Versatile Disk) drive I / O (Input / Output Device) 37, HDD (Hard Disk Drive) 38, NIC (Network Interface Card) 39, etc. Yes. M is a medium (recording medium) such as a CD / DVD in which a program or data is stored.

  FIG. 5 is a flowchart illustrating an example of processing according to the first embodiment.

  5A, when the process is started (step S101), the search log acquisition unit 302 of the query segment position learning unit 301 of the search device 3 acquires a search query from the search log raw data of the search log 314 (step S101). S102). FIG. 6A shows an example of the acquired search query.

  5A, the separator deletion unit 303 of the query segment position learning unit 301 deletes a separator such as a space from the search query acquired by the search log acquisition unit 302 (step S103). FIG. 6B shows a state where the separator is deleted from the search query of FIG.

  Next, returning to FIG. 5A, the query dividing unit 304 of the query segment position learning unit 301 divides the search query from which the separator is deleted into a plurality of character strings within a meaningful range (step S104). As a search query dividing method, for example, there are the following three methods.

  (1) Method of dividing by morphological analysis: This method has a merit that it can be divided with a certain accuracy, but has a demerit that it is weak against unknown words that are not registered in the dictionary.

  (2) A method of selecting a pattern with the largest number of divisions from the search log: In other words, all the same search queries are extracted from the search log 314 with a separator such as a space removed, and among them, the number of divisions is the largest. Choose one. Also, search queries with a relatively low search frequency (frequency) and search queries with a significantly lower probability than the connection probability model of character strings are rejected. This technique has the merit of being strong in segmentation because it is highly likely to have data for specific proper nouns such as video game names and idol group names that are not so much in the morphological analysis dictionary, but rejected. There is a demerit that the frequency that becomes the standard to do is not clear.

  (3) A method of dividing by a character string concatenation probability model calculated from a search log: This method has a merit of dealing with unknown words, but has a demerit of including meaningless character strings.

  The selection criteria for adopting each of the above methods are (1) when it is quickly implemented and appropriate accuracy is obtained, and (2) or (3) when unknown words are divided and search accuracy is considered. ) Is good. (2) is easy to implement because it only selects an optimal division pattern. (3) is a costly implementation because it is necessary to analyze whether to divide between character strings. Although (2) and (3) are small in accuracy, there is little difference in accuracy, but in (3), the character string is divided in finer units, so that further improvement in accuracy is expected.

  FIG. 6C shows a state where the division is performed from the state of FIG.

  Next, returning to FIG. 5A, the search unit 305 of the query segment position learning unit 301 searches the search DB 315 based on the individual character strings of the search query divided by the query division unit 304 (step S105).

  Next, the search result analysis unit 306 of the query segment position learning unit 301 analyzes the search result obtained by the search of the search unit 305 and, in addition to the number of hits of each character string, a summary obtained as a search result ( In the snippet), the number of concatenated co-occurrence existing on the page in a state where each of the divided character strings is connected to the adjacent character string is counted, and a score representing the concatenated co-occurrence frequency is calculated (step) S106). As the score, a Simpson coefficient or mutual information is used.

Simpson coefficient score = (number of hits when word1 and word2 are connected) / min (number of word1 hits, number of word2 hits)
Score based on mutual information = log (P (word1, word2)) / (P (word1) P (word2))
Here, min represents the smaller one of the numerical values in parentheses. P (word1, word2) is a simultaneous distribution function of word1 and word2, P (word1) is a marginal probability distribution function of word1, and P (word2) is a marginal probability distribution function of word2.

  Instead of counting the number of connected co-occurrence from the summary (snippet) by the search result analysis unit 306, the search DB 315 may be searched with the search unit 305 connecting adjacent character strings.

  Next, the query segment position determination unit 307 of the query segment position learning unit 301 determines an appropriate query segment position based on the score calculated by the search result analysis unit 306 (step S107). In FIG. 6D, if the score of “Manga □ Standing” is lower than the score of the set threshold value, an appropriate query segment position is determined between “Manga” and “Standing” as shown in FIG. To do.

  5A, the learning data generation / learning request unit 308 of the query segment position learning unit 301 generates learning data indicating the characteristics of the query segment position determined by the query segment position determination unit 307. Based on the learning data, the pattern recognition unit 309 is requested to perform learning (step S108).

  FIG. 7 shows an example of learning data of “Manga □ Browsing”, in which the window width is “3”. That is, each line of the learning data takes the form of “evaluation value: character string feature”. “1-gram”, “2-gram”, and “3-gram” represent N-gram types and, for example, “1” as an evaluation value. “Qcount” indicates the number of searches (frequency) of the search query, and “Wcount” indicates the number of hits in the Web search. “L_All /” indicates all character strings to the left of the query segment position, and “R_All /” indicates all character strings to the right of the query segment position. “I * /” indicates a character string straddling the query segment position of interest.

  Next, returning to FIG. 5A, the process is terminated (step S109).

  Next, an actual search process using the above-described learned pattern recognition unit 309 will be described.

  In FIG.5 (b), if a process is started (step S111), the search query reception part 310 of the search device 3 will receive a search query at the time of the search request from the browser 21 of the user terminal 2 (step S112).

  Next, the search query correction unit 311 of the search device 3 determines whether the query segment position of the search query received by the search query reception unit 310 is appropriate or not by using the pattern recognition unit 309 to sequentially set the query segment position between all character strings. The search query is corrected by shifting (step S113). Candidates whose query segment positions have been changed in advance are created, and which is appropriate may be determined by pattern recognition.

  FIG. 8 shows an example of processing for correcting the query segment position from the accepted search query. “1-gram: ˜”, “2-gram: ˜”, “3-gram:” showing the characteristics of the accepted search query. Are generated while shifting the query segment position for each character string and input to the pattern recognition unit 309. In this case, since the query segment position is not appropriate, the pattern recognition result is inappropriate at an inappropriate place. Then, the query segment position is determined at a position where the pattern recognition result indicates appropriateness, and the search query is corrected.

  Next, returning to FIG. 5B, the search unit 312 of the search device 3 searches the search DB 315 with the search query after correction (may not be corrected because it is appropriate) (step S <b> 114). The result response unit 313 returns the search result of the search unit 312 to the browser 21 of the requesting user terminal 2 (step S115). Then, the process ends (step S116).

<Second Embodiment>
FIG. 9 is a diagram showing a configuration example of a system according to the second embodiment of the present invention. In the first embodiment described above, the query segment position is learned by batch processing and the received search query is corrected based on the learning result. In the second embodiment, the received search query is I am trying to correct it sequentially.

  9, the search device 3 includes a search query receiving unit 310, a separator deleting unit 303, a query dividing unit 304, a search unit 305, a search result analyzing unit 306, a query segment position determining unit 307, a search query correcting unit 311, and a search unit. 312, a search result response unit 313, and a search DB 315. Although the arrangement of each part is slightly different from that in FIG. 1, functional parts having the same names and the same reference numerals have substantially the same functions.

  FIG. 10 is a flowchart illustrating an example of processing according to the second embodiment.

  In FIG. 10, when processing is started (step S201), the search query receiving unit 310 of the search device 3 receives a search query when a search request is issued from the browser 21 of the user terminal 2 (step S202).

  Next, the separator deleting unit 303 of the search device 3 deletes a separator such as a space from the search query received by the search query receiving unit 310 (step S203).

  Next, the query dividing unit 304 of the search device 3 divides the search query from which the separator has been deleted into a plurality of character strings within a meaningful range (step S204).

  Next, the search unit 305 of the search device 3 searches the search DB 315 based on the individual character strings of the search query divided by the query division unit 304 (step S205).

  Next, the search result analysis unit 306 of the search device 3 analyzes the search result obtained by the search of the search unit 305, and in addition to the number of hits of each character string, the summary (snippet) obtained as the search result Then, the number of concatenated co-occurrence existing on the page in a state where each of the divided character strings is connected to the adjacent character string is counted, and a score representing the concatenated co-occurrence frequency is calculated (step S206). Instead of counting the number of connected co-occurrence from the summary (snippet) by the search result analysis unit 306, the search DB 315 may be searched with the search unit 305 connecting adjacent character strings.

  Next, the query segment position determination unit 307 of the search device 3 determines an appropriate query segment position based on the score calculated by the search result analysis unit 306 (step S207).

  Next, the search query correction unit 311 of the search device 3 corrects the search query based on the query segment position determined by the query segment position determination unit 307 (step S208).

  Next, the search unit 312 of the search device 3 searches the search DB 315 with the search query after correction (it may not be corrected because it is appropriate) (step S209), and the search result response unit 313 searches the search unit 312. The result is returned to the browser 21 of the requesting user terminal 2 (step S210). Then, the process ends (step S211).

<Summary>
As described above, according to the present embodiment, the search accuracy can be improved by correcting the query segment position of the search query input by the user to an appropriate position.

  The present invention has been described above by the preferred embodiments of the present invention. While the invention has been described with reference to specific embodiments, various modifications and changes may be made to the embodiments without departing from the broad spirit and scope of the invention as defined in the claims. Obviously you can. In other words, the present invention should not be construed as being limited by the details of the specific examples and the accompanying drawings.

DESCRIPTION OF SYMBOLS 1 Network 2 User terminal 21 Browser 3 Search apparatus 301 Query segment position learning part 302 Search log acquisition part 303 Separator deletion part 304 Query division part 305 Search part 306 Search result analysis part 307 Query segment position determination part 308 Learning data generation / learning request Unit 309 pattern recognition unit 310 search query reception unit 311 search query correction unit 312 search unit 313 search result response unit 314 search log 315 search DB

Claims (10)

A search log acquisition means for acquiring a search log;
Separator deletion means for deleting the separator from the search query of the acquired search log,
Query splitting means for splitting the search query with the separator removed into multiple strings,
First search means for searching a search database based on the divided character strings;
An analysis means for calculating, as a score, the frequency at which one divided character string is connected and co-occurring with an adjacent character string, based on the search results;
A query segment position determining device comprising: query segment position determining means for determining a query segment position based on the calculated score. The query segment position determination apparatus according to claim 1, wherein
The query segment position determining device divides a search query into a plurality of character strings by morphological analysis. The query segment position determination apparatus according to claim 1, wherein
The query segment position determining device divides a search query into a plurality of character strings by selecting a pattern having the largest number of divisions from a search log. The query segment position determination apparatus according to claim 1, wherein
The query segment position determining device divides a search query into a plurality of character strings by generating a plurality of character string combination models from a search log. A search log acquisition means for acquiring a search log;
Separator deletion means for deleting the separator from the search query of the acquired search log,
Query splitting means for splitting the search query with the separator removed into multiple strings,
First search means for searching a search database based on the divided character strings;
An analysis means for calculating, as a score, the frequency at which one divided character string is connected and co-occurring with an adjacent character string, based on the search results;
Query segment position determining means for determining a query segment position based on the calculated score;
Learning means for learning the pattern recognition unit based on the determined query segment position;
Query correction means for determining whether or not a query segment position is appropriate by the pattern recognition unit for a search query input by a user, and correcting the query query position to an appropriate query segment position;
A search device comprising: second search means for searching the search database based on a corrected search query. A receiving means for receiving a search query from a user;
A separator deleting means for deleting the separator from the received search query;
Query splitting means for splitting the search query with the separator removed into multiple strings,
First search means for searching a search database based on the divided character strings;
An analysis means for calculating, as a score, the frequency at which one divided character string is connected and co-occurring with an adjacent character string, based on the search results;
Query segment position determining means for determining a query segment position based on the calculated score;
Query modifying means for modifying the search query based on the determined query segment position determining means;
A search device comprising: second search means for searching the search database based on a corrected search query. A search log acquisition step in which the control unit of the search device acquires the search log;
The control unit deletes a separator from the search query of the acquired search log;
A query dividing step for dividing the search query from which the separator is deleted into a plurality of character strings;
A first search step in which the control unit searches a search database based on the divided character string;
An analysis step in which the control unit calculates, as a score, the frequency at which one divided character string is connected and co-occurs with an adjacent character string, based on a search result;
A query segment position determining step in which the control unit determines a query segment position based on the calculated score;
A learning step in which the control unit learns the pattern recognition unit based on the determined query segment position;
A query correcting step in which the control unit determines whether or not a query segment position is appropriate by the pattern recognition unit for a search query inputted by a user, and corrects the query query to an appropriate query segment position;
A search control method comprising: a second search step in which the control unit searches the search database based on a corrected search query. The control unit of the search device accepts a search query from the user;
A separator deletion step in which the control unit deletes the separator from the accepted search query;
A query dividing step for dividing the search query from which the separator is deleted into a plurality of character strings;
A first search step in which the control unit searches a search database based on the divided character string;
An analysis step in which the control unit calculates, as a score, the frequency at which one divided character string is connected and co-occurs with an adjacent character string, based on a search result;
A query segment position determining step in which the control unit determines a query segment position based on the calculated score;
A query correcting step for correcting the search query based on the determined query segment position determining means;
A search control method comprising: a second search step in which the control unit searches the search database based on a corrected search query. The computers that make up the search device
Search log acquisition means for acquiring search logs,
Separator deletion means for deleting the separator from the search query of the acquired search log,
Query splitting means that splits the search query with the separator removed into multiple strings,
First search means for searching a search database based on the divided character string;
An analysis means for calculating, as a score, a frequency at which one divided character string is connected and co-occurring with an adjacent character string, based on the search result;
Query segment position determining means for determining a query segment position based on the calculated score;
Learning means for learning the pattern recognition unit based on the determined query segment position;
Query correction means for determining whether or not a query segment position is appropriate by the pattern recognition unit for a search query input by a user, and correcting the query segment position to an appropriate query segment position;
A search control program that functions as second search means for searching the search database based on a corrected search query. The computers that make up the search device
Accepting means for receiving search queries from users,
Separator deletion means for deleting separators from received search queries,
Query splitting means that splits the search query with the separator removed into multiple strings,
First search means for searching a search database based on the divided character string;
An analysis means for calculating, as a score, a frequency at which one divided character string is connected and co-occurring with an adjacent character string, based on the search result;
Query segment position determining means for determining a query segment position based on the calculated score;
Query correcting means for correcting the search query based on the determined query segment position determining means;
A search control program that functions as second search means for searching the search database based on a corrected search query.

Images