JP2006099427A - Full-text retrieval system and method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To attain the high speed of full text retrieval while securing retrieving precision to some extents. <P>SOLUTION: A retrieval result number approximation level deciding part 183 decides the approximation level between the hit number of items of primary retrieval by an N gram index to be obtained by an N gram retrieval engine 13, and the hit number of items of morpheme retrieval by a morpheme index to be obtained by a morpheme retrieval engine 16. A full text retrieval execution control mechanism 18 controls the N gram retrieval engine 13 so that secondary retrieval is omitted by the N gram index when it is decided that the hit number of items of the primary retrieval approximates to the hit number of items of the morpheme retrieval by a retrieval result number approximation level deciding part 183. In this case, the full text retrieval execution control mechanism 18 adopts the result of the primary retrieval or the result of the morpheme retrieval as a retrieval result. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a full-text search system and method suitable for searching a document that matches a specified search condition at a high speed from a large amount of digitized document information using a full-text search technique.

  Various search systems that search for documents that meet specified search conditions from a large amount of digitized document information have been developed. As a typical search method of document search applied in this type of search system, a search method using an N-gram index used for full-text search (full-text search) or natural language search (concept search) A search method based on morphemes used in Japanese is known. The outline of these search methods is as follows.

[Search method using N-gram index]
All characters appearing in the document are registered in the index as a continuous character string (gram) having a predetermined length N. Similarly, at the time of search, a search character string (search word) to be searched is divided into character strings (grams) having a length N, and search is performed by obtaining appearance information of the same character string from the index.

  In the search using the N-gram index, first, a primary search is performed in which candidate documents are selected based only on the presence / absence of a character string that matches the character string of length N divided from the search word (that is, the presence / absence of a hit). Thereafter, a secondary search for narrowing down to documents including the search word is performed by checking the adjacency relationship between words. As described above, in the search using the N-gram index, a full-text search without omission is realized by a two-stage search including a primary search and a secondary search.

  It is known that the N value of the N-gram should be increased in order to increase the search accuracy of the primary search. However, if the N value is increased, the index becomes extremely large, so that it may take a long time for the search. On the other hand, if the N value is reduced, search noise increases and search accuracy decreases. Since the secondary search targets all hit documents, the greater the number of hits (regardless of actual noise), the lower the efficiency.

[Search method by morpheme index]
The document is analyzed, morphemes (words) to be indexed are extracted from the document within the range of the smallest meaningful language unit (morpheme), and document information is assigned to each extracted morpheme and registered in the index. Similarly, at the time of retrieval, the retrieval character string is divided into morphemes, and retrieval is performed by obtaining document information that matches the same morphemes.

  In the search based on the morpheme index, there is no overlapping part between morphemes, so the size of the index is small and high-speed search is possible. However, a search omission occurs when the target document and the morphological result of the search condition do not match.

  Thus, there is no omission in the search using the N-gram index, and the temporary search is fast, but there is a problem that the secondary search for removing noise is slow. On the other hand, a search using a morpheme index can perform a high-speed search, but there is a problem that a search is omitted. That is, the search method using the N-gram index used for full-text search (full-text search) and the search method using morpheme index used for natural language search (concept search) have their advantages and disadvantages.

  Therefore, two types of searches, full-text search and natural language search, are executed to realize document search with few omissions by taking advantage of full-text search (full-text search) and natural language search (concept search). A document search method for merging both search results has been proposed (see, for example, Patent Document 1). In the document search technique described in Patent Document 1 (hereinafter referred to as the first prior art), a search including a specified text is performed by narrowing down the search result by a natural language search from the search results of a full-text search. From the search results obtained by performing a natural language search roughly by narrowing down the search results by full-text search from the search results of documents that are close to the question sentence from the target or the search results of natural language search A technique for searching a document including designated text is applied.

  Thus, the first prior art treats full-text search and natural language search as independent ones, and merges both search results. Specifically, the search result is narrowed down by the other of the full-text search and the natural language search from the search result by one of the full-text search and the natural language search. Therefore, in the first prior art, full text search and natural language search are always executed. However, full-text search is slower than natural language search. For this reason, if the search method using the N-gram index is applied to the full-text search, the full-text search requires a total search execution time (primary search time + secondary search time) using the N-gram index. That is, the first prior art does not have a mechanism for eliminating the drawbacks of full text search and speeding up the full text search itself. Therefore, the first prior art becomes a problem when the number of search hits is large.

  In addition, the “search expression” is evaluated in advance, and if the search expression is “keyword type (full text search)”, the full text search using the N-gram index is used. A document search method using the above has also been proposed (see, for example, Patent Document 2).

In the document search technique described in Patent Document 2 (hereinafter referred to as the second prior art), if it is determined as “keyword type”, the total search execution time (primary search) using the N-gram index is included in the search process. Secondary search) is required, and as with the first prior art, the full text search itself is not accelerated.
JP 2001-92831 (paragraphs 0005 to 0010, FIG. 5) JP 2003-308335 A (paragraph 0011)

  In the first prior art described above, a full text search is always executed. Also in the second prior art, if the search expression is “keyword type (full-text search)”, full-text search using the N-gram index is executed. This full-text search using the N-gram index takes a lot of time. However, neither of the first and second prior arts has a mechanism for speeding up the full text search itself.

  The present invention has been made in consideration of the above circumstances, and its purpose is to perform a primary search using an N-gram index and a search using a morpheme index, and perform a secondary search using an N-gram index when both search results are approximate. It is an object of the present invention to provide a full-text search system and method capable of executing full-text search at a high speed while omitting the search accuracy to some extent.

  According to one aspect of the present invention, a first search unit that executes a search based on an N-gram index by a primary search and a secondary search for the primary search result, a morpheme analysis unit that performs a morphological analysis on a search condition statement, There is provided a full-text search system including a second search unit that executes a morpheme search using a morpheme index based on a morpheme analysis result by the morpheme analysis unit. The system includes an approximation determining means for determining a degree of approximation between the number of primary search hits using the N-gram index and the number of morpheme search hits using the morpheme index, the number of primary search hits using the N-gram index and the morpheme When it is determined by the degree-of-approximation determining means that the number of morpheme hits by index is approximate, the first search means is controlled so that the secondary search by the N-gram index is omitted, A full-text search execution control unit that employs the result of the primary search or the result of the morpheme search as a search result.

  In the above configuration, when the number of hits as a result of the search using the morpheme index is used as an evaluation value, the number of hits for the search using the morpheme index is close to the number of hits for the primary search using the N-gram index. By omitting the secondary search by the gram, a high-speed search can be realized while ensuring a certain degree of search accuracy.

  Here, based on the morpheme analysis result by the morpheme analysis unit, a morpheme analysis result determination unit that determines whether the search condition sentence can be divided into words that can be searched for morpheme is added, and the search condition sentence is divided into words that can be searched for morpheme. Only when it is determined that the search has been completed, the full-text search execution control unit may control the second search unit so that the morpheme search using the morpheme index is executed.

  The fact that the search condition sentence can be divided into words that can be searched for morpheme is expected to be the same in many cases in how words are divided between the morpheme analysis result and the morpheme included in the morpheme index. Therefore, when a morpheme search using the morpheme index is performed using the morpheme analysis result at this time, the accuracy (reliability) of the morpheme index search result (number of hits) as an evaluation value, that is, the approximation determination by the approximation determination unit Accuracy (reliability) can be guaranteed to some extent.

  Whether the result of the primary search or the result of the morpheme search is adopted as the search result is determined based on the number of words in the analysis result of the morpheme analysis unit (the number of words into which the search condition sentence is divided) and the number of reference words. It is preferable that the configuration is determined based on the above.

  In such a configuration, when the number of words as a result of the morphological analysis is small (for example, one word), it is expected that there is almost no omission in the search for the morpheme search, so the result of the morpheme search is the result of the primary search. Can be considered more accurate. Therefore, based on the number of words in the morphological analysis result and the number of reference words, it is possible to obtain a more accurate search result by determining which of the primary search result or the morpheme search result is adopted as the search result. Can be adopted.

  Further, it is preferable that the number of primary search hits is compared with the reference hit count, and if the number of primary search hits is smaller than the reference hit count, a secondary search using the N-gram index is executed. If the number of hits in the primary search is small, the secondary search will not affect the search performance. Therefore, when the number of hits in the primary search is small, a secondary search is performed to realize a search with higher accuracy without affecting the search performance.

  Here, if a user interface for allowing the user to designate the number of reference words is added, tuning according to the search condition sentence becomes possible. Further, if a user interface for allowing the user to designate the number of hits of the reference is added, tuning according to the user's environment becomes possible. Further, if a user interface is added to allow the user to specify the degree of approximation that is the criterion for determination by the degree-of-approximation determination means, it corresponds to the search condition sentence or the characteristics of the document group to be searched. Tuning becomes possible.

  In either case, either a standard search up to a secondary search using the N-gram index or a high-speed search in which the secondary search using the N-gram index may be omitted based on the determination result by the approximation degree determination means. It is preferable to add a user interface that allows the user to specify the above. In this way, it is possible to reflect the user's intention to prioritize search speed or search accuracy.

  The full-text search execution control means controls the first search means and the second search means so that the primary search using the N-gram index and the morpheme search using the morpheme index are executed in parallel. And good. In this way, a faster search is possible.

  According to the present invention, a primary search using an N-gram index and a search using a morpheme index are performed, and when both search results are approximate, a secondary search using an N-gram index is omitted, thereby ensuring a certain degree of search accuracy. The full text search can be executed at high speed.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
FIG. 1 is a block diagram showing the configuration of a full-text search system according to an embodiment of the present invention. This full-text search system is a system that performs full-text search using an N-gram index and natural language search (search using a morpheme index) in response to a user request. However, the full-text search system of FIG. 1 omits a part of the full-text search (secondary search by N-gram index) when a certain condition is satisfied.

  The full-text search system of FIG. 1 includes a user interface 11, a search execution / response server 12, an N-gram search engine 13, an N-gram index database 14, a morpheme analysis mechanism 15, a morpheme search engine 16, and a morpheme index database. 17 and a full-text search execution control mechanism 18.

  The user interface 11 has an interface function that accepts a search request from the user and presents the search result to the user. In the present embodiment, the user interface 11 constitutes a part of the full-text search system, but is not limited to this. For example, the user interface 11 may be provided in a client terminal connected to the full-text search system of FIG. 1 via a communication line (for example, a network).

  The search execution / response server 12 uses the N-gram search engine 13 or the morpheme for the search condition indicated by the search request received by the user interface 11 for the search by the N-gram index or the search by the morpheme (the necessary morpheme analysis). It passes to the analysis mechanism 15. The search execution / response server 12 also causes the user interface 11 to present the search result by the N-gram search engine 13 or the morphological analysis mechanism 15 to the user.

  The N-gram search engine 13 performs a full text search using the N-gram index stored in the N-gram index database 14. The N-gram search engine 13 includes a primary search execution unit 131 and a secondary search execution unit 132. The primary search execution unit 131 performs a primary search using an N-gram index. The secondary search execution unit 132 performs a secondary search using an N-gram index. The N-gram index stored in the N-gram index database 14 includes all characters appearing in a document that can be searched as continuous character strings (grams) having a predetermined length N, for each character string. The position information of the character string is registered.

  The morphological analysis mechanism 15 performs morphological analysis on the search condition. The morpheme search engine 16 performs a morpheme search using the morpheme index stored in the morpheme index database 17 according to the morpheme analysis result by the morpheme analysis mechanism 15. In the morpheme index stored in the morpheme index database, document information including position information assigned to each morpheme extracted from a document that can be searched is registered.

  The full-text search execution control mechanism 18 controls the N-gram search engine 13 and the morpheme search engine 16 according to the setting contents of the setting information file 19 in order to execute full-text search using the N-gram index at high speed. In the setting information file 19, information such as conditions necessary for controlling execution of full text search by the full text search execution control mechanism 18 is set.

  The full-text search execution control mechanism 18 includes a morphological analysis result determination unit 181, a primary search result number determination unit 182, and a search result number approximation degree determination unit 183. The morpheme analysis result determination unit 181 determines whether to perform a search using a morpheme index (morpheme search) or a secondary search using an N-gram index based on a morpheme analysis result for a search condition by the morpheme analysis mechanism 15. The primary search result number determination unit 182 determines whether to perform a secondary search using the N-gram index based on the primary search result using the N-gram index. The search result number approximation degree determination unit 183 determines whether to perform a secondary search using the N-gram index based on the primary search result using the N-gram index and the morpheme search result.

  Next, the procedure of full text search processing (high speed search processing) in the high speed search mode executed in the full text search system of FIG. 1 will be described with reference to the flowchart of FIG. In this embodiment, a standard search mode is prepared as a search mode in addition to the high-speed search mode, and can be selected by the user as will be described later.

  Now, it is assumed that a user who desires full-text search performs an input operation using a client terminal, and a search request for designating full-text search is sent from the terminal to the full-text search system of FIG. The user interface 11 receives this search request and extracts search conditions indicated by the search request. The user interface 11 sends the extracted search condition to the search execution / response server 12. Further, the user interface 11 notifies the search execution / response server 12 of the search type (for example, full-text search) indicated by the search request. The search execution / response server 12 sends the search condition passed from the user interface 11 to the N-gram search engine 13 in order to execute the full text search when the full text search is designated.

  The primary search execution unit 131 in the N-gram search engine 13 performs a well-known primary search using the N-gram index stored in the N-gram index database 14 in accordance with the search conditions sent from the search execution / response server 12. Execute (Step S1). The primary search execution unit 131 holds the primary search result in the N-gram search engine 13 and sends the number of hits (number of hits) N1 to the full-text search execution control mechanism 18 together with the corresponding search conditions.

  The primary search result number determination unit 182 in the full-text search execution control mechanism 18 compares the hit number N1 sent from the primary search execution unit 131 with the hit number threshold value K set in the setting information file 19, and determines the magnitude Determine (step S2). If the number of hits N1 is less than or equal to K, the full-text search execution control mechanism 18 requests the N-gram search engine 13 for a secondary search. K can be changed (tuned) by a user operation as will be described later.

  On the other hand, when the hit count N1 is larger than K, the primary search result count determination unit 182 holds the hit count N1 in the full-text search execution control mechanism 18 and then sends the search condition to the morphological analysis mechanism 15. When receiving the search condition from the primary search result number determination unit 182, the morpheme analysis mechanism 15 performs morphological analysis on the search condition (step S 3). Then, the morpheme analysis mechanism 15 sends the result of the morpheme analysis to the full-text search execution control mechanism 18.

  The morpheme analysis result determination unit 181 in the full-text search execution control mechanism 18 determines the morpheme analysis result by the morpheme analysis mechanism 15 (step S4). That is, the morpheme analysis result determination unit 181 uses words that can be searched using a morpheme index (morpheme search), that is, words that are meaningful by themselves (for example, independent words represented by nouns, verbs, and adjectives). Judge whether it was divided. If the search condition cannot be divided into words that can be searched for morpheme, the full-text search execution control mechanism 18 requests the N-gram search engine 13 to perform a secondary search.

  In contrast, when the search condition can be divided into words that can be searched for morpheme, the morpheme analysis result determination unit 181 sends the morpheme analysis result by the morpheme analysis mechanism 15 to the morpheme search engine 16. When receiving the morpheme analysis result from the morpheme analysis result determination unit 181, the morpheme search engine 16 performs a morpheme search using the morpheme analysis result and the morpheme index database 17 (step S 5). The morpheme analysis mechanism 15 holds the morpheme search result in the morpheme analysis mechanism 15 and sends the number of hits (number of hits) N2 to the full-text search execution control mechanism 18.

  The search result number approximation degree determination unit 183 in the full-text search execution control mechanism 18 is a hit in the primary search by the N-gram search engine 13 (the primary search execution unit 131) held in the full-text search execution control mechanism 18. It is determined whether the number N1 and the number of hits N2 in the morpheme search sent from the morpheme search engine 16 are approximate (N1≈N2) (step S6). Here, the search result number approximation degree determination unit 183 determines whether the approximation degree (%) between N1 and N2 is within the approximation setting value P (%) set in the setting information file 19. In this embodiment, the degree of approximation between N1 and N2 is | N1-N2 | × 100% / N1 or | N1-N2 | × 100% / N2, that is, N1 or N2 of the absolute value of the difference between N1 and N2. The smaller the value, the closer the N1 and N2 are. The closeness setting value P can be tuned by a user operation as will be described later.

  When the approximation degree between N1 and N2 exceeds P, the search result number approximation degree determination unit 183 determines that N1 and N2 are not approximate. In this case, the full-text search execution control mechanism 18 requests the N-gram search engine 13 for a secondary search.

  On the other hand, when the degree of approximation between N1 and N2 is within P, the search result number approximation degree determination unit 183 determines that N1 and N2 are approximate. In this case, the full-text search execution control mechanism 18 determines whether to give priority to N-gram search or morpheme search according to the information set in the setting information file 19. In the case of N-gram search priority, the full-text search execution control mechanism 18 requests the N-gram search engine 13 to return the primary search result held internally to the search execution / response server 12. On the other hand, when priority is given to morpheme search, the full-text search execution control mechanism 18 requests the morpheme search engine 16 to return a morpheme search result stored internally to the search execution / response server 12. That is, the search result number approximation degree determination unit 183 causes the search execution / response server 12 to respond to the search result by the primary search or the morpheme search (step S7).

  When the search execution / response server 12 receives the search result by the primary search or the morpheme search from the full-text search execution control mechanism 18 or the morpheme search engine 16, the search execution / response server 12 notifies the user of the search result via the user interface 11 (and the search application). To do. Information indicating what kind of determination the search is executed is added to the search result.

  Thus, in the present embodiment, (a1) the number of hits N1 in the primary search using N-grams exceeds K, (a2) the search condition can be divided into words that can be searched for morpheme, and (a3) N-grams When the number of hits N1 in the primary search by N and the number of hits N2 in the morpheme search are similar (N1≈N2), the execution of the secondary search by N-gram is omitted, and the search result for the search request As a result, a search result by primary search or morphological search is adopted. However, as long as the condition (a3) among the above conditions (a1), (a2), and (a3) is satisfied, even if the execution of the secondary search using N-grams is omitted, a certain degree of search accuracy is secured. obtain. Therefore, when at least the condition (a3) is satisfied, the secondary search using N-grams may be omitted. In any case, the speed of the full text search can be increased by omitting the secondary search using N-grams while suppressing a decrease in search accuracy. However, if the above condition (a1) is not satisfied, even if a secondary search using N-grams is performed, the effect on performance is small, so there are few advantages of omitting the secondary search. Further, when the condition (a2) is not satisfied, the reliability of determining whether the condition (a3) is satisfied is lowered.

  The secondary search execution unit 132 in the N-gram search engine 13 executes a secondary search for the primary search result of the N-gram index only when a secondary search request is received from the full-text search execution control mechanism 18. (Step S8). As described above, in this embodiment, the secondary search is requested from the full-text search execution control mechanism 18 to the N-gram search engine 13 because (b1) the number of hits N1 in the primary search by N-gram is K or less. In the case of (b2) when the search condition cannot be divided into words that can be searched for morpheme, or (b3) the number of hits N1 in the primary search by N-gram and the number of hits N2 in the morpheme search are not approximate One of them. In the case of (b1), even if a secondary search using N-grams is performed in order to ensure a sufficiently high search accuracy, the influence on the search performance is small. In the case of the above (b2) or (b3), there is no guarantee that a certain degree of search accuracy can be ensured only by a morpheme search or a primary search by N-gram. For this reason, in this embodiment, even if there is a possibility of affecting the search performance, a secondary search using N-grams is performed in order to ensure sufficiently high search accuracy.

  In the present embodiment, the user interface 11 includes a first search interface that allows the user to select the accuracy of full-text search, and a second search interface that allows the user to specify tuning parameters used when executing the above-described high-speed search. And have. The user interface 11 presents a search interface screen for realizing these first and second search interfaces to the user. FIG. 3 shows an example of a search interface screen when full-text search is designated. This search interface screen is one of search execution screens. In addition to a search condition field 31 for specifying, for example, a keyword as a search condition, and a search button 32 for instructing search execution, a search accuracy selection area 33 and And tuning parameter area 34.

  In the search accuracy selection area 33, a “high speed” selection button 331 for instructing a high-speed search mode and a “standard” selection button 332 for instructing a standard search mode are arranged.

Standard When the “standard” selection button 332 is selected and a search is instructed, a full-text search process (standard search process) in the standard search mode is executed. Here, a search (primary and secondary search) using an N-gram index is performed. In this case, the search result is perfect, but the search speed decreases.

High-speed When the “high-speed” selection button 331 is selected and a search is instructed, a full-text search process (high-speed search process) in the high-speed search mode is executed. Here, when the search according to the flowchart of FIG. 2 described above is performed and the primary search result based on the N-gram index and the search result based on the morpheme are approximate, a high-speed search can be performed while ensuring a certain degree of search accuracy. It becomes possible.

  In the tuning parameter area 34, a hit count field 341 for designating a hit count threshold (reference hit count) K, an approximate ratio field 342 for specifying an approximation setting value (approximate ratio) P, and an adoption are adopted. An adoption condition field 343 for designating a condition is arranged.

-Number of hits In the primary search using the N-gram index, when the number of search hits exceeds the specified value (hit-number threshold K) in the hit number field 341, one of a plurality of conditions that can omit the secondary search using the N-gram index is Treated as established. If the number of search hits is less than the value specified in the hit number field 341, a secondary search is performed on the primary search result of the N-gram index. In other words, when the number of hits in the primary search of the N-gram index is small, the secondary search has little effect on the search performance, so the secondary search is performed to obtain a complete search result with high accuracy. It is done.

-Approximate ratio When the degree of approximation between the number of hits in the primary search using the N-gram index and the number of hits in the search using the morpheme index is less than the specified approximate ratio (approximation set value P), the secondary using the N-gram index One of a plurality of conditions that can be omitted is treated as being satisfied. If the degree of approximation exceeds the specified approximation ratio, a secondary search is performed on the primary search result of the N-gram index. That is, when the result of the primary search using the N-gram index is far from the result of the search using the morpheme index, it is determined that the search accuracy is poor, and the secondary search using the N-gram index is performed.

-Adopting conditions If the number of hits in the primary search using the N-gram index is close to the number of hits in the search using the morpheme index, the result is reasonable regardless of which search result is used. However, when the number of words obtained as a result of morphological analysis of the search condition (keyword) is less than the minimum number of words, the search using the morpheme index becomes more accurate. Therefore, by performing a morphological analysis on the search condition as a condition for determining whether to adopt the primary search result based on the N-gram index or the search result based on the morpheme index, that is, a condition for determining whether the N-gram search priority or the morphological search priority. A reference value (reference word number) of the number of divided words can be designated by the adoption condition field 343. In addition, when the number of words obtained as a result of morphological analysis of the search condition exceeds the reference number of words, the accuracy of the result of the morpheme analysis is low, and a certain degree of accuracy can be secured for the result of the search by the morpheme index. There is no guarantee. Therefore, in addition to the above step S4, for example, if the number of words obtained as a result of the morphological analysis of the search condition is less than the reference word number, if the reference word number is exceeded, the secondary by the N-gram index It is recommended that the search be performed. On the other hand, when the number of words obtained as a result of morphological analysis of the search condition is equal to or less than the reference word number and the above conditions (a1), (a2), and (a3) are satisfied, the secondary search using the N-gram index is omitted. With this configuration, it is effective in terms of search time particularly when a long search condition is specified.

  FIG. 4 shows an example of a search result screen notified to the user via the user interface 11 by the search execution / response server 12. This search interface screen is one of the search execution screens. In addition to the search condition field 31, the search button 32, the search accuracy selection area 33, and the tuning parameter area 34 that are common to the search interface screen shown in FIG. , A search accuracy area 41 and a search result area 42.

As described above, when the high-speed search process is executed, according to the flowchart of FIG. 2, the search result area 42 of the search result screen is used based on both the search results based on the N-gram index and the morpheme index and the tuning parameter information. Search results to be notified to the user. The search results notified to this user, that is, the search results displayed in the search result area 42 are the following three types of search results: (a) Results of search (primary search and secondary search) by N-gram index (b) Result of search by morpheme index (c) One of the results of search by N-gram index (primary search only). On the other hand, when the standard search process is executed, the above result (a) is always adopted as the search result displayed in the search result area 42.

  When the high-speed search processing is executed, which search result (a), (b), (c) is adopted is, for example, a term that expresses the corresponding “search accuracy” in an abstract manner. 41. Here, “appropriate”, “slightly coarse”, and “rough” are used as terms that abstractly express “search accuracy” corresponding to (a), (b), and (c), respectively.

[First Modification]
Next, a first modification of the above embodiment will be described with reference to the flowchart of FIG. In FIG. 5, the same reference numerals are given to the parts equivalent to the flowchart of FIG. 2 showing the procedure of the high-speed search process.

  The feature of the first modified example is that the processes of steps S1 and S2 (primary search using N-gram index) and the processes of steps S3 to S5 (search using morpheme index) are executed in the reverse order of the flowchart of FIG. Is in the point to be. Here, when the search condition cannot be divided into words that can be searched for morpheme (step S4), a primary search using an N-gram index corresponding to step S1 is executed (step S11).

[Second Modification]
Next, a second modification of the above embodiment will be described with reference to the flowchart of FIG. In FIG. 6, the same reference numerals are given to the parts equivalent to the flowchart of FIG.

  The feature of the second modified example is that the processing of steps S1 and S2 (primary search using N-gram index) and the processing of steps S3 to S5 (search using morpheme index) are executed in parallel. In this way, both searches are executed in parallel, thereby enabling a faster search.

  Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. Further, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment.

The block diagram which shows the structure of the full text search system which concerns on one Embodiment of this invention. 9 is a flowchart showing a procedure of full-text search processing (high-speed search processing) in the high-speed search mode in the embodiment. The figure which shows an example of a search interface screen. The figure which shows an example of a search result screen. The flowchart which shows the procedure of the high-speed search process in the 1st modification of the said embodiment. The flowchart which shows the procedure of the high-speed search process in the 2nd modification of the said embodiment.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 11 ... User interface, 12 ... Search execution / response server, 13 ... N-gram search engine (first search means), 14 ... N-gram index database, 15 ... Morphological analysis mechanism, 16 ... Morphological search engine (second search) Means), 17 ... morpheme index database, 18 ... full text search execution control mechanism, 19 ... setting information file, 33 ... search accuracy selection area, 34 ... tuning parameter area, 131 ... primary search execution section, 132 ... secondary search execution section , 181 ... Morphological analysis result determination unit, 182 ... Primary search result number determination unit, 183 ... Search result number approximation degree determination unit, 331 ... "High speed" selection button, 332 ... "Standard" selection button, 341 ... Hit number field, 342... Approximate ratio field, 343.

Claims (10)

Based on a first search means for executing a search based on an N-gram index by a primary search and a secondary search for the primary search result, a morpheme analysis means for morphological analysis of a search condition statement, and a morpheme analysis result by the morpheme analysis means A full-text search system comprising second search means for executing a morpheme search using a morpheme index,
Approximation degree determination means for determining an approximation degree between the number of hits of the primary search using the N-gram index and the number of hits of the morpheme search using the morpheme index;
When the approximation degree determining unit determines that the number of hits of the primary search using the N-gram index is close to the number of hits of the morpheme search using the morpheme index, the secondary search using the N-gram index is omitted. And a full-text search execution control means for controlling the first search means and adopting the result of the primary search or the result of the morpheme search as a search result. Further comprising morpheme analysis result determination means for determining whether the search condition sentence can be divided into morpheme searchable words based on the morpheme analysis result by the morpheme analysis means;
The full-text search execution control means executes the morpheme search using the morpheme index only when the morpheme analysis result determination means determines that the search condition sentence has been divided into words that can be searched for morpheme. 2. The full-text search system according to claim 1, wherein the search means is controlled.   The full-text search execution control means, based on the number of words obtained by dividing the search condition sentence indicated by the analysis result of the morpheme analysis means and the number of reference words, the result of the primary search or the result of the morpheme search The full-text search system according to claim 1, wherein which is adopted as a search result is determined.   4. The full-text search system according to claim 3, further comprising a user interface for enabling a user to designate the number of reference words. A primary search result number judging means for judging whether the number of hits of the primary search is large or small by comparing the number of hits of the primary search with the N-gram index with a reference hit number;
The full-text search execution control means is configured to execute the first search so that a secondary search using the N-gram index is executed when the primary search result number determination means determines that the number of hits of the primary search is small. 2. The full-text search system according to claim 1, wherein said second search result is adopted as a search result by controlling means.   6. The full-text search system according to claim 5, further comprising a user interface for enabling a user to designate the reference hit count.   The full-text search system according to claim 1, further comprising a user interface that allows a user to specify an approximation that is a criterion for determination by the approximation determination unit.   Either of the standard search for performing the secondary search using the N-gram index in any case, or the high-speed search in which the secondary search using the N-gram index may be omitted based on the determination result by the approximation degree determination unit. The full-text search system according to claim 1, further comprising a user interface for enabling the user to specify the password.   The full-text search execution control means controls the first search means and the second search means so that a primary search using the N-gram index and a morpheme search using the morpheme index are executed in parallel. The full-text search system according to claim 1. Based on a first search means for executing a search based on an N-gram index by a primary search and a secondary search for the primary search result, a morpheme analysis means for morphological analysis of a search condition statement, and a morpheme analysis result by the morpheme analysis means In a full-text search method applied to a system comprising second search means for executing a morpheme search using a morpheme index,
Determining the degree of approximation between the number of primary search hits by the N-gram index and the number of morpheme hits by the morpheme index;
When it is determined that the number of hits of the primary search by the N-gram index is close to the number of hits of the morpheme search by the morpheme index, the second search by the N-gram index is omitted. Controlling the search means;
And adopting the result of the primary search or the result of the morpheme search as a search result when a secondary search using the N-gram index is omitted.

Images