JP3924899B2 - Text search apparatus and text search method - Google Patents

Description

【００７５】
また、誤り文字列登録部２４に登録された誤り文字列パターンには、「タリ→列」、「タ（カタカナ）→夕（漢字）」 が含まれているものとする。
【００７６】
ランキング部２６は、候補文字列「アルタリア」に対しては完全一致の場合のポイント「１００」を付与し、「アル列ア」は２文字不一致かつ誤り文字列「タリ→列」が誤り文字列登録部２４に登録されているので、１０＋６０＝７０ポイントを付与される。また、候補文字列「アル夕リア」（“夕”は漢字）は１文字不一致かつ誤り文字列「タ→夕」が誤り文字列登録部２４に登録されているので、５０＋３０＝８０ポイントを付与される。そして、ランキング部２６は、例えば、ランキング結果として、候補文字列とその位置情報とポイントの組、例えば「アルタリア（1,10,100）」、「アル列ア（2,5,70）」、「アル夕リア（3,21,80）」を出力部１４へ出力する。
【００７７】
出力部１４は、ランキング部２６からのランキング結果を得ると、例えば、ポイントが高い順に、候補文字列を画面表示する。また、出力部１４は、ある値以上のポイントを得た候補文字列のみを表示してもよいし、ポイントが指定された範囲内にあるものをグループ化して表示してもよい。
【００７８】
なお、誤り文字列登録部２４を用いたランキングではなく、簡単に、完全一致、１文字曖昧、２文字曖昧というランキングを行うことも可能である。
【図面の簡単な説明】
【図１】 本発明の第一の実施形態であるテキスト検索装置の概略のブロック構成図である。
【図２】 第一の実施形態に係る検索対象テキストのイメージを示す模式図である。
【図３】 第一の実施形態に係る検索対象テキストのインデックスのイメージを示す模式図である。
【図４】 対象文字位置情報記憶部に格納される対象文字位置情報のイメージを示す模式図である。
【図５】 本発明の第二の実施形態であるテキスト検索装置の概略のブロック構成図である。
【図６】 第二の実施形態に係る検索対象テキストのイメージを示す模式図である。
【図７】 第二の実施形態に係る検索対象テキストのインデックスのイメージを示す模式図である。
【図８】 誤り文字列登録部に登録された誤り文字列パターンの一例を示す模式図である。
【図９】 第三の実施形態に係る検索対象テキストのイメージを示す模式図である。
【図１０】 検索対象テキストに対して生成され、インデックス記憶部に格納されているインデックスのイメージを示す模式図である。
【符号の説明】
２ インデックス記憶部、４ 入力部、６ 検索部、８ 対象文字位置情報記憶部、１０ 検索文字列展開部、１２，２２ マッチング部、１４ 出力部、２０ テキスト記憶部、２４ 誤り文字列登録部、２６ ランキング部。[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a text search apparatus for text that may have an inaccurate character string, and particularly relates to reduction of search omission and weighting in search results.
[0002]
[Prior art]
Conventionally, there has been a text search device that specifies a search character string and searches for the search character string included in a document or character string. The character string search function installed in the word processor is an example of such a case.
[0003]
It is assumed that the documents and character strings to be searched are basically free from errors. In the search, it is determined that the character string of interest exists in the search target text only when the character string included in the search target text completely matches the search character string.
[0004]
On the other hand, when the text to be searched is text data read by an optical character reader (OCR), there is a probability that an incomplete text including an inaccurate character string is generated due to a recognition error in the reading. high. Japanese OCR has a low accuracy, so it is particularly likely. If this incomplete text is searched with a complete match with the search character string as described above, there is a possibility that a search omission may occur. In other words, if the search target text is correctly read, the character string portion that should have been hit may not be hit due to a recognition error.
[0005]
In order to prevent such a search omission, there is a technique (hereinafter referred to as an ambiguous search) for performing a search with a certain degree of ambiguity of the search target. The prior art of the first fuzzy search disclosed in Japanese Patent Laid-Open No. 62-44878 is to search by embedding candidate characters in the search target text when a plurality of candidates are obtained as a result of recognition. (Example: [Honoki] sentence by sentence [Language] recognition [Shiori] ...). The prior art of the second fuzzy search disclosed in Japanese Patent Application Laid-Open No. 8-7033 is to leave them in the index when a plurality of candidates are obtained for each character that has been subjected to character recognition. In this case, in the index storing the recognition result for each character, a plurality of recognition result character candidates can be stored for one character to be recognized. These two techniques provide ambiguity to the recognition result, that is, the search target text. On the other hand, the third prior art of ambiguous search disclosed in Japanese Patent Laid-Open Nos. 6-195387, 7-152774, and 8-63487 has ambiguity on the side of the search character string. Is. This method creates an incomplete search string by replacing the erroneously recognized part of the search string with a character pattern that may be misrecognized (misrecognition pattern). The search is also performed by the incomplete search character string. The types of misrecognition patterns include character error, misdivision, and miscombination. For example, “character” is easily recognized as “study”, but such a type is a character error. In addition, “K” is easily recognized as “Ihi”, but such a type is misdivided. On the other hand, “51” is easily recognized as “Draw”, but such a type is erroneously combined.
[0006]
While fuzzy search has the advantage that search omissions can be reduced, conversely, there is a possibility that a search error that originally matches a character string that is different from the search character string is included. is there.
[0007]
[Problems to be solved by the invention]
The prior art of the first fuzzy search described above has a problem that the capacity of the search target text that is the recognition result increases, and the search cannot be performed unless it remains in the recognition result. In addition, there is a problem in that misrecognition patterns and miscombination cannot be handled.
[0008]
Even in the second prior art of the fuzzy search, there is a problem that the capacity of the index that is the recognition result increases, and the search is not performed unless it remains in the recognition result.
[0009]
The third prior art of fuzzy search has a problem that it is necessary to prepare a misrecognition pattern separately from the search character string, and its capacity increases. Also, for example, misconnection occurs depending on the combination of consecutive characters, so that many patterns can exist. It is difficult to prepare in advance all the erroneous recognition patterns that can occur in this way. If an erroneous recognition pattern that is not prepared in advance is generated, there is a problem that a search is omitted. Increasing the capacity of search target texts, indexes, misrecognition patterns, etc. used for search has caused not only a problem that the storage device requires a large capacity but also a problem that it takes a long time for the search process.
[0010]
Also, if you try to reduce search omissions and register multiple candidates in the index or enrich the misrecognition pattern, on the other hand, there is a risk of hitting a character string that is not originally related to the search character string. is there. That is, there is a problem that the search result includes a lot of “garbage” (search error), and the reliability of the search result is lowered.
[0011]
The present invention has been made to solve the above problems, and provides a text search device that reduces data used for ambiguous search while reducing search omission and reducing the influence of search errors. With the goal.
[0012]
[Means for Solving the Problems]
The text search device according to the present invention is a text search device that performs a search process on a search target text based on a search character string. Replacement string And an input means for accepting an input of the degree of matching specified between the search character string and the search character string inputted according to the degree of matching Replacement string Generate Replacement string Generating means and said generated Replacement string Candidate search means for searching the search target text for a candidate character string including a character string pattern that matches An error character string registration unit that registers an error character string pattern that may occur in the generation of the search target text, and a registration error that is registered in the error character string registration unit in a portion different from the search character string in the candidate character string An error character string detection means for detecting a character string pattern and a priority according to the possibility of matching with the search character string are determined based on the detection frequency of the registered error character string pattern, and the registration in the candidate character string Corresponding to detection of an error character string pattern, the priority is given to the candidate character string Is.
[0013]
The “matching degree” is the matching degree of the replacement character string of the search character string with respect to the search character string. For example, Based on the ratio of the number of characters in the search string and the number of characters remaining in the replacement string that are not replaced Can be defined. The specified degree of coincidence may be a numerical range specification or a threshold value. The “replacement character string” is obtained by masking some characters of the search character string, and the characters constituting the character string hold position information in the original search character string. For example, the replacement character strings “key ##” and “key ##” (# represents a masked character) of the search character string “keyword” are composed of the same set of characters, but are masked. It is treated as a different replacement character string at different positions. Further, as shown in this example, not only the characters constituting a certain replacement character string are continuous with each other, but also a masking position may be arranged between the constituting characters. The candidate search means extracts, from the search target text, a character string in which the same character is arranged at the position of each character constituting the replacement character string as a candidate character string. That is, in the extraction of the candidate character string, the match / mismatch of characters at the masked position in the search character string is not questioned.
[0015]
The search target text generated by character recognition or the like may include an erroneous character string, but the erroneous character string pattern is not random, and there may be a pattern that is likely to occur with respect to the original correct character or character string. . The error character string registration unit mainly stores such error character string patterns that are likely to occur. The present invention detects an error pattern stored in the error character string registration unit in a portion different from the search character string in the candidate character string. For example, if the detected error pattern is for the corresponding character string in the search character string, it is determined that the error pattern portion is likely to be a correct character string before the search target text is generated. Thus, the priority according to the possibility of matching can be set high.
[0016]
According to a preferred aspect of the present invention, the search character string and the search character string Replacement string Depending on the specified degree of match between and Replacement string Generate Replacement string Generating means, wherein the candidate character string is generated Replacement string Is a character string including a character string pattern that matches.
[0017]
In the text search device according to the present invention, the candidate search means in the invention described above Replacement string Having a means for considering the misrecognized allowable character included in the text as an arbitrary one character in the search target text, and a means for considering the character as an arbitrary two characters in the search target text Performing the search and also the occurred Replacement string One or more of performing the search with means for regarding two consecutive recognition-acceptable characters included in the text as one arbitrary character in the search target text. It is what.
[0018]
As a result, it is possible to search for candidate character strings having an error pattern of character error, misconnection, and error division, respectively.
[0019]
Book The text search apparatus according to the invention is characterized in that the priority assigning means determines the priority according to the detection frequency of the registration error character string pattern. According to the present invention, for example, it is determined that an original character string for a registered error character string pattern with a high detection frequency is likely to cause such an error, and a high priority can be given.
[0020]
In a preferred aspect of the present invention, the error character string registration unit further stores the detection frequency in addition to the registration error character string pattern.
[0021]
The text search device according to the present invention includes candidate character string display means for displaying the candidate character string in accordance with the priority. According to the present invention, the user can grasp the possibility of matching a search character string among a plurality of candidate character strings based on the priority, and is convenient when, for example, checking the result of search processing. .
The text search method according to the present invention is a text search method for performing search processing on a search target text on a computer based on a search character string, wherein the input means included in the computer uses a search character string as a search condition. And the search string and its search string Replacement string Receiving the input of the degree of coincidence specified between and the computer Replacement string The generation means determines the search character string according to the input search character string and the matching degree. Replacement string And candidate search means possessed by the computer are generated. Replacement string Searching for a candidate character string including a character string pattern matching ,in front The character string registration means possessed by the computer registers an error pattern that may occur in the generation of the search target text, and the error character string detection means possessed by the computer differs from the search character string in the candidate character string In part, a step of detecting a registration error character string pattern registered in the character string registration means, and a priority giving means that the computer has, A priority according to the possibility of matching with the search character string is determined based on the detection frequency of the registration error character string pattern, and in response to the detection of the registration error character string pattern in the candidate character string, the priority Give degree to the candidate string It is a step.
[0022]
DETAILED DESCRIPTION OF THE INVENTION
Next, embodiments of the present invention will be described with reference to the drawings.
[0023]
[Embodiment 1]
FIG. 1 is a schematic block diagram of a text search apparatus according to an embodiment of the present invention. This apparatus uses text recognized by OCR as a search target, index storage unit 2, input unit 4, search unit 6, target character position information storage unit 8, search character string expansion unit 10, matching unit 12, and output. The unit 14 is configured.
[0024]
The index storage unit 2 stores the search target text obtained by OCR in the form of an index prior to the search. The index is obtained by associating the appearance position of the character with the character appearing in the search target text as a key.
[0025]
The input unit 4 receives a search condition such as a search character string and a matching degree from the user.
[0026]
The search unit 6 obtains a search character string from the input unit 4, searches the index stored in the index storage unit 2 with each character included therein, and determines the appearance position of each character in the search character string as the target character position It outputs to the information storage part 8, and the object character position information storage part 8 stores this.
[0027]
The search character string expansion unit 10 is a partial string generation unit that generates a substring of the search character string. The search character string development unit 10 obtains a matching degree with the search character string from the input unit 4, and from the search character string according to the matching degree, Expand and generate a replacement string that includes a substring. The replacement character string is generated by replacing some characters of the search character string with, for example, the symbol “#” and masking the original character. The part other than the part replaced with “#” in the replacement character string is a partial string composed of the original characters of the search character string.
[0028]
The matching unit 12 performs matching with the target character position information stored in the target character position information storage unit 8 using the replacement character string output from the search character string expansion unit 10. The matching result is output to the output unit 14 and displayed on the screen as a search result on a display device such as a CRT.
[0029]
Next, the operation of each component will be described using a specific example. FIG. 2 is a schematic diagram showing an image of the search target text. Here, the search target texts used in the example are document A, document B, and document C. In the document A, the character string “Persia” exists from the 10th character from the beginning. Similarly, the document B includes the character string “Bersha” from the fifth character from the beginning, and the document C includes the character string “Penoresha” from the 21st character from the beginning.
[0030]
FIG. 3 is a schematic diagram illustrating an index image generated for these search target texts and stored in the index storage unit 2. The index is obtained by classifying the position in the document in which the character type appears in each key using the type of character appearing in the search target text (shown at the left end in the figure) as a key. The appearance position of the character is a number Ndoc (document A is “1”, document B is “2”, document C is “3”) and the number of characters from the beginning of each document. It is expressed in the form of a pair (Ndoc, Nchar) with Nchar.
[0031]
As a search condition, the user inputs a search character string “Persia” and a matching degree of 70% to the input unit 4. Here, the degree of matching η is defined by the ratio of the number of characters in the substring to the search character string. That is, if the number of characters in the search character string is M and the number of characters remaining in the replacement character string that are not replaced is m, the degree of matching is η = m / M × 100 [%]. The matching degree input to the input unit 4 is a threshold value ηth of η, and this apparatus searches for a character string having a matching degree exceeding ηth in the search target text. Note that, if the threshold value ηth of the matching degree is low, “dust” included in the search result increases. Therefore, a preferable value of the threshold value ηth is generally about 70% or more. On the other hand, if the threshold ηth is higher than necessary, there is a high possibility that a search omission will occur. Considering this point, ηth = 70% is set here.
[0032]
The input unit 4 notifies the search unit 6 of the search character string “Persia”. When the search unit 6 obtains the search character string, the search unit 6 searches the index storage unit 2 using each of the characters “pe”, “le”, “si”, and “ja” as a key, and the result is the target character position. The information is stored in the information storage unit 8. Specifically, in this example, the appearance positions (1,10) and (3,21) for the character “pe”, the appearance positions (1,11) and (2,6) for the character “le”, and the character “si” Appearance position (1,12), (2,7), (3,24), and appearance position (1,13), (2,8), (3,25) for the character " It is stored in the storage unit 8. FIG. 4 is a schematic diagram showing an image of target character position information stored in the target character position information storage unit 8.
[0033]
The search character string expansion unit 10 receives the threshold value ηth of the matching degree with the search character string from the input unit 4, and replaces the number of characters corresponding to the matching degree among the search character strings with the misrecognized allowable characters. Is generated. A misrecognized allowable character (an ambiguous character) is represented by a symbol “#” here. The characters in the search character string where the erroneous recognition allowable characters are placed are masked. The masking means that the difference between the replacement character string and the search target text, which will be described later, does not matter.
[0034]
Here, since the number M of characters in the search character string is 4, the number m of characters in the substring satisfying the matching threshold ηth = 70% is 3 or 4. Therefore, the search character string expansion unit 10 generates a replacement character string that does not include any misrecognized characters (this is equal to the search character string) and a replacement character string that includes only one misrecognized character. Specifically, in this example, five replacement strings “Persia”, “#Lucia”, “Pécha”, “Persia”, and “Persi #” are generated.
[0035]
The matching unit 12 is candidate search means for matching the replacement character string with the search target text and searching for a candidate character string that has a possibility of matching the search character string. Matching is performed by collating the relative character position in the replacement character string with the appearance position stored in the target character position information storage unit 8. Hereinafter, α and β are normal characters appearing in the replacement character string.
[0036]
For example, the matching of the two-character part “αβ” appearing in the replacement character string is performed as follows. First, the matching unit 12 searches the target character position information storage unit 8 using α and β as keys. Here, the appearance positions corresponding to α and β are (Ndoc (α), Nchar (α)) and (Ndoc (β), Nchar (β)), respectively. The matching unit 12 finds the appearance position of Ndoc (α) = Ndoc (β) with respect to the document number and Nchar (β) = Nchar (α) +1 with respect to the character position. Detect presence.
[0037]
Next, the matching process for character string portions “α # β”, “α ## β”, “α #### β”, “α ####... This is done as follows. There are three basic matching rules for the misrecognized character “#” as follows.
[0038]
(i) "#" is considered to be the same as any single character,
(ii) "#" is considered the same as any two characters,
(iii) “##” is regarded as the same as any one character.
[0039]
(i) is a rule corresponding to a character error. Also, (ii) and (iii) are rules corresponding to erroneous division and incorrect coupling, respectively.
[0040]
Rule (i) is realized as a search for an appearance position where Ndoc (α) = Ndoc (β) and Nchar (β) = Nchar (α) +2. Rule (ii) is realized as a search for the appearance position where Ndoc (α) = Ndoc (β) and Nchar (β) = Nchar (α) +3. The rule (iii) is realized by searching for an appearance position where Ndoc (α) = Ndoc (β) and Nchar (β) = Nchar (α) +2. Through these searches, the matching unit 12 detects the presence of a character string pattern such as “α # β”.
[0041]
The matching unit 12 performs the above-described matching process for each part of the replacement character string, and detects the presence of the replacement character string in the search target text. For example, for the replacement character string “Persia”, “Pe (1,10)”, “Le (1,11)”, “Shi (1,12)”, “ja (1,13)” In accordance with the basic matching rule, the matching unit 12 outputs a pair “Persia (1, 10)” of the replacement character string and the appearance position of the first character as a matching result. Then, once matched, it is removed from the target so as not to match with another replacement character string. For the replacement character string “#Lucia”, an arbitrary one character preceding “Le (2,6)” and “Le (2,6)”, “ (2,7) "and" (2,8) "are detected, and the matching unit 12 outputs"#Lucia (2,5) "as a matching result. Also, for the replacement character string “Pesha”, “Pe (3, 21)”, any two characters following it, and “ “(3, 24)” and “(3, 25)” conform to the above matching rule, and the matching unit 12 outputs “Pesha (3, 21)” as a matching result. Although the search is also performed for the replacement character strings “Per #” and “Persi #”, in this example, there is no character string (candidate character string) that hits them.
[0042]
The output unit 14 displays the search result on the screen based on the matching result obtained by the matching unit 12. The above-mentioned matching unit 12 outputs the position of the candidate character string as a matching result, and the output unit 14 outputs it as, for example, “Document A: (1,10) Document B: (2,5) Document C: (3, 21) "can be displayed. In addition, ranking may be performed such as complete match, one character ambiguity, and two character ambiguity depending on the number of misrecognition allowable characters, and the classification may be displayed for each group.
[0043]
Further, the output unit 14 may access the search target text based on the document number and the character position obtained from the matching unit 12 to obtain a candidate character string and display it. Further, the matching unit 12 itself may be configured to access the index storage unit 2 based on the position information of the candidate character string and output the candidate character string together with the position information to the output unit 14. With such a configuration, the output unit 14 includes contents including the candidate character string, for example, “Document A: Persia (1,10) Document B: Bersha (2,5) Document C: Penolesha (3, 21)”. Can be displayed.
[0044]
Incidentally, this apparatus can be configured using a general-purpose computer, and in particular, the functions of the search unit 6, the search character string expansion unit 10, and the matching unit 12 can be executed by a central processing unit (CPU). .
[0045]
According to the present invention described in this embodiment, for example, when a character or character string is erroneously recognized, a certain character or character string is converted into any wrong character or character string, and a search target text is generated. Regardless of whether or not information is used, it is possible to deal with character errors, misdivisions, and miscombinations, and search for all candidate character strings that may match the search character string.
[0046]
[Embodiment 2]
FIG. 5 is a schematic block diagram of a text search apparatus according to the second embodiment of the present invention. Of the components of this apparatus, the same components as those in the above embodiment are given the same reference numerals to simplify the description. The main difference of this apparatus is that it further includes a text storage unit 20, an error character string registration unit 24, and a ranking unit 26 in addition to the configuration of the above apparatus.
[0047]
The text storage unit 20 stores search target text, and each document is assigned a document number and can be distinguished from each other.
[0048]
The matching unit 22 performs the same processing as that of the apparatus of the above embodiment, and obtains position information of the candidate character string. The matching unit 22 of this apparatus further accesses the text storage unit 20 based on the position information, and acquires and outputs a candidate character string. At this time, position information can also be output.
[0049]
The error character string registration unit 24 stores an error character string pattern that is a character or a character string that is easily misrecognized in character recognition.
[0050]
When the ranking unit 26 obtains a candidate character string from the matching unit 22, the ranking unit 26 searches for an error character string pattern registered in the error character string registration unit 24 in the candidate character string. Then, the ranking unit 26 determines priority according to the possibility of matching between the candidate character string and the search character string according to the result (priority giving means). The ranking unit 26 outputs the candidate character string and its priority to the output unit 14.
[0051]
Next, the features of the operation of this apparatus will be described using a specific example. FIG. 6 is a schematic diagram illustrating an image of a search target text. Here, the search target texts used in the example are document A, document B, and document C. Document A has a character string “scanner” from the 10th character from the beginning. Similarly, the document B includes the character string “scan” from the fifth character from the top, and the document C includes the character string “scanner” from the 21st character from the top.
[0052]
FIG. 7 is a schematic diagram showing an image of an index generated for these search target texts and stored in the index storage unit 2.
[0053]
As a search condition, the user inputs a search character string “scanner” and a matching degree of 70% to the input unit 4.
[0054]
The input unit 4 notifies the search unit 6 of the search character string “scanner”. When the search unit 6 obtains this search character string, it searches the index storage unit 2 using each of the characters constituting the key as a key, and stores the result in the target character position information storage unit 8 as in the above embodiment.
[0055]
The search character string expansion unit 10 receives a threshold value ηth of coincidence with the search character string from the input unit 4 and generates a replacement character string corresponding to the threshold value ηth.
[0056]
Here, based on the number of characters M = 4 in the search character string and the matching threshold ηth = 70%, the search character string expansion unit 10 sets a replacement character string that does not include any misrecognition allowable character and a misrecognition allowable character as one. Generate a replacement string containing only one. Specifically, in this example, five “scanner”, “# canna”, “scan #”, “ski #na”, and “scan #” are generated as replacement character strings.
[0057]
The matching unit 22 performs matching between the replacement character string and the search target text, and detects the presence of the replacement character string in the search target text. For example, "su (1,10)", "ki (1,11)", "ja (1,12)", "na (1,13)" matches the replacement string "scanner" . Based on this position information, the matching unit 22 acquires the candidate character string “scanner” from the search target text stored in the text storage unit 20, and sets the combination “scanner (1, 1, 10) "is output to the ranking unit 26. Also, for the replacement character string “Sca #”, based on the basic rule and rule (i) described in the above embodiment, “S (2,5)”, “Ki (2,6)”, “ 2,7) ”and any single character that follows it. Based on this position information, the matching unit 22 accesses the text storage unit 20 to acquire the candidate character string “scan”, and outputs “scan (2, 5)” as a matching result. Also, for the replacement character string “skin #na”, based on the basic rule and rule (i), “su (3,21)”, “ki (3,22)”, any one character following this, "NA (3, 24)" following this one character matches. Based on this position information, the matching unit 22 accesses the text storage unit 20 to obtain the candidate character string “scanner”, and outputs “scanner (3, 21)” as a matching result.
[0058]
When the ranking unit 26 obtains a matching result from the matching unit 22, the ranking unit 26 performs ranking of the candidate character strings. Ranking is a process that determines the priority according to the possibility that the candidate character string matches the search character string. The number of characters in the part (error character string) where the candidate character string is different from the search character string and the error character string registration It is determined based on whether or not it is registered in the section 24 as an error character string pattern.
[0059]
FIG. 8 is a schematic diagram illustrating an example of an error character string pattern registered in the error character string registration unit 24. The figure shows that the character or character string on the left side of “→” is easily mistaken for the character or character string on the right side in character recognition when generating the search target text. For example, “S” is easily recognized as “I”, “A” as “Ya” or “N”, “N” as “Me”, and “Le” as “Nore”. Yes.
[0060]
For example, when the candidate character string completely matches the search character string, the ranking unit 26 assigns the point “100” as a numerical value indicating the priority, and assigns the point “10” when the candidate character string does not match. . Then, the ranking unit 26 checks whether or not an error character string that is a difference between the candidate character string and the search character string is registered in the error character string registration unit 24 as an error character string pattern. In this case, for example, “40” points are added to the points already acquired.
[0061]
Therefore, for example, since the candidate character string “scanner” is a perfect match, the point “100” is obtained, the candidate character string “scanner” does not match one character, and the error character string registration unit 24 further receives “a → ya”. Are registered, the point “50” obtained by adding the respective points “10” and “40” is obtained. On the other hand, although the candidate character string “scan” does not match one character, since the error character string is not registered in the error character string registration unit 24, only the point “10” is obtained. Then, the ranking unit 26, for example, as a ranking result, sets of candidate character strings and their positional information and points, for example, “scanner (1,10,100)”, “scanner (2,5,50)”, “scan (3 , 21, 10) ”is output to the output unit 14.
[0062]
When the output unit 14 obtains the ranking result from the ranking unit 26, the output unit 14 can perform display using the points included therein. For example, the candidate character strings can be displayed on the screen in the order from the highest point, that is, the highest possibility of matching with the search character string. Further, the output unit 14 may display only candidate character strings that have obtained a point equal to or greater than a certain value, or may display grouped items that have points within a specified range.
[0063]
The ranking unit 26 described above gives certain points to the error character string patterns registered in the error character string registration unit 24, but the points to be given are not necessarily uniform. For example, the error character string registration unit 24 stores the degree of error ease represented by the detection frequency of each error character string pattern and reflects this in the ranking, whereby more detailed ranking can be performed. For example, there is a method in which the ease of error is set with an adjustment coefficient of 0 to 1 and the point common to the error character string pattern is multiplied by the adjustment coefficient of the ease of error. In such a method, for example, if the adjustment coefficient of the candidate character string “scanner” is 0.8 in the above example, the point is 10 + 40 × 0.8 = 42. Further, the user can increase or decrease the detection frequency of the error character string pattern based on the search result.
[0064]
According to the present invention described in the present embodiment, as in the invention described in the first embodiment, the matching itself in the search process requires the error character string pattern registered in the error character string registration unit 24. Therefore, it is possible to deal with character errors, misdivision, and miscombination, and search for all candidate character strings that may match the search character string. It is undeniable that by performing this search without fail, those that have a low possibility of matching with the search character string are also detected as candidate character strings, and the percentage of “dust” (search error) included in the matching result increases. The present invention makes it possible to search without fail and to display the search results in a more probable order, thereby grasping the importance (priority) of each candidate character string when the user uses the search results. Even if a search error occurs, the influence on actual use can be reduced. In a conventional method of performing a search by developing a search character string using an error character string pattern, if the dictionary in which the error character string pattern is registered is not sufficiently enhanced, the search leaks increase and the reliability is lowered. On the other hand, in the present invention, even if there is no data in the error character string registration unit 24, it is possible to search without fail, and the accuracy of the ranking can be improved by enriching the data in the error character string registration unit 24. .
[0065]
[Embodiment 3]
The third embodiment of the present invention relates to another search processing example using another search character string, and the configuration of the text search device according to this embodiment is the same as that of the above-described second embodiment. It is the same.
[0066]
FIG. 9 is a schematic diagram showing an image of the search target text. Here, the search target texts used in the example are document A, document B, and document C. In document A, there is a character string “Altalia” from the 10th character from the beginning. Similarly, in document B, the character string “Al string A” exists from the fifth character from the beginning, and in document C, the character string “Al Yuria” (“Even” is kanji) from the 21st character from the beginning. Exists.
[0067]
FIG. 10 is a schematic diagram showing an index image generated for these search target texts and stored in the index storage unit 2.
[0068]
As a search condition, the user inputs a search character string “Altalia” and a matching degree of 60% to the input unit 4.
[0069]
The input unit 4 notifies the search unit 6 of the search character string “Altalia”. When the search unit 6 obtains this search character string, it searches the index storage unit 2 using each of the characters constituting the key as a key, and stores the result in the target character position information storage unit 8 as in the above embodiment.
[0070]
The search character string expansion unit 10 receives a threshold value ηth of coincidence with the search character string from the input unit 4 and generates a replacement character string corresponding to the threshold value ηth.
[0071]
Here, from the number of characters M = 5 in the search character string and the threshold value ηth = 60% of the matching degree, two erroneous recognition allowable characters are allowed. Specifically, in this example, the search character string expansion unit 10 includes, as replacement character strings, “Altalia”, “## Talia”, “# Le # Rear”, “# Luta # A”, “# Rutari #”, “A ## Rear”, “A # T ##”, “A # Tari #”, “Al ## A”, “Al # L #”, “Alter ###” are generated and output to the matching unit 22. .
[0072]
The matching unit 22 performs matching between the replacement character string and the search target text, and detects the presence of the replacement character string in the search target text. For example, for the replacement string “Altalia”, “a (1,10)”, “le (1,11)”, “ta (1,12)”, “li (1,13)”, “a” (1,14) "matches. Based on the position information, the matching unit 22 acquires the candidate character string “Altalia” from the search target text stored in the text storage unit 20, and sets the combination “Altalia (1,1, 10) "is output to the ranking unit 26. For the replacement character string “# le # rear”, any one character, followed by “le (3, 22)”, followed by any one character, “li (3, 24)”, “ A (3,25) "matches. Based on this position information, the matching unit 22 accesses the text storage unit 20 to obtain the candidate character string “Al Yuria” (“Yu” is kanji), and the matching result is “Al Yuria (3, 21)”. Is output. Further, for the replacement character string “al # a”, rules (iii) to “a (2,5)”, “le (2,6)”, and the like described in the first embodiment are followed. One character and "a (2,8)" match. Based on this position information, the matching unit 22 accesses the text storage unit 20 to acquire the candidate character string “al string a” and outputs “al string a (2, 5)” as a matching result.
[0073]
When the ranking unit 26 obtains a matching result from the matching unit 22, the ranking unit 26 performs ranking of the candidate character strings. In the present apparatus, the points given by the ranking unit 26 are expanded when there are two erroneous recognition allowable characters, and are determined for each case that can occur in that case. For example, it can be determined as follows.
[0074]

[0075]
Further, it is assumed that the error character string pattern registered in the error character string registration unit 24 includes “Tari → column” and “Ta (katakana) → evening (kanji)”.
[0076]
The ranking unit 26 assigns the point “100” in the case of a perfect match to the candidate character string “Altalia”, “Al string A” does not match two characters, and the error string “Tari → string” is an error string. Since it is registered in the registration unit 24, 10 + 60 = 70 points are given. Further, since the candidate character string “Al Yuria” (“Yu” is Kanji) does not match one character and the error character string “Ta → Yu” is registered in the error character string registration unit 24, 50 + 30 = 80 points are given. Is done. Then, for example, as a ranking result, the ranking unit 26 sets a candidate character string, its position information, and a point set, for example, “Altalia (1,10,100)”, “Al Sequence A (2,5,70)”, “Al "Evening rear (3, 21, 80)" is output to the output unit 14.
[0077]
When the output unit 14 obtains the ranking result from the ranking unit 26, for example, the output unit 14 displays candidate character strings on the screen in descending order of points. Further, the output unit 14 may display only candidate character strings that have obtained a point equal to or greater than a certain value, or may display grouped items that have points within a specified range.
[0078]
Instead of ranking using the error character string registration unit 24, it is also possible to simply perform ranking of perfect match, one character ambiguity, and two character ambiguity.
[Brief description of the drawings]
FIG. 1 is a schematic block diagram of a text search apparatus according to a first embodiment of the present invention.
FIG. 2 is a schematic diagram showing an image of a search target text according to the first embodiment.
FIG. 3 is a schematic diagram showing an image of a search target text index according to the first embodiment.
FIG. 4 is a schematic diagram showing an image of target character position information stored in a target character position information storage unit.
FIG. 5 is a schematic block diagram of a text search apparatus according to a second embodiment of the present invention.
FIG. 6 is a schematic diagram showing an image of a search target text according to the second embodiment.
FIG. 7 is a schematic diagram showing an image of a search target text index according to the second embodiment.
FIG. 8 is a schematic diagram illustrating an example of an error character string pattern registered in an error character string registration unit.
FIG. 9 is a schematic diagram showing an image of a search target text according to a third embodiment.
FIG. 10 is a schematic diagram showing an image of an index generated for a search target text and stored in an index storage unit.
[Explanation of symbols]
2 index storage unit, 4 input unit, 6 search unit, 8 target character position information storage unit, 10 search character string expansion unit, 12, 22 matching unit, 14 output unit, 20 text storage unit, 24 error character string registration unit, 26 Ranking section.

Claims (8)

In a text search device that performs a search process based on a search character string for a search target text,
As a search condition, an input means for receiving an input of a matching degree specified between a search character string and the search character string and its replacement character string;
A replacement character string generating means for generating a replacement character string of the search character string in accordance with the input search character string and the degree of matching;
Candidate search means for searching the search target text for a candidate character string including a character string pattern that matches the generated replacement character string;
An error character string registration means for registering an error character string pattern that may occur in the generation of the search target text;
An error character string detection means for detecting a registered error character string pattern registered in the error character string registration means in a portion different from the search character string in the candidate character string;
A priority according to the possibility of matching with the search character string is determined based on the detection frequency of the registration error character string pattern, and in response to the detection of the registration error character string pattern in the candidate character string, the priority Priority giving means for giving a degree to the candidate character string;
A text search apparatus comprising: The text search apparatus according to claim 1, wherein the degree of coincidence is expressed based on a ratio between the number of characters in the search character string and the number of characters remaining in the replacement character string that are not replaced .   The replacement character string includes a misrecognition allowable character, and the candidate search unit includes a unit that regards the misrecognition allowable character included in the generated replacement character string as an arbitrary character in the search target text. The text search apparatus according to claim 1, wherein the search is performed.   The replacement character string includes a misrecognition allowable character, and the candidate search unit includes a unit that regards the misrecognition allowable character included in the generated replacement character string as two arbitrary characters in the search target text. The text search apparatus according to claim 1, wherein the search is performed.   The replacement character string includes a misrecognition allowable character, and the candidate search means sets two consecutive misrecognition allowable characters included in the generated replacement character string as an arbitrary character in the search target text. The text search apparatus according to claim 1, wherein the search is performed with a means to be regarded.   The error character string registration means further stores the detection frequency of the registered error character string pattern, and the priority assigning means determines the priority according to the detection frequency stored in the error character string registration means. The text search apparatus according to claim 1.   4. The text search apparatus according to claim 3, further comprising candidate character string display means for displaying the candidate character string in accordance with the priority. In a text search method for performing search processing on a search target text on a computer based on a search character string,
The step of accepting input of the degree of coincidence designated between the search character string and the search character string and its replacement character string, as an input means included in the computer;
The replacement character string generating means of the computer generates a replacement character string of the search character string according to the input search character string and the degree of matching;
A candidate search means included in the computer for searching the search target text for a candidate character string including a character string pattern that matches the generated replacement character string;
A step of registering an error pattern that may occur in the generation of the text to be searched by the character string registration means of the computer;
An error character string detection means that the computer has, a step of detecting a registered error character string pattern registered in the character string registration means in a portion different from the search character string in the candidate character string;
The priority giving means of the computer determines priority according to the possibility of matching with the search character string based on the detection frequency of the registration error character string pattern, and the registration error character string in the candidate character string In response to pattern detection, assigning the priority to the candidate character string;
A text search method characterized by comprising:

Images