JP2010044597A - Information processor, information processing method, information processing system, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To estimate similarity of phrases expressed by a plurality of types of characters. <P>SOLUTION: An information processor 110 includes: a character decision part 118 which receives a character sequence showing a full name, and decides the type of characters configuring the received character sequence; a different notation acquisition part 120 which generates different notations in which the character sequence is described in the different type of characters including ideogram or phonogram from the character sequence, and which generates notation vectors including at least two different types of full name notations including the phonogram; a similarity calculation part 122 which executes different similarity determination in response to the type of characters, and which calculates scores giving the scale of similarity for the elements of the notation vectors; and a similarity score calculation part 124 which calculates similar scores for full name candidates by using the scores calculated by the similarity calculation part 122. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

The present invention relates to a technique for determining similarity between first and last names, and more specifically, from first and last names written in Indo-European, Afro-African, Japanese, Chinese, Korean, etc. The present invention relates to an information processing apparatus, an information processing method, an information processing system, and a program for determining sex.

In recent years, international human activities have become active due to advances in transportation and network technology. When an individual or a company conducts various activities across multiple countries, the person's first and last name and the name of the company are described and various procedures are performed. As international activities increase, it may be necessary to perform various searches using the information in countries other than the countries where individuals have processed. As an example, in the case of international money laundering, etc., the identity of the person who created, remittance, withdrawal, etc. in the bank account etc. is shared in multiple countries, and different cultures The case where it is necessary to recognize the first and last name corresponding to the category without losing the identity of the first and last name can be given.

Last name search is not always necessary for international action. For example, in Japan, it is necessary to determine whether bank accounts, health insurance, etc. indicate the same person, so what kind of similar surname can be generated from the acquired surname, etc. Therefore, it is necessary to generate the accuracy and the likelihood of the possibility with high accuracy and efficiency.

Surnames written in Indo-European and Afro-Asian languages, including English, Cyrillic, Greek, etc., are written only in phonograms such as alphabets, so you can search using strings. The search is relatively easy including a similar search using a character string. Note that a method of comparing names by comparing the names of the names described in alphabets and calculating the similarity score of the names is used for name search. As a system / method for searching for a person name written in alphabet, for example, an automated person name for searching for a person name written in alphabet is described in US Pat. No. 6,963,871B1 (Patent Document 1). List search systems.

In addition, the URI specified in http://publibfp.boulder.ibm.com/epubs/pdf/c1912860.pdf (Non-patent Document 1) uses the similarity of person names written in Indo-European languages. A Global Name Analytics (GNA) system that searches for a person's name is also disclosed.

By the way, in the cultural sphere using so-called ideographs, which can be exemplified as CJKV (Chinese, Japanese, Korean, Vietnam), there are allo-characters, and even the same character depends on the character string sequence. The phonetic characteristics may be different. For this reason, there is a problem different from the first name search in the Indo-European cultural sphere when the similar determination of the first name and the second name including the ideogram is automatically executed with sufficient accuracy. As such problems, for example, the following problems can be exemplified.

(B) Furthermore, in the cultural sphere referred to as CJKV, there are old, new, and variant fonts that originally represent the same ideogram and have the same phonetic characteristics. For example, “Sai”, “ There may be a plurality of ideographic characters for the first and last names written in the same phonetic character, such as “齊”, “sai”, “齋”. Therefore, it is not possible to automatically determine the identity of phonetic characters and the identity and similarity of ideographic characters.

(B) In addition to the above-mentioned problems, for example, when Japanese is written in Roman letters, it corresponds to the phonetic character string “Taro”, “Taro”, “Taroh”, “Tarou”, etc., or “Tomoro” and “ There are fluctuations in the phonetic notation exemplified as “Tomoroo”, “Tomoe”, “Tomomo”, and “Tomomo”.

(C) Furthermore, when processing a large amount of character information at present, an optical reading system such as OCR has been introduced, and automatic character recognition has become widespread by comparing images input from image readers with character templates. is doing. In this case, in consideration of misrecognition by OCR, the appearance of similar characters such as “K”, “K”, “」 ”and“ 萩 ”,“ wealth ”and“ 冨 ”, etc. are considered. There is a need. In addition, these characters that are similar in appearance need to take account of transcription mistakes and misprints by business staff, etc., and automatically determine the first and last names of languages that have both ideograms and phonograms using a computer That was difficult.

  Up to now, various techniques for searching for first and last names for kanji strings, such as Japanese, are known. For example, Japanese Patent Application Laid-Open No. 11-338859 (Patent Document 2) is a name input device that converts a name input in kana to romaji and inputs the name, and manages the correspondence between the name spelled in kanji and the kana. In addition, a name dictionary for managing the reading attribute information indicating whether or not it is necessary to perform prescribed romaji conversion for each character of the reading, and the name indicated by the input reading by searching the name dictionary Search means for acquiring the reading attribute information of the name, selection means for selecting one name from the names searched by the search means, and the reading attribute of the name selected by the selection means Name conversion, characterized by comprising conversion means for converting the input furigana to romaji while performing the prescribed romaji conversion according to the instructions of the information It discloses an apparatus.

  Further, in Japanese Patent Laid-Open No. 2000-251017 (Patent Document 3), a first storage means storing a “name” character string representing the name of an organization or an individual, a “last name” character string representing a surname of the individual's first name, and Second storage means for storing a "name" character string representing a name together with type data representing its type; and a third storage means for storing a "general name" character string representing a general name together with type data representing the type; A fourth storage means for storing a word dictionary for collation used for collation of word search, and a constituent element of a “name” character string in the first storage means is a character string in the second storage means or the second 3. It is determined whether or not it matches the character string in the storage means, and the “name” character string for which the matching condition is satisfied is stored in the fourth storage means as a matching word dictionary together with the type data corresponding to the matching destination character string. Remarks A generation unit describes the word dictionary creation device being characterized in that comprises a.

Further, Japanese Patent Laid-Open No. 2007-305046 (Patent Document 4) discloses a first example in which a kanji is associated with a phonetic character representing a reading of the kanji.
Means for storing the information in advance, a kanji representing a person's first and last name, and a second included in the person's e-mail address
The acquisition means for acquiring the information, the generation means for generating a kanji reading candidate representing the first and last names based on the first information, and the second information and the candidate are collated An information processing apparatus is disclosed that includes a determination unit for determining the reading of the kanji representing the first and last names based on the result.

  Japanese Patent Laid-Open No. 2004-318699 (Patent Document 5) searches for a word that matches the word in the surname dictionary from the beginning of the input name. As a result of the search, if the type is a corporate type or a job type, the process is terminated because it is not an individual at that time. When the type is a surname, the name dictionary is searched for a word that matches the input name. Next, narrow down the candidates for surnames and surnames to one. Finally, a nominal analysis method for determining whether the narrowed solution is an individual is described.

  Furthermore, Japanese Patent Laid-Open No. 10-171799 (Patent Document 6) discloses a surname / name analysis method for assigning a surname / name delimiter and a character delimiter for each kanji to an individual name and a reading that are input without a delimiter , Using the surname dictionary where surnames are registered and the surname dictionary where surnames are registered, the entered surname is divided into surnames and surnames, and the surname or one of the surnames is registered using a character dictionary in which the readings for the characters are registered However, if it is not in the surname dictionary, the surname dictionary, or both are not in the surname / name dictionary, the kana is assigned to each kanji by reading the kanji for each kanji. Describes a surname analysis method that determines whether a solution is correct based on character delimiter information for resolving ambiguity.

US Pat. No. 6,963,871B1 JP 11-338859 A Japanese Unexamined Patent Publication No. 2000-251017 JP 2007-305046 A JP 2004-318699 A JP-A-10-171799 http://publibfp.boulder.ibm.com/epubs/pdf/c1912860.pdf

Patent Documents 2 to 6 each disclose a technique for automatically analyzing first and last names by a computer. However, in the technique described in Patent Document 2, the operator executes association between phonograms and ideograms, and the computer automatically associates the ideogram sequences with the phonogram sequences. It does not disclose technology. Further, although the technique described in Patent Document 3 makes it possible to select a customer name by using a dictionary in consideration of the possibility of similarity and how to apply the similarity, It does not describe at all whether the first name and the last name are automatically selected, and is not intended to associate ideograms and phonograms.

Further, Patent Document 4 selects a phonetic character sequence that is applied to kanji used for first and last names as the one that is most frequently applied to kanji, and associates the ideogram with the phonetic character. Is what you do. However, no determination is made as to whether or not the ideographic character sequence to be processed at that time should be represented by the most frequent ideographic character sequence. Moreover, since patent document 4 requires the input of the information which suggests a full name, such as e-mail address, with the surname in kanji notation, about the correspondence between a mute character and an ideogram, a phonetic character or an ideogram When a first name, last name, or last name is entered alone, the correspondence between phonograms and ideograms cannot be uniquely determined.

Furthermore, there is no correspondence as to which type of character should be used when there is a mistake in the first name, transcription mistake, or difference in character type. For example, "Shun Takada: Takashi Takada" and "Takada Ken : No reference is given to the similarity between ": Takada Takeshi" and it is not intended for ideographic character-phonetic character conversion including the possibility of erroneous conversion in OCR. The sequence also does not associate different first and last names as possible candidates.

Furthermore, also in Patent Documents 5 to 6, kanji and reading are registered in association with frequencies, and conversion between them is performed, and correspondence between ideograms and phonograms using only the appearance frequency. It does not generate surname candidates that may not be associated with the input ideographic character sequence or phonetic character sequence at the appearance frequency.

In addition, the techniques described in Patent Literature 1 to Patent Literature 6 can be used even when the first and last names are input in any phonetic character including Indo-European, Afro-Asia, Hiragana, Katakana, Hangul, etc. Even when input is performed using ideograms, it is not an issue to determine a surname that includes the most likely ideogram sequence or phonogram sequence and that may be the input surname.

In other words, up to now, the most likely first name surname candidate is generated for the first name surname input in the ideogram sequence or phonetic character sequence, and the notation of other languages is considered for the last name input. In addition to the same or similar candidate names, there is a need for a technique that can determine dissimilar names in relation to phonetic characteristics.

The present invention has been made in view of the above-mentioned problems of the prior art. In the present invention, the information processing apparatus receives a character sequence describing a person's name, and displays a first and last name composed of at least two different phonetic characters. Or a notation vector including at least two types of phonetic characters and ideograms.

For the elements of the notation vector, the degree of similarity is calculated with the kanji sequence or kana sequence managed by the information processing apparatus, and a score sub-matrix including the scores calculated for a plurality of first and last names included in the notation vector is generated. In the present invention, for the score sub-matrix, a similarity score is calculated in consideration of the similarity between at least a plurality of first and last names, and the similarity is crossed across languages for the phrases constituting the first and last names. Use with the judged similarity measure.

In the present invention, first and last name representations of at least two phonetic characters are generated from a character sequence, and in a specific embodiment, a kana sequence and an alphabet sequence. When the character sequence is a kana sequence, a surname or first name notation by ideographic characters such as kanji is further given.

The score sub-matrix is integrated with respect to last name or first name phrases to form a scoring matrix. In the present invention, the scoring matrix is used to determine the likelihood of transliteration for first name and last name, and the first name and last name candidates are integrated by integrating the last name and first name registered in the scoring matrix. Used to generate similar first and last name candidates that are neither the same sound nor the same character.

According to the present invention, it is possible to determine similarities for a plurality of different notations using an input character sequence to generate first and last name candidates. Information processing device, information processing method, and information processing capable of generating similar first and last name candidates including the possibility of mistyping, transcription mistakes, abbreviations, conversion errors, etc. Systems and programs can be provided.

Also, the likelihood in multiple languages is determined for surnames entered in phonetic character sequences or ideogram sequences, and weighted candidates for surnames in the target phonogram sequence or ideogram sequence are determined. An information processing apparatus, an information processing method, and a program that can be provided can be provided.

In addition, perform the same / similarity of surnames across multiple languages such as Indo-European, Afro-Asian, Japanese, Chinese, Korean, Vietnamese, etc., and also determine dissimilar surnames with phonetic characteristics An information processing apparatus, an information processing method, and a program capable of performing the above can be provided.

  Hereinafter, although this invention is demonstrated using embodiment, this invention is not limited to embodiment mentioned later. FIG. 1 is a functional block diagram of an information processing system 100 according to this embodiment. The information processing system 100 shown in FIG. 1 includes a server 110, a client computer (hereinafter simply referred to as a client) 150, a server computer (hereinafter simply referred to as a server) 110, and a client 150. The client 150, which is configured to include the network 112 to be connected, receives input of a kana, kanji, or a surname, first name, or surname surname written in Indo-European, Afro Asian, etc. Accept, for example, send a search request to server 110 using HTTP protocol or the like.

  In this embodiment, the term “kana” is defined as a thing including katakana and hiragana. The server 110 receives a full name query request via a network 112 including a LAN, the Internet, etc., and executes a full name generation process. When the server 110 receives other character notation corresponding to the received character sequence using the character sequence included in the received first and last name query request, for example, the kana sequence, A kanji or kanji sequence, which is an example of a sequence and ideogram, is generated and used for first name surname search.

  The server 110 may use a CISC architecture microprocessor such as PENTIUM (registered trademark), a PENTIUM (registered trademark) compatible chip, or a RISC architecture microprocessor such as POWER PC (registered trademark). A single-core or multi-core configuration can be adopted. Examples of the operating system (OS) implemented by the server 110 include WINDOWS (registered trademark) 200X, UNIX (registered trademark), and LINUX (registered trademark).

  In addition, the server 110 executes server programs such as CGI, servlet, APACHE, and IIS, which are implemented using programming languages such as C ++, JAVA (registered trademark), JAVA (registered trademark) BEANS, PERL, and RUBY. Then, various requests are processed through a network (not shown). The server 110 can be implemented as a web server, or can be a dedicated server that enables distributed computing using CORBA (Common Object Resource Broker Architecture).

  The client 150 is implemented using a personal computer or a workstation. The microprocessor (MPU) implemented by the client 150 may include any single-core processor or dual-core processor that has been known so far. it can. The client 150 can be installed with an OS such as WINDOWS (registered trademark), UNIX (registered trademark), LINUX (registered trademark), or MAC OS (registered trademark).

  If the client functions as a web client, it uses browser software such as Internet Explorer (registered trademark), Mozilla, Opera, and Netscape (registered trademark) Navigator to access the web server using the HTTP protocol. To do. In addition, when using a distributed computing environment such as CORBA, the client 150 can also implement a dedicated client program for mutual communication with the server 110.

  Hereinafter, the functional block configuration of the server 110 will be described in detail. The functional block of the server 110 is a functional unit realized on the server 110 by expanding the program in the RAM mounted on the server 110 and causing the CPU or MPU to execute the program and cause the server 110 to function as each functional block. It is.

  The server 110 includes a network adapter 114, a character sequence acquisition unit 116, and a character determination unit 118. The network adapter 114 receives the full name query request, processes the transaction at the TCP / IP level, and passes the full name query request to the processing unit that executes the full name candidate generation process of the present embodiment. The character sequence acquisition unit 116 acquires from the payload of the received packet a character sequence including a surname notation, a surname notation or a surname notation and a name notation included in the surname surname query request, and registers it in an appropriate memory as a character sequence to be processed. To do. In this embodiment, the surname notation means a character string corresponding to the surname (sir Name) of the name, and the name notation means a character string corresponding to the name of the name (Second Name). Character sequences that may contain either or both are referred to as full name notation.

  The character determination unit 118 analyzes the characters that constitute the registered character sequence, and determines whether the character sequence is, for example, katakana, hiragana, kanji, or alphabet. Character analysis can be executed by determining a character code, and any code system such as SJIS, GB, or UNICODE can be used. However, the use of UNICODE in performing a global first and last name search can appropriately correspond to the character type to be determined, and hereinafter, when performing character code comparison, UNICODE (such as UNICODE 2.0) is used. The comparison shall be performed. The character determination unit 118 performs character determination of the received character sequence, identifies the language area of the received character sequence, and uses it for subsequent processing.

  Further, the server 110 is implemented including an other notation acquisition unit 120, a similarity calculation unit 122, and a similarity score calculation unit 124. The other notation acquisition unit 120 determines the received character sequence, and when it is determined that the received character sequence is kana, in the present embodiment, transliteration processing is executed to acquire alphabet and kanji notation. Alphabetic notation includes Hebon, ceremonial, Roman characters such as ISO 3602, and surnames or first names in Indo-European speaking. In the present embodiment, other notations such as Greek letters, Cyrillic letters, and Arabic letters can be acquired in addition to the alphabet depending on the specific purpose of the information processing system 100.

  For example, when the other notation acquisition unit 120 determines that the character sequence is a kana sequence, the other notation acquisition unit 120 refers to the phonogram conversion table 130 and converts the character sequence to an ideogram corresponding to the received character sequence, more specifically. In the embodiment, the intention processing for converting to kanji is executed. The phonetic character conversion table 130 can be configured as a first name surname dictionary that intends first name and last name and registers kana notation in association with possible kana notation.

  Furthermore, the phonetic character conversion table 130 according to the present embodiment registers kanji characters that may be used for personal names regardless of cultural spheres, and in addition to the phonetic character-ideographic character correspondence table, the corresponding ideographic characters are displayed. It includes cultural area judgment data that may be attributed. For example, when the input character sequence is “Hari”, the other notation acquisition unit 120 refers to the phonetic character conversion table 130 to acquire “Zhang” as the personal name kanji, and uses the cultural zone determination data as It is also possible to generate a code indicating that the “Zhang” may belong to a Japanese-speaking area, a Chinese-speaking area, or a Korean-speaking area, and assign it to a kanji.

  Further, when the received character sequence is kana, the other notation acquisition unit 120 generates an alphabetic sequence in addition to acquiring kanji. This process is executed by kana-alphabet conversion using a hidden Markov model when the character sequence is kana, and when it is determined that the character sequence is kanji code, the phonetic character conversion table 130 is used. The kana notation is acquired, and the kana notation is converted into an alphabet, thereby registering a character sequence as a notation vector of {kana, alphabet, kanji}.

  Further, when the character sequence is alphabet, the server 110 does not execute the conversion process to kanji, and generates a notation vector while leaving the {kanji} field of the notation vector blank. When the character sequence is alphabetic, the kana notation is generated by referring to an alphabet-kana correspondence table used when generating a transition probability table, which will be described later, and converting within a range where kana corresponding to the alphabet sequence can be searched. be able to. Moreover, it can produce | generate by processing kana as a target series using the hidden Markov model mentioned later.

  When an alphabetic sequence is input as a character sequence, the similarity is calculated by the GNR process. Therefore, even if no kana or kanji is generated, the process using the similarity can be executed without any inconvenience.

  After the other notation acquisition unit 120 acquires a set of notation vectors that may correspond to the character sequence, the notation vector is passed to the similarity calculation unit 122. The similarity calculation unit 122 calculates a similarity score for each of alphabets, kana, and kanji included in the notation vector using a plurality of methods.

  Among the character types included in the notation vector, the similarity of kanji is determined by referring to the phonetic character conversion table 130 and the like so that the kanji included in the notation vector matches the character code registered in the phonetic character conversion table. For example, when the character codes are completely matched, 1.0 is given, and when they are not completely matched, a binary reference value giving 0 is given. be able to.

  Furthermore, the server 110 generates a similar first name surname, a homonym character, a homonym character, or a homonym character that is neither a homonym nor a homonym, but includes a transcription mistake, a conversion error due to OCR, and the like, The different character table 134 and the same / dissimilar Kanji table 136 are managed. The above-mentioned similar model table 132 and variant script table 134 provide multi-value reference values from 0 to 1 in consideration of variant characters and error incentives other than binary reference values for the same / dissimilarity of kanji. Can also be used to

  The similarity calculation unit 122 also acquires the similarity for the alphabet sequence. When calculating the similarity of the alphabet sequence, the similarity calculation unit 122 extracts the alphabet sequence and passes it to the GNR processing unit 160 configured as the processing unit of the server 110 in the specific embodiment. The similarity with the alphabet name registered in 160 is acquired. For the similarity determination process executed by the GNR processing unit 160, for example, methods disclosed in detail in Patent Document 1 and Non-Patent Document 1 can be used.

  In other embodiments, the GNR processing unit 160 instead of the GNR processing unit 160 can be configured as a GNR server 140 that exclusively executes GNR processing. In this case, the similarity calculation unit 122 sends an alphabet sequence for determining the similarity to the GNR server 140 via the network 112 and acquires the similarity as a response from the GNR server 140.

  Furthermore, the similarity calculation unit 122 calculates the similarity between the kana sequence and the kana sequence. For this purpose, the similarity calculation unit 122 extracts a kana sequence code from the notation vector, and performs comparison with the kana sequence registered in the phonetic character conversion table 130. In still another embodiment, the alphabet sequence generated from the kana sequence is sent to the GNR processing unit 160 or the GNR server 140, and the similarity value provided by the GNR processing is acquired to obtain the kana sequence. Similarity can be acquired.

  When performing kana sequence comparison, the similarity determination unit 122 refers to the phonogram conversion table 130 for a character string included in the character sequence, and generates a possible reference kana sequence. For example, when “Takeshi” is input as the character sequence, the associated kanji character “Ken” is extracted, the “Ken” that is the kana sequence associated with each kanji code is extracted, and the reference kana sequence is extracted. Generate as Then, the similarity determination unit 122 passes the generated “ken” to the other notation acquisition unit 120, generates “Ken” that is a corresponding alphabet sequence, and can calculate the score of the target sequence and the reference sequence. To do.

  Thereafter, the code of the target kana sequence is executed with respect to the code of the reference kana sequence using sequential forward comparison or the like, and for example, (total number of matched character codes) / ( The score is calculated as the total number of character codes of the target kana sequence and the reference kana sequence. Each calculated set of scores is registered as a score sub-matrix.

  In this embodiment, when a corresponding character code such as kanji or kana cannot be found, a character template is prepared as image data or the like so as to be used in, for example, OCR. On the other hand, a bitmap comparison with a character template or the like can be performed to determine the degree of similarity. The similarity calculation unit 122 generates score vectors for all sets of reference notation vectors generated by the other notation acquisition unit 120 and stores them in an appropriate memory such as a RAM, an L2 cache, or an L3 cache.

  Server 110 further includes a similarity score calculation unit 124 and a first and last name candidate generation unit 126. The score sub-matrix generated by the similarity calculation unit 122 is associated with the notation vector and sent to the similarity score calculation unit 124. The similarity score calculation unit 124 registers the received sub-score matrix in units of phrases constituting the first name or last name, and generates a scoring matrix. In the described embodiment, the scoring matrix is registered as a set of 3 × 3 score sub-matrices, and the score sub-matrix extracted for the surname and the score sub-matrix extracted for the first name are the surname-first name Such. It is defined as a square matrix arranged in the set order.

  Thereafter, the similarity score calculation unit 124 calculates a similarity score using each element value of the generated score sub-matrix. The calculation of the similarity score can be obtained by calculating the inner product using the elements of the score vector and an appropriate weight vector, or by extracting and calculating the elements according to a set rule. More details will be described later.

  The server 110 also includes a first and last name similarity calculation unit 128 implemented as an independent processing unit of the server 110 or a functional module of the first and last name candidate generation unit 126 described later. The first and last name similarity calculation unit 128 extracts the first and last name candidate having the maximum total similarity score from the first and last name candidates generated by the similarity score calculation unit 124, or the top ten first and last name candidates in other embodiments. Are generated as similar first and last name candidates that have different phonetic characteristics but are similar in appearance or appearance.

  The first and last name candidate generation unit 126 receives the output of the first and last name similarity calculation unit 128, generates a first and last name candidate list including the first and last name candidates or similar first and last name candidates corresponding to the received character sequence, and returns it to the client 150 as a table format. Execute the process.

  The full name candidate list at this stage can take into account the first name and the second name of the same sound and different letters corresponding to the similarity determination criteria, and can provide sufficient results for a specific application. Further, when the first and last name candidate generation unit 126 is configured to include the first and last name similarity calculation unit 128 as its module, the first and last name similarity calculation unit 128 determines the first and last name candidate having a high total similarity score from the first and last name candidate list. First and last name candidates that are similar in character and appearance and may have transcription mistakes, recognition errors, typographical errors, and the like can be generated and added to the first and last name candidate list.

  In still another embodiment, the first and last name similarity calculation unit 128 can be implemented as an independent processing unit of the server 110. In the other embodiment to be described, the first and last name similarity calculation unit 128 receives the first and last name in kanji from the similarity score calculation unit 124 or the kanji sent from the client 150, and may have a variant or appearance similar. A process of generating a given first and last name candidate and returning it to the client 150 together with the similarity index value is executed. The processing of the present embodiment will be described in detail later so that it can be performed by those skilled in the art. However, Japanese Patent Application No. 2008-117538 (IBM Atonie Docket No. JP9080054), Japanese Patent Application No. 2008-141209 (IBM Atonie Docket Number: JP9080081) and Japanese Patent Application No. 2008-168087 (IBM Atonie Docket Number: JP9080115) may be incorporated in their entirety as part of this specification. .

  FIG. 2 is a diagram illustrating a processing sequence 200 for similarity generation executed by the server 110 according to the present embodiment. A character sequence that may be input from the client 150 corresponds to a specific application and may be any one of alphabet, kana, and kanji. In yet another embodiment, there is a possibility of an Afro-Asian-speaking character string such as Arabic. The server 110 determines the character code of the character sequence received from the client, and generates another notation expressed by a character code other than itself, for example, the alphabet sequence 230 and the kanji 220 when the character sequence is Kana 210. And sent to the similarity calculation unit 122.

  The similarity calculation unit 122 calls the GNR similarity search 240 and requests processing of the alphabet sequence 230 as indicated by a box 240. As for the kana sequence 210, after conversion to an alphabetic sequence, as shown in box 250, a GNR similarity search can be executed, a kana-kana character code comparison can be performed, and In other embodiments, a bitmap comparison with a character template can be applied.

  In the case of the Chinese character 220, as shown by a box 260, a binary reference value can be given by comparing character codes, or the respective values 132, 134, and 136 shown in FIG. A score indicating can also be calculated. Further, as indicated by a box 260, when character codes cannot be compared, similarity can be determined by performing bitmap comparison using a character template or the like. By using the processing sequence 200 shown in FIG. 2, it is possible to execute an efficient and accurate name search using an appropriate similarity calculation process regardless of the character sequence received from the client 150.

  FIG. 3 shows a flowchart of an information processing method for generating first and last name candidates according to this embodiment. The process shown in FIG. 3 starts from step S300, and a character sequence is received from the client 150 in step S301. In step S302, the character code determination of the received character code is executed. In the embodiment to be described, it is assumed that one of kanji, kana, and alphabet is judged, but other character sequences to be received such as Korean, Arabic, Greek, Cyrillic, etc. correspond to the purpose and application of generating first and last name candidates. There is no particular limitation.

  In step S303, other notations are acquired and registered as kanji, katakana, and alphabet notation vectors. If there is no notation, for example, if the character sequence is an alphabet notation, kanji is supported without generating kanji from the alphabet sequence. Leave the element to be blank. In step S304, the character type of the character sequence is determined, and kanji, alphabet (kana), or kana is determined. If the character sequence is kanji, the process is passed to step S310, and if it is kana, the process is passed to step S305 or step S309 corresponding to the process sequence to be adopted.

  If the character sequence is alphabetic, the process is passed to step S306. In step S305, the kana sequence is converted to alphabet, and the alphabet sequence is transferred to step S306 and sent to the GNR processing unit 140 or 160. In step S307, a process of receiving similarity from the GNR processing unit 140 or 160 and passing the process to step S308 is executed.

  On the other hand, in step S309, character codes comparison between Chinese characters is performed, or bitmap comparison with character templates is executed, and the similarity based on the similarity determination is passed to step S308. In step S308, a score is calculated using the similarity for at least two notations corresponding to the cultural sphere of the character sequence. This is because, for example, when the character sequence received from the client 150 is an alphabet, it is not forcibly converted to Kanji, and the Kanji field of the notation vector is left blank. Further, in step S308, a similarity score that gives an index of similarity with respect to the score sub-matrix is calculated from the calculated score to enable extraction of first and last name candidates, and a first and last name candidate list is generated. When the generation of the first and last name candidate list is completed, the process is passed to step S310 to end the process.

  The process of FIG. 3 allows the user of the client 150 to acquire the first and last name candidates corresponding to the inputted character sequence together with the certainty, and determines what kind of first and last names are candidates from the character sequence. It becomes possible to do.

  FIG. 4 shows a flowchart of the kana-alphabet conversion process executed by the other notation acquisition unit 120 of this embodiment. The process in FIG. 4 starts from step S400, and a kana sequence is received in step S401. In step S402, the Kana sequence is phoneme decomposed. The phoneme decomposition is performed corresponding to the phonetic character sound, and the long sound and the prompt sound are assumed to be included in the immediately preceding phoneme. For example, when the kana sequence = “infomersion”, the phoneme decomposition is “i” + “n” + “fo” + “mae” + “sho” + “n”.

  In step S403, the conversion likelihood χ is calculated using the transition probability model for the phoneme decomposition result, and in step S404, the alphabet sequence having the highest conversion likelihood χ is acquired. The transition probability model can be given by language analysis because it analyzes phonemes and phoneme sequences, applies a hidden Markov model with kana as an observation series and alphabet as a target series. In step S405, the alphabet conversion result is passed to the similarity calculator 122 to request similarity calculation, and the process ends in step S406.

FIG. 5 shows a state transition diagram of alphabet conversion using the hidden Markov model described in FIG. The character sequence to be converted is “information”, and the other notation acquisition unit 120 registers the phoneme-phoneme transition probability πi and the transition probability p _ij between phonemes for each phoneme as a transition probability table. Apply alphabet conversion. As shown in FIG. 5, the conversion likelihood is given by log likelihood, and each log likelihood is accumulated over phonemes and calculated using the conversion likelihood χ given by the following equation (1): The alphabet with the highest conversion likelihood is output to the similarity calculation unit 122.

  FIG. 6 shows an embodiment of the similar model table 132, the variant character table 134, and the same / dissimilar kanji table 136 described in FIG. The value shown in FIG. 6 is the cost corresponding to the risk of determining that the associated kanji is similar. Each table shown in FIG. 6 is used by the first and last name similarity calculation unit 128 to calculate a path cost of a difference for extracting similar first and last name candidates.

  In another embodiment, the similarity model table 132 and the variant font table 134 can also be applied to allow the similarity calculation unit 122 to calculate a multi-value reference value related to the similarity between Chinese characters. In this case, for example, using the cost value shown in FIG. 6, the score for the similarity of the kanji is set to score = 1 / (cost + α) (α is smaller than the minimum value of the cost and underflow overflow. Is an appropriate constant that can be avoided.). It should be noted that any other more appropriate formulation may be used to provide a multi-level reference value. Furthermore, in the embodiment shown in FIG. 6, each table 132, 134, 136 is configured as a single integrated table, but each table can be individually created and registered. it can.

  In the table of FIG. 6, the non-diagonal element of the table 610 is an area for registering the dissimilar cost of the same / dissimilar Kanji table 136, and the non-diagonal element of the table 620 corresponds to the variant character table 134. Further, the non-diagonal elements of the table 630 are areas corresponding to the similar model table 132 in which Kanji characters that are similar in appearance are registered. In the embodiment shown in FIG. 6, in order to explicitly describe the data structure, the cost for the same / dissimilar kanji is shown as being registered. The diagonal elements in the table of FIG. 6 correspond to the same table of the identical / dissimilar Kanji table 136.

  For the purpose of explanation, FIG. 6 will be described assuming that cost values are registered for all entries. In the preferred embodiment of the present embodiment, only the similar model table 132 and the variant character table 134 are explicitly mounted, and the cost corresponding to the same / dissimilar kanji table 136 is reduced in the program in order to reduce the data amount of the table. It can be set as a fixed value or the like to reduce the memory consumed for the table. In the case of this embodiment, the server 110 determines that the same / dissimilar kanji are the same / dissimilar based on the fact that they are not registered in the similar model table 132 and the variant script table 134, and executes the processing. Can do.

  Further, when the Chinese characters are dissimilar, 2 is assigned as the cost, and when the Chinese characters are the same, the cost is 0. In addition, Kanji characters that are similar in appearance are highly likely to cause transcription mistakes, misprints, OCR conversion mistakes, and the like, and thus there is a range in the conversion. For this reason, “萩” and “荻” are not completely different, but may be erroneously written, there is a possibility of conversion error due to OCR, and the phonetic characteristics are similar, so that cost = 0.3 is given. ing. On the other hand, although the Chinese characters “朋” and “Ming” are similar in appearance, the phonetic characteristics are completely different, so that even if they are similar in appearance, a higher cost = 0.4 is given.

  Also, for “wealth”, “冨”, “high”, “▲ high ▼” (under the hot pot, two horizontal lines between two vertical lines), “Sai”, “齊”, etc. Cost = 0.1 is assigned as a variant that tends to be abbreviated.

  FIG. 7 illustrates an embodiment of a score sub-matrix 700 that corresponds to a character sequence that the server 110 may enter in this embodiment. The row is a score vector corresponding to the notation vector generated by the other notation acquisition unit 120, and in the embodiment shown in FIG. 7, the element value is {Takeshi, Takeshi, Ken}. The columns are scores calculated corresponding to kana sequences and kanji extracted from the phonetic character conversion table 130 when the similarity calculation unit 122 executes score calculation.

  For example, in the case of the Chinese character “Ken”, kana “Ken” and “Takeshi” are assigned. As a result, the score vectors are reference notation vectors of {Takeshi, Takeshi, Ken} and {Ken, Ken, Ken}. Are generated, and a score sub-matrix 710 is generated. In the score sub-matrix 710, since the alphabet sequence and the kana sequence do not match, the degree of match = 0 is given. In addition, the kanji is matched in the registered reading range with respect to the character code of the kanji “Ken” and is also generated by alphabet conversion, so the degree of coincidence = 1.0 is given.

  On the other hand, the score sub-matrix 720 shows an embodiment of the score sub-matrix 720 when the server 110 receives the alphabet as a character sequence. The score sub-matrix 720 is when the server 110 receives “James” as a character sequence. In the score sub-matrix 720, “James” which is a kana sequence corresponding to the alphabet sequence is found, so that a notation vector is generated as {James, James, null}.

  In this case, for the reference notation vector, only the kana sequence “James” is registered for the alphabet sequence “James” in the described embodiment. = 1.0 is assigned, and the degree of similarity with the Chinese character is also left blank because the Chinese character is left blank.

  The score sub-matrix 730 shows an embodiment in the case where the server 110 receives “hair” of a Chinese character as a character sequence. For the Chinese character “hair”, “mo”, “ke”, “ge”, “mao”, etc. are registered as kana. On the other hand, in the alphabet conversion process, Mou, Mao, Ge, Ke and the like are extracted. In this case, in the illustrated score sub-matrix 730, the matching score = 1.0 is always assigned for the kanji characters, and for the others, the score / number generated from the notation vector including the corresponding alphabet and kana and the reference notation vector Unless it is a sub-matrix, the degree of coincidence = 0 is given.

  Further, in the present embodiment, a similarity score is calculated for each of the score sub-matrices 710, 720, and 730, and the similarity of the entire score sub-matrix is generated. FIG. 8 shows an assignment map 800 for explaining the assignment of scores corresponding to each notation when calculating a similarity score. In this embodiment, the similarity can be simply calculated as an average value of each score. However, paying attention to the fact that a unique judgment can be performed on the first and last names corresponding to each notation, in the present embodiment, a similarity score can be generated using a different process for each element. In the first embodiment, a similarity vector is generated from a score sub-matrix.

  The similarity vector is obtained by extracting the elements of the score vector in the set order. For example, in the embodiment described below, the similarity vector is given by the following expression (2).

また、本実施形態では、各要素値の割当てマップ８００の位置に対応して重み付けが与えられ、Ｓ１，．．．，Ｓ９にそれぞれ対応して、Ｗ１，．．．，Ｗ９が与えられ、重みベクトルについては、下記式（３）で与えられる。

In the present embodiment, weighting is given in correspondence with the position of each element value assignment map 800, and S1,. . . , S9, W1,. . . , W9, and the weight vector is given by the following equation (3).

スコア・サブマトリックス全体としての類似スコアは、各ベクトルの内積を使用して、下記式（４）で与えることができる

The similarity score for the entire score sub-matrix can be given by the following equation (4) using the inner product of each vector.

The weighting values are, for example, kana-kana match, kana-kanji match, alphabet-
It can be set as appropriate according to the specific application, such as when emphasizing kana matching. For example, when generating surname matching candidates in kana notation for Japanese, the weight of S5 is increased, -When generating surname candidates for kana match and kana-kanji match, the weights of S5, S6, S8, and S9 are increased. On the other hand, for foreigners who should place importance on alphabet-alphabet match and alphabet-kana match, the weighting of S1, S2, and S4 can be set high, corresponding to the generation of first and last name candidates as appropriate. It can be set by system setting or user selection.

  Hereinafter, the calculation of the similarity score of the score sub-matrices 710, 720, and 730 may be calculated using other methods. Although other embodiments will be described below, the mapping of element allocation is as shown in FIG. In particular, the difference between Japanese and foreigners including Chinese and Koreans depends on the match / mismatch between kana and kanji.

Therefore, it is possible to determine an element that reflects the feature of each element value without calculating the overall similarity score of the score sub-matrix, and generate a similarity score that enables cultural zone determination. For example, as described above, each element value of S5, S6, S8, and S9 including the correspondence relationship with the kanji notation is the feature value. For this reason, a threshold value TH _j is set for each element value, and if the degree of similarity in S8 and S9 exceeds the threshold value TH _j , it is determined that the person is Japanese and a similarity score is calculated.

Furthermore, in the second embodiment for calculating the similarity score, attention is paid to not generating kanji notation for character sequences related to foreigners, especially ethnic groups of Indo-European languages and Afro-African languages. Judgment is possible. For example, in FIG. 8, when S3 and S7 are null, similar scores for foreigners excluding Japanese, Chinese, and Korean nationals such as Japan, China, Taiwan, South Korea, Vietnam, and Malaysia are calculated. If the similarity between Kanji and Kanji is higher than the set threshold TH _k and the Kana-Kana similarity is lower than the set threshold TH _c , the similarity score of the Chinese / Korean people is calculated. Can be made. In the following formula (5), pseudo code for processing when the above judgment is used is described.

  In the above formula, cult_category is a cultural sphere identification variable for setting a cultural sphere identification value. The server 110 can pass the value of cult_category to another processing unit in accordance with the process implementation format, and execute the culture area identification process. In yet another embodiment, the result of the cultural sphere judgment that is executed by detecting characters that are characteristic of the cultural sphere such as name likelihood / national character, akusan decu, umlaut is used as auxiliary information, and the similarity score calculation process is changed. You can also

  FIG. 9 illustrates an embodiment of a scoring matrix 900 configured with the score sub-matrix as a partial matrix. The score sub-matrix constitutes a surname or first name generated by the server 110 by referring to an appropriate dictionary or the like, and is configured with a phrase that can be processed by the GNR server 140 or the GNR processing unit 160 as a unit. The scoring matrix 900 is generated by arranging score sub-matrices that may be used as phrases such as the first name and last name of the evaluation target in the order of recognition of the character sequence.

A similarity score for a particular character or phrase can be calculated using the above equation (4) or (5) using, for example, the 3 × 3 element of the score sub-matrix 910 of the scoring matrix 900. In order to indicate the position of the 3 × 3 score sub-matrix, the score sub-matrix is referred to as (Name _i , Name _j ), and the phrase units are shown in FIG. 9 as Name1, Name2, Name3,.・ As shown. In the above notation, Name1 corresponds to the surname “齊 Rattan”, Name2 corresponds to the surname “Saito”, and Name3 corresponds to “Nishito”, for example. Furthermore, names can also be registered in the scoring matrix 900 in a similar sequence, and in other embodiments, a scoring matrix for names can be generated separately, and similarity determination can be performed using the Western book method. . If the character sequence to be received is an alphabet, “James” can be used as the unit of Name1.

In the embodiment shown in FIG. 9 of the present embodiment, the score submatrix indicated by (Name _i , Name _{j ± 1} ) adjacent to the score submatrix 910 and the subelements 920 and 930 of the score submatrix are used. Thus, it is possible to determine the consistency of the culture name judgment of the first and last name sequence. That is, the sub-elements 920 and 930 correspond to the Kana-alphabet similarity of S2 and S3 as shown in the above formulas (4) and (5) corresponding to the element values corresponding to S2, S3, and S6 in FIG. , S6 Kanji-Kana matching similarity, for example, enables not only the similarity when generating a surname sequence, but also a join process, and a more accurate surname candidate can be generated.

With respect to the scoring matrix 900 shown in FIG. 9, the above formula (4) or (5) is used, and the surname and first name such as top 3 in descending order of similarity score for the score sub-matrix (Name _i , Name _j ) And the first and last phrases are concatenated to generate a total similarity score, and the total similarity score is the highest first-name candidate or the top K of the total similarity score (K is a positive integer) .) Is extracted and registered in the result list. At this stage, the server 110 can also pass it to the client 150 as a first and last name candidate list.

  Further, by using the scoring matrix 900 shown in FIG. 9, it is possible to determine whether or not the first and last names indicate the same person. For example, even if the kana sequences are different, if they are similar to the same kanji sequence, the probability Pr (kana sequence 1 | kana sequence 2) that is inherently another person is statistically determined. In addition, first and last name candidates can be displayed with a probability (1-Pr) of being the same person.

  In the present embodiment, it is possible to adopt a configuration in which a first name surname candidate including abbreviations, transcription mistakes, and possibility of misrecognition by OCR is generated and added to the first name surname candidate list. This full name candidate is an embodiment that can generate full name candidates that are neither the same sound nor the same character. In the present embodiment, a first and last name sequence is generated that is a candidate for a first name and last name that is highly likely to be abbreviated at the time of writing, is likely to cause a transcription mistake, or has a possibility of erroneous recognition of OCR.

  FIG. 10 is a flowchart of a process for extracting similar first and last name candidates that are neither the same sound nor the same character in the present embodiment. The process of FIG. 10 starts from step S1000. Note that the processing is not performed for the phrases constituting the first and last name sequence, but for each Chinese character in order from the beginning to the end of the character sequence. However, the determination may be performed in reverse order. In step S1001, the Chinese characters constituting the surname sequence are checked from the beginning to the end, the table shown in FIG. 6 is referred to, the kanji included in the surname sequence is referred to the variant table 134 and the similar model table 132, A reference first name surname candidate is generated and a route map is generated.

  At this time, the first and last name candidates are comprehensively extracted with last names and first names that may include kanji included in the first name and last name generated as the first and last name candidates, and with reference to the table of FIG. A surname that can be similar in appearance and replaced the similar kanji can be set as a similar surname candidate. Specifically, the surname = “Hagiwara” can be assumed to be “Hagiwara”, as well as “Sagano”, “Hagi” and other surnames. ”,“ 荻 ”, etc. are possible candidates. If there are kanji characters that cannot be matched between the first and last name candidates and the reference first and last name candidates, for example, by setting an appropriate large number such as “Hakubara” for “Hagiwara”, it is similar by increasing the cost. It can be excluded from first and last name candidates.

  The processing of FIG. 10 will be described using a specific surname / name sequence. The surname / name sequence with the highest similarity score explained in FIG. 9 is “Akira Sugano ▲ High ▼”. The case where is generated will be specifically described. For the surname sequence “Hagiwara Takataka”, “荻”, “Ming”, “萩”, and “朋” are costs less than or equal to the threshold values set in the similar model table 132 (0.3, 0.00, respectively). 4) Similar, “High” and “▲ High ▼” are assigned as cost variants of 0.1 as a variant and have a value that is less than or equal to the set threshold value. Is done.

  Also, for surname = “Sagano”, surname = “Sagano”, “Kashihara” is extracted, and the kanji characters “Sano” and “Hara” are neither similar nor variant, so the cost = 2.0 Is assigned. Note that the above-mentioned threshold value for costs can be set as appropriate in consideration of a specific application, accuracy, and processing efficiency. If the threshold is generally too low, similar name surname candidates are limited. If it is lowered and too high, the processing efficiency is lowered.

  When the first and last name candidates “Akira Sugano ▲ High ▼” and the generated “Hagiwara 朋 高” are determined by comparing the character codes in the sequence ranking step S1001, whether or not they are the same Kanji characters ( no) The process is passed to step S1002. If the same kanji is used, the cost = 0 is set, and the process is passed to step S1002. In step S1002, it is determined whether the kanji being processed is a variant. If it is not a variant character (no), the process is passed to step S1003. If it is a variant character (yes), the variant character table 134 is looked up in step S1008, the cost is acquired, and the process is passed to step S1003. . In step S1003, it is determined whether or not the character to be compared is similar in appearance and is registered as an entry that generates a transcription error, a misprint, or a misrecognition.

  If the kanji being processed is not registered in the entry, the process passes to step S1005, and if it is registered (yes), the similar model table 132 is looked up and registered in step S1009. The cost is acquired and the process is passed to step S1005. In step S1005, the registered costs are summed along the route map to calculate the route cost, the lowest name sequence with the lowest cost is selected and registered as a similar first name surname candidate for the first name surname candidate, and step S1006. To end the process. In another embodiment, the registered similar first name surname candidates are sent to the first name surname candidate generation unit 126 and added to the first name surname candidate list as additional information for the first name surname candidates generated by the processing described in FIG.

  FIG. 11 shows an embodiment of a route cost calculation process 1100 used in the process of FIG. In the route map 1110 shown in FIG. 11, the column direction is a first and last name candidate extracted by the process described with reference to FIG. 9, and the row direction is a first and last name sequence of similar first and last names generated by the process described with reference to FIG. 10. is there. As shown in FIG. 6, the cost between the kanji characters “０．３” and “荻” is 0.3, and the sounds are similar, but “hara” and “field” are Since it is dissimilar, it passes through the rectangular side of the route map 1110. Note that. For the rectangular piece, a path cost = 1.0 is given, and it can be associated with the table of FIG. In addition, the Chinese characters “Ming” and “設定” are set as 0.4, and the Chinese characters “High” and “▲ High ▼” are different character forms, so the cost = 0.1 is assigned. .

  In the route map shown in FIG. 11, when a cost equal to or less than the set threshold is registered, a route cost along the minimum cost route 1120 is calculated by traveling along a diagonal route. On the other hand, when the cost equal to or less than the set threshold is not registered, the side of the rectangle connecting the endpoints of the diagonal line is advanced. When the kanji strings do not match at all, the maximum cost route 1130 is used. As shown, it passes only the outermost side. For this reason, route cost = 2.8 is given by selecting a route in relation to the similarity of kanji from the first character to the last character and executing the cumulative calculation of the cost. The route cost can be used as a similarity index value indicating the similarity between first and last names.

  In addition, when the first and last name candidates are different from the reference first and last name candidates, a cost given by a large number is assigned at the stage where unmatchable characters are detected, so that they are excluded from similar first and last name candidates. Candidates such as “Hara” can be excluded. In addition, in the embodiment described, since the lower cost is a more likely surname candidate in the described embodiment, among the generated surname candidates, for example, a similar cost having a cost equal to or less than a set threshold value. Add the first and last name candidates to the first name and last name list to construct the final result list.

  In the embodiment shown in FIG. 11, the description has been given assuming that the similar first and last name candidates and the route map 1100 to be associated are generated in advance using the table of FIG. 6 and the route cost is calculated. In yet another embodiment, a transition probability that causes an error or transcription error between kanji and kanji is assigned, and a first name surname candidate registered in the column direction using the Vitabi algorithm and a first name surname candidate registered in the row direction. The first and last name candidates that give the route having the smallest log likelihood are generated as similar first and last name candidates and added to the result list.

  The processing shown in FIGS. 10 and 11 is configured as a module of the first and last name candidate generation unit 126 of the server 110, as well as a first and last name similarity calculation unit 128 that determines the first and last name similarity, and an independent processing unit. Can do. The first name surname similarity determination unit 128 receives the kanji sequence, executes each process, sets the received kanji character sequence in the column of FIG. 11, generates similar first name surname candidates, and sets them in the row. The route cost can be calculated and the result returned to the client 150.

  FIG. 12 shows a GUI 1200 when inputting a Chinese character sequence. For example, the client 150 may wish to receive a list of first and last name candidates from the server 110 and know if there are other potential first and last name candidates. In such a case, the client 150 can display the GUI 1200 shown in FIG. 12, input a kanji sequence to be targeted, and obtain possible first and last name candidates.

  In the GUI 1200 shown in FIG. 12, a display 1220 indicating that the GUI is used for determining the Kanji character string similarity is displayed in the display window 1210. The user of the client inputs the target kanji sequence in the input field 1230 and clicks the “OK” button to send the kanji sequence to the server 110. The server is configured to generate similar first and last name candidates similar to the first and last name = “Akira Sugano” sent from the client 150 by applying the above-described processing, and return the list to the client as a similar candidate list 1250. be able to.

  Whether the first and last name similarity calculation unit 128 is configured as a module of the first and last name candidate generation unit 126 or as an independent processing unit is appropriately selected according to the purpose in a specific application and mounted on the server 110. can do.

Further, by inputting only the first name, last name, or last name of the ideogram sequence or phonetic character sequence, in addition to the same or similar first name surname candidates related to the phonetic characteristics, the dissimilar first name surname candidates regarding the phonetic characteristics can be changed to CJKV. In addition, an information processing apparatus, an information processing method, and a program capable of determining similarity in consideration of multiple languages such as Indian and European languages can be provided.

  As described above, according to the present invention, it is possible to generate similar first and last name candidates by judging the similarity of a plurality of different notations using the inputted character sequence, and the generated first and last name candidates are not the same sound. It is possible to generate similar first name surname candidates including the possibility of misprints that are not the same characters, transcription mistakes, abbreviations, conversion errors, etc., and further, an information processing apparatus, information processing method, An information processing system and program can be provided.

  The functions of this embodiment are described in an object-oriented programming language such as C ++, Java (registered trademark), Java (registered trademark) Beans, Java (registered trademark) Applet, Java (registered trademark) Script, Perl, and Ruby. The program can be realized by a program executable by the apparatus, and the program can be stored in a device-readable recording medium such as a hard disk device, CD-ROM, MO, flexible disk, EEPROM, EPROM, and distributed. It can be transmitted over the network in a possible format.

  Although the present embodiment has been described so far, the present invention is not limited to the above-described embodiment, and other embodiments, additions, changes, deletions, and the like can be conceived by those skilled in the art. It can be changed, and any aspect is within the scope of the present invention as long as the effects and effects of the present invention are exhibited.

The functional block diagram of the information processing system 100 of this embodiment. The figure explaining the processing sequence 200 for the similarity generation which the server 110 of this embodiment performs. The flowchart of the information processing method which produces | generates the full name candidate of this embodiment. The flowchart of the kana-alphabet conversion process which the other description acquisition part 120 of this embodiment performs. FIG. 5 is a state transition diagram of alphabet conversion using the hidden Markov model described in FIG. 4. The figure which showed embodiment of the similar model table 132 demonstrated in FIG. 1, the variant character table 134, and the same and dissimilar kanji table 136. FIG. FIG. 6 is a diagram illustrating an embodiment of a score sub-matrix 700 using a notation vector corresponding to a character sequence that may be input by the server 110 in the present embodiment. The figure which showed the element allocation map 800 for demonstrating the production | generation process of the similarity vector produced when calculating a similarity score. FIG. 9 illustrates an embodiment of a scoring matrix 900 that includes a score sub-matrix as a partial matrix. The flowchart of the process which extracts the similar first name surname candidate which is not the same sound and the same character in this embodiment. The figure which showed embodiment of the route cost calculation process 1100 used by the process of FIG. The figure which showed GUI1200 in the case of inputting a Chinese character sequence.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 ... Information processing system, 110 ... Server, 112 ... Network, 114 ... Network adapter, 116 ... Character sequence acquisition part, 118 ... Character determination part, 120 ... Other notation acquisition part, 122 ... Similarity calculation part, 124 ... Similarity Score calculation unit 126 ... First name and last name candidate generation unit 128 128 First name and last name similarity calculation unit 130 130 Phonetic character conversion table 132 132 Similar model table 134 134 Different character table 136 136 Same / dissimilar kanji table 140 ... GNR server, 150 ... Client computer, 160 ... GNR processing section

Claims (21)

An information processing apparatus for calculating a first name surname similarity,
A character determination unit that receives a character sequence indicating a first and last name, and determines a character type constituting the received character sequence;
From the character sequence, other notations in which the character sequence is described in other character types including ideograms or phonograms are generated, and a notation vector including at least two different first and last name notations including phonograms is generated. A notation acquisition unit;
A similarity calculation unit that performs a different similarity determination in response to the character type and calculates a score that gives a measure of similarity for the elements of the notation vector;
A similarity score calculation unit that calculates a similarity score for the first and last name candidates using the score calculated by the similarity calculation unit.   If the character type of the character sequence is determined to be alphabet, the other notation acquisition unit generates the other notation using a phonetic character corresponding to the character sequence, and the character sequence is kana The information processing apparatus according to claim 1, wherein when the determination is made, the other notation is generated using ideograms.   The information processing apparatus manages a similar model table that associates ideographic characters having similar appearances, and a different character table that registers different characters of the ideographic characters. The information processing apparatus according to claim 1, further comprising: a first name surname similarity calculation unit that calculates a route cost by looking up the table and the different character table to include appearance similarities and character differences.   Further, the first and last name candidates are selected using the similarity score, and sent to the sender of the character sequence as a first and last name candidate list, and the first and last name similarity calculation unit includes: The information processing apparatus according to claim 3, wherein the information processing apparatus is a module or an independent module of the information processing apparatus.   The similarity calculation unit generates a score sub-matrix in which scores for the elements of the notation vector to be evaluated are registered, and the similarity score calculation unit uses the score of the score sub-matrix to generate the score The information processing apparatus according to claim 1, wherein a similarity score that gives an index of similarity of the sub-matrix is calculated. An information processing method for determining similarity between first and last names executed by an information processing device, wherein the information processing device includes:
Receiving a character sequence indicating a first and last name and determining a character type constituting the received character sequence;
Generating another notation in which the character sequence is described with other character types including ideograms or phonetic characters from the character sequence, and generating a notation vector including at least two different first and last name notations including phonetic characters When,
Performing different similarity decisions in response to the character type and calculating a score to give a measure of similarity for the elements of the notation vector;
And a step of calculating a similarity score for the first and last name candidates calculated by the similarity calculation unit.   In the step of generating the notation vector, when it is determined that the character type of the character sequence is alphabet, the other notation using a phonetic character corresponding to the character sequence is generated, and the character sequence is The information processing method according to claim 6, further comprising the step of generating the other notation with ideographic characters when it is determined that   Further, it includes a step of calculating a path cost including appearance similarities and font differences by looking up a similar model table that associates the ideographic characters having similar appearances and a different character table that registers the different characters of the ideographic characters. The information processing method according to claim 6.   The information processing method according to claim 8, further comprising: selecting the first and last name candidates using the similarity score, and sending the selected first and last name candidates as a first and last name candidate list to a sender of the character sequence. The step of calculating the score includes generating a score sub-matrix in which scores for the elements of the notation vector to be evaluated are registered,
The information processing method according to claim 6, wherein the step of calculating the similarity score includes a step of calculating a similarity score that gives an index of the similarity of the score sub-matrix using the score of the score sub-matrix. . An information processing executable program for an information processing device to execute an information processing method for determining first name surname similarity, the information processing device comprising:
Receiving a character sequence indicating a first and last name and determining a character type constituting the received character sequence;
Generating another notation in which the character sequence is described with other character types including ideograms or phonetic characters from the character sequence, and generating a notation vector including at least two different first and last name notations including phonetic characters When,
Performing different similarity decisions in response to the character type and calculating a score to give a measure of similarity for the elements of the notation vector;
A computer-executable program for executing the step of calculating a similarity score for the first and last name candidates calculated by the similarity calculation unit.   In the step of generating the notation vector, when it is determined that the character type of the character sequence is alphabet, the other notation using a phonetic character corresponding to the character sequence is generated, and the character sequence is The program according to claim 11, including a step of generating the other notation with an ideogram when it is determined that   Further, it includes a step of calculating a path cost including appearance similarities and font differences by looking up a similar model table that associates the ideographic characters having similar appearances and a different character table that registers the different characters of the ideographic characters. The program according to claim 12.   The program according to claim 13, further comprising: selecting the first and last name candidates using the similarity score and sending the selected first and last name candidates to a sender of the character sequence as a first and last name candidate list. The step of calculating the score includes generating a score sub-matrix in which scores for the elements of the notation vector to be evaluated are registered,
12. The program according to claim 11, wherein the step of calculating the similarity score includes calculating a similarity score that gives an index of the similarity of the score sub-matrix using the score of the score sub-matrix. An information processing system for calculating first and last name similarity, the information processing system comprising:
A client computer connected via a network;
A server computer that receives a character sequence indicating first and last names from the client computer and determines similarity of the first and last names, the server computer comprising:
A character determination unit that receives a character sequence indicating a first and last name, and determines a character type constituting the received character sequence;
From the character sequence, other notations in which the character sequence is described in other character types including ideograms or phonetic characters are generated, and a notation vector including at least two different first and last name notations including phonetic characters is generated. A notation acquisition unit;
A similarity calculation unit that performs a different similarity determination in response to the character type and calculates a score for giving a measure of similarity for the elements of the notation vector;
A similarity score calculation unit that calculates a similarity score for the first and last name candidates using the score calculated by the similarity calculation unit.   If the character type of the character sequence is determined to be alphabet, the other notation acquisition unit generates the other notation using a phonetic character corresponding to the character sequence, and the character sequence is kana The information processing system according to claim 16, wherein when the determination is made, the other notation is generated with ideographic characters.   The information processing system manages a similar model table for associating ideographic characters with similar appearance, and a different character table for registering different characters of the ideographic characters, and for the other name surname candidate, the similar model The information processing system according to claim 16, further comprising a first and last name similarity calculation unit that calculates a path cost including look-up similarity and font difference by looking up the table and the variant font table.   Further, the first and last name candidates are selected using the similarity score, and sent to the sender of the character sequence as a first and last name candidate list, and the first and last name similarity calculation unit includes: The information processing system according to claim 18, wherein the information processing system is a module or an independent module of the information processing apparatus.   The similarity calculation unit generates a score sub-matrix in which scores for the elements of the notation vector to be evaluated are registered, and the similarity score calculation unit uses the score of the score sub-matrix to generate the score The information processing system according to claim 16, wherein a similarity score that gives an index of similarity of sub-matrix is calculated. An information processing apparatus for calculating a first name surname similarity,
A character determination unit that receives a character sequence indicating a first and last name, and determines a character type constituting the received character sequence;
From the character sequence, other notations in which the character sequence is described in other character types including ideograms or phonograms are generated, and a notation vector including at least two different first and last name notations including phonograms is generated. A notation acquisition unit;
A similarity calculation unit that performs a different similarity determination in response to the character type and calculates a score that gives a measure of similarity for the elements of the notation vector;
A similarity score calculation unit that calculates a similarity score for the first and last name candidates using the score calculated by the similarity calculation unit;
The other notation acquisition unit generates the other notation using a phonetic character corresponding to the character sequence when it is determined that the character type of the character sequence is alphabet, and the character sequence is kana If it is determined, generate the other notation with ideograms,
The information processing apparatus manages a similar model table that associates ideographic characters having similar appearances, and a different character table that registers different characters of the ideographic characters. A first and last name similarity calculator that looks up the table and variant font table to calculate the path cost including appearance similarity and font difference;
Selecting the first and last name candidates using the similarity score, and sending the first and last name candidates to the sender of the character sequence as a first and last name candidate list,
The full name similarity calculation unit is a module of the full name candidate generation unit or an independent module of the information processing device,
The similarity calculation unit generates a score sub-matrix in which scores for the elements of the notation vector to be evaluated are registered, and the similarity score calculation unit uses the score of the score sub-matrix to generate the score An information processing apparatus that calculates a similarity score that gives an index of similarity of sub-matrix

Images