JP5466376B2 - Information processing apparatus, first and last name identification method, information processing system, and program - Google Patents

Description

  The present invention relates to an information processing technique for identifying first and last names, and more specifically, an information processing apparatus, a first and last name identifying method, an information processing system, and a program that enable a search for a person name including multibyte characters including its pronunciation characteristics About.

  A method of comparing the names of persons described by single-byte characters such as alphabets and comparing the similarity of the names by calculating a similarity score is used for searching for names. For example, US Pat. No. 6,963,871B1 (Patent Document 1) discloses an automated personal name search system for searching personal names described in alphabets as a system for searching personal names described in single-byte characters. Is disclosed.

In addition, the URI specified in http://publibfp.boulder.ibm.com/epubs/pdf/c1912860.pdf (Non-patent Document 1) uses the similarity of personal names in alphabets, Search, Global
A Name Analytics (GNA) system is disclosed.

  On the other hand, an information search apparatus for searching a document by dividing a document described in multibyte characters such as Japanese into words (tokens) of a set unit is disclosed in, for example, Japanese Patent Application Laid-Open No. 2004-206473 (Patent Document 2). )It is described in. Furthermore, an information search device technique for searching by assigning different weights using surnames and first names among the names described in multibyte characters is also described in, for example, Japanese Patent Application Laid-Open No. 2004-295797 (Patent Document 3). Yes.

  The information retrieval disclosed in Patent Documents 1 to 3 and Non-Patent Document 1 can provide sufficient accuracy and searchability to retrieve personal names described in alphabets or other single byte codes. In Patent Documents 2 and 3, it is also possible to search for a document in which one character is described by a multibyte character defined by a multibyte code, and extract a person name from the document and reflect it in weighting. However, there is a possibility that different pronunciations may be assigned to the same character in a cultural area using a language described in multibyte characters such as Japanese. For example, when the Chinese character “Large” in Japanese is used as a personal name, it is also assumed that a plurality of pronunciations such as “dai” and “hiroshi” are assigned.

  Also, the first and last names greatly depend on differences in cultural spheres. If the cultural spheres are different, even if the first and last names are the same, the order of (Sir Name, Given Name) is different, and multiple multibyte characters such as kanji are different. May be derived, resulting in different alphabetic notations. In such a case, the first and last name search techniques described in Patent Literatures 1 to 3 and Non-Patent Literature 1 do not perform efficient first and last name identification. For this reason, it can be said that a search for a person name described in multibyte characters requires a technique different from a person name search technique for single-byte characters.

  For example, consider a person name written in Japanese kanji, which is a typical multibyte character. Various techniques for converting a Chinese character string into a Roman character can be envisaged. For example, a technique for executing a search by applying a morphological analysis technique to a kanji string is known. In morphological analysis, a dictionary including part-of-speech and frequency information is referred to, and an input character string such as a surname and a first name is subjected to morphological analysis, and decomposed into entry units registered in the dictionary. If there are multiple morpheme decomposition patterns, use a method that preferentially selects the one with the smallest number of divisions or a method that selects the most likely candidate using grammatical information and frequency information. The first name and the last name are searched by decomposing.

Also, when dealing with differences in kanji readings, that is, differences in pronunciation characteristics, there is also known a method for dealing with the variety of pronunciations for kanji using a dictionary registered corresponding to kanji. Yes. In recent years, along with the globalization of economic activities, the development of transportation facilities, and the advancement of network technology, various activities such as people in different cultural areas living in different cultural areas, conducting economic activities, accessing servers, etc. Is done normally. For this reason, there is an increasing need to search for first and last names across cultural spheres. Furthermore, in recent years, if the cultural spheres are the same or non-identical, such as name collation and money laundering for the purpose of integrating pensions, bank accounts, etc. The need to determine whether or not it is increasing is also increasing.
US Pat. No. 6,963,871B1 JP 2004-206473 A JP 2004-295797 A http://publibfp.boulder.ibm.com/epubs/pdf/c1912860.pdf

  As described above, as a method for executing a search for first and last names including multibyte characters such as kanji, a method using a morphological analysis method and a reading dictionary is known. However, unlike the first and last names in the alphabet, the first and last names written in kanji are not divided by a delimiter such as a space or a comma, but are described in one line. For this reason, it is impossible to determine directly where the last name is and where the last name is. In addition, in the first and last names, a title, title, etc. may be added before and after. Dividing a person's name written in Kanji into a surname / first name / honorific component is considered easy for a person with a lot of knowledge, but there may be multiple division methods. In addition, there is a problem of reproducibility and accuracy. On the other hand, in the case of automatically executing first and last name division using a computer, it is necessary to resolve the ambiguity that a division pattern must be selected. Furthermore, there are new fonts, old fonts, variant fonts, abbreviated fonts, etc., and it has been necessary to identify these names by integrating these differences.

  Furthermore, considering the transfer of each component such as the surname and first name of a person name to a kana or alphabet (Roman character), the diversity of how to read kanji and the variety of patterns that convert the pronunciation characteristics of kanji into the alphabet There are many candidates due to For this reason, in addition to the diversity of first and last name identification, there is also a problem in that the diversity of pronunciation is added, and sufficient accuracy and efficiency cannot be provided for performing a person name search.

  Furthermore, even with names that use the same kanji, there are differences in phonetic characteristics due to differences in cultural spheres. For example, Chinese pronunciation characteristics and Japanese pronunciation characteristics are completely different even for the same kanji, so that a correct pronunciation characteristic cannot be estimated by simple dictionary lookup. When the phonetic character is written in Kanji or Korean, it will be directly linked to the first and last names of the alphabet and kana, so multi-byte characters-single-byte characters can also be used to deal with the diversity of personal names when searching for names. It was important to conduct a search including the transcription of

  The present invention has been made in view of the above-described problems of the prior art, and may give a plurality of pronunciation characteristics such as kanji, so that first and last names written in kanji, hiragana, katakana, hangul, etc. It is an object to provide an information processing apparatus, a first and last name identification method, an information processing system, and a program for generating first and last names written in alphabet from multi-byte first and last names.

  In order to solve the above problems, the present invention performs morphological analysis on a character string including a first name and a last name. In the morpheme analysis, a morpheme token is assigned to a character string to obtain a morpheme token string. In the morpheme dictionary used for morpheme analysis, in addition to morpheme tokens, attribute identifiers to indicate attributes such as “kun” and “representative director” that are attached to the surname, first name, or last name. The morpheme tokens are categorized with the attribution of any cultural sphere, such as Japanese-speaking, Korean-speaking, or Chinese-speaking.

  All the results of the morphological analysis are listed and sent to the first and last name candidate creation unit. The surname candidate creation section creates a surname candidate list. The first and last name candidate list registers at least a morpheme token, a connection identifier indicating a connection relationship between the morpheme tokens, and a culture area weight value specified by a first and last name characteristic determined in the culture area to which the morpheme token belongs. Cultural area weighting values are used for the morphological division of first and last names in the case where the order of the first and last name sequence of the relevant cultural area is not included, and when the characters unique to the cultural area, for example, Japanese characters are included, include national characters. The first and last name candidates given the uniqueness are values indicating an unusual classification.

  In another embodiment, the first and last name candidate list can also register a cultural area identification value that explicitly indicates the cultural area of the first and last name candidates, specifies the most likely first and last name candidate for each cultural area, and for each cultural area. Possible alphabetical character string output.

  The value registered in the field of the first and last name candidate list is given a score corresponding to the frequency used as the first name and last name of the morpheme token. A score is also assigned to an attribute identifier between morpheme tokens. In the case of the present embodiment, the culture sphere weight value can be registered as a set value for, for example, an SQL sentence or a script for summing up the scores. In another embodiment, a score for the cultural sphere weight value may be assigned.

  The score calculation unit calculates the total score value by adding up the score values of the morpheme token, the connection identifier, and the culture area weight value registered in the surname candidate list. The total score value is used as a measure of the first and last name distance indicating the distance from the beginning to the end of the first and last name, and the first and last name candidate with the smallest first and last name distance is sent to the notation conversion unit and converted to single-byte characters such as alphabetic notation .

  In another embodiment, a set number of first and last name candidates can be extracted from those having the smallest first and last name distance, and the notation conversion can be performed collectively on these. In yet another embodiment, for the first and last name candidates generated for each cultural zone identifier, the first and last name candidates with the minimum first and last name distance can be extracted for each cultural zone identification value, and the notation conversion can be performed.

  According to the present invention, the most likely first name surname candidate can be converted into a single-byte character string for each cultural area, and according to the present invention, even if the first and last name spans multiple cultural areas. It is possible to provide an information processing apparatus, a first and last name identification method, an information processing system, and a program that enable conversion into a single-byte character string corresponding to the most likely pronunciation characteristics.

  Furthermore, according to the present invention, an information processing apparatus, first name and last name identification that can use a single byte character string that is most likely for a first name and last name in a cultural area that uses multibyte characters as an input character string to an existing first name and last name identification system. A method, an information processing system, and a program can be provided.

  Hereinafter, although this invention is demonstrated with embodiment, this invention is not limited to embodiment mentioned later. FIG. 1 shows functional blocks of the information processing apparatus 100 of the present embodiment. The information processing apparatus 100 can be preferably configured as a server. In other embodiments, if the first and last name identification processing is possible in terms of processing capacity, application size, etc., it can be implemented as a workstation or a personal computer.

  Hereinafter, a hardware configuration of the information processing apparatus 100 will be schematically described. The information processing apparatus 100 includes a single-core or multi-core CPU that manages various processes under an operating system (OS), a RAM that provides an application execution space, a ROM that stores a boot code for initial setup, a BIOS, and the like. And a hard disk device.

  As the CPU, for example, 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) is used as a single core or multi-core. Can be implemented. Furthermore, a hierarchical cache such as an L1 cache, an L2 cache, or an L3 cache can be mounted for the purpose of speeding up the instruction fetch and data fetch of the CPU processing.

  As the OS, for example, WINDOWS (registered trademark) 200X, UNIX (registered trademark), LINUX (registered trademark), AIX (registered trademark), or the like can be used. The information processing apparatus 100 executes various applications described using a programming language such as C ++, JAVA (registered trademark), JAVA (registered trademark) BEANS, PERL, RUBY, and processes input data. The processing result is generated. When the information processing apparatus 100 is configured as a server, the information processing apparatus 100 can be a server for providing a distributed computing environment using a web server or CORBA (Common Object Resource Broker Architecture).

  When the information processing apparatus 100 is configured as a server that provides data via a network, the information processing apparatus 100 uses a file transfer protocol such as HTTP or HTTPS via a network (not shown) such as a LAN, WAN, or the Internet. Using a procedure call or method call such as RPC or RMI, a client computer (not shown) connected via a network receives data such as a kanji character string corresponding to the first and last name, and executes processing by the application .

  When the information processing apparatus 100 is configured as a web server, the information processing apparatus 100 executes a server program such as CGI, servlet, or APACHE described in the programming language described above, and performs a web client. The processing result, such as the first and last name identification value, the bank account referred to by the first and last name identification value, and specific information such as pension information is returned.

  The functional blocks of the information processing apparatus 100 will be described with reference to FIG. Each function processing unit of the information processing apparatus 100 is a computer device 110 including the hardware resources described above, and can be realized as a functional unit on the computer by developing the program on the RAM. The information processing apparatus 100 includes an input interface 112 configured using CGI or the like. The input interface 112 receives a character string that is considered to be a first and last name input from the outside, and calls each functional block for executing subsequent processing. In the present embodiment, the character string considered as the first and last name includes multi-byte characters, and may be an ideographic character string such as a kanji character string, a katakana character string, a kana character character string, a hangul character string, a phonogram character, an ideographic character, for example. And a mixed character string of phonetic characters.

  The information processing apparatus 100 further includes a Chinese character normalization unit 114, a morpheme analysis unit 116, and a first and last name candidate creation unit 118. When the character string includes a kanji character, the kanji normalization unit 114 refers to the kanji dictionary 124 and corrects the different characters, the old characters, the abbreviated characters, and the like into characters suitable for processing by the computer. Appropriate fonts can include standard fonts for each cultural sphere. For example, Japanese is corrected to a font set according to JIS 1st level, JIS 2nd level, etc. Execute processing in a unified manner in analysis. For this purpose, the kanji dictionary 124 is configured as a table in which different characters, old characters, abbreviations, and the like are associated with kanji registered by the information processing apparatus 100.

  The morpheme analysis unit 116 receives the normalized character string, refers to the morpheme dictionary 126 in which the morpheme token is registered, and decomposes the character string into a morpheme token string. In addition, the morpheme dictionary 126 uses an identification value indicating a cultural area to which the morpheme token belongs, and the morpheme token is used as a surname (SN: Sir Name) or as a first name (GN: Given Name). Registers an attribute identifier that indicates whether it is used. The morpheme analysis unit 116 acquires the cultural zone identification value to which the morpheme token belongs, the attribute identifier that indicates the surname, the surname, or the attached word from the morpheme dictionary 126, and the surname and surname described later as a pair It is sent to the candidate creation unit 118.

  In addition, the morpheme dictionary 126 may be added before and after a character string that is determined to be a first name, last name, “representative director”, “department manager”, “section manager”, “section chief”, “professor”, “you”, “ An identifier “SX (Suffix)” indicating that it is an attached word is assigned to a morpheme token that may be an attached word and registered with respect to morpheme tokens such as “sama” and “dono”. Hereinafter, an identifier for indicating a surname, a first name, and an attached word is referred to as an attribute identifier.

  Note that the morpheme dictionary 126 can also be implemented with a synonymous kanji dictionary (not shown) that registers morpheme tokens that can be processed as the same kanji, such as different characters, old characters, and abbreviated characters. In this case, the morpheme dictionary 126 can also create first and last name candidates including differences in kanji such as variant, old font, and abbreviation. In the case of this embodiment, the functions of the kanji normalization unit 114 and the kanji dictionary 124 can be integrated with the functions of the morpheme dictionary 126.

  In the present embodiment, even if the character string of the kanji is different, the alphabet notation corresponding to the character string is output, so that the kanji normalization unit is more preferable than creating a first and last name candidate consisting of a plurality of different characters. 114, it is preferable to use the kanji dictionary 124 to unify and process the character strings before the morphological analysis from the viewpoint of improving the efficiency of the processing after the first and last name candidate creation unit 118 described later.

  The morpheme analysis unit 116 analyzes the character string, classifies the character string into a morpheme token corresponding to the surname, a morpheme token corresponding to the first name, and a morpheme token corresponding to the attached word, and assigns the attribute identifier assigned to the morpheme token. , Extracted in association with the morpheme token, and in the order corresponding to the character string (morpheme token, attribute identifier, morpheme token, attribute identifier, ...) A combination is generated and passed to the first and last name candidate creation unit 118.

  The surname candidate creation unit 118 receives surname / name combination information extracted by the surname / name combination generated by the morpheme analysis unit 116, and analyzes the connection of the attribute identifier attached to the morpheme token. In the analysis of attribute identifiers, attribute identifiers attached to successive morpheme tokens are extracted and determined as connection identifiers. The first and last name candidate creation unit 118 is configured to insert a connection identifier characterizing the connection of the morpheme token between the morpheme tokens, and includes a field of (morpheme token-connection identifier-morpheme token-connection identifier,...). A full name candidate list (600, 800) including a record is created for all generated morpheme token strings.

  As a result, the first and last name candidate list (600, 800) created by the first and last name candidate creating unit 118 constitutes a field in which a morpheme token and a connection identifier for a specific first and last name candidate constitute one record. Also, in certain embodiments, the first and last name candidate list (600, 800) may be configured to include a field for registering a total score value calculated for the first and last name candidates. In the field for registering the total score value, the total score value calculated by using the attribute identifier and the connection identifier of the morpheme token by the score calculation unit 120 is registered, and a first name surname candidate to be output as a processing result is selected. used.

  Further, if the input character string includes kanji, hangul, katakana, hiragana, etc., the first and last name candidate creation unit 118 performs cultural sphere judgment and generates a cultural sphere weight value. If the character string includes kanji, hangul, katakana, hiragana, etc., the cultural attribute weight value is SN or the morpheme token's personal name attribute identification value at the beginning of the first and last name candidates excluding attached words is SN or GN Determine if there is. Further, the first name surname candidate creation unit 118 determines to which cultural area each morpheme token belongs to the morpheme token of the first name surname candidate and the morpheme token of the first name candidate, and the first name surname and first name are assigned to the same cultural sphere. In the case of belonging to different cultural spheres, different cultural sphere weighting values are generated and registered in the first and last name candidate lists (600, 800) as in the case of the first and last name sequence.

  Furthermore, in another embodiment, the first and last name candidate creation unit 118 generates a culture area identification value for each first name and last name candidate, and makes it possible to use the most likely first and last name candidate for each culture area for the score calculation unit 120 described later. Can do.

  The score calculation unit 120 acquires, for a specific record in the surname candidate list (600, 800), the value of a field whose record value is not NULL. After that, the score calculation unit 120 refers to the score table 128, acquires the score value assigned for the field value, and sums the score values acquired for the non-NULL field value of the processing record. Furthermore, the score calculation unit 120 calculates the total score value by adding the score values for the cultural sphere weight values, and describes them in the corresponding records of the first and last name candidate lists (600, 800).

  Note that, when the cultural area weight value is registered in the score table 128, the score calculation unit 120 is referred to for calculation, but is set as a SQL statement or script constant describing the score calculation unit 120. be able to. Each score value registered in the score table 128 is used to indicate the first and last name distance from the beginning to the end of the character string of the first and last name candidate given by the morpheme token string.

  In a preferred embodiment, the non-NULL values of the first and last name candidate list records are summed to calculate a total score value, and the total score value is a value indicating the distance of the sequence from the beginning to the end of the first name and last name including the cultural zone judgment. Use as In other embodiments, although less accurate, it may be used as an index of the distance from the beginning to the end of the first and last name candidates using the total number of connection identifiers included in the cultural area weighted surname and first name candidates. it can.

  The notation conversion unit 122 extracts the first and last name candidate having the smallest total score value from the first and last name candidate list (600, 800) as the most likely first and last name candidate. The notation conversion unit 122 refers to the conversion dictionary 130 for the most likely first name surname candidate, assigns the alphabet notation registered for the most likely first name surname and the first name, and assigns the last name and first name as the maximum likelihood alphabet notation for the first name surname candidate. Output. The alphabetical notation conversion dictionary 130 uses a statistical probability that the character included in the most likely first name surname candidate is used for the last name or first name, or in the case of a kanji, whether the kanji is a national character or a non-national character, Different alphabets are registered for each cultural area. Note that the first and last name candidate list (600, 800) can be configured as, for example, a view generated during processing in a storage device such as a RAM, and the detailed configuration thereof will be described later.

  The notation conversion unit 122 inquires the conversion dictionary 130 that registers a table for registering surname / candidate character strings for surnames and surnames. The notation conversion unit 122 extracts the alphabet notation set for the character string, and can cope with the possibility that the notation is different for each cultural sphere even for the same ideogram.

  Each functional unit described in FIG. 1 can be implemented as a server program in a programming language such as JAVA (registered trademark), PERL, or RUBY. In another embodiment, when a relational database such as DB2 (registered trademark), MYSQL, or ORACLE (registered trademark) or an object-oriented database (OODB) can be used, an SQL statement (Structured Query Language) that executes various processes is used. ) Can be implemented as a function module created using. Which form to implement can be appropriately selected according to the convenience of programming and the usability of the database.

  FIG. 2 is a flowchart of the first and last name identification method of the information processing apparatus 100 according to the present embodiment. The process shown in FIG. 2 starts from step S200, and in step S201, for a character string that is considered to be a first and last name, it is determined whether a first and last name identification request is received together with the character string, and a first and last name identification request is not received. (No) The process waits in step S201 until a first and last name identification request is received.

  When the first and last name identification request is received in step S201 (yes), the character string is acquired in step S202, and the kanji normalization unit 114 is called in step S203 to normalize the different character, old character, and abbreviated character for the character string. Execute. When a plurality of character strings requesting first and last name identification are sent, the received character strings are registered in a queue, and processing is executed in the order of reception. The normalization process simultaneously executes normalization of Japanese characters, as well as normalization of Chinese characters that should be classified as synonyms of different abbreviations such as Chinese and Taiwanese.

  In step S204, the morpheme analyzer 116 is called to refer to the morpheme dictionary 126 for the normalized character string, and determine morpheme tokens registered as last names and first names and corresponding attribute identification value candidates. In step S205, the first and last name candidate creating unit 118 is called to register the determined morpheme token and attribute identification value in a record of the first and last name candidate list to create a first and last name candidate list.

  In step S206, the score calculation unit 120 is called to execute score calculation. The calculated total score value is registered in the total score value field of the records to be summed in the surname candidate list as the surname / name distance for the character string. In step S206, the notation conversion unit 122 searches for a first name surname candidate with the shortest surname surname name registered in the surname surname candidate list, and uses the notation conversion dictionary 130 as a most likely single-byte character string. In certain embodiments, the alphabetic notation is determined.

  In step S207, the determined alphabetical notation is output as a processing result, and in step S208, it is determined whether or not an unprocessed character string remains in the queue. If no unprocessed character string remains (no), the process returns to step S201, and the process is repeated until there is no unprocessed character string.

  On the other hand, if an unprocessed character string remains in the queue (yes), the process returns to step S202 to determine a first and last name candidate for another character string, and the process is performed until alphabetical notation is output in step S207. Repeat.

  In another embodiment, a plurality of first and last name candidates can be extracted from those having a short last name distance by the processing of the score calculation unit 120 in step S205, and alphabetical notation can be assigned to each. In still another embodiment, a case is also assumed in which a plurality of possible first and last name candidates for a character string are obtained for different cultural areas, such as “Hayashi Kohei”, for example. In this case, by extracting the most likely first and last name candidates for each cultural sphere and repeating the processing of step S206 and step S207, it is possible to output a plurality of alphabetic notations across the cultural spheres. In this embodiment, for each cultural area, a set number of first and last name candidates may be extracted in ascending order from the most likely first and last name candidates to increase the number of possible alphabetic selections.

  FIG. 3 shows an embodiment of the morpheme token string generation process generated by the morpheme analysis process executed by the morpheme analysis unit 116 of the present embodiment in step S204 of the process described in FIG. In the embodiment of FIG. 3, the normalized character string is assumed to be “Yama Shimotsuruma”.

  If the character string is “Yamamoto Shimotsuruma”, if you are a Japanese who has knowledge about first and last names, it is important to know where the last name is, where it is the first name, and what the appendix is. It can be classified to some extent by using it. In addition, for example, up to the first character in the character string can be registered in the morpheme dictionary 126 by statistical analysis of the first name and last name. In addition, for example, “Ma” as the short last name, “Seimon Saburo”, “Saemon Saburo”, “Kano Koji” as the long last name can be exemplified. In the morpheme dictionary 126, the first name and the last name are divided so as to give the most likely combination from the combination of the morpheme tokens registered as the last name and the first name. Can be executed.

  Furthermore, in another embodiment, particularly when dividing Japanese surnames, surnames and surnames to be assigned in the morpheme dictionary 126, combined with mora analysis, which is generally used in speech-language processing, The most likely last name split may be determined.

  In the embodiment shown in FIG. 3, when the character string is “Yamoto Shimotsuruma”, it is determined from the determination of the character string that the cultural area is any one of Japanese, Chinese, and Korean. In the described embodiment, as a result of the morphological token analysis, “lower”, “crane”, “lower crane”, “tsuruma”, “lower crane”, “large”, “wa”, “you”, “Yamato ”,“ Wa ”, and“ Wa-kun ”morpheme tokens are extracted.

  When reproducing the first and last names using these morpheme tokens, the morpheme analysis unit 116 determines the cultural sphere by determining whether the cultural sphere is registered as a first and last name in each cultural sphere. . In the embodiment shown in FIG. 3, surnames including “lower”, “crane”, “ma”, “shita crane”, “tsuruma”, “shimazuru” are assigned to surnames in the Japanese cultural sphere. It is a morpheme token. Note that the morpheme tokens “lower” and “crane” will be described as morpheme tokens that are also assigned to names in the Japanese cultural sphere. FIG. 3 shows representative attribute identifiers and score values assigned to the attribute identifiers.

  The morpheme analysis unit 116 of the information processing apparatus 100 according to the present embodiment executes a process of graphing a morpheme token string in the morpheme token analysis shown in FIG. The graphing of morpheme token strings is that the specified morpheme tokens are used as nodes for the morpheme token strings, and the connection identifiers are generated in the sequential order from the attribute identifiers given to the consecutive nodes. It is executed by registering the identifier.

  In the embodiment shown in FIG. 3, the character string “Shimotsuru Yamato-kun” is searched by the morpheme dictionary 126 for “Crane”, “Shizuru”, “Tsuruma”, “Shimotsuruma”, “Dai”, “Wa”, The morpheme tokens of “Kimi”, “Yamato”, “Wa” and “Wa Kimi” have been extracted. Then, the extracted morpheme token string is assigned to the character string “Yamata Shimotsuruma” from the first “lower” to the last “kun”. Specifically, for the morpheme token “Shimotsuruma”, six morpheme tokens of “lower”, “crane”, “ma”, “shitazuru”, “tsuruma”, and “shimotsuruma” are extracted, and these morphemes are extracted. Allocate the token to the string “Shimotsuruma”.

  In the embodiment shown in FIG. 3, {"lower (SN)"-"crane (SN)"}, {"lower (GN)"-"crane (GN)"-"between" Shimotsuruma " (SN) "}, {" Shizuru (SN) "-" Between (SN) "}, {" Shimozuru (SN) "} are assigned morpheme token strings. In the embodiment shown in FIG. 3, the morpheme token string is arranged such that the morpheme token string including the morpheme token that is likely to be used as the surname is located higher. For example, a score of 1 is given for “lower” and “Tsuruma”, and a score of 4 is given for “Shimotsuruma”.

  The morpheme dictionary 126 statistically analyzes whether the morpheme token indicates a surname (SN) or a first name (GN), and the attribute identifier of the morpheme token corresponds to the morpheme token. Add and register. In the embodiment shown in FIG. 3, the morpheme tokens “lower” and “crane” will be described assuming that an attribute identifier indicating a name (GN) is adopted. Note that a surname candidate with the morpheme token “lower” as the last name and “crane” as the first name (GN) can be assumed, but even in this case, only one series is added to the morpheme token string of FIG. Thus, the processing can be performed in the same manner. In the embodiment shown in FIG. 3, the total score value is given by the morpheme token string 310 surrounded by a broken line to give the minimum value. ”, Name =“ Yamato ”, and adjunct =“ Kimi ”are output as the most likely result.

  After the morpheme token sequence shown in FIG. 3 is generated, the morpheme analysis unit 116 associates the attribute identifiers “SN”, “GN”, and “SX” of the morpheme token with each morpheme token in the surname candidate list. For example, “lower”-“Tsuruma” is registered in the sequence {lower, SN-Tsuruma, SN}. In addition, “lower”-“crane”-“between” are registered while adding attribute identifiers in the sequence {below, GN-crane, GN-between, SN}, and a morpheme token string representing the first and last names is a surname candidate Defined in the list.

  FIG. 4 is a flowchart of processing executed by the first and last name candidate creation unit 118 and the score calculation unit 120 of this embodiment. The process of FIG. 4 starts from step S400, receives the result of the morpheme analysis unit 116 for the character string to be processed after normalization in step S401, and obtains a possible morpheme token and attribute identification value for the character string. Go to the first name surname list.

  In step S402, a connection identification value between morpheme tokens is generated using attribute identifiers of adjacent morpheme tokens, and registered in the first and last name candidate list. For example, it is SS when morpheme tokens configured as surnames (SN) are continuous, SG when surnames (SN) and surnames (GN) are contiguous, GS when surnames (GN) surnames (SN) are contiguous, and so on. Furthermore, a culture area weight value is generated using the determination of the culture area, and is registered in the first and last name candidate list. In the embodiment shown in FIG. 3, the character string is configured to include Chinese characters in the determination of the cultural area. For this reason, when a character string corresponds to a surname, the last name is usually written first (N: Normal), and the sequence in which the first name is written first is unusual (AN: Abnormal). it can. Cultural weight is a value that is used to identify normal or non-normal weights when considering cultural spheres, such as specific identification characters, flags, or other suitable characters. Can be used.

  Also, in another cultural area determination embodiment, the first and last name candidate creation unit 118 executes the cultural area determination of the morpheme tokens belonging to the first name and the last name when the morpheme token division of the first name and the first name is completed. In this embodiment, when it is determined that the morpheme token belonging to the surname and the morpheme token belonging to the first name are a combination of different cultural spheres, the value AN is added as the cultural sphere weighting value, and the same morpheme is assigned. When the first and last names are decomposed only with the token, the value N is added and registered in the first and last name candidate list. If the cultural zone weighting value is set first, the previous cultural zone weighting value and the subsequent cultural zone weighting value are referred to. If AN exists, AN If there is no AN, the process of adding N is executed.

  Thereafter, in step S403, the score table 128 is referred to, and the score values of the morpheme token and the connection identifier are acquired. In step S404, from the beginning to the end of the morpheme token forming the character string, the score value of the record and the score value or the value of the culture area weight value for the specific first and last name candidate are summed to calculate the first and last name distance. Thereafter, the one having the morpheme token string having the smallest calculated first and last name distance is determined as the most likely first and last name candidate.

  Thereafter, in step S405, the data is sent to the notation conversion unit 122, and the process ends in step S406. In step S404, as described above, a plurality of candidate morpheme token strings can be sent to the notation conversion unit 122 as first and last name candidates according to rank order or culture sphere according to a specific purpose and implementation form.

  FIG. 5 shows an embodiment of the score table 128 referred to by the score calculation unit 120 of the present embodiment. The score table 128 includes a morpheme token score table 500 and a sequence score table 550. The morpheme token score table 500 is configured as a data structure in which the higher the frequency when a morpheme token is used as a first name or last name, the lower the score is assigned. For example, the use of the morpheme tokens “below” and “between” as surnames can be said to have a high frequency, so score = 1 is assigned.

  On the other hand, although it cannot be said that “lower” is frequently used for the name, it can be said that the frequency is not so high. Therefore, in the embodiment shown in FIG. 3, score = 2 is assigned. Also, since the appearance frequency is considered to be lower for the three-letter surname “Shimotsuruma”, a score = 4 is assigned. be able to. In the morpheme token score table 500 shown in FIG. 5, a surname (SN) score and a first name (GN) score are registered as a pair for a specific morpheme token to constitute a record.

  In addition, the sequence score table 550 uses a statistical judgment as to whether or not the order of the morpheme tokens assigned to the character strings is appropriate in relation to the cultural sphere and the first and last name structure. Is a data structure used for scoring. In the sequence score table 550 shown in FIG. 5, connection relationships of morpheme tokens such as surname-first name, surname-surname, surname-surname, etc. are registered as connection identifiers SG, SS, GS, etc. A connection score is registered with respect to the connection identifier based on the determination of whether it is appropriate and the judgment of the cultural area.

  When a value is registered in the connection identifier field registered in the first and last name candidate list, the score calculation unit 120 acquires a connection score value corresponding to the value from the sequence score table 550, and obtains a morpheme token score. Used together with the score value of the morpheme token registered in the table 500 for the score calculation. Note that the cultural sphere weight value can be registered in the score table 128 as described above, or can be set in the program as an appropriate large number in the case of 0 for N and AN. It can be implemented in consideration of the convenience of programming.

  FIG. 6 shows an embodiment of the first and last name candidate list 600 created by the first and last name candidate creating unit 118 in this embodiment. In the embodiment of the first and last name candidate list 600 shown in FIG. 6, all combinations of three kinds of first name candidates are listed for each of the four kinds of last name candidates. In the embodiment shown in FIG. 6, the record set 610 corresponds to the surname combination “lower + crane”, and the record set 620 corresponds to the surname combination “lower + crane + ma”. Further, the record set 630 corresponds to the surname combination “Shimotsuru + ma”, and the record set 640 further corresponds to the surname combination “Shimotsuru ma”.

  The first and last name candidate list 600 includes a field 650 for registering a culture sphere weight value, a preceding morpheme token connection list field 660 indicating connection of morpheme tokens from the start of the character string, and remaining morpheme tokens up to the end of the character string. A subsequent morpheme token connection list field 680 indicating the connection of the preceding morpheme token, a front / rear connection list field 670 indicating the connection before and after the preceding morpheme token and the subsequent morpheme token, and a total score value field 690.

  In the field 650, if a morpheme token having an SN attribute identifier is assigned to the head of the first and last name in response to the judgment as a Japanese name in the cultural sphere judgment, the first and last name candidate creation unit 118 performs N: Sets the normal identifier. In addition, since the attribute identifier “GN” is assigned to “below” in the record set 620, it is not a normal description of Japanese first and last names in the cultural zone judgment. Therefore, the identifier of AN: Abnormal is used. It is set.

  The score calculation unit 120 acquires the score values of the fields that are not NULL for each record in the first and last name candidate list 600 of FIG. 6 with reference to the score table 128, and in a preferred embodiment, the following formula (1) is obtained. Use to calculate the total score value.

上記式（１）中、ｉは、レコードを指定するサフィックスであり、ｊは、レコードｉのｊ番目のフィールドを示すサフィックスである。また、SUM_SCORE(i)は、合計スコア値であり、Culture_weight(i)は、文化圏重付け値である。なお、上述したように文化圏重付け値は、スコア・テーブル１２８の登録項目としておくこともできるし、またＳＱＬ文中の条件値として指定しておくこともできる。なお、「君」、「さん」、「殿」などの付属語については、姓名のスコア計算に影響を与えない適切なスコアを割当てることができ、図６に示した実施形態では、付属語「君」についてスコア＝１を割当てて合計スコア値であるSUM_SCORE(i)を計算している。なお、付属語については、スコア＝０として設定することもできるし、上記式（１）の計算から除外する処理を採用することもできる。

In the above formula (1), i is a suffix that designates a record, and j is a suffix that indicates the jth field of the record i. SUM_SCORE (i) is a total score value, and Culture_weight (i) is a culture area weight value. As described above, the culture sphere weight value can be set as a registered item in the score table 128, or can be specified as a condition value in the SQL sentence. Note that an appropriate score that does not affect the score calculation of the first and last names can be assigned to the ancillary words such as “you”, “san”, and “dono”, and in the embodiment shown in FIG. SUM_SCORE (i) which is a total score value is calculated by assigning score = 1 for “you”. Note that the attached word can be set as score = 0, or a process of excluding it from the calculation of the above formula (1) can be adopted.

In the first and last name candidate list 600 shown in FIG. 6, the value of SUM_SCORE _ij , which is the total score value calculated by applying the above formula (1), is registered in the field 690. Note that the values of N and AN in the total score value can be appropriately set to values such as N = 0 and AN = large number, for example, in connection with the cultural zone determination. The score calculation unit 120 compares the values in the field 690, adopts the record with the smallest value as a first name surname candidate, and passes it to the notation conversion unit 122.

  Furthermore, as another embodiment of the score calculation process of the score calculation unit 120, the following formula (2) is used to extract only the cultural area weighting value and the connection identifier, and the scores are summed to give the surname as the total score value It can also be registered in the candidate list 600.

上記式（２）を使用して合計スコア値を計算する場合、形態素トークンが姓を示すか、または名を示すかの判断とは別に、文字列をいくつの形態素トークンで接続したか、および文化圏の判断結果を使用して姓名候補を抽出することが可能となる。この実施形態は、使用される姓、名の頻度を使用するものではないが、姓に利用される文字が比較的限定される場合、姓名候補を効率的に抽出するためには好ましく利用することができる。なお、上記式（２）の計算においても、付属語についての取扱いは、上記式（１）の場合と同様に実行することができる。

When calculating the total score value using equation (2) above, the number of morpheme tokens connected to the string and the culture, apart from determining whether the morpheme token represents a last name or a first name It becomes possible to extract first and last name candidates using the judgment result of the area. This embodiment does not use the last name and the frequency of the first name used, but preferably used in order to efficiently extract the first and last name candidates when the characters used for the last name are relatively limited. Can do. In addition, also in the calculation of the above formula (2), the handling of the attached word can be executed in the same manner as in the case of the above formula (1).

  In the embodiment shown in FIG. 6, regarding the character string = “Yamata Shimotsuruma”, the minimum total score value is to decompose the surname = “Shimotsuruma”, first name = “Yamato”, and adjunct = “you”. give. The notation conversion unit 122 searches the first name surname candidate list, and extracts the first name surname candidate with the smallest total score value as the most likely first name surname candidate using an appropriate extraction form.

  FIG. 7 shows an embodiment of a table set 700 registered by the conversion dictionary 130 used by the notation conversion unit 122. The table set shown in FIG. 7 is configured as a last name table 710 and a first name table 750. The surname table 710 is further allocated with a JP section 720, a CN section 730, and a KR section 740 for each cultural area. In each section, a morpheme token corresponding to the surname belonging to the cultural area is displayed in alphabetical form. Registered as a pair. In the surname table 710, even when the same character is used as a surname in a plurality of cultural spheres, it is registered in each section together with its alphabetical notation.

  Note that single-byte character strings such as alphabetic characters can use Roman characters for Japanese names. In addition to the ceremonial notation method defined in ISO 3602, the Roman character notation method includes Hebon and Japanese Roman characters. Alphabet notation of a notation system can be used, and these can be registered alone, or can be converted into a single-byte character string in parallel as an alternative notation form candidate and output.

  When the conversion dictionary 130 according to this embodiment is implemented, for example, for the last name “willow”, it is possible to search the alphabet notation longitudinally using the last name table. On the other hand, it is possible to search for alphabetical expressions longitudinally using the name table. On the other hand, when combining the alphabet notation acquired for the surname and the alphabet notation acquired for the first name, there is a possibility that the surname and the first name of the alphabet notation of different cultures will be combined.

  However, even in the above-described case, the score calculation unit 120 assigns a high score to the surname and surname candidates in which surnames and surnames belonging to different cultural spheres are combined using the cultural sphere weighting value. For this reason, if a character string can only be described with last names and first names belonging to different cultural spheres (for example, a man and a woman belonging to different cultural spheres are married and unified to one of the last names) In general, the notation conversion unit 122 is unlikely to combine first and last names belonging to different cultural spheres. As a result, the notation conversion unit 122 can return the alphabet notation corresponding to the cultural sphere as a single-byte character string. As a result, even when only a combination of first and last names in different cultures is generated, a corresponding single-byte character string can be output.

  Further, in another embodiment, the conversion dictionary 130 includes cultural zone tables such as a JP table, a CN table, and a KR table that include morpheme tokens used as first and last names, first names, and attribute identifiers for each cultural zone. Can be implemented as follows.

  In the other embodiment described, which implements a culture table rather than a first name table, a first name table, single-byte strings belonging to the same cultural sphere are searched preferentially in first name and last name identification. More efficient processing can be performed. In addition, when either a surname or a surname is searched only in a specific cultural area table, the search target table is changed from the JP table to the CN table or the KR table for the surname or surname not searched. By performing the search, it is possible to output alphabetical representations of first and last names belonging to different cultures as single-byte character strings.

  FIG. 8 shows an embodiment of processing executed by the first and last name candidate creation unit 118 of the information processing apparatus 100 when there is a possibility that a character string that is considered to be a first and last name spans a plurality of cultural spheres. As shown in FIG. 8, for example, when the input character string is “Ezawa Higashi”, as shown in FIG. 8, “Ezawa-Higashi”, “Ezawa-Higashi”, “E-zawa-Higashi” 3 It can be divided into one morpheme token sequence. In this case, first and last name candidates indicated by three morpheme token strings of E SS Sawa SG East, E SG Sawa Higashi, and Ezawa SG East are created in the first and last name candidate list 800. “Jiang” as a surname is extremely rare as a Japanese surname, but a Chinese surname is a commonly used surname, and “Sawadong” may also be used as a surname in Chinese.

  On the other hand, the surname “Ezawa” is a surname commonly used in Japan, and the name “East” may also be used as a first name. In such a case, the cultural sphere weights for the later-explained Jiang-Sawahigashi and Ezawa-Higashi cannot be distinguished from each other, and both are determined as N. In addition, selecting a lower surname candidate in terms of the total score value will forcefully determine whether the string “Ezawa Higashi” is either a Japanese surname or a Chinese surname and discard the other. Absent. As for “E-Sawa-Higashi”, the total score value is higher than “Ezawa-Higashi” and “E-Sawa Higashi” based on the weight of the cultural area and the number and score of connection identifiers. The notation conversion unit 122 does not select and is not extracted, and therefore will not be described in detail.

  In the case of explaining the first and last name candidate “Ezawa Higashi”, for example, a field for registering a cultural sphere identifier is provided in the first and last name candidate list 800, and when each morpheme token string is generated, the first morpheme token is assigned. A field for entering an identifier for identifying a cultural area, for example, JP, CN, KR, etc., can be defined. The full name candidate creation unit 118 calls the score calculation unit 120 at the time when the entry of the value in each field is completed, and calculates the total score value by summing the values of each record in the full name candidate list 800. After that, the notation conversion unit 122 extracts the maximum likelihood morpheme token string that gives the minimum of the total score value for each identifier of JP, CN, and KR.

  The notation conversion unit 122 refers to the first and last name candidate list 800, extracts the most likely form and the token string for each cultural area, and temporarily stores the extracted first and last name candidates in the queue buffer. Further, the display conversion unit 122 acquires alphabetical representation for each of JP, CN, and KR until the queue buffer becomes empty, and outputs it as a single-byte character string. Note that the notation conversion unit 122 can also pass a plurality of morpheme token strings to the notation conversion unit 122 in order from the smallest total score value for each cultural area, and output more single-byte character string conversion candidates.

  In the embodiment shown in FIG. 8, the character string “Kanazawa Kaoru” is a morpheme token and can be divided into Kana-Kawazawa, Kanazawa-Kan, and Kana-Sawa-Kan. In this embodiment, “Kin” as a surname is a surname commonly used as a Korean surname, and the surname “Kanazawa” is a surname commonly used as a Japanese surname. On the other hand, “Kin-Sawa-Kin” uses the number and type of connection identifiers, and the total score becomes high anyway. Therefore, the same processing as “E-Sawa-Higashi” is executed, and the notation conversion unit Since it is not passed to 122, detailed description is abbreviate | omitted.

  By the way, among the Chinese characters included in the character string “Kanazawa Kaoru”, “雫” is a Japanese-made Chinese character, a so-called national character. For this reason, the first and last name candidate creation unit 118 identifies “Kana-Sawa 雫” as KR and “Kanazawa- 澤” as JP. The cultural zone weight is N as the cultural zone identifier = KR because the first morpheme token is "gold" for "Kin-Sawa 雫", but the character attribute includes the cultural zone identifier = JP. Therefore, the value of AN is detected, and as a result, AN is given as the cultural area weighting value, and the total score value is increased. Even in this case, the total score value is lower for “Kana-Sawa Kaoru”. Therefore, “Ezawa Higashi”, “Kanazawa Higashi” and “Ezawa Higashi” are selected for JP, KR, and CN, respectively. Extracted by the unit 122.

  FIG. 9 shows an embodiment of an information processing system 900 that executes the first and last name identification method of the present embodiment. The information processing system 900 shown in FIG. 9 includes a first name / name identification server 910 that is an information processing apparatus that performs first name / name identification processing, and a database that returns a first name / first name identification result to the first name / first name identification server 910 in association with alphabetical names and first name / name candidates. A server 960.

  The first and last name identification server 910 receives a first and last name identification request from the client computer via the network 920 and a character string including at least multibyte characters considered to be a first and last name using an HTTP protocol or the like. The received first and last name identification request is sent to the alphabet conversion unit 940 a via the network adapter 930. The alphabet conversion unit 940a executes the first and last name identification processing of the present embodiment, and is implemented as a front processor of the first and last name identification server. The alphabet conversion unit 940a converts a character string including multi-byte characters into a single or a plurality of single-byte character strings, and sends it to the first and last name identification unit 950.

  The first and last name identifying unit 950 can be implemented as a Global Name Analytics system manufactured by International Business Machines Corporation described in Non-Patent Document 1, for example. The full name identifying unit 950 acquires the output from the alphabet converting unit 940a as the input character string. In the embodiment shown in FIG. 9, the first and last name identifying unit 950 issues a search query to the database server 960 using, for example, an SQL sentence, and instructs a search for a multibyte character string corresponding to a single byte character. . The database server 960 manages a relational database 970 such as DB2, for example, and the relational database 970 associates a first and last name described by a single-byte character string with a first and last name identifier that uniquely identifies the first and last name. Table 980a and a table 980b in which the first and last name identifiers and the first and last names in the multibyte character string are associated with each other.

  The database server 960 refers to the single byte character to identify the first and last name identifier, searches the table 980b using the identified first and last name identifier as a search key, and obtains a character string described in the corresponding multibyte character as the first and last name . Thereafter, the database server 960 returns the acquired first and last names to the first and last name identifying unit 950. The first and last name identifying unit 950 passes the obtained first and last name of the multibyte character string to the output processing unit 940b and creates a processing result to be returned to the sender of the first and last name identifying request. For example, the output processing unit 940b can send the first and last name identified by the first and last name identifying unit 950 as it is to the sender of the first and last name identification request, or, for example, a space between the last name (SN) and the first name (SN). , Medium black (•), hyphen, and other characters can be inserted and returned as a search result. In still another embodiment, as a form, the search results can be described separately in the last name field, the first name field, etc., and sent to the client computer that is the transmission source.

  In yet another embodiment, the database server 960 manages an information table 990 that manages information associated with first and last names, such as addresses, offices, pension information, bank accounts, careers, etc. corresponding to first and last names. In this embodiment, it is also possible to search for various types of information referred to by the multi-byte character specified by the first and last name identifier and the single-byte character used therefor, and to return the search result to the first and last name identifying unit 950 together with the first and last name identification result. . The full name identification unit 950 sends the full name identification result obtained from the database server 960 and other information searched using the full name identification result to the output processing unit 940b, and sends the full name through the network adapter 930. By returning to the client computer from which the identification request was sent, the client computer is made available to the first and last name identification and other information tagged with the first and last name.

  In still another embodiment, the process is performed when a character string that is considered to include a first and last name includes a single-byte character and a multi-byte character. When a character string including single-byte characters and multi-byte characters is considered to include a first name and last name, for example, a preprocessor that executes processing for deleting a single-byte character from the character string can be implemented. Thereafter, the information processing apparatus 100 or the information processing system 900 can execute the same search by using the character string as the character string to be searched. In addition, in the case of a character string in which an alphabet notation, for example, an alphabet notation such as Jane or Johnson is described in multibyte characters such as katakana, a preprocessor is implemented that deletes the code corresponding to katakana and makes it a character string to be searched. As a result, it is possible to execute the same processing as that described above and generate a processing result.

  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 figure which showed the functional block about the information processing apparatus 100 of this embodiment. The flowchart of the full name identification method of the information processing apparatus 100 of this embodiment. The figure which showed embodiment of the morpheme token sequence generation process produced | generated by the morpheme analysis process which the morpheme analysis part 116 of this embodiment performs in step S204 of the process demonstrated in FIG. The flowchart of the process which the full name candidate creation part 118 and the score calculation part 120 of this embodiment perform. The figure which showed embodiment of the score table 128 which the score calculation part 120 of this embodiment refers. The figure which showed embodiment of the full name candidate list 600 which the full name candidate creation part 118 produces in this embodiment. The figure which showed embodiment of the table set 700 which the conversion dictionary 130 which the notation conversion part 122 uses registers. The figure which showed the embodiment of the process which the surname candidate creation part 118 of the information processing apparatus 100 performs when the character string considered as a surname may extend over several culture spheres. The figure which showed embodiment of the information processing system 900 which performs the full name identification method of this embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 ... Information processing apparatus, 110 ... Computer apparatus, 112 input interface, 114 ... Kanji normalization part, 116 ... Morphological analysis part, 118 ... First name surname candidate creation part, 120 ... Score calculation part, 122 ... Notation conversion part, 124 ... Kanji Dictionary, 126 ... Morphological dictionary, 128 ... Score table, 130 conversion dictionary, 600, 800 ... First and last name candidate list, 900 ... Information processing system

Claims (10)

An information processing apparatus that converts a character string including a first and last name described in multibyte characters into a single byte character string, wherein the information processing apparatus includes:
A kanji normalization unit that normalizes multibyte characters included in the character string into a font registered by the information processing apparatus;
The character string after normalization by the kanji normalization unit is divided into morpheme tokens, and the cultural sphere identification value indicating the cultural sphere to which the morpheme token belongs and the attribute identifier assigned to the morpheme token are obtained from the morpheme dictionary. A morphological analyzer that
A culture area that generates a connection identifier from the attribute identifier between the morpheme token divided by the morpheme analysis unit, the attribute identifier, and the morpheme token, and indicates whether the combination of the morpheme token representing the first and last name is normal or unusual A weighted value is generated and registered as a first and last name candidate list, and a first name and last name candidate creating unit having the cultural area to which the first morpheme token of each first name and first name candidate belongs as the cultural area of the first and last name candidate,
Get the assigned score values for the morpheme token, the connection identifier, and the culture weight value, and calculate the total score value to generate a surname distance giving a measure of the distance from the beginning to the end of the surname And a score calculation unit that uses the short name distance as a first name surname candidate,
Refers to a dictionary in which a morpheme token and a single-byte notation are registered as a pair for each cultural area, extracts the first and last name candidates, and the last name of the morpheme token included in the first and last name candidates according to the cultural area of the first and last name candidates And a notation conversion unit that outputs a single-byte character string corresponding to the name. A kanji dictionary that provides registered kanji corresponding to different character forms in association with the kanji normalization unit;
Providing the morpheme token and attribute identification value for configuring the first and last name to the morpheme analysis unit, and registering the morpheme token and the attribute identifier in association with the cultural sphere to which the first and last name belongs;
A score table for providing the score calculation unit with a score value assigned to the morpheme token and the connection identifier;
The information processing apparatus according to claim 1, further comprising: a single-byte character string corresponding to the surname of the morpheme token and a conversion dictionary that registers a single-byte character string corresponding to the first name.   The conversion dictionary includes a surname table in which a single-byte character string for the surname is registered for each cultural sphere and a name table in which a single-byte character string for the first name is registered for each cultural sphere, and the cultural spheres are the same The information processing apparatus according to claim 2, wherein the information is converted into the single-byte character string for a combination of first and last names and a combination of first and last names of different culture areas.   The information processing apparatus according to claim 1, wherein the notation conversion unit outputs a single-byte character string for the last name and the first name that are most likely for each different cultural area. A first and last name identifying method executed by an information processing apparatus that converts a character string including a first and last name described in multibyte characters into a single-byte character string, wherein the first and last name identifying method includes:
Normalizing a multi-byte character included in the character string with reference to a Chinese character dictionary into a character style registered by the information processing device;
Referring to the morpheme dictionary, the character string after normalization by the kanji normalization unit is divided into morpheme tokens, and the culture sphere identification value indicating the culture sphere to which the morpheme token belongs and the attribute assigned to the morpheme token A connection identifier is generated from the attribute identifier between the morpheme token obtained by dividing the morpheme token obtained by the morpheme analyzer, the attribute identifier, and the attribute identifier between the morpheme tokens. A culture area weight value indicating the culture area, the morpheme token, the connection identifier, and the culture area weight value are described in a surname candidate list, and a culture to which the first morpheme token of each surname candidate belongs Making the zone a cultural zone for the first and last candidate ,
Using the score values assigned for the morpheme token, the connection identifier, and the culture weight value, a total score value is calculated to give a first and last name distance that gives a measure of the distance from the beginning to the end of the first and last name. Generating a first and last name candidate using the short first and last distance,
Refers to a dictionary in which a morpheme token and a single-byte notation are registered as a pair for each cultural area, extracts the first and last name candidates, and the last name of the morpheme token included in the first and last name candidates according to the cultural area of the first and last name candidates And a step of outputting a single-byte character string corresponding to the first name and the first and last name identification method. The step of outputting the single-byte character string includes:
Referring to a surname table in which a single-byte character string for the surname is registered for each cultural area and a name table in which a single-byte character string for the first name is registered for each cultural area;
6. The method for identifying a first and last name according to claim 5 , wherein the first and last character combinations of the first and last names having the same cultural sphere and the first and last names having different cultural spheres are converted. An information processing system for performing first and last name identification,
A first and last name identification server connected to a network and receiving a first and last name identification request and a multibyte string including the first and last name via the network;
The single-byte character string corresponding to the single-byte character string converted from the character string received by the full-name identification server is searched, and the first and last names of the multi-byte characters associated with the searched single-byte character string are obtained. A database server returning to the first and last name identification server,
The first and last name identification server is
A kanji normalization unit that normalizes the character string to a font registered by the information processing apparatus;
The character string after normalization by the kanji normalization unit is divided into morpheme tokens, and the cultural sphere identification value indicating the cultural sphere to which the morpheme token belongs and the attribute identifier assigned to the morpheme token are obtained from the morpheme dictionary. A morphological analyzer that
A culture area that generates a connection identifier from the attribute identifier between the morpheme token divided by the morpheme analysis unit, the attribute identifier, and the morpheme token, and indicates whether the combination of the morpheme token representing the first and last name is normal or unusual A weighted value is generated, registered as a first and last name candidate list, and a first name and last name candidate creation unit having a culture area to which the first morpheme token of each first name and first name attribute belongs as a culture area of the first and last name candidates ,
Get the assigned score values for the morpheme token, the connection identifier, and the culture weight value, and calculate the total score value to generate a surname distance giving a measure of the distance from the beginning to the end of the surname And a score calculation unit that uses the short name distance as a first name surname candidate,
Refers to a dictionary in which a morpheme token and a single-byte notation are registered as a pair for each cultural area, extracts the first and last name candidates, and the last name of the morpheme token included in the first and last name candidates according to the cultural area of the first and last name candidates And a notation conversion unit that outputs a single-byte character string corresponding to the name,
A name identifier that receives the output of the notation converter and issues a query to the database server;
An output processing unit that executes a process of receiving a query result of the database server as a first and last name of a multibyte character string and displaying the first and last names separately. The database server includes a table for registering a first and last name identifier for associating the multi-byte character string with the single-byte character string;
An information table for managing information associated with the first and last names, searching for the last name and the first name from the table, searching for the related information from the information table, and passing the result to the first name search server as a search result The information processing system according to claim 7 . A computer-executable program for executing an information processing method for converting a character string including a first and last name described in multibyte characters into a single-byte character string,
A kanji normalization unit that normalizes multibyte characters included in the character string into a font registered by the information processing apparatus;
The character string after normalization by the kanji normalization unit is divided into morpheme tokens, and the cultural sphere identification value indicating the cultural sphere to which the morpheme token belongs and the attribute identifier assigned to the morpheme token are obtained from the morpheme dictionary. A morphological analyzer that
A culture area that generates a connection identifier from the attribute identifier between the morpheme token divided by the morpheme analysis unit, the attribute identifier, and the morpheme token, and indicates whether the combination of the morpheme token representing the first and last name is normal or unusual A weighted value is generated, registered as a first and last name candidate list, and a first name and last name candidate creation unit having a culture area to which the first morpheme token of each first name and first name attribute belongs as a culture area of the first and last name candidates ,
Get the assigned score values for the morpheme token, the connection identifier, and the culture weight value, and calculate the total score value to generate a surname distance giving a measure of the distance from the beginning to the end of the surname And a score calculation unit that uses the short name distance as a first name surname candidate,
Refers to a dictionary in which a morpheme token and a single-byte notation are registered as a pair for each cultural area, extracts the first and last name candidates, and the last name of the morpheme token included in the first and last name candidates according to the cultural area of the first and last name candidates And a notation converter that outputs a single-byte character string corresponding to a name. The notation conversion unit includes a surname table in which a single-byte character string for the surname is registered for each cultural sphere, and a name table in which a single-byte character string for the first name is registered for each cultural sphere, The program according to claim 9 , wherein conversion is performed to the single-byte character string for a combination of the same first and last names and a combination of first and last names of different culture areas.

Images