JP5583230B2 - Information search apparatus and information search method - Google Patents

Description

  The present invention relates to an information retrieval apparatus and an information retrieval method for retrieving data from a predetermined database using a character string.

  2. Description of the Related Art Conventionally, a technique is known in which a user's speech is input, a speech recognition result in a text format is obtained by recognizing the speech, and a predetermined database is searched using the speech recognition result as a search keyword (described below) Patent Document 1).

  In Patent Document 1 below, when an example sentence search is performed based on a speech recognition result, the speech recognition result is first presented to the user, and the user himself / herself performs an operation for removing the misrecognized word to thereby detect the misrecognized word. A method for preventing a search with a character string including the text has been proposed.

JP 2003-330925 A

  However, in the above method, since confirmation and operation by the user are required, it takes time and effort for the user. In addition, there is a problem that it takes more time to complete a series of search processes due to the process in which the user once confirms and operates the voice recognition result. In the above method, no countermeasure is taken when a segmentation error occurs when a search is performed for each of a plurality of segmented character strings. Hereinafter, this will be described using a specific example.

  For example, the input speech “Call Yoko Saito” is recognized by morphological analysis, etc., and five character strings segmented by “/” like “Saito / Yoko / Ms. / Ni / Telephone” (speech Consider the case where a recognition result is obtained. Each character string thus obtained is used as a search keyword, for example, by searching a telephone directory database that stores data in which information indicating a name and a telephone number is associated with each other, the information indicating the name becomes “Yoko Saito”. Can be obtained and presented to the user.

  However, in the above example, if an erroneous recognition (conversion error, segmentation error, etc.) occurs as “Sa / Ito / Yoko / Ni / Telephone” in the speech recognition process, an inappropriate search keyword (“sa”) Is obtained. When a search is executed using such an inappropriate search keyword, for example, search results unintended by the user such as “Sato” and “Sasaki” are presented to the user, which may cause a problem that the user's convenience is impaired.

  The present invention has been made in view of the above problems, and provides an information search apparatus and an information search method that can efficiently reduce search errors when searching a predetermined database using a character string. With the goal.

The information search apparatus according to the present invention includes, as one search information, a character string information acquisition unit that acquires a plurality of character string information indicating a character string, and for each character string information acquired by the character string information acquisition unit, According to the number of characters for each character string information counted by the character number counting means, using the character number counting means for counting the number of characters of the character string indicated by the character string information and the search information obtained by the character string information obtaining means. Search means for acquiring a search result by executing a search, and search result output means for outputting the search result acquired by the search means.

The information search apparatus according to the present invention acquires a plurality of pieces of character string information as one piece of search information, and executes a search according to the number of characters for each piece of character string information. Thereby, for example, character string information that is likely to cause a search error is extracted based on the number of characters, and the search can be executed after performing appropriate processing such as excluding the character string information from the information for search. Therefore, it is possible to prevent a search error that occurs when a search is performed using the acquired character string information as it is. That is, it is possible to efficiently reduce search errors when searching a predetermined database using a character string.

In the information search device, the search means excludes character string information having one character and performs a search using character string information having two or more characters.

  According to the above configuration, the search error is made efficient by excluding from the search information character string information (character string information having one character) that is likely to cause a search error when used as a search character string. It can be reduced well.

In the information search device, the character string information acquisition unit acquires a plurality of pieces of character string information that are ordered in advance, and the search unit adjoins two character string information having different numbers of characters based on the ordering. If the two character string information is adjacent to each other, the character strings indicated by the two character string information are connected to generate a new character string, and the new character string is characters containing information indicating performs a search using the character string information of more than one letter.

  For example, the character string information acquisition means is a part where one character string should not be originally separated (for example, between “K” and “G” for a character string indicating a name “Kago”). Character string information indicating each character string generated by being divided (segmented) may be acquired. In such a case, according to the above configuration, “K” and “G” are connected to generate “Kago” (new character string), and the new character string is used as a search character string for searching. It becomes possible to do. Accordingly, occurrence of a search error can be prevented and a search can be executed using a new character string (“cage”) more suitable for the search, so that it can be expected to improve the search accuracy.

  In the information search apparatus, the search means executes a search based on different search conditions for a search using character string information having one character and a search using character string information having two or more characters. May be.

  For example, when searching a database including character string information indicating a person's name or the like, a character string with one character is likely to be a character string obtained by segmentation error, and is used as a search character string. Probably not appropriate. Therefore, for example, when a fuzzy search such as a partial match search that extracts a character string with one character as a search result and extracts the search result if the search character string is partially included is performed. A large amount of search errors may occur. On the other hand, there may be a surname (Mr.) having one character such as “Li” or “Tsu”. According to the above configuration, for example, the above-mentioned fuzzy search is executed for a character string with two or more characters, and a character string with a single character is searched only when the character string is included without excess or deficiency. It is possible to execute a search based on an appropriate search condition according to the number of characters, such as executing a complete match search extracted as a result. As a result, it is possible to increase the possibility of acquiring information desired by the user and to efficiently reduce search errors.

  The information search apparatus further includes voice recognition means for inputting a series of user's utterance voices and generating character string information by performing voice recognition processing on the utterance voices. The character string information generated by the recognition unit may be acquired.

  According to the said structure, the character string information acquisition means can acquire character string information from a user's speech voice by a speech recognition means. Further, even when the character string information is character string information as a result of a segmentation error caused by speech recognition, search errors can be efficiently reduced by the above-described means.

  The information search apparatus further includes second search means for acquiring a search result by executing a search using character string information indicated by a second character type different from the first character type, and the character string The information acquisition means includes a first character string information indicated by a first character type as search information, and a second character string corresponding to the first character string information and indicated by a second character type Information, the character number counting means and the search means perform each process on the first character string information, the second search means performs a search using the second character string information, The search result output means may output the search result based on the first search result acquired by the search means and the second search result acquired by the second search means.

  According to the above configuration, for example, the above-described search is performed on character string information (first character string information; for example, “Kago”) indicated by katakana (first character type), and a search result (first search result) ) To get. At the same time, for example, a search result (second search result) obtained by executing a search by an arbitrary method for character string information (second character string information; for example, “protection”) indicated by kanji (second character type). get. Thereby, the search result output means can output the search result presented to the user based on each search result. That is, it is expected that the search quality is improved by executing a search for each of a plurality of character types, comprehensively evaluating each search result, and outputting the search result.

  In the information search device, the second search unit is configured to search for the second character string information, and the search unit corresponds to the first character string information corresponding to the second character string information. The same process as the process executed according to the number of characters in the column information may be executed, and the search may be executed using the second character string information.

  According to the above configuration, when the search means performs a search after performing some process (for example, an exclusion process) on the first character string information according to the number of characters, the second search means The second character string information corresponding to the first character string information is subjected to the same processing, and the search is executed by the same search method as the search means. Thereby, for example, the second character string information corresponding to the first character string information excluded from the search information can also be excluded from the search information, and search errors can be reduced more efficiently.

  By the way, the present invention can be described as an invention of an information search apparatus as described above, and can also be described as an invention of an information search method as follows. Since these inventions are substantially the same inventions only in different categories, they exhibit the same operations and effects.

That is, the information search method according to the present invention includes, as one search information, a character string information acquisition step for acquiring a plurality of character string information indicating character strings, and for each character string information acquired in the character string information acquisition step. The character count step for counting the number of characters in the character string indicated by the character string information and the search information acquired in the character string information acquisition step are used to calculate the number of characters for each character string information counted in the character number count step. A search step for obtaining a search result by executing a corresponding search, and a search result output step for outputting the search result obtained in the search step. In the search step, character string information having one character is stored. A search is performed using character string information with two or more characters excluded. The information search method according to the present invention includes, as one search information, a character string information acquisition step for acquiring one or more character string information indicating a character string, and for each character string information acquired in the character string information acquisition step. The number of characters for each character string information counted in the character number counting step using the character number counting step for counting the number of characters of the character string indicated by the character string information and the search information acquired in the character string information acquiring step A search step for acquiring a search result by executing a search according to the search result, and a search result output step for outputting the search result acquired in the search step. In the search step, two character string information having different character numbers based on the above ordering are obtained. If the two character string information are adjacent to each other, the character strings indicated by the two character string information are connected to generate a new character string, and the new character string A search is executed using character string information including two or more characters including information indicating a simple character string.

  ADVANTAGE OF THE INVENTION According to this invention, the search error at the time of searching a predetermined database using a character string can be reduced efficiently.

It is a block diagram which shows the function structure of the information search device which concerns on one Embodiment of this invention. It is a figure which shows the hardware constitutions of the information search device which concerns on one Embodiment of this invention. It is a figure which shows the example of the speech recognition result with respect to a user's speech sound. It is a figure which shows an example (long sound reading conversion) about the conversion about a specific character. It is a figure which shows the example of the information used in order to demonstrate operation | movement of an information search device. It is a figure which shows operation | movement of the information search device in a 1st example. It is a figure which shows operation | movement of the information search device in a 2nd example. It is a figure which shows operation | movement of the information search device in a 3rd example.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the description of the drawings, the same or equivalent elements are denoted by the same reference numerals, and redundant description is omitted.

  FIG. 1 is a configuration diagram of an embodiment of an information search apparatus according to the present invention. The information search apparatus 1 according to the present embodiment is a user terminal such as a portable terminal carried by a user, for example, and can provide a search service based on the user's uttered voice in order to improve user convenience and operability. It is what. Specifically, the information search apparatus 1 acquires one or more character string information obtained by voice recognition of a user's uttered voice as search information, and searches a predetermined database using the search information. Device.

  In the present embodiment, a telephone directory database that stores character string information indicating a person's name and information indicating a telephone number in association with each other is assumed as the predetermined database. That is, the information search apparatus 1 uses the search information as a search keyword and searches the phone book database for character string information indicating a person's name stored in the phone book database. It is a device that provides a user with a phone book search function for acquiring relevant information (name and phone number of a person).

  However, the form of the information search device 1 is not limited to the above. For example, each function of the information search device 1 may be provided on a server that can be used from a user terminal via a communication network. In this case, the user can use the function provided by the information search device 1 via the user terminal. Moreover, each function of the information search device 1 may be realized by, for example, being distributed between a user terminal and a server, and the user terminal and the server operating in cooperation with each other.

  As shown in FIG. 1, the information search apparatus 1 according to the present embodiment includes a voice recognition unit 11, a character string information acquisition unit 12, a character number counting unit 13, a character string information database 14 that is a phone book database, A search unit 15 and a second search unit 16 that execute search processing, and a search result output unit 17 that outputs a search result are provided.

  FIG. 2 is a hardware configuration diagram of the information search apparatus 1 according to the present embodiment. A hardware configuration of the information search apparatus 1 will be described with reference to FIG. As shown in FIG. 2, the information retrieval apparatus 1 includes a CPU 101 that executes an operating system, application programs, and the like, a main storage unit 102 that includes a ROM and a RAM, an auxiliary storage unit 103 that includes a hard disk memory, and the like. , A communication control unit 104 that performs data communication, an output unit 105 including a liquid crystal monitor, an input unit 106 including a keyboard, a mouse, and a microphone as input devices, a USB memory, a CD-ROM, a DVD And a recording medium reading unit 107 that reads the recording medium 108.

  Each function of the information search apparatus 1 shown in FIG. 1 is realized by reading and executing a predetermined software program in the main storage unit 102 under the control of the CPU 101. At that time, the CPU 101 controls data reading and writing operations in the main storage unit 102 and the auxiliary storage unit 103 according to the processing procedure of the software program, and controls operations of the input unit 106, the output unit 105, and the communication control unit 104. . Hereinafter, each functional element of the information search apparatus 1 shown in FIG. 1 will be described.

  The voice recognition unit 11 inputs a series of user's utterance voices, executes voice recognition processing on the uttered voices using a conventional voice recognition technology, and generates character string information as a voice recognition result. Means. The voice recognition unit 11 performs, for example, hardware that performs voice input, such as a microphone built in the information search apparatus 1, and performs voice recognition processing on voice input through the hardware to obtain character string information. It implement | achieves as what is provided with the voice recognition middleware to produce | generate.

  For example, when a voice that the user intentionally utters the content of “call to Yoko Saito” is input, the voice recognition unit 11 performs a voice recognition process on the uttered voice using morphological analysis or the like. As a result, the voice recognition unit 11 uses, for example, character string information (first / same / yoko / san / ni / denwa) indicated by katakana (first character type) as the character string information indicating the reading (first character). Get string information). Here, “/” indicates a location that the speech recognition unit 11 determines to be a word boundary by speech recognition. The voice recognition unit 11 acquires a group of character string information segmented (divided) with “/” as a boundary (one word string information each word separated by “/”) as a voice recognition result. Note that the speech recognition result acquired by the speech recognition unit 11 does not necessarily have to be segmented, and may be one character string information without a break.

  In addition, the voice recognition unit 11 corresponds to the first character string information, and as character string information (second character string information) indicating notation, for example, a character type (second character) in which hiragana, katakana and kanji are mixed. Character string information (second character string information) “difference / Ito / Yoko / san / ni / telephone” indicated by (character type) may be acquired together. Such second character string information is obtained by using, for example, a conversion dictionary (a dictionary storing a character string indicating a reading and a character string indicating a kana-kanji notation of the word) held in the information search apparatus 1. It can be obtained by converting the first character string information.

  Here, since the accuracy of the speech recognition process depends on ambient noise and the clarity of the pronunciation of the user, a correct speech recognition result is not always obtained. Therefore, the speech recognition unit 11 may acquire a plurality of speech recognition result candidates (i-Best recognition results). FIG. 3 shows an example of a speech recognition result for the user's speech. As illustrated in FIG. 3, for example, the voice recognition unit 11 performs a voice recognition process on the voice that the user intentionally utters the content of “call Yoko Saito”, and the first candidate of the voice recognition result. As “1-Best recognition result”, “Sa / Ito / Yoko / San / Ni / Denwa” and “Difference / Ito / Yoko / Mr. / Ni / Telephone” are acquired and the second candidate (2-Best recognition result) “Site / Yoko / San / Ni / Denwa” and “Saito / Yoko / Ms. / Ni / Telephone” may be acquired. Here, the smaller the value of “i” in the “i-Best recognition result”, the higher the speech recognition rank (rank of the speech recognition estimation accuracy). Note that the above example of speech recognition result candidates is an example.

  The character string information acquisition unit 12 is a character string information acquisition unit that acquires one or more character string information as one piece of search information. “One search information” is, for example, one or more pieces of character string information included in one candidate (i-Best recognition result) of a speech recognition result. Specifically, one piece of search information includes first character string information indicating a reading (for example, “sa / itou / yoko / san / ni / denwa”) and second character string information indicating notation ( For example, “Difference / Ito / Yoko / Ms. / Ni / Telephone”) is included. One search result (ranked search result) is obtained for one search information by a search performed by the search unit 15 and the second search unit 16 described later.

  The character string information acquisition unit 12 acquires the voice recognition result (character string information) generated by the voice recognition unit 11. As illustrated in FIG. 3, when the speech recognition unit 11 generates two speech recognition result candidates, character string information (yomi / notation) corresponding to “1-Best recognition result” and “2-Best recognition” Each piece of character string information (yomi / notation) corresponding to “result” is acquired as one piece of search information. The first character string information (yomi) acquired by the character string information acquisition unit 12 is used for search by the search unit 15 described later, and the second character string information (notation) is used for the second search unit described later. 16 is used for search. The ranked search results obtained for each speech recognition result candidate are comprehensively evaluated by the search result output unit 17 described later, and correspond to one utterance by the user that is finally output (presented to the user). One search result is generated and output.

  Moreover, the character string information acquisition part 12 may perform processes, such as title removal and noun acquisition, with respect to the character string information acquired as mentioned above by using techniques, such as a morphological analysis. By performing such processing, it is possible to efficiently extract character string information that can be an appropriate search keyword when searching the character string information database 14. For example, in the case where the title removal and noun acquisition processing is performed on the search information corresponding to the “1-Best recognition result” shown in FIG. 3, “san” indicating the title is identified and deleted. . Furthermore, noun phrases (for example, “sa (difference)”, “Ito (Ito)”, “yoko (Yoko)”) satisfying a certain criterion are specified and acquired from the remaining character string information. However, the above-described process is not an essential process in the information search apparatus 1 and may be omitted.

  The character number counting unit 13 is a character number counting unit that counts the number of characters of the character string indicated by the character string information for each first character string information acquired by the character string information acquiring unit 12. The character number counting unit 13 obtains the number of characters for each first character string information by counting the number of characters constituting the character string. Specifically, the character count unit 13 counts, for example, 1 character for “sa” and 3 characters for “Ito”. The number of characters for each first character string information counted by the character number counting unit 13 is used by the search unit 15 described later.

  The character string information database 14 is a storage unit that stores data (record) including character string information to be searched in the information search apparatus 1. Specifically, the character string information database 14 is a phone book database that stores character string information indicating a person's name and information indicating a telephone number in association with each other. In the character string information database 14, for example, character string information (for example, “Yoko Saito”) that indicates the name in hiragana, katakana and kanji, and character string information (for example, “Saito Yoko”) that indicates the name's reading in katakana. ) And information indicating the telephone number are stored in advance by a user operation (information input and registration) or the like.

  The search unit 15 uses the first character string information of the search information acquired by the character string information acquisition unit 12 and uses the first character string information counted by the character number counting unit 13 according to the number of characters for each first character string information. Search means for acquiring a search result (first search result) by executing the search. The second search unit 16 acquires a search result (second search result) by executing a search using the second character string information among the search information acquired by the character string information acquisition unit 12. Search means (second search means). The search result output unit 17 is a search result output unit that outputs a search result that is finally presented to the user based on a result of comprehensive evaluation of the first search result and the second search result. Specific functions of the search unit 15, the second search unit 16, and the search result output unit 17 will be described in detail in the specific examples shown below.

(First example)
In the information search device 1 in the first example, the search unit 15A excludes character string information having one character. Specifically, the search unit 15A acquires the number of characters counted by the character number counting unit 13 for the first character string information among the character string information acquired by the character string information acquisition unit 12 as the search information. When it is determined that the number of characters is one, the first character string information is excluded (deleted / discarded) from the search information.

  For example, when the user speaks with the intention of “protection (carriage)”, the first character string information “ta / go” acquired by erroneous recognition by the voice recognition unit 11 is the character string information acquisition unit 12. Consider the case where the information is obtained as search information. In this case, the search unit 15A acquires the number of characters of each first character string information (“Ta” and “Go”) counted by the character number counting unit 13, and the number of characters of these first character string information is one character. It is determined whether or not. Since both “ta” and “go” are one character, the search unit 15A excludes “ta” and “go” from the search information. The search unit 15A does not exclude the first character string information (for example, “Kago”, “Tanaka”, etc.) having two or more characters, and executes the search using the character string information as a search keyword.

  The search unit 15A executes a search by “fuzzy search”, for example. Specifically, the search unit 15A sets the individual first character string information as individual search keywords (search character strings), and stores the search target character string information (character string information database 14) for each search keyword. If it is determined whether or not a certain criterion is satisfied, and it is determined that the criterion is satisfied, the record information including the character string information to be searched is used as a search result. Search method to obtain.

  As a specific example of the fuzzy search, there is a method of searching by complete match search or partial match search. The exact match search is a search method that determines whether or not the search keyword and the character string information to be searched match completely, and if it is determined that the search keyword matches completely, the character string information is acquired as a search result. is there. The partial match search is a search method for determining whether or not a search keyword is partially included in character string information to be searched, and acquiring the character string information as a search result when it is determined that the search keyword is included. It is. In partial match search, a method for determining whether or not a search keyword is included in the front part, rear part, and other parts of the character string information to be searched (front part match, rear part match, and part match) There are different types. In the partial match search, only one of these types may be used, or a plurality may be used in combination.

  For example, when the search keyword is “protection”, if the search target character string information is “protection”, it is acquired as a search result by an exact match search. In addition, when the search target character string information is “Kago-chan”, “Tokyo Kago-chan”, and “Tokyo Kago-chan”, partial matches classified as front partial match, rear partial match, and partial match, respectively. Acquired as a search result by search.

  Another example of fuzzy search is a search method based on consonant matching. In the comparison between the search keyword and the character string information to be searched, the consonant matching means that, for example, the vowels of the first word (first character) match and the consonants of the second word (second character) This is a method of determining whether or not vowels match each other, and obtaining character string information to be searched as a search result when it is determined that they match. However, the determination method used here is not limited to the above. For example, when the above conditions are relaxed and the vowels of the first word and the vowels of the second word match each other, the character string information to be searched May be acquired as a search result.

  For example, if the search keyword is “kago” and the search target character string information is “tago”, the vowels (a) of the first word match and the second word Since the consonants (g) and the vowels (o) match each other, the search target character string information is acquired as a search result by any of the above-described search methods using consonant matching. On the other hand, if the character string information to be searched is “kako”, the vowels (a) of the first word and the vowels (a) of the second word match, but the consonant of the second word is Does not match (g and k). Therefore, in this case, the search target character string information is not acquired as a search result by the former search method using consonant matching, but is acquired as a search result by the latter search method using consonant matching.

  The search unit 15A converts a partial character string included as part or all of the first character string information into a specific character before executing an ambiguous search using the first character string information of two or more characters. You may perform the process to do. Specifically, for example, the search unit 15A holds in advance dictionary data that stores correspondence information between a character string before conversion and a character string after conversion. The search unit 15A refers to the dictionary data to determine whether or not all partial character strings included in the first character string information match the character strings before conversion stored in the dictionary data. To do. The search unit 15A converts the partial character string determined to match the character string before conversion by the above determination into a character string after conversion corresponding to the character string before conversion. Thereby, the search unit 15A acquires the converted first character string information.

  FIG. 4 is a diagram illustrating an example of conversion (long sound conversion) for a specific character. By such conversion, for example, first character string information that does not include a long sound symbol such as “Oono” or “Saito” is obtained from first character string information that includes a long sound symbol such as “Ono” or “Saito”. can do. Usually, it is assumed that the data registered as the reading of the person's full name (character string information indicating yomi) does not contain a long clef symbol, and therefore, by using the above conversion process, a search keyword more suitable for the search is acquired. Can do. The search unit 15A may execute the search using only the first character string information after conversion, or may execute the search using both the first character string information before conversion and after conversion. Also good.

  The search unit 15A searches the character string information database 14 using the first character string information having two or more characters as a search keyword (searches for each character string information), and finds a search hit for each search information ( A search score (initial value is “0”) corresponding to the number of search hits is calculated in association with the record on the character string information database 14 (obtained by the search). The “search score” may be a value obtained by adding a fixed value (for example, “1”) to a single search hit, for example, the first candidate corresponding to the first candidate (1-Best recognition result). “1” is added when a search hit is made using character string information as a search keyword, and “1” is added when a search hit is made using the first character string information corresponding to the second candidate (2-Best recognition result). It may be set such that more search scores are added when a search hit is made using character string information having a higher voice recognition rank, such as adding “0.8”.

  In this embodiment, the search score is calculated by the latter method. Here, information indicating a search score for each search information associated with each record is stored, for example, in a storage area temporarily secured on the information search apparatus 1, and the search result output unit 17 described later searches the information. Referenced when outputting the result.

  The second search unit 16 uses the second character string information acquired as the search information to perform a search by “fuzzy search”, similar to the search unit 15A. However, in the case where the second character string information does not include reading information (information indicating consonants and vowels), fuzzy search by consonant matching is excluded. Further, the second search unit 16 calculates a search score for each search information in association with each record acquired by the search by the same method as the search unit 15A.

  The search result output unit 17 searches the search score calculated by the search using the first character string information by the search unit 15A and the search using the second character string information by the second search unit 16 for each search information. The sum with the search score calculated by the above is calculated. And the search result output part 17 arranges the character string information (for example, information which shows a name and telephone information) contained in the record information memorize | stored in the character string information database 13 in the order with the largest sum of search scores, and outputs it as a search result. To do.

  Here, when a search is performed using a plurality of search information (a plurality of speech recognition result candidates), as described above, the search score of each search result (record) is calculated for each search information. Will be. In this case, for example, the search result output unit 17 extracts, for each record, the maximum search score among the search scores calculated for each search information as a search score (representative score) associated with each record. This will be described below using a specific example.

  For example, the search information corresponding to the voice recognition result of the first candidate is “Ta / go (ta / go)”, and the search information corresponding to the voice recognition result of the second candidate is “ka / go (ka / go). I think about the case where In this case, when the search score for the first candidate of the record including “bago” is “1” and the search score for the second candidate is “1.6”, The search result output unit 17 acquires “1.6” as the representative score associated with the record including “protection (cage)”. Thereby, for each record, the search score with the highest search score (having the highest search fitness) can be extracted as the representative score. Such a representative score is considered to be one of appropriate indexes for comparing search ranks (search suitability ranks) between records.

  By performing such processing, the search result output unit 17 acquires a representative score uniquely associated with each record. The search result output unit 17 outputs and displays the search results (for example, information indicating name and telephone information) in the order of the search score on the output unit 105 such as a display provided in the information search device 1 in descending order of the representative score.

  Here, the search result output unit 17 displays both the first search result (search result based on the first character string information indicating the reading) and the second search result (search result based on the second character string information indicating the notation). The reason for outputting the search result based on the search score will be described using an example different from the above.

  For example, consider a case in which records including character string information such as “Yoko Saito”, “Hiroshi Saito”, “Yoko Ono”, and the like are stored in the character string information database 14. In this case, when “Saito / Yoko” is acquired as the first character string information indicating the reading and “Saito / Yoko” is acquired as the second character string information (notation) indicating the notation, In the search based on character string information, the same search score is attached to both “Yoko Saito” and “Yoko Ono”, and in the search based on the second character string information, the same is applied to both “Yoko Saito” and “Hiroshi Saito”. A case where a search score is attached is assumed.

  As described above, when a search is performed using only one of the first character string information and the second character string information, even if there is no difference in the search score, the first character string information By determining both the second character string information and the total of the second character string information (summing the search score), the search score of “Yoko Saito” intended by the user can be increased, and can be expected to be displayed at the top.

  Subsequently, an operation of processing executed by the information search apparatus 1 in the first example will be described with reference to FIGS. 5 and 6. FIG. 5 is a diagram illustrating an example of information used for explaining processing executed by the information search apparatus 1. FIG. 6 is a flowchart showing the operation of the information search apparatus 1 in the first example. In the following description, the search unit 15A and the second search unit 16 execute the above-described complete match search and partial match search (including all of the front partial match, the rear partial match, and the partial match) as a fuzzy search. Shall.

  First, the voice recognition unit 11 inputs a voice uttered by the user with the intention of “calling Kago-san” and executes a voice recognition process (step S101). Subsequently, the character string information acquisition unit 12 generates character string information generated by the voice recognition processing (“Ta / Go / San / Ni / Denwa” and “Ta / go / san / san / N / telephone ”and“ Kago / San / Ni / Denwa ”and“ Kago / san / ni / telephone ”) are acquired as the second candidate speech recognition results (step S102). The character string information acquisition unit 12 acquires the speech recognition result of the first candidate (step S103), and executes search processing (“TA / GO” and “TA / GO” by executing the title deletion and noun acquisition processing. ") Is acquired (step S104, character string information acquisition step). Here, when the process of step S104 is not performed, step S103 corresponds to a character string information acquisition step.

  Subsequently, the character number counting unit 13 counts the number of characters of each character string information included in the search information (“t / go”) indicating the reading obtained by the character string information obtaining unit 12 (step S105, character number counting step). ). Thereby, the character number counting unit 13 acquires the number of characters (one character) of each of “T” and “G”.

  Subsequently, the search unit 15A executes a search process based on the number of characters counted in step S105 (step S106, search step). Specifically, the search unit 15A acquires the character string information “ta” acquired by the character string information acquisition unit 12 (step S106A), and determines whether the number of characters is one character (step S106B). Here, since the number of characters of “ta” is one, the search unit 15A skips the search process (fuzzy search) using “ta” (excludes “ta” from the search information) (step S106B: YES) Thereafter, the same processing is performed by the search unit 15A for the next character string information “go” (step S106E: NO, S106A, S106B: YES).

  On the other hand, the second search unit 16 acquires the character string information “ta” from the character string information (“TA / GO”) indicating the notation acquired by the character string information acquisition unit 12 (step S107). A fuzzy search is executed using the information (step S108). Here, the second search unit 16 adds 1 to the search score for the first candidate of the record including “Tahara” and “Tanabe” that has been searched for by the fuzzy search.

  Subsequently, the second search unit 16 performs an ambiguous search using the character string information “after” of “ta” (steps S109: NO, S107, S108). Here, the second search unit 16 adds 1 to the search score for the first candidate of the record including “Goto” that has been searched for by the fuzzy search.

  When the search processing by the search unit 15A and the second search unit 16 for all character string information included in the search information corresponding to the speech recognition result of the first candidate is completed (steps S106E: YES, S109: YES), respectively. The search scores for each record obtained by the above search are added together, and the search score of each record related to the search information corresponding to the speech recognition result of the first candidate is determined (step S110). That is, the search score for the first candidate of the record including “Goto”, the record including “Tahara”, and the record including “Tanabe” is determined as “1”.

  Subsequently, the character string information acquisition unit 12 acquires the speech recognition result of the second candidate (steps S111: NO, S103), and executes the process of deleting the title and acquiring the noun, thereby obtaining the search information (“Kago”). And “Protect”) (step S104, character string information acquisition step).

  Subsequently, the character number counting unit 13 counts the number of characters of each character string information included in the search information (“cage”) indicating the reading acquired by the character string information acquiring unit 12 (step S105, character number counting step). Thereby, the character count unit 13 obtains the number of characters “2” (two characters). Subsequently, the search unit 15A acquires “basket” (step S106A) and determines whether the number of characters is one (step S106B). Since the number of characters of “Kago” is 2, the search unit 15A executes a process of converting a specific character for “Kago” (for example, the long sound conversion shown in FIG. 4) as necessary (Step S106B: NO). , S106C), a fuzzy search is executed using “basket” (step S106D). The search unit 15A adds 0.8 to the search score for the second candidate for the record including “Kago” that has been searched for by the fuzzy search.

  On the other hand, the second search unit 16 acquires search information (“protection”) indicating the notation acquired by the character string information acquisition unit 12 (step S107), and executes an ambiguous search using the character string information (step S107). Step S108). Here, the second search unit 16 adds 0.8 to the search score for the second candidate of the record including “protect” that has been hit by the fuzzy search.

  When the search processing for all character string information included in the search information is completed by the search unit 15A and the second search unit 16 (step S106E: YES, S109: YES), search for each record obtained by each search The scores are added together to determine the search score of each record related to the search information corresponding to the second candidate speech recognition result (step S110). That is, the search score for the second candidate for the record including “protection” is determined as “1.6”.

  When the processing is completed for all speech recognition result candidates (step S111: YES), the search result output unit 17 extracts a representative score for the search results acquired by the search unit 15A and the second search unit 16. The search result output unit 17 displays the search results in the order of representative scores (“Kago” (1.6) → “Goto” (1) → “Tahara” (1) → “Tanabe” (1) →... The numbers in parentheses are representative scores.) Are output to the output unit 105 such as a display provided in the information search device 1 (step S112, search result output step).

  In the above processing, if the processing of step S106 and step S107A by the character count unit 13 and the search unit 15A is not performed, the search processing for “ta” and “go” is not skipped, and an ambiguous search is performed in step S107C. The As a result, the search score (representative score) of the record including “Goto”, the record including “Tahara”, and the record including “Tanabe” is “2”, and the search score (representative score) of the record including “Kago” ) Which is “1.6”. Therefore, the order arranged by the search result output unit 17 in step S113 is “Goto” → “Tahara” → “Tanabe” → “protection”, and the record including “protection” intended (desired) by the user. However, there arises a problem that a record unintended by the user is displayed at the top. In particular, as in the above example, when a fuzzy search is performed using a one-character keyword, it is considered that there is a high risk that a large number of records that are not intended by the user will be searched and the search score will be calculated high.

  According to the information search device 1 in the first example, as described above, by removing character string information (character string information with one character) that is likely to cause a search error from the search keyword, The impact can be reduced. That is, it is possible to reduce a possibility that a search error unintended by the user is displayed (displayed at a higher level) due to a search error and a search result desired by the user is not displayed (or displayed at a lower level).

(Second example)
The information search device 1 in the second example will be described mainly regarding differences from the first example. In the information search device 1 in the second example, the character string information acquisition unit 12 acquires character string information that has been ordered in advance as search information. The character string information acquisition unit 12 can acquire each segmented character string in order from the top, for example, from the speech recognition result as shown in FIG. 3, and can associate the acquired order as the order of each character string. In addition, the search unit 15B determines whether two character string information having different character numbers are adjacent to each other based on the ordering. An example of a specific determination processing procedure will be described in the description of the flow described later. When the search unit 15B determines that the two character string information is adjacent to each other, the search unit 15B generates a new character string by connecting the character strings indicated by the two character string information, and The search is executed using character string information including two or more characters including the information to be indicated.

  Subsequently, an operation of processing executed by the information search apparatus 1 in the second example will be described with reference to FIGS. 5 and 7. FIG. 7 is a flowchart showing the operation of the information search apparatus 1 in the second example. However, steps S201 to S205 and S207 to S212 are the same as steps S101 to S105 and S107 to S112 executed by the information search apparatus 1 in the first example shown in FIG. Omitted. In the following description, the search unit 15B and the second search unit 16 execute the above-described complete match search and partial match search (including all of the front partial match, the rear partial match, and the partial match) as an ambiguous search. Shall.

  Through the processing in steps S201 to S204, the character string information acquisition unit 12 acquires search information ("ta / go" and "ta / go"). Subsequently, by the processing in step S205, the character count unit 13 obtains the number of characters (one character) of each of the character string information “ta” and “go” by counting.

  Subsequently, the search unit 15B performs a search process based on the number of characters counted in step S205 (step S206, search step). Specifically, the search unit 15B acquires the character string information “ta” acquired by the character string information acquisition unit 12 (step S206A), and determines whether the number of characters is one (step S206B). Here, since the number of characters of “Ta” is one character, the search unit 15B determines whether or not the first character string information (next segment) is one character (step S206C). Here, since the character string information “go” next to the character string information “tag” is also one character (step S206C: YES), “ta” and “go” are connected to form a new character string “tag”. Is generated (step S206D).

  Subsequently, the search unit 15B executes a process of converting a specific character for the new character string “tag” (for example, the long sound conversion shown in FIG. 4) as necessary (step S206E). The fuzzy search is executed using the information (step S206F). In the character string information stored in the character string information database 13 shown in FIG. 5, there is no record that makes a search hit by an ambiguous search using “tag”, so the search score of any record is not added.

  After that, when the processing for all speech recognition result candidates including the search processing (steps S207 to S209) by the second search unit 16 is completed (step S211: YES), the search result output unit 17 includes the search unit 15B and the search unit 15B. A representative score is extracted for the search result acquired by the second search unit 16. The search result output unit 17 displays the search results in the order of representative scores (“Kago” (1.6) → “Goto” (1) → “Tahara” (1) → “Tanabe” (1) →... The numbers in parentheses indicate representative scores.) Are output to the output unit 105 such as a display provided in the information search device 1 (step S212, search result output step).

  As described above, according to the information search apparatus 1 in the second example, the search unit 15B connects character string information (“tag” and “go”) of adjacent one character to “tag” (new (Character string) is generated, and the new character string is used as a search character string to perform a search, so that the influence of a search error can be reduced as in the first example.

  Further, for example, a case where the user utters intentionally with the content “call Kago-san”, and character string information segmented as “ka / go” is obtained due to a voice recognition error is considered. In this case, the search unit 15B does not exclude “f” and “go”, which are character string information of one character adjacent to each other, but creates new character string information “ By performing a search using “cage”, it is possible to acquire a record including “protection (cargo)” intended by the user as a search result and add the search score of the record. Therefore, it is possible to increase the possibility that a record including “protection (carriage)” intended by the user is output (displayed) by the search result output unit 17 and to improve the search accuracy.

  In the second example, the search unit 15B sequentially processes the character string information, and determines whether or not the character string information is linked by determining whether or not the first character string information is one character. Although described as a determination, the determination method is not limited to this. For example, the search unit 15B may determine whether or not to link the character string information by sequentially processing the character string information and determining whether or not the previous character string information is one character. Further, the search unit 15B may first scan all the ordered character string information, and connect the character string information when detecting portions where the character string information of one character is adjacent.

(Third example)
The information search device 1 in the third example will be described mainly regarding differences from the first example. In the information search apparatus 1 in the third example, the search unit 15C uses different search conditions for a search using character string information having one character and a search using character string information having two or more characters. Perform a search based on Specifically, the search unit 15C performs the above-described fuzzy search for the first character string information having two or more characters, and performs only an exact match search for the first character string information having one character. Run.

  Subsequently, an operation of processing executed by the information search apparatus 1 in the third example will be described with reference to FIGS. 5 and 8. FIG. 8 is a flowchart showing the operation of the information search apparatus 1 in the third example. However, since the processes of steps S301 to S305 and S307 to S312 are the same as the processes of steps S101 to S105 and S107 to S112 executed by the information search apparatus 1 in the first example shown in FIG. Omitted. In the following description, the search unit 15C and the second search unit 16 execute the above-described complete match search and partial match search (including all of the front partial match, the rear partial match, and the partial match) as an ambiguous search. Shall.

  Through the processing in steps S301 to S304, the character string information acquisition unit 12 acquires search information ("ta / go" and "ta / go"). Subsequently, by the processing in step S305, the character count unit 13 obtains the number of characters (one character) of each of the character string information “ta” and “go” by counting.

  Subsequently, the search unit 15C executes a search process based on the number of characters counted in step S305 (step S306, search step). First, the search unit 15C acquires the character string information “ta” acquired by the character string information acquisition unit 12 (step S306A), and needs a process of converting a specific character (for example, the long sound conversion shown in FIG. 4). The process is executed accordingly (step S306B). Subsequently, the search unit 15C determines whether or not the number of characters for the character string information “ta” is one (step S306C). Here, since the number of characters of “ta” is one character, the search unit 15C performs a complete match search using the character string information “ta” (step S306D). Here, since the character string information database 13 does not store character string information that completely matches “ta”, there is no record that makes a search hit, and the search score of any record is not added. Thereafter, the same processing is performed on the next character string information “go” by the search unit 15C (steps S306F: NO, S306A, S306B, S306C: YES, S306D).

  After that, when processing for all speech recognition result candidates including the search processing (steps S307 to S309) by the second search unit 16 is completed (step S311: YES), the search result output unit 17 includes the search unit 15C and the search unit 15C. A representative score is extracted for the search result acquired by the second search unit 16. The search result output unit 17 displays the search results in the order of representative scores (“Kago” (1.6) → “Goto” (1) → “Tahara” (1) → “Tanabe” (1) →... The numbers in parentheses indicate the representative score.) And output to the output unit 105 such as a display provided in the information search device 1 (step S312, search result output step).

  As described above, the search unit 15C does not allow fuzzy search for single-character string information ("ta" and "go"), and uses the single-character string information by executing a complete match search. Thus, it is possible to reduce the influence of a large number of search errors that may occur when a fuzzy search is performed.

  Further, for example, a case where the user utters intentionally with the content of “Call Mr. Lee” and the character string information “Li” is obtained by correctly recognizing the speech. In this case, the search unit 15C does not simply exclude “Li”, but performs a complete match search using “Li”, so that a record including information desired by the user (“Li (Li)”) is obtained. ) Can be obtained (the search score of the record is increased, and the search result output unit 17 displays it at the top).

  The information search apparatus 1 according to the present embodiment has been described from the first example to the third example. In these examples, the second search unit 16 acquires the second character string information acquired as the search information. On the other hand, the search unit 15 performs the same process as the process executed on the first character string information corresponding to the second character string information according to the number of characters of the first character string information. The search may be executed using the second character string information.

  For example, in the example illustrated in FIG. 5, for the second character string information “ta” corresponding to the first character string information “ta” for which the search unit 15 has performed the exclusion process, the second search unit 16 uses the search unit. Exclusion processing may be performed in the same manner as 15. For example, when the search unit 15 determines that it is not suitable for use as search information as it is and executes processing such as exclusion and concatenation of the first character string information, it corresponds to the first character string information. The second character string information is also likely to be unsuitable for use as search information. Therefore, according to the above configuration, the same method (exclusion / concatenation processing, etc.) based on the same concept is used for the first character string information and the second character string information corresponding to the first character string information. And a search method (exact match search, partial match search, etc.)), search errors can be reduced more efficiently.

  According to the above configuration, in the above first to third examples, the fuzzy search using “field” and “go” is not executed by the second search unit 16, so that “Goto”, “Tahara”, And a record including “Tanabe” is not acquired as a search result, and the search score is not added. As a result, the search result output unit 17 displays only “protection” acquired as the search result. That is, according to the above configuration, it can be understood that display of search results unintended by the user (number of erroneous data items) can be suppressed.

  Further, the information search device 1 may be realized as a configuration combining the first example and the second example. That is, in the processing by the search unit 15, for the first character string information acquired as the search information, the two character strings information adjacent to each other are linked, and the character string information of one character not adjacent to each other is linked. It is good also as a structure to exclude.

  In addition, the information search device 1 may be realized as a configuration in which the second example and the third example are combined. That is, in the processing by the search unit 15, for the first character string information acquired as the search information, the two character strings information adjacent to each other are linked, and the character string information of one character not adjacent to each other is linked. It may be configured to execute an exact match search.

  In the present embodiment, an example in which the character string information database 14 is a phone book database and the information search device 1 is a device that provides a user with a phone book search function is shown, but the present invention is not necessarily limited thereto. It is not something. The character string information database 14 may be anything as long as it includes character string information that can be searched (for example, information indicating proper nouns such as person names and place names), and the information search device 1 is similar to, for example, telephone directory search. The present invention can be applied to a search device that performs mail search (search for a destination email address by a sender) and schedule search (search for a schedule of a specific individual) using character string information indicating a person's name as a search keyword. Further, the present invention can also be applied to a search device that performs a station name search, a map (place name) search, and the like using character string information indicating proper nouns such as station names and place names as search keywords.

  DESCRIPTION OF SYMBOLS 1 ... Information retrieval apparatus, 11 ... Voice recognition part, 12 ... Character string information acquisition part, 13 ... Character number counting part, 14 ... Character string information database, 15 (15A, 15B, 15C) ... Search part, 16 ... 2nd search Part, 17 ... search result output part.

Claims (8)

Character string information acquisition means for acquiring a plurality of character string information indicating character strings as one search information,
For each of the character string information acquired by the character string information acquisition means, a character number counting means for counting the number of characters of the character string indicated by the character string information;
Search means for acquiring a search result by executing a search according to the number of characters for each character string information counted by the character number counting means, using the search information acquired by the character string information acquiring means; ,
Search result output means for outputting the search results acquired by the search means,
The information search apparatus, wherein the search means excludes character string information having one character and performs a search using character string information having two or more characters. Character string information acquisition means for acquiring one or more character string information indicating a character string as one search information,
For each of the character string information acquired by the character string information acquisition means, a character number counting means for counting the number of characters of the character string indicated by the character string information;
Search means for acquiring a search result by executing a search according to the number of characters for each character string information counted by the character number counting means, using the search information acquired by the character string information acquiring means; ,
Search result output means for outputting the search results acquired by the search means ,
The character string information acquisition means acquires a plurality of character string information ordered in advance,
The search means determines whether or not two character string information having a different number of characters are adjacent to each other based on the ordering, and if the two character string information are adjacent to each other, the two characters An information search apparatus that generates a new character string by concatenating character strings indicated by the column information, and executes a search using character string information including two or more characters including information indicating the new character string . The searching means, for the search character is using the character string information search and the number of characters is more than 2 characters using the character string information of a character, performs a search based on different search criteria, according to claim 2, wherein Information retrieval device. Further comprising speech recognition means for inputting the user's series of uttered speech and generating the character string information by performing speech recognition processing on the uttered speech;
The character string information obtaining means obtains the character string information generated by the speech recognition means, information retrieval apparatus of any one of claims 1-3. A second search means for acquiring a search result by executing a search using character string information indicated by a second character type different from the first character type;
The character string information acquisition means corresponds to the first character string information indicated by the first character type as the search information and the first character string information and indicated by the second character type. Second character string information to be acquired,
The character number counting means and the search means execute each process on the first character string information,
The second search means executes a search using the second character string information,
The search result output means outputs the search result based on the second search result obtained by the first search result and the second search unit obtained by the search unit, according to claim 1-4 The information search device according to any one of the above. The second search means for the second character string information, the search means for the first character string information for the first character string information corresponding to the second character string information. The information search device according to claim 5 , wherein the same processing as the processing executed according to the number of characters is executed, and the search is executed using the second character string information. As one piece of search information, a character string information acquisition step for acquiring a plurality of character string information indicating character strings,
For each of the character string information acquired in the character string information acquisition step, a character number counting step for counting the number of characters of the character string indicated by the character string information;
A search step for acquiring a search result by executing a search according to the number of characters for each character string information counted in the character number counting step, using the search information acquired in the character string information acquisition step; ,
A search result output step for outputting the search result acquired in the search step,
An information search method for an apparatus, wherein in the search step, character string information having one character is excluded, and search is performed using character string information having two or more characters. A character string information acquisition step of acquiring one or more character string information indicating a character string as one piece of search information;
For each of the character string information acquired in the character string information acquisition step, a character number counting step for counting the number of characters of the character string indicated by the character string information;
A search step for acquiring a search result by executing a search according to the number of characters for each character string information counted in the character number counting step, using the search information acquired in the character string information acquisition step; ,
See containing and a search result output step of outputting the acquired search result in the search step,
In the character string information acquisition step, a plurality of pre-ordered character string information is acquired,
In the search step, based on the ordering, it is determined whether or not two character string information having a different number of characters are adjacent to each other. If the two character string information are adjacent to each other, the two characters A device information search in which character strings indicated by column information are concatenated to generate a new character string, and a search is performed using character string information having two or more characters including information indicating the new character string . Method.

Images