JP2004171222A - Information extracting device and method and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an information extracting device, method and a program capable of precisely extracting information associated with desired contents from an inputted character string. <P>SOLUTION: A corresponding relation between a plurality of rule morphemes and information contents classifications indicating the classifications of the information contents of the rule morphemes is shown by rule data. One or more input morphemes are extracted from an input morpheme 21 obtained by decomposing an input character string 20 per part of speech by an information extracting means so that an input morphemic column 22 can be configured. The input morphemic column 22 is collated with the respective rule morphemes of the rule data so that the input morphemic column 22 of the specific information contents classification can be retrieved based on the information contents classification corresponding to the rule morphemes made coincident with the input morphemic column 22. Thus, the obtained input morphemic column 22 of the specific information contents classification can be extracted as desired information. <P>COPYRIGHT: (C)2004,JPO

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to an information extracting apparatus, method, and program, and more particularly to an information extracting apparatus, method, and program for extracting desired information relating to specific contents from an input character string.
[0002]
[Prior art]
2. Description of the Related Art With the spread of computers, a technique for realizing a man-machine interface between a human and a computer has attracted attention. These technologies aim to be able to interact with computers without burden using basic human communication methods.
In such a technology, when a computer automatically analyzes a natural language used daily by a human, a speech recognition technology that automatically converts a word spoken by a human into a character string is used. Along with the voice recognition processing, an information extraction technique for extracting desired information from a character string is also important.
[0003]
Conventionally, many techniques using morphological analysis techniques have been proposed as techniques for extracting desired information from such a character string of a natural language.
The morphological analysis is to analyze a component of the character string by decomposing a character string composed of a natural language into a plurality of words for each part of speech (for example, see Non-Patent Document 1).
On the other hand, natural languages have a feature in the syntax (sentence expression pattern) at the part of speech level. In the conventional information extraction technology, a desired sentence is extracted from a natural language by performing a morphological analysis on a character string composed of a natural language and extracting the features of the obtained part-of-speech-level syntax (for example, Japanese Patent Technical document 1 etc.).
[0004]
The applicant has not found any prior art documents related to the present invention other than the prior art documents specified by the prior art document information described in this specification by the time of filing.
[Patent Document 1]
JP-A-8-77196
[Non-patent document 1]
Yuji Matsumoto et al., "Morphological Analysis System ChaSen", Nara Institute of Science and Technology, [Search November 11, 2002], Internet <URL: http // chasen. aist-nara. ac. jp / index. html. ja>
[0005]
[Problems to be solved by the invention]
However, in such a conventional information extraction technology, since the matching with the input character string is performed by using the syntax feature at the part of speech level, that is, the sentence expression pattern, the sentence expression pattern of the sentence prepared in advance is used. Although a close sentence can be extracted, there is a problem that it is not possible to accurately extract only information relating to specific contents, such as a date and time, a partner, a place, and an action, included in an input character string.
An object of the present invention is to solve such a problem, and an object of the present invention is to provide an information extracting device, an information extracting method, and an information extracting method capable of accurately extracting information about desired contents from an input character string.
[0006]
[Means for Solving the Problems]
In order to achieve such an object, an information extraction device according to the present invention decomposes an input character string into morphemes in a part of speech unit, and extracts desired information regarding specific information content from a character string based on the obtained morpheme. In the information extracting device to extract, a correspondence relation between a plurality of rule morphemes obtained by previously decomposing an arbitrary character string including specific information content into morphemes and an information content type indicating a type of information content of the rule morpheme is determined. Rule data and one or more input morphemes taken out of the input morpheme obtained by decomposing the input character string by the part of speech unit to form an input morpheme sequence. The input morpheme sequence and each rule morpheme of the rule data are By matching, an input morpheme string of a specific information content type is searched for based on the information content type associated with the rule morpheme that matches the input morpheme string, and obtained. Input morpheme string of specific information content type that is intended and a data extracting means for extracting a desired information.
[0007]
When constructing the input morpheme string, the information extraction means may constitute the input morpheme string from a plurality of input morphemes successively extracted from the original of the input character string and thus from the input morpheme.
When searching for a morpheme string, if the information extraction means cannot obtain an input morpheme string corresponding to a specific information content type, a new input morpheme string that has been shortened by reducing the number of input morphemes constituting the input morpheme string May be used for re-matching.
[0008]
When collating the morpheme string, the information extraction means may collate each input morpheme constituting the input morpheme string with each rule morpheme of the rule data based on the sequence of the respective morphemes. Alternatively, each input morpheme constituting the input morpheme sequence may be collated with each rule morpheme of the rule data based on the part of speech of each morpheme.
When collating the morpheme string, the information extraction unit collates each input morpheme constituting the input morpheme string with each rule morpheme of the rule data based on the character string of each morpheme, and matches the input morpheme string. When there is no rule morpheme, each input morpheme constituting the input morpheme sequence and each rule morpheme of the rule data may be collated based on the part of speech of each morpheme.
[0009]
Regarding the configuration of rule data, it shows a plurality of rule morphemes obtained by decomposing a case character string prepared in advance for each part of speech, and the type of information content to which the rule morpheme belongs, which is associated with each rule morpheme. Rule data including the information content type may be used.
[0010]
Further, the information extraction method according to the present invention is used in an information extraction device that decomposes an input character string into morphemes in a part of speech unit and extracts desired information on specific information content from the character string based on the obtained morpheme. In the information extraction method, a first step of extracting one or more input morphemes from an input morpheme obtained by decomposing an input character string into parts of speech to form an input morpheme string, and an optional step including a specific information content Each rule morpheme of rule data indicating a correspondence relationship between a plurality of rule morphemes obtained by previously decomposing a character string into morphemes and an information content type indicating a type of information content of the rule morpheme is obtained in the first step. By comparing the input morpheme string with the input morpheme string, a specific information content type is input based on the information content type associated with the rule morpheme that matches the input morpheme string. Those comprising a second step of searching for morphemes column, and a third step of extracting the input morphemes obtained by the search as the desired information.
[0011]
When constructing the input morpheme sequence, in the first step, the input morpheme sequence may be constructed from a plurality of input morphemes that are successively extracted from the original sequence of the input character string and thus from the input morpheme.
When searching for a morpheme string, if the input morpheme string corresponding to the specific information content type cannot be obtained in the second step, a new input morpheme that has been shortened by reducing the number of input morphemes constituting the input morpheme string Re-matching may be performed using a column.
[0012]
When collating the morpheme sequence, in the second step, each input morpheme constituting the input morpheme sequence and each rule morpheme of the rule data may be collated based on the character sequence of each morpheme. Alternatively, each input morpheme constituting the input morpheme sequence may be collated with each rule morpheme of the rule data based on the part of speech of each morpheme.
When matching the morpheme string, in the second step, each input morpheme constituting the input morpheme string and each rule morpheme in the rule data are matched based on the sequence of the respective morphemes, and match with the input morpheme string. If there is no rule morpheme to be executed, each input morpheme constituting the input morpheme sequence may be collated with each rule morpheme of the rule data based on the part of speech of each morpheme.
[0013]
Regarding the configuration of rule data, it shows a plurality of rule morphemes obtained by decomposing a case character string prepared in advance for each part of speech, and the type of information content to which the rule morpheme belongs, which is associated with each rule morpheme. Rule data including the information content type may be used.
[0014]
Further, the program according to the present invention is a computer of an information extraction device that decomposes an input character string into morphemes in a part of speech unit and extracts desired information related to specific information content from character data based on each obtained morpheme. Is a program for executing any one of the information extraction methods described above.
[0015]
BEST MODE FOR CARRYING OUT THE INVENTION
Next, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 is a block diagram showing a configuration of an information extraction device according to one embodiment of the present invention.
The information extraction device 10 decomposes the input character string of the input natural language into morphemes in the unit of part of speech and, based on the obtained information content type corresponding to each morpheme, obtains desired information relating to specific content from the input character string. Is a device for extracting
The information extraction device 10 is generally composed of a computer such as a server device, and has an input / output interface unit (hereinafter, referred to as an input / output I / F unit) 1, an operation input unit 2, a screen display unit 3, and a storage unit 4. , And a control unit 5 are provided.
[0016]
The input / output I / F unit 1 receives an input character string or an input character string from an information processing device (not shown) connected via a communication line 6 or a recording medium 9 such as a CD-ROM or a flexible disk. It is a circuit section for exchanging information extracted from character strings and various data such as programs.
The operation input unit 2 includes a keyboard, a mouse, and the like, and is an input device for performing operation input of various data such as input character strings and instructions for various processes.
The screen display unit 3 is a screen display device that includes an LCD, a CRT, and the like, and that displays information extracted from an input character string and a processing state on a screen.
[0017]
The storage unit 4 is composed of a hard disk or a memory, and various types of information used for processing operations in the control unit 5, such as a large number of rule data 4A used for information extraction processing in the control unit 5 and programs 4B executed by the control unit 5. Is a storage device for storing.
The rule data 4A is data indicating a correspondence between a plurality of rule morphemes obtained by morphological analysis of a character string in a natural language and an information content type indicating information content of the rule morpheme.
The program 4B is prefetched from the recording medium 9 or the communication line 6 via the input / output I / F unit 1 and stored in the storage unit 4.
[0018]
The control unit 5 includes a microprocessor such as a CPU and its peripheral circuits. The control unit 5 reads and executes the program 4B in the storage unit 4, thereby causing the program 4B to cooperate with the hardware resources of its own device to extract information. Implement functional means for performing processing.
As this function means, there are an information extraction means 5A, a morphological analysis means 5B, and a rule generation means 5C.
[0019]
The information extraction unit 5A analyzes a character string of a natural language input from the input / output I / F unit 1 or the operation input unit 2 into a morpheme by the morpheme analysis unit 5B, and obtains an input composed of one or more input morphemes obtained. This is a functional unit that extracts desired content information from an input character string by collating the morpheme string with each rule data 4A in the storage unit 4.
The morphological analysis unit 5B is a functional unit that decomposes a character string in a natural language into a plurality of morphemes for each part of speech such as a noun, a particle, or a verb.
[0020]
The rule generation unit 5C analyzes a case character string composed of a character string of a natural language prepared in advance into a morpheme by the morphological analysis unit 5B, and the type of information content to which the rule morpheme belongs to a plurality of obtained rule morphemes. That is, it is a functional unit that generates the rule data 4A by associating the information content types, and stores the rule data 4A in the storage unit 4.
[0021]
Next, the operation of the information extraction device 10 according to the present exemplary embodiment will be described with reference to the drawings.
First, the morphological analysis performed by the morphological analysis unit 5B of the control unit 5 will be schematically described with reference to FIG. FIG. 2 is an explanatory diagram illustrating a morphological analysis process performed by the morphological analysis unit 5B.
The information extraction device 10 according to the present embodiment aims to extract information of specific content as desired information 30 from an input character string 20 of a natural language as shown in FIG. Here, an example will be described in which information relating to a date and time, a partner, a place, and an action is extracted as information having specific contents.
[0022]
Morphological analysis is a process of decomposing a character string in a natural language into morphemes in units of parts of speech such as nouns, particles, and verbs. The part of speech is a name when a character string having a meaning is classified by its property, and the unit of the character string decomposed in the unit of part of speech, that is, a morpheme is the shortest character string having a meaning.
For the morphological analysis performed by the morphological analysis unit 5B, the above-described known morphological analysis method may be used (see Non-Patent Document 1, etc.). In a general morphological analysis process, a character string is decomposed into morphemes using a dictionary in which many specific examples of part of speech are registered in advance.
[0023]
For example, as shown in FIG. 2, when the input character string 20 of “meeting with Mr. Murakami on October 18 at Fujisawa” is subjected to morphological analysis, a plurality of morphemes including character data 20A and its parts of speech 20B and 20C (hereinafter referred to as “morphological”) , A morpheme obtained from the input character string is referred to as an input morpheme) 21.
For example, parts of speech 20B and 20C of “noun” and “numeral” are assigned to character data 20A of “10”, and these constitute one input morpheme 21 as a set.
As for the part of speech, for example, “noun” includes detailed classifications such as “numerical”, “personal name”, and “place name”. The deeper the classification, the higher the matching accuracy, but the longer the matching time. In this example, a two-stage depth classification is used, but the classification depth may be arbitrarily adjusted in consideration of the matching accuracy and the matching required time.
[0024]
Here, from the input character string 20 of “meeting with Murakami on October 18 at Fujisawa”, the date and time “October 18”, the partner “Murakami”, the place “Fujisawa”, and the action “meeting” When extracting desired information 30, it is necessary to know which character string constituting the input character string 20 indicates which information, that is, its information content type 20M.
[0025]
In the present embodiment, an input morpheme sequence 22 that is obtained by extracting one or more input morphemes from a plurality of input morphemes 21 obtained by decomposing with such parts of speech forms a character string that is useful information. It focuses on doing.
Then, by using the input morpheme string 22 as a unit, by comparing the rule data 4A having a plurality of rule morphemes set in relation to the information content type with each input morpheme string 22 included in the input character string 20, An input morpheme string of a specific information content type is searched, and the obtained input morpheme string of the specific information content type 20M is extracted as desired information.
[0026]
Here, the rule morpheme is a morpheme obtained by morphological analysis of an example sentence including an arbitrary character string, preferably, information of a type to be extracted. The rule data 4A is obtained by associating rule morphemes obtained from a plurality of example sentences, ie, case character strings, with information contents of the character strings.
Focusing on useful information that has a part-of-speech sequence, when comparing the input character string with the rule data, in addition to the method of matching the arrangement of characters constituting the morpheme string, the part-of-speech arrangement Is collated.
[0027]
Next, the rule data will be described with reference to FIGS. FIG. 3 is a configuration example of the rule data. FIG. 4 is a flowchart showing the rule generation processing in the rule generation means 5C.
A large number of rule data 41, 42,... Are registered in the storage unit 4 as rule data 4A. The rule data 41 stores information constituting each rule morpheme 40, that is, a character (character data) 41A, parts of speech 41B, 41C... And an information content type 41M of information included in the character data 41A as a set. The other rule data 42 have the same configuration as the rule data 41.
[0028]
Such rule data 4A is generated by the rule generation means 5C of the control unit 5 and stored in the storage unit 4.
The rule generation means 5C executes a rule generation process shown in FIG. 4 according to an instruction from the operation input unit 2.
[0029]
First, the input character string input for the case is morphologically analyzed by the morphological analysis means 5B to decompose it into a plurality of rule morphemes (step 200). The content type is set (step 201).
The setting of this information content type may be made by the user with respect to the input character string for the case, or a character string whose character position indicating the information of the information content type is known as the input character string for the case. May be used.
[0030]
Then, each rule morpheme is associated with the information content type of the rule morpheme, and registered in the storage unit 4 as the rule data 4A in the configuration as shown in FIG. 4 (step 202), and a series of rule generation processing To end.
Although FIG. 4 illustrates an example in which all morphemes included in the input character string for the case are registered as the rule data 4A regardless of the presence or absence of the information content type, the present invention is not limited to this.
For example, only a rule morpheme having a clear information content type or a column thereof may be registered as the rule data 4A, and the size of the rule data 4A can be reduced, and the time required for collation can be reduced.
[0031]
Next, the information extraction processing in the information extraction means 5A will be described with reference to FIGS. FIG. 5 is a table showing the matching mode in the information extraction process. FIG. 6 is a flowchart showing the information extraction processing in the information extraction means 5A. FIG. 7 is a flowchart showing the collation processing between the input morpheme string and the rule data.
The information extraction means 5A of the control unit 5 starts the information extraction processing of FIG. 6 in response to an instruction from the operation input unit 2 or an external instruction via the input / output I / F unit 1.
First, a morphological analysis is performed using a morphological analysis unit 5B on an input character string input from the outside via the operation input unit 2 or the input / output I / F unit 1 as a processing target (step 100).
[0032]
Then, an unprocessed information content type is selected from desired information to be extracted from the input character string (step 101), and a collation mode is selected based on a predetermined priority (step 102). The collation mode is a rule for collating an input morpheme string with rule data, and is set for each information content type.
Here, each matching is determined by a combination of a matching level indicating whether the two morpheme strings are compared with “character data” or “part of speech” and the length of the morpheme string to be matched, that is, the number of morphemes constituting the morpheme string. The mode is configured.
[0033]
In each matching mode, the matching accuracy and required matching time obtained by the combination of the matching level and the morpheme sequence length are different. Therefore, the order of using the matching mode is set as a priority order according to these performances. If the desired information cannot be obtained from the input character string by the collation based on the mode, the collation mode of the next order is selected.
[0034]
For example, FIG. 5 shows an example of a collation mode table for date and time information. According to the priorities of the collation mode table, collation using “character data” has priority over “part of speech” as the matching level. The morpheme sequence length is used first for each matching level as a morpheme sequence length, and thereafter the shorter morpheme sequence length is used up to “1”.
In general, the shorter the morpheme length, the higher the success rate of matching, but at the same time the noise increases. Therefore, in matching, if the matching level is the same, it is only necessary to start from a certain morpheme length, and if the matching is not successful, the matching may be repeated while shortening the morpheme and narrowing the matching width.
[0035]
If the morpheme lengths are the same, the matching level, that is, the depth of classification of each morpheme may be reduced from a deep place to a shallow place, and finally the collation using only the part of speech may be performed.
Note that the order of the matching modes may be referred to at the time of the information extraction processing, or may be incorporated in the information extraction processing program.
[0036]
In this manner, after selecting the collation mode in step 102 described above, the collation processing of FIG. 7 is executed based on the collation mode. An input morpheme string is searched (step 103).
Here, when the search of the input morpheme string of the information content type is successful (step 104: YES), the characters of the input morpheme string are extracted as desired information corresponding to the information content type (step 105). .
[0037]
If there is an unprocessed information content type (step 106: YES), the process returns to step 101 to repeatedly execute a search process for desired information on the unprocessed information content type.
Here, if there is no unprocessed information content type (step 106: NO), the desired information extracted so far is displayed and output on the screen display unit 3, or is output via the input / output I / F unit 1. The information is output to the outside (step 107), and a series of information extraction processing ends.
If the search of the input morpheme string of the information content type fails in step 104 described above (step 104: NO), it is determined whether or not there is an unprocessed collation mode (step 108).
[0038]
If there is an unprocessed matching mode (step 108: YES), the process returns to step 102 to select a matching mode of the next priority and change the matching level and the morpheme length of the input morpheme string. The extraction processing of the desired information is repeatedly executed.
If there is no unprocessed collation mode (step 108: NO), it is determined that the information corresponding to the information content type does not exist in the input character string, and the process proceeds to step 106 described above. Performs processing related to the information content type.
[0039]
Next, the collation processing in step 103 will be described in detail with reference to FIG.
The information extraction means 5A first sets a matching level and a morpheme string length based on the collation mode selected in the above-mentioned step 102 (step 110), and from the input character string, the input morpheme which has not been processed by the morpheme string length. Is obtained as an input morpheme sequence (step 111).
At this time, of the morphemes obtained from the input character string, for example, the head of the input character string is extracted along the original sequence and the morpheme of the morpheme string length is extracted as the extraction start position. May be moved backward to sequentially extract morphemes for the morpheme sequence length.
[0040]
Next, unprocessed rule data is selected from the rule data 4A (step 112), and it is checked whether or not a rule morpheme string that matches the input morpheme string is included in the rule data 4A (step 113).
At this time, collation is performed based on the matching level of the collation mode.
That is, when the matching level is “character data”, the arrangement of the character data of the input morpheme string is compared with the arrangement of the character data of the rule morpheme in the rule data. On the other hand, when the matching level in the matching mode is “part of speech”, the arrangement of the part of speech of the input morpheme sequence is compared with the arrangement of the part of speech of the rule morpheme in the rule data.
[0041]
In this way, the collation is performed in step 113, and if a rule morpheme sequence that matches the input morpheme sequence is not found in the rule data (step 113: NO), if there is unprocessed rule data, (Step 114: YES), the process returns to the above-described step 112, selects unprocessed rule data, and repeatedly performs collation with the rule data.
[0042]
On the other hand, if there is no unprocessed rule data (step 114: NO), and if there is an unprocessed input morpheme string (step 115: YES), the process returns to step 111 to acquire an unprocessed input morpheme string. Then, the matching for the input morpheme sequence is repeatedly performed.
If there is no unprocessed input morpheme string (step 115: NO), it is determined that the search for the input morpheme string of the information content type in the matching mode has failed (step 116), and a series of matching in the matching mode is performed. The process ends.
[0043]
When the same rule morpheme sequence as the input morpheme sequence is found in the rule data in step 113 described above (step 113: YES), the information content type associated with each found rule morpheme is checked. (Step 117).
[0044]
Here, when the information content type of each rule morpheme does not match the desired information content type (step 117: NO), the process proceeds to step 114 described above, and processing is performed on unprocessed rule data.
On the other hand, if all the information content types associated with each found rule morpheme match the desired information content type (step 117: YES), it is determined that the search is successful (step 118), and a series of collation processing is performed. To end.
[0045]
As described above, rule data indicating a correspondence relationship between a plurality of rule morphemes obtained by previously decomposing an arbitrary character string including specific information content into morphemes and an information content type indicating a type of information content of the rule morpheme 4A, one or more input morphemes are taken out from the input morpheme obtained by decomposing the input character string into parts of speech by the information extraction means 5A, and an input morpheme sequence is constructed. By collating each rule morpheme, the input morpheme string of the specific information content type is searched based on the information content type associated with the rule morpheme that matches the input morpheme string, and the obtained specific information is obtained. Since the input morpheme string of the content type is extracted as the desired information, information relating to the desired content can be accurately extracted from the input character string.
[0046]
At this time, the information extracting means 5A constructs the input morpheme sequence from a plurality of input morphemes which are successively extracted from the original sequence of the input character strings and, hence, from the input morpheme. It is possible to extract desired information with higher accuracy.
When the input morpheme sequence corresponding to the specific information content type cannot be obtained, the information extracting means 5A re-uses the input morpheme sequence constituting the input morpheme sequence by reducing the number of input morpheme sequences to shorten the input morpheme sequence. Since search is performed, desired information can be extracted flexibly while giving priority to collation accuracy.
[0047]
In addition, the information extraction means 5A matches each input morpheme constituting the input morpheme string and each rule morpheme of the rule data based on the character sequence of each morpheme at the time of matching. Information is extracted only when a morpheme string that matches the character arrangement of the character string exists in the rule data, and desired information can be extracted with high accuracy.
In addition, the information extraction means 5A matches each input morpheme constituting the input morpheme string and each rule morpheme of the rule data based on the part of speech of each morpheme at the time of matching. If a morpheme string that matches the part-of-speech sequence exists in the rule data, the information will be extracted. Information can be extracted.
[0048]
In addition, at the time of matching, the information extracting means 5A checks each input morpheme constituting the input morpheme string and each rule morpheme of the rule data based on the sequence of the respective morphemes, and matches the input morpheme string. If the rule morpheme does not exist, each input morpheme constituting the input morpheme sequence and each rule morpheme of the rule data are collated based on the part of speech of each morpheme. When the rule data does not have exactly the same character sequence as the input character string, the matching with the part-of-speech sequence automatically performs flexible matching over a wide range. As a result, desired information can be flexibly extracted in a wide range while considering high accuracy, and the probability that desired information can be extracted can be improved.
[0049]
Further, the rule generation unit 5C uses a plurality of rule morphemes obtained by decomposing the case character string prepared in advance for each part of speech, and the type of information content to which the rule morpheme belongs, which is associated with each rule morpheme. Since the rule data consisting of the information content type indicating the information content is generated, by using this rule data for collation, even if the input character string has an unknown information content type, the desired information can be efficiently and accurately obtained. Can be extracted.
[0050]
Next, a specific example of the information extraction operation will be described with reference to FIG. FIG. 8 shows a specific example of the information extraction operation.
Here, as an example, it is assumed that date and time information is extracted from the input character string as desired information, the matching mode is set, the matching level is “character data”, and the morpheme string length is “4”. explain.
When a character string “Met with Mr. Murakami on October 8 at Fujisawa” is input as the input character string 20, this input character string is decomposed into 11 morphemes “10” to “meeting” . In this example, since the matching level is "character data", character data 20A of each morpheme is to be processed.
[0051]
In the collation processing of FIG. 7 described above, morphemes with a morpheme string length = 4 are extracted from the character data 20A, and an input morpheme string 51 is first generated from the extracted morpheme character data (step 111). In addition, according to the mismatch of the collation, the input morpheme sequence 52,..., 5n is sequentially generated by moving the extraction start position by one morpheme.
[0052]
Then, the extracted input morpheme string 51 is collated with each rule morpheme included in the character data 41A, 42A,... Of each rule data of the rule data 4A (step 113). In this example, the input morpheme string 51 of “October 18” exists in the character data 42A of the rule data, and the arrangement of both character data matches.
At this time, each rule morpheme found in the character data 42A of the rule data is associated with the information content type 42M of “date and time”, which matches the desired information content type (step 117). The character data string of the input morpheme string is extracted as desired date and time information (FIG. 6: step 105).
[0053]
Next, another specific example of the information extracting operation will be described with reference to FIG. FIG. 9 shows another specific example of the information extracting operation.
Here, it is assumed that the other party information is extracted from the input character string as desired information, the matching mode is set, the matching level is “part of speech”, and the morpheme string length is “2”. I do.
First, the input character string 20 is decomposed into morphemes. In this example, since the matching level is "part of speech", the input strings "noun,""noun,""noun,""noun,""particle,""noun,""suffix,""noun,""noun,""particle," and "verb" Is the processing target.
[0054]
In the collation processing of FIG. 7 described above, morphemes are extracted from the part of speech 20B by two morpheme string lengths, and an input morpheme string 61 is first generated from the part of speech of the extracted morpheme (step 111). It should be noted that the input morpheme strings 62,..., 6k,.
Then, the extracted input morpheme string 61 is collated with each rule morpheme included in the part of speech 41B, 42B,... Of each rule data of the rule data 4A (step 113). In this example, the input morpheme sequence 6k of “noun” and “suffix” generated thereafter is present in the part of speech 41B of the rule data, and the arrangement of the two parts of speech matches.
[0055]
At this time, each rule morpheme found in the part-of-speech 41B of the rule data is associated with the information content type 41M of “partner”, which matches the desired information content type (step 117). A character data string of a morpheme string is extracted as desired partner information (FIG. 6: step 105).
Note that the comparison of the morpheme strings at the part of speech level is performed at a predetermined depth of the classification of the part of speech. At this time, the comparison may be started from the deepest level of the classification depth prepared as the rule data 4A, and the comparison may be performed at the shallower level in order according to the mismatch. It is possible to extract desired information flexibly while doing so.
[0056]
【The invention's effect】
As described above, the present invention relates to a method in which a plurality of rule morphemes obtained by preliminarily decomposing an arbitrary character string including specific information content into morphemes and an information content type indicating a type of information content of the rule morpheme. Rule data indicating a correspondence relationship is provided, and information extraction means extracts one or more input morphemes from input morphemes obtained by decomposing the input character string in units of part of speech to form an input morpheme sequence. By matching the rule morphemes of the rule data with the rule morphemes of the rule data, the input morpheme string of the specific information content type is searched based on the information content type associated with the rule morpheme that matches the input morpheme string, and Since the input morpheme string of the specific information content type is extracted as desired information, information on desired contents can be accurately extracted from the input character string.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of an information extraction device according to an embodiment of the present invention.
FIG. 2 is an explanatory diagram showing a morphological analysis process.
FIG. 3 is a configuration example of rule data.
FIG. 4 is a flowchart illustrating a rule data generation process.
FIG. 5 is an example of a collation mode table.
FIG. 6 is a flowchart illustrating information extraction processing.
FIG. 7 is a flowchart illustrating a collation process.
FIG. 8 is a specific example of an information extraction operation.
FIG. 9 is another specific example of the information extracting operation.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 10 ... Information extraction device, 1 ... Input / output I / F part, 2 ... Operation input part, 3 ... Screen display part, 4 ... Storage part, 4A ... Rule data, 4B ... Program, 5 ... Control part, 5A ... Information extraction Means, 5B: Morphological analysis means, 5C: Rule generation means, 6: Communication line, 9: Recording medium, 20: Input character string, 20A: Character data, 20B, 20C: Part of speech, 20M: Information content type, 21: Input Morpheme, 22 ... input morpheme sequence, 30 ... desired information, 40 ... rule morpheme, 41, 42 ... rule data, 41A, 42A ... character data, 41B, 41C, 42B, 42C ... part of speech, 41M, 42M ... information content type , 51, 52, 53, 5n ... input morpheme strings (character data), 61, 62, 6k, 6n ... input morpheme strings (part of speech).

Claims (15)

In an information extraction device that decomposes an input character string into morphemes in a part of speech unit and extracts desired information related to specific information content from the character string based on the obtained morpheme,
Rule data indicating a correspondence relationship between a plurality of rule morphemes obtained by previously decomposing an arbitrary character string including the specific information content into morphemes and an information content type indicating a type of information content of the rule morpheme,
By extracting one or more input morphemes from input morphemes obtained by decomposing an input character string into parts of speech, constructing an input morpheme sequence, and comparing the input morpheme sequence with each rule morpheme of the rule data, Searching for an input morpheme string of a specific information content type based on the information content type associated with the rule morpheme that matches the input morpheme string, An information extraction device, comprising: an information extraction unit that extracts the information as the information. The information extraction device according to claim 1,
The information extracting device, wherein the information extracting means constructs the input morpheme string from a plurality of input morphemes which are successively extracted from the original of the input character string and thus the input morpheme. The information extraction device according to claim 1,
When the input morpheme sequence corresponding to the specific information content type cannot be obtained, the information extracting unit re-uses the input morpheme sequence constituting the input morpheme sequence by reducing the number of input morpheme sequences to shorten the input morpheme sequence. An information extraction device characterized by performing collation. The information extraction device according to claim 1,
The information extracting device, wherein the information extracting means performs a collation between each input morpheme constituting the input morpheme string and each rule morpheme of the rule data based on a character string of each morpheme. . The information extraction device according to claim 1,
The information extracting device, wherein the information extracting means performs a collation between each input morpheme constituting the input morpheme string and each rule morpheme of the rule data based on a part of speech of each morpheme. . The information extraction device according to claim 1,
The information extracting means compares each input morpheme constituting the input morpheme string and each rule morpheme of the rule data based on a character string of each morpheme at the time of the matching, and matches the input morpheme string. An information extracting apparatus characterized in that, when a rule morpheme does not exist, each input morpheme constituting the input morpheme sequence is collated with each rule morpheme of the rule data based on a part of speech of each morpheme. The information extraction device according to claim 1,
The rule data includes a plurality of rule morphemes obtained by decomposing a case character string prepared in advance for each part of speech, and information indicating the type of information content to which the rule morpheme belongs, which is associated with each rule morpheme. An information extraction device characterized by comprising a content type. An information extraction method used in an information extraction device that decomposes an input character string into morphemes of a part of speech unit and extracts desired information related to specific information content from the character string based on the obtained morpheme,
A first step of extracting one or more input morphemes from an input morpheme obtained by decomposing an input character string in units of part of speech to form an input morpheme sequence;
Each rule of rule data indicating a correspondence relationship between a plurality of rule morphemes obtained by previously decomposing an arbitrary character string including the specific information content into morphemes and an information content type indicating a type of information content of the rule morpheme By comparing the morpheme with the input morpheme string obtained in the first step, a specific information content type is identified based on the information content type associated with the rule morpheme that matches the input morpheme string. A second step of searching for an input morpheme sequence;
Extracting the input morpheme string obtained by the search as the desired information. The information extraction method according to claim 8,
The information extracting method according to claim 1, wherein the first step comprises constructing the input morpheme string from a plurality of input morphemes successively extracted from the original of the input character string and thus from the input morpheme. The information extraction method according to claim 8,
In the second step, when an input morpheme string corresponding to a specific information content type cannot be obtained, the input morpheme string is reduced again by using a new input morpheme string that is reduced in number to form the input morpheme string. An information extraction method characterized by collating. The information extraction method according to claim 8,
The second step is characterized in that at the time of the collation, each input morpheme constituting the input morpheme sequence is collated with each rule morpheme of the rule data based on a character string of each morpheme. Method. The information extraction method according to claim 8,
The second step is characterized in that at the time of the collation, each input morpheme constituting the input morpheme sequence is collated with each rule morpheme of the rule data based on a part of speech of each morpheme. Method. The information extraction method according to claim 8,
In the second step, at the time of the collation, each input morpheme constituting the input morpheme string is collated with each rule morpheme of the rule data based on a character string of each morpheme, and matches with the input morpheme string. If there is no rule morpheme to be executed, each input morpheme constituting the input morpheme string is collated with each rule morpheme of the rule data based on the part of speech of each morpheme. The information extraction method according to claim 8,
The first step includes, as the rule data, a plurality of rule morphemes obtained by decomposing a case character string prepared in advance for each part of speech, and the rule morphemes associated with each of the rule morphemes belong. An information extraction method characterized by using rule data including an information content type indicating a type of information content. 15. A computer of an information extracting device for decomposing an input character string into morphemes in a part of speech unit and extracting desired information relating to specific content from the character data based on each of the obtained morphemes. A program for executing the information extraction method according to one aspect.

Images