JP4468608B2 - Semantic information estimation device, semantic information estimation method, and program - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
  The present invention provides a semantic information estimation device, a semantic information estimation method,And programsAbout.
[0002]
[Prior art]
In a word dictionary used for so-called language processing, information on the meaning (classification) of words is registered in addition to information on notation, reading, and utilization of words. Thus, the meaning information of the words registered in the word dictionary is very effective at the time of document search processing and document classification processing.
[0003]
By the way, when searching for a predetermined document from a large-scale document database, a search using a keyword is common. In such a document search using keywords, in actuality, for example, in addition to a case where a search is performed using a specific keyword “Ox Electric” which is a manufacturer that manufactures and sells OA equipment, a concept higher than “Ox Electric” is used. There is a case where the search is performed by a keyword such as “OA device manufacturer”.
[0004]
However, in ordinary documents, although there is a proper noun such as “○ × Electricity” which is the company name, it is normal that the word representing the superordinate concept of “○ × Electricity” (“OA equipment manufacturer”) is not expressed. is there.
[0005]
Therefore, as one measure for solving such a problem, there is a method of giving semantic information such as a business type to which the company belongs to a proper noun such as a company name. However, the assignment of information related to meaning and knowledge to the word dictionary and the creation of the word dictionary require considerable labor, knowledge, and technology. For this reason, various methods have been considered for giving various information related to meaning and knowledge to the word dictionary.
[0006]
Japanese Patent Laid-Open No. 6-223109 discloses an efficient method for acquiring knowledge information by analyzing a natural language sentence and registering a database for a specific object.
[0007]
Japanese Patent Laid-Open No. 7-85071 discloses a method for extracting information including a relationship between words using syntax information in a fixed format as a method for estimating a word based on syntax information. According to the technique disclosed in Japanese Patent Laid-Open No. 7-85071, when extracting information including relations between words in a certain format using syntax information, the words of unregistered words and their meanings are estimated. Yes. For example, with respect to the expression “XX Electric has developed a sputtering device”, even if “XX Electric” is an unknown word, the method of recognizing as a device developer from the co-occurring expression “Developing the device” Has proposed.
[0008]
[Problems to be solved by the invention]
However, according to the technique disclosed in Japanese Patent Laid-Open No. 6-223109, it is assumed that a certain amount of semantic information is registered in advance in the word dictionary. That is, it is necessary to preliminarily define the meaning of the target area and the expression associated therewith, and this setting itself is very labor intensive.
[0009]
In addition, according to the technique disclosed in Japanese Patent Laid-Open No. 7-85071, information is obtained by performing detailed syntax analysis using syntax analysis rules. The analysis itself may be difficult if it is not registered or if there is no semantic information. Further, in this method, for example, a similar result can be obtained in a sentence “XX electricity has developed a communication device”, but detailed semantic information such as what device is developed is not known.
[0010]
An object of the present invention is to automatically assign semantic information of a word to a word registered in a word dictionary without using a technique such as manual operation or syntax analysis.
[0011]
An object of the present invention is to automatically acquire a word not registered in a word dictionary and its semantic information by using input document information.
[0012]
[Means for Solving the Problems]
  The semantic information estimation apparatus according to claim 1 is a word dictionary storage means for storing a plurality of word expressions, and a character string constituting at least a part of one word expression stored in the word dictionary storage means. A common character string that is common to the first word is searched from the notation of words other than the one word, and when the predetermined number or more of the common character string is searched, the common character string is a notation including the common character string. Meaning estimation means for estimating the meaning information of the word, and meaning information storage for storing the meaning information estimated by the meaning estimation means in the word dictionary storage means in association with the notation of the word including the common character string Means,The first co-occurrence pattern defining the relationship between co-occurring words and the semantic information of each word of the first co-occurrence pattern are stored in association with each other, and any word of the first co-occurrence pattern A co-occurrence pattern dictionary storage means for storing a second co-occurrence pattern that defines a hierarchical relationship or an association relationship between the second co-occurrence word and another word, and semantic information of the other word of the second co-occurrence pattern in association with each other A document receiving unit that receives an input of a document from the input unit, and each of the documents that is received by the document receiving unit and that satisfies the first co-occurrence pattern stored in the co-occurrence pattern dictionary storage unit Co-occurrence information extracting means for extracting words, and the semantic information storage means is further stored in the notation of each word extracted by the co-occurrence information extraction means and the co-occurrence pattern dictionary storage means. Said meaning of the word Information is stored in the word dictionary storage unit in association with each other, and a hierarchical relationship or association relationship between any one of the words extracted by the co-occurrence information extraction unit and another word is the co-occurrence. If the second co-occurrence pattern stored in the pattern dictionary storage means is defined, the notation of the other words and the other words stored in the co-occurrence pattern dictionary storage means Semantic information is associated and stored in the word dictionary storage means.
[0013]
  Therefore, since it is highly likely that character strings that are common in words represent a common concept, it is possible to easily give meaning by using this. As a result, it becomes possible to automatically give semantic information of a word to a word registered in the word dictionary without using a technique such as manpower or syntax analysis.Further, by using the co-occurrence pattern appearing in the document, it becomes possible to automatically acquire a word not registered in the word dictionary and its semantic information. In addition, it is possible to reduce the labor of pattern description by associating or hierarchically describing the co-occurrence patterns. Further, by being associated, even if there is an unregistered word, it is possible to estimate semantic information based on related known information. Furthermore, since the associated co-occurrence patterns allow not only semantic information but also semantic information to be associated with each other, more detailed semantic information can be estimated.
[0014]
  According to a second aspect of the present invention, in the semantic information estimation device according to the first aspect, the semantic estimation means includes:The common character string is searched from the last character string or the first character string of the notation of the word other than the one word..
[0015]
Therefore, by limiting the common character string positions, more accurate meaning estimation can be performed.
[0026]
  Claim3The semantic information estimation method according to the invention is characterized in that the common character common to the character string constituting at least a part of the notation of one word stored in the word dictionary storage means in which the meaning estimation means stores a plurality of word expressions. When a string is searched from notations of words other than the one word and the common character string is searched for a predetermined number or more, the common character string is used as the meaning information of the notation word including the common character string. And a semantic information storage unit that stores the semantic information estimated in the semantic estimation step in the word dictionary storage unit in association with a word notation including the common character string.FirstSemantic information storage process;A document receiving unit that receives a document input from the input unit; and a co-occurrence information extracting unit that defines a relationship between co-occurring words from the documents received in the document receiving step. One co-occurrence pattern and semantic information of each word of the first co-occurrence pattern are stored in association with each other, and a hierarchical relationship or an association relationship between any one word of the first co-occurrence pattern and another word is defined. Each word satisfying the first co-occurrence pattern stored in the co-occurrence pattern dictionary storage means for storing the second co-occurrence pattern and the semantic information of the other words of the second co-occurrence pattern in association with each other A co-occurrence information extracting step for extracting the word, and the semantic information storage means further includes the notation of each word extracted by the co-occurrence information extraction step and the word stored in the co-occurrence pattern dictionary storage means. The semantic information of The word dictionary storage means associates and stores the hierarchical relationship or association relation between any one of the words extracted by the co-occurrence information extraction step and another word, and the co-occurrence pattern dictionary storage means. If the second co-occurrence pattern stored in is defined in the second co-occurrence pattern, the notation of the other word and the semantic information of the other word stored in the co-occurrence pattern dictionary storage means A second semantic information storage step of storing the word dictionary storage means in association with each other;Comprising.
[0027]
  Therefore, since it is highly likely that character strings that are common in words represent a common concept, it is possible to easily give meaning by using this. As a result, it becomes possible to automatically give semantic information of a word to a word registered in the word dictionary without using a technique such as manpower or syntax analysis.Further, by using the co-occurrence pattern appearing in the document, it becomes possible to automatically acquire a word not registered in the word dictionary and its semantic information. In addition, it is possible to reduce the labor of pattern description by associating or hierarchically describing the co-occurrence patterns. Further, by being associated, even if there is an unregistered word, it is possible to estimate semantic information based on related known information. Furthermore, since the associated co-occurrence patterns allow not only semantic information but also semantic information to be associated with each other, more detailed semantic information can be estimated.
[0028]
  Claim4The described invention is claimed.3In the semantic information estimation method described above, in the semantic estimation step, the semantic estimation means calculates the common character string from the last character string or the first character string of words other than the one word. Search for.
[0029]
Therefore, by limiting the common character string positions, more accurate meaning estimation can be performed.
[0040]
  Claim5The program of the invention described is a program for causing a computer to estimate word semantic information, and the computer uses one word notation stored in word dictionary storage means for storing a plurality of word notations in the computer. When a common character string that is common to at least a part of the character string is searched from notations of words other than the one word, and the common character string is searched for a predetermined number or more, the common character string Is estimated as semantic information of a notation word including the common character string, and the semantic information estimated by the meaning estimation function is associated with the notation of the word including the common character string. Store in dictionary storageFirstSemantic information storage function,A document reception function for receiving a document input from the input unit; a first co-occurrence pattern that defines a relationship between co-occurring words from the documents received by the document reception function; and a first co-occurrence pattern The second co-occurrence pattern and the second co-occurrence pattern defining the hierarchical relationship or the association relationship between any word of the first co-occurrence pattern and other words while storing the semantic information of each word in association with each other A co-occurrence information extraction function for extracting each word satisfying the first co-occurrence pattern stored in the co-occurrence pattern dictionary storage means for storing the semantic information of the other words of the pattern in association with each other; The notation of each word extracted by the occurrence information extraction function and the semantic information of the word stored in the co-occurrence pattern dictionary storage unit are stored in the word dictionary storage unit in association with each other, and Co-occurrence information When the second co-occurrence pattern stored in the co-occurrence pattern dictionary storage unit defines a hierarchical relationship or an association relationship between any one of the words extracted by the output function and another word The second semantic information storage for storing the notation of the other word and the semantic information of the other word stored in the co-occurrence pattern dictionary storage means in association with each other in the word dictionary storage means Function andIs executed.
[0041]
  Therefore, since it is highly likely that character strings that are common in words represent a common concept, it is possible to easily give meaning by using this. As a result, it becomes possible to automatically give semantic information of a word to a word registered in the word dictionary without using a technique such as manpower or syntax analysis.Further, by using the co-occurrence pattern appearing in the document, it becomes possible to automatically acquire a word not registered in the word dictionary and its semantic information. In addition, it is possible to reduce the labor of pattern description by associating or hierarchically describing the co-occurrence patterns. Further, by being associated, even if there is an unregistered word, it is possible to estimate semantic information based on related known information. Furthermore, since the associated co-occurrence patterns allow not only semantic information but also semantic information to be associated with each other, more detailed semantic information can be estimated.
[0042]
  Claim6The described invention is claimed.5In the program described above, the meaning estimation function searches for the common character string from the last character string or the first character string in the notation of the word other than the one word.
[0043]
Therefore, by limiting the common character string positions, more accurate meaning estimation can be performed.
[0056]
DETAILED DESCRIPTION OF THE INVENTION
A first embodiment of the present invention will be described with reference to FIGS.
[0057]
FIG. 1 is a block diagram schematically showing a hardware configuration of the semantic information estimation apparatus 1. As shown in FIG. 1, the semantic information estimation device 1 includes a CPU (Central Processing Unit) 2 that centrally controls each unit of the semantic information estimation device 1. The CPU 2 stores BIOS and the like. A ROM (Read Only Memory) 3 that is a read-only memory and a RAM (Random Access Memory) 4 that stores various data in a rewritable manner are connected by a bus 5. Further, the bus 5 includes a HDD (Hard Disk Drive) 6 serving as an external storage, a CD-ROM drive 8 that reads a CD (Compact Disc) -ROM 7, and communication control that controls communication between the semantic information estimation device 1 and the network 9. A device 10, an input device 11 such as a keyboard and a mouse that function as an input unit, and an output device 12 such as a CRT (Cathode Ray Tube) and an LCD (Liquid Crystal Display) are connected via an I / O (not shown). ing.
[0058]
Since the RAM 4 has a property of storing various data in a rewritable manner, the RAM 4 functions as a work area for the CPU 2 and plays a role of, for example, an input buffer and an analysis buffer.
[0059]
The HDD 6 stores a word dictionary 13 in which word notation and semantic information are stored, in addition to program files for storing various programs. In the word dictionary 13 of the present embodiment, as shown in FIG. 2, for example, only the name of a company or the like (word notation) is stored as an initial state.
[0060]
A CD-ROM 7 shown in FIG. 1 implements the storage medium of the present invention, and stores a predetermined program. The CPU 2 reads the program stored in the CD-ROM 7 with the CD-ROM drive 8 and installs it in the HDD 6. Thereby, the semantic information estimation apparatus 1 is in a state in which various processes as described later can be performed.
[0061]
As the storage medium, not only the CD-ROM 7 but also various types of media such as semiconductor memory such as various optical disks such as DVD, various magnetic disks such as various magneto-optical disks and floppy disks, and the like can be used. Alternatively, the program may be downloaded from the network 9 such as the Internet via the communication control device 10 and installed in the HDD 6. In this case, the storage device storing the program in the server on the transmission side is also a storage medium of the present invention. Note that the program may operate on a predetermined OS (Operating System), in which case the OS may execute a part of various processes described later, or a word processor. It may be included as part of a group of program files that constitute predetermined application software such as software or an OS.
[0062]
Next, the contents of various processes executed by the CPU 2 of the semantic information estimation apparatus 1 based on the program will be described. The semantic information estimation apparatus 1 of the present invention generally estimates the semantic information of a word. As shown in FIG. 3, the CPU 2 operates based on a program, so that the semantic information estimation apparatus 1 A semantic information estimation unit 14 that functions to estimate the semantic information of a word based on the word dictionary 13 is formed.
[0063]
Next, the flow of semantic information estimation processing in the semantic information estimation unit 14 of the present embodiment will be described with reference to FIG. As shown in FIG. 4, the semantic information estimation processing according to the present embodiment performs processing in step S2 when there is an unprocessed word among the words registered in the word dictionary 13 (Y in step S1). The word is compared with other words in the word dictionary 13, and if there is a common character string, the character string is stored, the counter associated with the character string is counted up, and the word dictionary The matched character string portion in 13 is marked. In addition, when the mark already collated at the time of collation completely matches the common character string, it is not counted in order to avoid duplication count. Here, the function of the meaning estimation means is executed.
[0064]
On the other hand, when there is no unprocessed word among the words registered in the word dictionary 13, that is, when the processing for all the words registered in the word dictionary 13 is completed (N in step S1). In step S3, when the count of the common character string is equal to or larger than a certain number, the semantic information is given to the word in the word dictionary 13 having the character string, using the common character string as semantic information. The fixed number can be changed according to the number of words in the word dictionary 13 or the like. Here, the function of the semantic information storage means is executed.
[0065]
The semantic information estimation process will be described using a specific example. Here, description will be given below assuming that each word in the word dictionary 13 shown in FIG. 2 is processed in order.
[0066]
First, “AA Trading” is processed. A common character string of “AA Trading” and other words in the word dictionary 13 is searched. In this case, since “AA trading” is common to “BBB trading” and “ZZZ trading” in terms of “trade”, the word “trade” is stored as a common character string as a count “2”. At this time, the “trading” portion of “AA trading”, “BBB trading”, and “ZZZ trading” in the word dictionary 13 is marked as counted.
[0067]
“BBB Shoji” and “ZZZ Shoji” are also common to “Shosho” but are not counted up because they have already been counted.
[0068]
Next, “XY beer” is processed. A common character string of “XY beer” and other words in the word dictionary 13 is searched. In this case, since “XY beer” is common to “YYY beer” in the word “beer”, the word “beer” is stored as a common character string as a count “1”. At this time, the “beer” portion of “XY beer” and “YYY beer” in the word dictionary 13 is marked as counted.
[0069]
The following “YYY beer” is also common to “beer”, but since it has already been counted, it does not count up.
[0070]
When all the words registered in the word dictionary 13 are finished, the common character string count is checked. For example, the count of “1” or more is used as the semantic information, and this common character string (“trade” “beer”) is used. It is given in association with the notation of each word in the word dictionary 13.
[0071]
Here, FIG. 5 is an explanatory diagram showing an example of the word dictionary 13 to which the semantic information is given by the processing of the semantic information estimation unit 14. As shown in FIG. 5, each industry field is given as semantic information to the name of a company or the like.
[0072]
Here, since there is a high possibility that character strings that are common in words represent common concepts, it is possible to easily give meaning by using this, so without using techniques such as manual labor and syntax analysis. The semantic information of the word can be automatically given to the word registered in the word dictionary 13.
[0073]
Next, a second embodiment of the present invention will be described with reference to FIGS. Note that the same parts as those of the above-described embodiment are denoted by the same reference numerals, and description thereof is omitted. This embodiment is a modification of the semantic information estimation process in the semantic information estimation unit 14 of the semantic information estimation apparatus 1 described in the first embodiment.
[0074]
The flow of the semantic information estimation process in the semantic information estimation unit 14 of the present embodiment will be described with reference to FIG. As shown in FIG. 6, the semantic information estimation processing of the present embodiment is performed in step S12 when there is an unprocessed word among the words registered in the word dictionary 13 (Y in step S11). Proceed, compare the word with another word in the word dictionary 13, and if the common character string is located at the end of the word or at the beginning of the word, store the common character string and associate it with the character string The counter is incremented by 1, and the collated character string portion in the word dictionary 13 is marked. In addition, when the mark already collated at the time of collation completely matches the common character string, it is not counted in order to avoid duplication count.
[0075]
On the other hand, when there is no unprocessed word among the words registered in the word dictionary 13, that is, when the processing for all the words registered in the word dictionary 13 is completed (N in step S11). In step S13, when the count of the common character string is equal to or larger than a certain number, the semantic information is given to the word in the word dictionary 13 having the character string, using the common character string as semantic information. The fixed number can be changed according to the number of words in the word dictionary 13 or the like.
[0076]
That is, the semantic information estimation process of the present embodiment is compared with the semantic information estimation process of the first embodiment to determine the position of the common character string in the word to be collated when determining whether or not it is a common character. The difference is that it is restricted to the end of the word or the beginning of the word.
[0077]
The semantic information estimation process will be described using a specific example. Here, description will be made below assuming that each word in the word dictionary 13 shown in FIG. 7 is processed in order.
[0078]
First, “(stock) XX” is processed. A common character string of “(stock) XX” and other words in the word dictionary 13 is searched. In this case, “(share) XX” is common to “(share) XX securities” and “(AA) securities” with the word “(share)”, so the word “(share)” Is stored as a count “2” as a common character string. At this time, the “(stock)” portion of “(stock) XX”, “(stock) XX securities”, and “(AA) securities” in the word dictionary 13 is marked as counted.
[0079]
Next, “(stock) xx securities” is processed. A common character string of “(stock) XX securities” and other words in the word dictionary 13 is searched. In this case, since “(stock) xx securities” is common to “(stock) AA securities” in the word “securities”, the word “securities” is stored as a common character string as a count “1”. .
[0080]
Regarding “(AA) Securities”, “(stock)” and “securities” are common, but they are not counted up because they have already been counted.
[0081]
Next, “XY beer” is processed. A common character string of “XY beer” and other words in the word dictionary 13 is searched. In this case, since “XY beer” is common to “ZZZ beer” in the word “beer”, the word “beer” is stored as a common character string as a count “1”. At this time, the “beer” portion of “XY beer” and “ZZZ beer” in the word dictionary 13 is marked as counted.
[0082]
As for “ZZZ beer”, “beer” is common, but it is not counted up because it has already been counted.
[0083]
When all the words registered in the word dictionary 13 are finished, the common character string count is checked. For example, the count of “1” or more is used as the semantic information, and the common character string (“(stock)” “securities” “Beer”) is assigned in association with the notation of each word in the word dictionary 13.
[0084]
Here, FIG. 8 is an explanatory diagram showing an example of the word dictionary 13 to which the semantic information is given by the processing of the semantic information estimation unit 14. As shown in FIG. 8, by restricting the common character string to the end of the word, it is possible to extract only words / phrases that are in a higher class of words / phrases that represent the concept of the object for names of companies and the like. Here, the business fields “beer” and “securities” are given as semantic information. In addition, by limiting the common character string to the beginning of the word, an expression characterizing the meaning of the entire word such as “(stock)” can be extracted for the name of a company or the like. Here, the company form “(stock)” is given as semantic information.
[0085]
Here, by limiting the common character string positions, more accurate meaning estimation can be realized.
[0086]
Next, a third embodiment of the present invention will be described with reference to FIGS. Note that the same parts as those of the above-described embodiment are denoted by the same reference numerals, and description thereof is omitted. This embodiment is a modification of the semantic information estimation process in the semantic information estimation unit 14 of the semantic information estimation apparatus 1 described in the first embodiment. Unlike the state where the word dictionary 13 does not have any semantic information as described in the first embodiment and the second embodiment, the present embodiment includes words that have no semantic information registered in the word dictionary 13. This is semantic information estimation processing when there are several cases or when a new word is registered.
[0087]
The flow of the semantic information estimation process in the semantic information estimation unit 14 of the present embodiment will be described with reference to FIG. As shown in FIG. 9, when there is a word whose semantic information is not registered in the word dictionary 13 (Y in step S21), the semantic information estimation process of the present embodiment proceeds to step S22, and the word is Character strings are compared with other words in the word dictionary 13, and if there is a common character string, the character string is stored, and the counter associated with the character string is incremented by one. Here, the function of the second meaning estimating means is executed.
[0088]
On the other hand, when the word whose semantic information is not registered disappears from the word dictionary 13 (N in Step S21), the process proceeds to Step S23, and the common character string whose count is a certain number or more, or the common character string is selected. The semantic information of the possessed word is assigned to the word in the word dictionary 13 having the character string. The fixed number can be changed according to the number of words in the word dictionary 13 or the like. Here, the function of the second semantic information storage means is executed.
[0089]
The semantic information estimation process will be described using a specific example. Here, a description will be given below assuming that the words in the word dictionary 13 shown in FIG. 10 are processed.
[0090]
Since there is a word (“ZZZ Shoji” “XYZ Beer”) whose semantic information is not registered in the word dictionary 13, first, “ZZZ Shoji” is processed. A common character string of “ZZZ Shoji” and other words in the word dictionary 13 is searched. In this case, since “ZZ Trading” is common to “AA Trading” and “BBB Trading” in terms of “Shosho”, the word “Shosho” is stored as a common character string as a count “2”.
[0091]
Next, “XYZ beer” is processed. A common character string of “XYZ beer” and other words in the word dictionary 13 is searched. In this case, since “XYZ beer” is common to “XY beer” in the word “beer”, the word “beer” is stored as a common character string as a count “1”.
[0092]
After checking the common character string for the word whose semantic information in the word dictionary 13 is not registered (“ZZZ Shoji” “XYZ Beer”), the common character string count is checked. This common character string (“trade” and “beer”) is assigned in association with the notation of each word in the word dictionary 13 as meaning information of an unregistered word.
[0093]
Here, FIG. 11 is an explanatory diagram showing an example of the word dictionary 13 to which the semantic information is given by the processing of the semantic information estimation unit 14. As shown in FIG. 11, each industry field is assigned as semantic information to a word whose semantic information is not registered.
[0094]
Here, semantic information can be automatically estimated from existing dictionary information for an unregistered word having no semantic information.
[0095]
In the present embodiment, the case where there is a word in the word dictionary 13 whose semantic information is not registered has been described, but the same processing can be performed when a word itself is newly registered.
[0096]
Next, a fourth embodiment of the present invention will be described with reference to FIGS. Note that the same parts as those of the above-described embodiment are denoted by the same reference numerals, and description thereof is omitted. This embodiment does not estimate the semantic information of words in the word dictionary 13 as described in the first embodiment to the third embodiment based on the word dictionary 13, but the input device 11 or the like. Is used to estimate semantic information of a word from a document input via.
[0097]
Here, FIG. 12 is a functional block diagram of the semantic information estimation apparatus 1 of the present embodiment. As shown in FIG. 12, the co-occurrence information extraction unit 15 and the semantic information estimation unit 16 are formed in the semantic information estimation device 1 by the CPU 2 operating based on the program. The co-occurrence information extraction unit 15 that implements the co-occurrence information extraction unit generally uses a co-occurrence pattern dictionary 17 (see FIG. 13) for a document input via the input device 11 or the like. The corresponding expression is extracted from the inside. As shown in FIG. 13, the co-occurrence pattern dictionary 17 describes a set of co-occurrence expressions (co-occurrence patterns) extracted from a document. In the example shown in FIG. 13, the co-occurrence expression is specified by notation or part of speech. On the other hand, the semantic information estimation unit 16 roughly corresponds to a case where the word itself is not registered in the word dictionary 13 or the semantic information is not registered for the corresponding expression extracted by the co-occurrence information extraction unit 15. A word and semantic information of the word are registered.
[0098]
First, the process in the co-occurrence information extraction unit 15 will be described. The co-occurrence information extraction unit 15 includes an unregistered word or sequence when there is a sequence of unregistered words or nouns in the word dictionary 13 after a predetermined word in the input document, and there is a particle after that. The predetermined word is given to the noun to be used as semantic information. That is, in the example of the co-occurrence pattern dictionary 17 shown in FIG. 13, for an unregistered word or a continuous noun (the underlined portion in FIG. 13: a co-occurrence word) after the word “PC” in the input document. , "PC" and "maker" are given as semantic information. It is also possible to assign different semantic information for each co-occurrence word using a table or the like. Here, the input document may or may not be analyzed by morphological analysis or the like. When the morphological analysis is not performed, the part of speech is specified by the word dictionary 13.
[0099]
Next, processing in the semantic information estimation unit 16 will be described. Here, FIG. 14 is a flowchart showing the flow of the semantic information estimation process in the semantic information estimation unit 16 of the present embodiment. As shown in FIG. 14, the semantic information estimation process of the semantic information estimation unit 16 collates the co-occurrence patterns in the co-occurrence pattern dictionary 17 one by one with the input document one by one (step S31: document accepting unit). If a predetermined co-occurrence word (underlined in FIG. 13) is not registered or meaning information is not registered in the word dictionary 13 (Y in step S32, Y in step S33), an instruction from the co-occurrence pattern dictionary 17 Accordingly, the semantic information is assigned or the description of the co-occurrence pattern dictionary 17 is converted into another expression and given (step S34). The processes in steps S32 to S34 are performed for all the co-occurrence patterns.
[0100]
The semantic information estimation process will be described using a specific example. Here, for example, consider a case where there is an expression “Large maker XX Electric has released PC A series” in the input document.
[0101]
First, based on the co-occurrence pattern of the co-occurrence pattern dictionary 17 shown in FIG. 13, there is a “computer” in the front, an unregistered word or a continuous noun (co-occurrence word) behind, and a particle after that. Find out if there is. Here, “PC A series” corresponds. Next, based on the co-occurrence pattern dictionary 17, “A series” is extracted as semantic information “PC”. This is searched in the word dictionary 13, and if it is not registered, it is registered in the word dictionary 13 as shown in FIG.
[0102]
Subsequently, based on the co-occurrence pattern of the co-occurrence pattern dictionary 17 shown in FIG. 13, there is a “maker” in the front, an unregistered word or a continuous noun (co-occurrence word) behind, and a particle after that. Find out if there is. Here, “Manufacturer XX Electric” corresponds. Next, based on the co-occurrence pattern dictionary 17, “XX electricity” is extracted as semantic information “maker”. This is searched in the word dictionary 13, and when it is not registered, “XX electricity” is registered in the word dictionary 13, and “maker” is registered as semantic information.
[0103]
Here, by using the co-occurrence information appearing in the document, it is possible to automatically acquire a word not registered in the word dictionary 13 and its semantic information. In particular, the meaning can be more reliably acquired by using the information of the word structure.
[0104]
Next, a fifth embodiment of the present invention will be described with reference to FIGS. Note that the same parts as those of the above-described embodiment are denoted by the same reference numerals, and description thereof is omitted. This embodiment is a modification of the processing in the co-occurrence information extraction unit 15 and the semantic information estimation unit 16 of the semantic information estimation device 1 described in the fourth embodiment.
[0105]
In the word dictionary 13 of the present embodiment, as shown in FIG. 16, a word and semantic information of the word are stored.
[0106]
Next, the co-occurrence pattern dictionary 17 used in the co-occurrence information extraction unit 15 of the present embodiment will be described. As shown in FIG. 17, the co-occurrence pattern dictionary 17 according to the present embodiment describes a co-occurrence pattern to be collated and a semantic relationship between words in the co-occurrence pattern. Note that the semantic relationship of words in the co-occurrence pattern may be an upper or lower relationship or another relationship. In the example shown in FIG. 17, “()” indicates a word having the semantic relationship, and “[]” specifies the notation of the word itself. Here, “[X | Y]” means that either X or Y may be used.
[0107]
In addition, in the co-occurrence pattern dictionary 17 of the present embodiment, as shown in FIG. 17, it is possible to describe a hierarchy or association in the co-occurrence pattern. In the example shown in FIG. 17, in the case of “(A) :( B) (C)”, the meaning “(A)” can be expressed by the whole chain of “(B)” and “(C)”. ing. Note that “(name: noun sequence or unregistered word)” indicates that the expression meaning the name is composed of a sequence of nouns or unregistered words, and is used as a condition for matching when matching. Yes. As described above, among the expressions in the co-occurrence pattern, an expression formed by combining a plurality of words can be described separately because, for example, in the case of a chain of internal constituent words in the word, it is easy to describe. The example shown in FIG. 17 indicates that the word meaning the product indicated by “(product)” in the co-occurrence pattern can be expressed by a combination such as (product) (name: noun continuous or unregistered word). ing.
[0108]
Next, processing in the semantic information estimation unit 16 will be described. Here, FIG. 18 is a flowchart showing the flow of the semantic information estimation process in the semantic information estimation unit 16 of the present embodiment. As shown in FIG. 18, the semantic information estimation processing of the semantic information estimation unit 16 collates expressions in the input document in order, and whether or not the co-occurrence pattern of the co-occurrence pattern dictionary 17 exists in the input document. Is to examine. At this time, if there is a word “()” having a semantic relationship in the input document, collation including the meaning by “()” is performed (step S 42), and the co-occurrence of the co-occurrence pattern dictionary 17 is included in the input document. If there is no pattern (N in step S42), the process proceeds to step S43, and collation is performed with the meaning part “()” removed.
[0109]
If there is a co-occurrence pattern in the co-occurrence pattern dictionary 17 in the input document (Y in step S43), the process proceeds to step S44, and the meaning part “()” is defined by another co-occurrence pattern in the co-occurrence pattern. Find out if there is. Then, if there is a regulation with other co-occurrence patterns, the semantic part “()” and the corresponding character string are collated with other co-occurrence patterns.
[0110]
When the semantic part “()” and the corresponding character string are collated with other co-occurrence patterns (Y in step S44), the co-occurrence pattern with the least unmatched co-occurrence patterns is selected ( If the unmatched part is “1” (step S46: Y in step S46), the unmatched part is recognized as the meaning, and the matched co-occurrence pattern and the corresponding character string (word) are stored (step S47). ).
[0111]
Also, when the co-occurrence pattern of the co-occurrence pattern dictionary 17 is present in the input document (Y in step S42), the matched co-occurrence pattern and the corresponding character string (word) are stored (step S47).
[0112]
On the other hand, when there is no co-occurrence pattern of the co-occurrence pattern dictionary 17 in the input document (N in step S43), the semantic part “()” and the corresponding character string are not collated with other co-occurrence patterns (step S44). N), if the unmatched part is not “1” (Y in step S46), the process returns to step S41.
[0113]
The processes in steps S42 to S47 are performed for all the expression patterns in the input document.
[0114]
When the processing is completed for all the expression patterns in the input document (N in step S41), the description of the semantic relationship of the corresponding co-occurrence pattern dictionary 17 is based on the matched co-occurrence pattern, the corresponding character string (word), and meaning. Is added to the word dictionary 13 (step S48).
[0115]
The semantic information estimation process will be described using a specific example. Here, for example, in the input document
"XX Electric has released PC A series"
"XX Electric has developed a sputtering system"
Consider the case where there is an expression.
[0116]
First, based on the co-occurrence pattern of the co-occurrence pattern dictionary 17 shown in FIG. 17, “XX Electric has released the personal computer A series” is collated. The top of the co-occurrence pattern dictionary 17 shown in FIG. 17, “(manufacturer) releases (product)” is collated including the meaning “()”, and then the meaning other than the meaning “()” Check only the part. Here, except for the semantic part (manufacturer) (product), it matches.
[0117]
Since it is unclear whether or not the portion in which the meaning “()” is defined is coincident, whether or not the other co-occurrence pattern dictionary 17 has the definition for each meaning is checked. As for (product), a co-occurrence pattern “(product) :( product) (name: noun continuous or unregistered word)” is present in the co-occurrence pattern dictionary 17 and collated with “PC A series” of the corresponding part. In the word dictionary 13 shown in FIG. 16, since “PC” is registered as the meaning “product”, a co-occurrence pattern “(product) :( product) (name: noun continuous or unregistered word)” is present. Applied, “PC A series” is recognized as (product) (name), and this entire part can be confirmed as (product). That is, in the co-occurrence pattern “(maker) releases (product)” at the top of the co-occurrence pattern dictionary 17, the (product) part matches.
[0118]
On the other hand, for the remaining “(maker)”, there is no character string matching condition in the word dictionary 13 shown in FIG. In this example, the co-occurrence pattern dictionary 17 at the top of the co-occurrence pattern dictionary 17 “(maker) sells (product)” is best collated and the unverified part is “1”. It is assumed that the first co-occurrence pattern “(maker) has released (product)” has been collated, and the unverified part “XX Electric” is estimated as (maker) and is not registered in the word dictionary 13. “XX Electric” is recognized as “Manufacturer”, “A Series” is recognized as “Name”, and the collated co-occurrence pattern and its word are stored.
[0119]
Next, collation is performed on “XX Electric has developed a sputtering system”. The second co-occurrence pattern “(maker) develops (technology)” in the co-occurrence pattern dictionary 17 shown in FIG. 17 is collated including the meaning “()”, and then the meaning other than the meaning “()”. Match only the part of. Here, except for the semantic part (manufacturer) (technique), they match.
[0120]
Since it is unclear whether or not the portion in which the meaning “()” is defined is coincident, whether or not the other co-occurrence pattern dictionary 17 has the definition for each meaning is checked. As for (technology), a co-occurrence pattern “(technology) :( technology) | (technology) [apparatus | system]” is present in the co-occurrence pattern dictionary 17 and collated with the “sputtering apparatus” of the corresponding part. The co-occurrence pattern of “(Technology) :( Technology) | (Technology) [Device | System]” is applied, and “Sputtering device” is recognized as (Technology) [Device | System]. It can be confirmed. That is, in the second co-occurrence pattern “(maker) develops (technique)” in the co-occurrence pattern dictionary 17, the (technical) part matches.
[0121]
On the other hand, for the remaining “(maker)”, there is no character string matching condition in the word dictionary 13 shown in FIG. In this example, the second co-occurrence pattern “(maker) develops (technology)” in the co-occurrence pattern dictionary 17 is best collated, and the unmatched part is “1”. It is assumed that the second co-occurrence pattern of 17 “(maker) developed (technology)” is collated, and the unverified part “XX Electric” is estimated as (maker) and is not registered in the word dictionary 13 Is recognized as “maker” and “sputtering” as “technology”, and the collated co-occurrence pattern and the word are stored.
[0122]
After all the co-occurrence patterns are collated, semantic information is given to the words in the word dictionary 13 for the collated co-occurrence patterns. Here, the meaning “maker”, the meaning “name” for the word “A series”, and the meaning “technology” for the word “sputtering” are registered in the word dictionary 13 as shown in FIG. be able to. Regarding the semantic relationship, the order of association can be changed according to the purpose (for example, the manufacturer is on the left side in FIG. 19). In addition, as shown in FIG. 20, it is possible to give meaning to the word dictionary 13 itself.
[0123]
As mentioned above, even if the word itself and the meaning of the word are not registered by the co-occurrence pattern, they can be guessed by the co-occurring expression and the meanings are related to each other. Detailed semantic information can be estimated from the results.
[0124]
Here, it is possible to reduce the labor of pattern description by associating or hierarchically describing the co-occurrence patterns. Further, by being associated, even if there is an unregistered word, it is possible to estimate semantic information based on related known information. Moreover, since not only the semantic information but also the semantic information can be associated with each other by the associated co-occurrence pattern, more detailed semantic information can be estimated.
[0125]
【The invention's effect】
  Claim 1,3,5According to the described invention, since common character strings in words are likely to represent a common concept, meaning can be easily given by using this, so that techniques such as manpower and syntax analysis can be used. The semantic information of the word can be automatically given to the word registered in the word dictionary without using.Further, by using the co-occurrence pattern appearing in the document, it is possible to automatically acquire a word not registered in the word dictionary and its semantic information. Moreover, the labor of pattern description can be reduced by associating or hierarchically describing the co-occurrence patterns. Further, by being associated, even if there is an unregistered word, it is possible to estimate semantic information based on related known information. Furthermore, since the associated co-occurrence patterns allow not only semantic information but also semantic information to be associated with each other, more detailed semantic information can be estimated.
[0126]
  Claim 2,4,6According to the described invention, it is possible to achieve more accurate meaning estimation by limiting the common character string positions.
[Brief description of the drawings]
FIG. 1 is a block diagram schematically showing a hardware configuration of a semantic information estimation apparatus according to a first embodiment of the present invention.
FIG. 2 is an explanatory diagram showing an initial state of a word dictionary.
FIG. 3 is a functional block diagram of a semantic information estimation apparatus.
FIG. 4 is a flowchart schematically showing a flow of semantic information estimation processing.
FIG. 5 is an explanatory diagram showing a word dictionary to which semantic information is assigned.
FIG. 6 is a flowchart schematically showing a flow of semantic information estimation processing in the semantic information estimation apparatus according to the second embodiment of the present invention.
FIG. 7 is an explanatory diagram showing an initial state of a word dictionary.
FIG. 8 is an explanatory diagram showing a word dictionary to which semantic information is assigned.
FIG. 9 is a flowchart schematically showing a flow of semantic information estimation processing in the semantic information estimation apparatus according to the third embodiment of the present invention;
FIG. 10 is an explanatory diagram showing an initial state of a word dictionary.
FIG. 11 is an explanatory diagram showing a word dictionary to which semantic information is assigned.
FIG. 12 is a functional block diagram of a semantic information estimation apparatus according to a fourth embodiment of this invention.
FIG. 13 is an explanatory diagram showing a co-occurrence pattern dictionary.
FIG. 14 is a flowchart schematically showing a flow of semantic information estimation processing.
FIG. 15 is an explanatory diagram showing a word dictionary to which semantic information is assigned.
FIG. 16 is an explanatory diagram showing an initial state of a word dictionary of the semantic information estimation apparatus according to the fifth embodiment of the present invention.
FIG. 17 is an explanatory diagram showing a co-occurrence pattern dictionary.
FIG. 18 is a flowchart schematically showing a flow of semantic information estimation processing.
FIG. 19 is an explanatory diagram showing a word dictionary to which semantic information is assigned.
FIG. 20 is an explanatory diagram showing another example of a word dictionary to which semantic information is assigned.
[Explanation of symbols]
1 Semantic information estimation device
7 Storage media
11 Input section
13 word dictionary
17 Co-occurrence pattern dictionary

Claims (6)

Word dictionary storage means for storing a plurality of word expressions;
A common character string common to a character string constituting at least a part of a notation of one word stored in the word dictionary storage unit is searched from notation of words other than the one word, and the common character Meaning estimating means for estimating the common character string as semantic information of a notation word including the common character string when a predetermined number or more of columns are searched,
Semantic information storage means for storing the semantic information estimated by the semantic estimation means in the word dictionary storage means in association with the notation of the word including the common character string;
The first co-occurrence pattern defining the relationship between co-occurring words and the semantic information of each word of the first co-occurrence pattern are stored in association with each other, and any word of the first co-occurrence pattern A co-occurrence pattern dictionary storage means for storing a second co-occurrence pattern that defines a hierarchical relationship or an association relationship between the second co-occurrence word and another word, and semantic information of the other word of the second co-occurrence pattern in association with each other ,
A document receiving means for receiving a document input from the input unit;
Co-occurrence information extracting means for extracting each word satisfying the first co-occurrence pattern stored in the co-occurrence pattern dictionary storage means from the documents received by the document receiving means;
With
The semantic information storage means further associates the notation of each word extracted by the co-occurrence information extraction means with the semantic information of the word stored in the co-occurrence pattern dictionary storage means. While being stored in the word dictionary storage unit, a hierarchical relationship or association relationship between any one of the words extracted by the co-occurrence information extraction unit and another word is stored in the co-occurrence pattern dictionary storage unit. If the second co-occurrence pattern is defined, the notation of the other word is associated with the semantic information of the other word stored in the co-occurrence pattern dictionary storage means. Semantic information estimation device stored in the word dictionary storage means .   2. The semantic information estimating apparatus according to claim 1, wherein the meaning estimating means searches for the common character string from a character string at the end of a notation of a word other than the one word or a character string at the beginning. A word other than the one word is a common character string common to a character string constituting at least a part of the notation of one word stored in a word dictionary storage means for storing a plurality of word expressions. When the common character string is searched for a predetermined number or more, the meaning estimation step of estimating the common character string as semantic information of a notation word including the common character string,
Semantic information storage means, said semantic information estimated by the meaning estimation step, a first semantic information storage step of storing in said word dictionary storage means in association with the representation of the word containing the common character string,
A document accepting step in which the document accepting means accepts an input of a document from the input unit;
A co-occurrence information extracting unit includes a first co-occurrence pattern that defines a relationship between co-occurring words from the documents received by the document receiving step, and semantic information of each word of the first co-occurrence pattern; Are stored in association with each other, and the second co-occurrence pattern defining the hierarchical relationship or the association relationship between any word of the first co-occurrence pattern and the other word and the other word of the second co-occurrence pattern A co-occurrence information extracting step of extracting each word satisfying the first co-occurrence pattern stored in the co-occurrence pattern dictionary storing means for storing the semantic information in association with each other;
The semantic information storage means further associates the notation of each word extracted by the co-occurrence information extraction step with the semantic information of the word stored in the co-occurrence pattern dictionary storage means. While being stored in the word dictionary storage means, a hierarchical relationship or association relationship between any one of the words extracted by the co-occurrence information extraction step and other words is stored in the co-occurrence pattern dictionary storage means. If the second co-occurrence pattern is defined, the notation of the other word is associated with the semantic information of the other word stored in the co-occurrence pattern dictionary storage means. A second semantic information storage step for storing in the word dictionary storage means,
A semantic information estimation method comprising: Wherein in the sense estimating step, said means estimating means from among the strings of the string or the top part of the partial end of notation of words other than the word of the 1, meaning according to claim 3, wherein searching for the common character string Information estimation method. A program for causing a computer to estimate word semantic information, wherein the computer
A common character string common to a character string constituting at least a part of one word notation stored in the word dictionary storage means for storing a plurality of word notations is selected from the word notations other than the one word. A semantic estimation function that searches and estimates the common character string as semantic information of a notation word including the common character string when the predetermined number or more of the common character strings are searched;
A first semantic information storage function for storing the semantic information estimated by the semantic estimation function in the word dictionary storage unit in association with a notation of a word including the common character string;
A document reception function for receiving input of a document from the input unit;
The first co-occurrence pattern defining the relationship between co-occurring words and the semantic information of each word of the first co-occurrence pattern are stored in association with each other from the documents received by the document receiving function. The second co-occurrence pattern defining the hierarchical relationship or the association relationship between any one word of the first co-occurrence pattern and another word is associated with the semantic information of the other word of the second co-occurrence pattern. A co-occurrence information extracting function for extracting each word satisfying the first co-occurrence pattern stored in the co-occurrence pattern dictionary storing means;
The notation of each word extracted by the co-occurrence information extraction function and the semantic information of the word stored in the co-occurrence pattern dictionary storage unit are stored in the word dictionary storage unit in association with each other. In the second co-occurrence pattern, a hierarchical relationship or association relationship between any one of the words extracted by the co-occurrence information extraction function and another word is stored in the co-occurrence pattern dictionary storage unit. If defined, the notation of the other word and the semantic information of the other word stored in the co-occurrence pattern dictionary storage unit are stored in the word dictionary storage unit in association with each other. A second semantic information storage function;
A program for running 6. The program according to claim 5 , wherein the meaning estimation function searches for the common character string from a character string at the end of a notation of a word other than the one word or a character string at the beginning.

Images