JP3603395B2 - Matching device and matching method - Google Patents

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a matching device and a matching method for calculating a degree of data matching, and more particularly to a matching device and a matching method based on a hierarchical relationship between data.
[0002]
[Prior art]
Conventionally, research on text search and automatic classification has been actively conducted. In these fields, dictionaries that define superordinate / subordinate relations and synonyms between words called thesaurus have become increasingly important.
[0003]
In a text search, the expression input by the user often does not match the expression in the search target. For this reason, in order to secure a hit rate of the searched content, it is necessary to perform a search including synonyms and upper / lower terms. Therefore, a method of expanding a word of a search request by using a thesaurus is often used.
[0004]
Generally, at the time of search, the hierarchy of the thesaurus is followed based on the search request, and the search is performed using the obtained synonyms or the upper / lower terms. At this time, if the hierarchy of the thesaurus is traversed every time the search is performed, the search speed decreases. For example, as described in “Comprehensive search method of database search system” in JP-A-2-280274, A technique has been proposed in which a comprehensive index including the upper / lower levels of a thesaurus is created in advance to prevent a reduction in speed at the time of search.
[0005]
However, even if such a comprehensive index is held, it is necessary to perform a search based on a plurality of database keys included in the comprehensive index, so that the speed is still lower than that of the simple match search. Further, since it is necessary to create such comprehensive index information in advance, there is a problem that the data amount and the cost for constructing the database are increased.
[0006]
For example, it is conceivable to perform matching in consideration of a thesaurus between a search key and a keyword, and determine whether or not to obtain a search result based on the degree of matching. It was not a technology that could be used with.
[0007]
On the other hand, in a field called automatic text classification, texts are classified by using a meaningful distance between texts using some criterion. At this time, a thesaurus is also used to determine the distance between the texts.
[0008]
For example, information processing, Vol. 36, no. 2, 1995.2, Iida, “Supercomputing in Artificial Intelligence”, pp. In 164-168, a method of calculating the distance between words using a concept hierarchy is shown as a part of a technique for obtaining the distance between phrases, and a semantic distance between an input and an example is given by 10 It is calculated by comparing hexadecimal thesaurus codes.
[0009]
However, since this calculation method is a calculation based only on the distance to a common upper node, it is a leaf (leaf) of a tree structure, and can only calculate the distance between words arranged in the same hierarchy. Since not all words are expressed as leaves at the same level, this cannot be applied when the concept in the middle of the hierarchy or the number of layers up to the leaf is different. Therefore, such a simple method is still not sufficient for comparing the distance between words. Further, since a parallel computer is assumed, a special process such as an associative memory is required for efficient distance calculation, which is not common at present.
[0010]
As described above, in the related art, there is still only an insufficient technique for efficiently performing matching between words in a thesaurus.
[0011]
[Problems to be solved by the invention]
The present invention has been made in view of the above-described circumstances, and has as its object to provide a matching device and a matching method for efficiently performing word-to-word matching in a thesaurus at high speed and with a small data size. It is.
[0012]
[Means for Solving the Problems]
According to the present invention, in a matching device that performs matching based on a hierarchical relationship between data, each data of a data group having a hierarchical relationship is held and input together with a value sequentially allocated to each data having the same parent of each hierarchy. Hierarchical data encoding means for converting values assigned to data of each layer when tracing the layer to data into a code arranged, and determining the closeness between data in a hierarchical relationship for at least two given data The logical distance between the exclusive OR of the codes obtained by the hierarchical data coding means from the two data and the mask code of the value of each layer as the conceptual distance is represented by the maximum mask code at which the logical product is non-zero. Characterized in that it has a distance calculating means for calculating a value associated with the above as the conceptual distance.
Further, according to the present invention, in a matching method for performing matching based on a hierarchical relationship between data, hierarchical data encoding is performed together with a value obtained by sequentially assigning each data of a data group having a hierarchical relationship to each data having the same parent of each hierarchy. Means for converting the values assigned to the data of each hierarchy when the hierarchy is traversed to the data by the hierarchical data encoding means for at least two given data, and The value associated with the mask code is calculated by the distance calculating means from the largest mask code in which the logical product of the exclusive OR of the obtained code and the mask code of the value of each layer becomes non-zero, Is a conceptual distance indicating the closeness between data in a hierarchical relationship.
[0013]
Further, as in the second and fifth aspects of the present invention, the logical product of the code obtained by the hierarchical data encoding means and the mask code of the value of each hierarchical level is given for at least two given data. It is possible to provide a hierarchy difference calculating means for calculating the hierarchy level of each data from the minimum mask code which becomes zero and calculating the difference in the hierarchy level between the data.
[0014]
Further, as in the third and sixth aspects of the present invention, a priority calculating means for calculating a priority based on the results of the distance calculating means and the hierarchical difference calculating means can be provided.
[0015]
[Action]
According to the present invention, a code corresponding to a hierarchical relationship is held for each data of a data group having a hierarchical relationship. Then, for at least two given data, the value assigned to the data of each layer when the data is traversed to each data is converted to a code in which the exclusive OR of the obtained code and each layer are calculated. The value associated with the mask code is calculated from the largest mask code that makes the logical product of the value of the value and the mask code non-zero, and the value is used as a conceptual distance indicating the closeness between data in a hierarchical relationship. . As a result, distance calculation can be performed without performing a time-consuming operation such as an arithmetic operation as in the related art, and matching based on a hierarchical relationship between data can be realized at high speed and easily. In addition, as in the inventions according to claims 2 and 3, various configurations such as providing a hierarchy difference calculating means and a priority calculating means depending on the application are used, so that a hierarchical level difference and a priority level can be obtained. This enables a more detailed comparison.
[0016]
DESCRIPTION OF THE PREFERRED EMBODIMENTS
FIG. 1 is a schematic configuration diagram showing one embodiment of the matching device of the present invention. In the figure, 1 is an input unit, 2 is a hierarchy relation coding unit, 3 is a matching unit, and 4 is an output unit. The input unit 1 provides data to be compared. The hierarchy relation coding unit 2 returns a code given in advance according to the position between the hierarchies in response to the input. The matching unit 3 performs matching of a plurality of data provided from the input unit 1 by a bit logical operation using a code obtained from the hierarchical relation coding unit 2. The output unit 4 outputs the result obtained by the matching unit 3.
[0017]
FIG. 2 is a block diagram showing an example in which one embodiment of the matching device of the present invention is applied to a similar word search system. In the figure, 10 is a search request input unit, 11 is a terminal, 12 is an OCR, 13 is a telephone and voice recognition unit, 14 is a storage device, 15 is an infrared / wireless receiving unit, 20 is a thesaurus coding unit, 21 is a thesaurus, 30 is a matching unit, 31 is a distance calculation unit, 32 is a hierarchy difference calculation unit, 33 is a priority calculation unit, 34 is an output instruction unit, 40 is an output unit, 41 is a terminal, 42 is a facsimile or printer, 43 is a telephone or Pager, 44 is an infrared / wireless transmitter, 50 is a database, and 51 is a database. The search request input unit 10 corresponds to the input unit 1 shown in FIG. 1. Similarly, the thesaurus coding unit 20 outputs the hierarchical relation coding unit 2, the matching unit 30 outputs the matching unit 3, and the output unit 40 outputs the output. It corresponds to the unit 4 respectively. The database section 50 is a section for electronically storing various documents to be searched, and the documents are stored in the database 51.
[0018]
The search request input unit 10 includes, for example, a terminal 11, an OCR 12, a telephone and a voice recognizing unit 13 equipped with a microphone for inputting voice, a storage device 14 such as a memory, a disk, and a tape, and infrared and wireless signals from a portable information device. And an infrared / wireless receiving unit 15 for receiving the data. Of course, some of these may be used, or other input devices may be used.
[0019]
The output unit 40 includes a display of a terminal 41, a facsimile / printer 42, a telephone or pager 43 output from a speaker by voice synthesis, and an infrared / radio transmission unit 44 for transmitting information to a portable information device using infrared or radio. Is done. Of course, it may be constituted by a part of these, or an output device other than the above may be used, or the output device may be temporarily stored in a storage device connected to a network or the like.
[0020]
The thesaurus coding unit 20 is a unit that converts the input word into a code stored according to the hierarchical structure of the thesaurus 21.
[0021]
The matching unit 30 is a unit that compares an input word with a keyword of a document in the database 51 and outputs a document having a keyword that matches or is similar to a search request. The distance calculation unit 31 is a part that performs a logical operation between the search request converted by the thesaurus coding unit 20 and the keyword codes of the document to calculate a distance in a hierarchical relationship. The hierarchy difference calculation unit 32 is a part that performs a logical operation between the search request converted by the thesaurus coding unit 20 and the keyword codes of the document to calculate a hierarchy level difference. The priority calculation unit 33 is a unit that calculates the priority of the matching result based on the results of the distance calculation unit and the hierarchy difference calculation unit. The output instruction unit 34 extracts a document having a keyword that matches or is similar to the search request from the database unit 50 based on the result obtained by the priority calculation unit 33, and instructs the output unit 40 to output.
[0022]
FIG. 3 is a flowchart showing an example of the overall operation in an example of the similar word search system, and FIG. 4 is a flowchart showing an outline of an example of the matching algorithm. First, in S61, when a user's search request RQ is input from the search request input unit 10, in S62, the thesaurus coding unit 20 performs conversion to a thesaurus code, and obtains a thesaurus code C1 of the search request.
[0023]
Subsequently, in S63, one document DQ is extracted from the search target database 51. Further, in S64, the keywords KW of the document are sequentially extracted. The keyword may be extracted from the document on the spot at the time of the search, or if a high-speed search is required, a keyword manually or automatically added to the document in advance may be used. In step S65, the keyword KW is code-converted by the thesaurus coding unit 20, similarly to the search request RQ, to obtain a keyword thesaurus code C2.
[0024]
When the thesaurus code C1 of the search request and the thesaurus code C2 of the keyword are both obtained, the code matching is performed in S66. First, in S81, the hierarchy difference calculation unit 32 obtains the hierarchy level of the thesaurus code C1 of the search request by a logical operation, and sets it as the hierarchy level L1 of the search request. Similarly, in S82, the hierarchical level of the keyword thesaurus code C2 is obtained by a logical operation, and is set as the keyword hierarchical level L2. In S83, a difference between the hierarchical level L1 of the search request and the hierarchical level L2 of the keyword is obtained, and is set as a hierarchical level difference LD. Subsequently, in S84, the distance calculation unit 31 calculates a bit exclusive OR of the thesaurus code C1 of the search request and the thesaurus code C2 of the keyword, and sets the result to the conceptual distance CD. Further, in S85, the priority calculation unit 33 calculates a priority PL according to the extracted hierarchy level difference LD and the conceptual distance CD.
[0025]
When the code matching is performed in this way, it is checked in S67 whether the obtained hierarchical level difference LD, concept distance CD, priority PL, and the like satisfy the criteria. If it is determined in S68 that the standard is satisfied, the output instruction unit 34 outputs the document DQ to the result buffer BF in S69. Of course, an output instruction may be directly given to the output unit 40.
[0026]
In S70, it is determined whether all the keywords assigned to the document currently being processed have been processed, and the processing of S64 to S70 is repeated until the processing of the last keyword is completed. When the processing of all keywords is completed for one document, it is determined in S71 whether or not processing has been performed for all documents, and the processing of S63 to S70 is repeated until the processing of the last document is completed.
[0027]
By performing such processing, a document having a keyword that matches or is similar to the search request is searched for and output. As described above, in the matching processing, the hierarchical difference and the conceptual distance are performed using the logical operation, so that the matching processing can be performed at high speed. Therefore, it is possible to realize a high-speed similar word search with a small data size without tracing the thesaurus hierarchy and using a comprehensive index as in the related art.
[0028]
Hereinafter, an example of the above-described operation will be described using a specific example. FIG. 5 is an explanatory diagram showing a specific example of the configuration of a thesaurus. For example, the word “private car” is a subordinate concept of “automobile” and represents a superordinate concept such as “sports car”, “sedan”, and “RV car”. As shown in square brackets “[]” in the figure, codes starting from level 1 are sequentially assigned to each word having the same parent in each hierarchy. Here, four bits are used for one layer, and are indicated by a hexadecimal number.
[0029]
Some examples of codes actually assigned to each word are also shown in parentheses "()" in the figure. For example, in the word "private car"
Concrete object [2] → machine [3] → vehicle [A] → vehicle [2] → car [4] → private car [4]
Since the data has a lower level of one level, it is represented as [0], and a 4-byte hexadecimal code "23A2440H" is assigned. Here, the suffix “H” indicates a hexadecimal number, and indicates a value “23A2440” in hexadecimal. Similarly, the code “23A2442H” is assigned to the word “sedan” which is located below “private car” and is the terminal data. Here, since each node has 15 layers or less and 8 layers, a 4-bit 7-level code is used. However, it is needless to say that this assignment is free and may be designed according to the structure of the thesaurus. Since this assignment can be easily automated by traversing the hierarchical structure of the thesaurus, it is possible to eliminate the trouble of manually assigning codes.
[0030]
The search request RQ input in S61 of FIG. 3 is converted to a thesaurus code C1 in S62. FIG. 6 is a conceptual diagram illustrating an example of conversion into a thesaurus code. Here, an example using hashing as a data structure described in “Data Access Method” of Japanese Patent Application Laid-Open No. 5-28194 will be described. The input data refers to a hash table based on a hash value obtained by a hash function. Here, a pointer to the chain index is stored in the hash table. The chain index is referred to based on the pointer, and after checking for collision with the chain index, the converted data is referred to. A code corresponding to the word is stored in the conversion data, and a code corresponding to the input data is obtained.
[0031]
For example, when data "sedan" is input, a hash value is obtained by a hash function, a hash table is referred to based on the hash value, and a collision is checked by a chain index. Is obtained. Also, as shown in FIG. 6, synonyms such as “motorcycle” and “motorcycle” are different as input data, but are codes that are made the same by conversion. In this way, the code can be obtained directly from the data without having to follow hierarchical relationships in the thesaurus.
[0032]
Although an example using the hash method has been described here, it goes without saying that another data structure such as BTTree may be used.
[0033]
In this way, the input search request RQ is converted by the hash function, and is converted into the thesaurus code C1 by following the hash table and the chain index. Subsequently, in S63, one document is extracted from the search target database 51, and further in S64, the keyword of this document is extracted. Here, it is assumed that “PC” has been extracted as the keyword KW. The keyword KW is code-converted by the thesaurus coding unit 20 in the same manner as described above in S65. For example, when the keyword is “PC”, as shown in FIG. 5, it is converted into a thesaurus code of “23EC220H”.
[0034]
When the thesaurus code C1 of the search request and the thesaurus code C2 of the keyword are both obtained, the codes are matched in S66. Matching is roughly divided into calculation of the level difference in the hierarchy difference calculation unit 32 and calculation of the conceptual distance in the distance calculation unit 31.
[0035]
First, in S81 and S82, the hierarchy difference calculation unit 32 calculates the hierarchy levels of the thesaurus code C1 of the search request and the thesaurus code C2 of the keyword. Specifically, a mask code in the following 4-bit unit is prepared, the AND of the thesaurus code C1 of the search request and the thesaurus code C2 of the keyword is sequentially obtained, and the level is determined by repeating until it becomes non-zero. Is done. When the ANDs up to level 1 are all 0, that is, when the code of C1 or C2 is 0, the level is 0 because it is the root of the thesaurus, in this case "concept".
Level 7: 000000FH
Level 6: 00000F0H
Level 5: 0000F00H
Level 4: 000F000H
Level 3: 00F0000H
Level 2: 0F0000H
Level 1: F0000H
[0036]
The code for “Sedan” was “23A2442H” as described above, so if you first AND with 000000FH,
000000FH AND 23A2442H = 2H
Since it is non-zero, the word is found to be a word of level 7, and the processing ends.
[0037]
Since the code related to “PC” was “23EC220H” as described above, first take AND with 000000FH,
000000FH AND 23EC220H = 0H
Since it is zero, AND with 00000F0H is further taken. Then
00000F0H AND 23EC220H = 20H
Since it is non-zero, it is determined that the word is a level 6 word, and the processing ends.
[0038]
Here, a method using a mask by AND has been described. However, a method of shifting the number of bits according to the level, for example, rightward by 4 bits in the case of level 7 and checking the lower-order overflow flag, is of course logically possible. Become the same. In any case, these bit-wise logical operations can generally be executed at a very high speed, so that a high-speed processing can be realized.
[0039]
The absolute value of the difference between the hierarchical levels thus obtained is calculated in S83, and is set as the hierarchical level difference LD. Here, | 7−6 | = 1.
[0040]
Subsequently, the distance calculating unit 31 calculates the conceptual distance between words. The concept distance is obtained based on a bit exclusive OR (XOR) between codes. here,
23A2442H XOR 23EC220H = 4E662H
It becomes. The value thus obtained indicates the closeness between words (concepts). The conceptual distance is given according to the range of the value obtained here. This addition can be easily obtained as the maximum value that becomes non-zero by ANDing with the following mask code, as in the above-described hierarchy level check. When the result of XOR is all 0s, the conceptual distance is 0 because the codes are the same.
Concept distance 7: F0000H
Concept distance 6: 0F00000H
Concept distance 5:00: 00F0000H
Concept distance 4: 000F000H
Concept distance 3: 0000F00H
Concept distance 2: 00000F0H
Concept distance 1: 000000FH
[0041]
Here, since the value of the bit exclusive OR is “4E662H”,
F000000H AND 4E662H = 0H
0F00000H AND 4E662H = 0H
00F0000H AND 4E662H = 40000H
It can be seen that the conceptual distance = 5.
[0042]
FIG. 7 is an explanatory diagram of a specific example of a matching result. FIG. 7 shows the value of the bit exclusive OR of each keyword, the thesaurus code C2 of the keyword, and the thesaurus code C1 of “Sedan” “23A2442H”, the concept distance, the level difference, and the priority. The degree is shown. The priority will be described later again with reference to FIG.
[0043]
As shown in FIG. 7, the conceptual distance is not directly related to the level difference. For example, regarding the "sedan", the value of the bit-exclusive OR with "private car" which is one level higher (parent-child relationship) is 2H, and "sports car" having the same broader word one level below is 3H, These are very close relationships, and the conceptual distance becomes 1.
[0044]
On the other hand, in the "automobile" which is two levels higher (grandparent relationship), the value of the bit exclusive OR is 42H, but the "bus" which is one level lower (uncle / aunt) is 52H, and one level lower ( Cousin), that is, 50H for the same level "bonnet bus", and the conceptual distance to them is 2. Further, for the above-mentioned “PC”, the value of the bit exclusive OR is 4E662H and the conceptual distance is 5.
[0045]
FIG. 8 is a conceptual diagram in which a part of the conceptual distance in the specific example is two-dimensionally mapped. In FIG. 8, ○ indicates a node, a solid line indicates a thesaurus hierarchy, and a height indicates a hierarchy level. The thesaurus hierarchy shown here is a part of the one shown in FIG. Now, if the concept distance is made to correspond to a two-dimensional distance based on “sedan”, nodes having the same concept distance can be arranged and shown on an equidistant arc as shown in FIG.
[0046]
In the above-described example, AND using a mask code is used in calculating the conceptual distance, but it can be realized by various methods as long as they are logically the same. FIG. 9 is an explanatory diagram of a method of calculating a conceptual distance by an OR gate. For example, as shown in FIG. 9, it is conceivable to OR the bits corresponding to the same layer and set the most significant bit that is turned on as the conceptual distance. In the example shown in FIG. 9, the value "4E662H" of the bit exclusive OR of the thesaurus codes of "Sedan" and "PC" is input to the OR circuit every 4 bits, and "0011111" is obtained as an output. Since the position of the most significant bit in the result is the fifth, the conceptual distance is 5. In this calculation method, since a circuit can be easily configured by wired logic, it can be said that this method is suitable for hardware. Of course, the calculation may be performed using another method.
[0047]
Further, in this example, a conceptual distance limited to 0 to 7 is introduced in order to calculate the priority thereafter. When the XOR operation between expressions having a certain conceptual distance is actually performed as described above, the result takes only a specific range (10H to FFH in the case of a conceptual distance of 2). When a rough distance is obtained, the result of XOR calculation may be used directly.
[0048]
Further, in this embodiment, a fixed length of 4 bits is assigned to all the hierarchies of the thesaurus. However, the structure of the thesaurus is various such that the number of nodes increases as approaching the leaf. In such a case, a different number of bits may be assigned to each layer, and even if the storage capacity is limited, this can be effectively used. In addition, in a thesaurus, not all leaf data is at the same hierarchical level, but the present invention is very effective in such a case because data across different hierarchical levels can be compared.
[0049]
Subsequently, in S85, the priority calculation unit 33 calculates the priority according to the conceptual distance calculated by the distance calculation unit 31 and the hierarchy level difference calculated by the hierarchy difference calculation unit 32. As described above, the concept distance and the hierarchy level difference have no direct relationship. Even if the conceptual distances are the same, the similarity between them is often different. As shown in FIG. 8 described above, both “car” and “bonnet bus” have a conceptual distance of 2 with respect to “sedan”, but the similarity is “automobile” because it is a direct grandchild. It is considered high. On the other hand, depending on the configuration of the thesaurus, synonyms may be arranged as lower nodes with respect to upper nodes. In such a case, the similarity level is higher at the same hierarchical level. In this way, by using the concept distance and the hierarchy level difference, these can be evaluated as different axes.
[0050]
In this example, the following formula is used as the priority.
Priority = total number of levels x 2-(concept distance x 2-level difference)
Here, in this example, the total number of levels is seven. For each of the keywords shown in FIG. 7, the following values are returned for “sedan”.
14: sedan (7 × 2 − (0 × 2 − 0) = 14)
13: Private car
12: Sports car, car
11: Bus, vehicle, ...
10: Bonnet bus, motorcycle, ...
...
5: PC, ...
...
Of course, the priority may be calculated by various calculation methods other than the above-described calculation formula.
[0051]
Subsequently, in S67 of FIG. 3, the extraction criterion is checked using the matching result obtained as described above. Here, if the value of the priority is simply 12 or more, the result is determined as the matching result. In the example of each keyword shown in FIG. 7 described above, if the document has any of the keywords “sedan”, “private car”, “sports car”, and “car”, the keyword is output to the result buffer BF. You.
[0052]
Of course, this reference value may be set as appropriate. Further, another extraction criterion may be used as the extraction criterion. For example, after the entire document is scanned as necessary, the extraction criterion is weighted by statistical processing.
[0053]
Thus, the matching process between the search request and one keyword and the extraction of the document are completed. The processing up to this point is repeated for all the keywords attached to one document, and further, the processing is performed by sequentially performing them for each document in the database 51.
[0054]
By the above processing, the result buffer BF includes, in the result buffer BF, a document including a keyword similar to the search request “Sedan” from the user, specifically, “Sedan”, “Sports car”, “RV car”, “Private car”, Documents including "car" and the like are accumulated. In response to this result, the output instructing unit 34 instructs the output unit 40 to output the document based on the priority. The output unit 40 outputs the search result, and the search processing ends.
[0055]
In the above-described similar word search system, the matching unit 30 calculates the concept distance, the hierarchy level difference, and the priority based on the concept distance. , A part of the calculation results may be used.
[0056]
The present invention can be applied to general comparison processing between data having a hierarchical relationship, and it is needless to say that the present invention is not limited to the application to the similar word search system described above. Hereinafter, a case where one embodiment of the matching device of the present invention is applied to an automatic document classification system will be described.
[0057]
At present, automatic classification using sample data is one of the general techniques. However, including a thesaurus deployment has traditionally been difficult due to cost considerations. By applying the matching device of the present invention to matching with standard data, automatic classification can be performed at low cost.
[0058]
For example, consider a case where a certain document is automatically classified into a sample data group (part) having the following keywords.
[Specimen 1]
Keywords: mobile, telephone, car, PHS
Category: Mobile Phone
[Specimen 2]
Keywords: terminal, reflection, resistance, delay
Category: Terminator
[Specimen 3]
Keywords: mobile, terminal, PDA, network
Category: Mobile Terminal
[0059]
It is assumed that the input document also has a keyword. This may be given in advance or extracted on the spot.
[Input Document]
Keywords: mobile, terminal, communication
[0060]
Here, the similarity between the input document including thesaurus expansion and the keywords of each sample is obtained. For the calculation of the similarity, the same method as in the above-described similar word search system can be used. Further, the sum of the keywords having the lowest similarity with each keyword of the input document is obtained. The example which calculated in the samples 1-3 is shown.
(Specimen 1)
Mobile: Mobile = 1, Terminal: PHS = 4, Communication: Telephone = 2 Total 7
(Specimen 2)
Mobile: delay = 6, terminal: terminal = 0, communication: delay = 3, 9 in total
(Specimen 3)
Mobile: Mobile = 1, Terminal: Terminal = 0, Communication: Network = 2 Total 3
[0061]
Since the sample with the smallest sum obtained in this way is considered to have the highest similarity to the input document, the input document is classified into the category. In this example, the input document is classified into the category “mobile terminal” of the sample 3 because the sum calculated in the sample 3 is the smallest. The above processing is sequentially performed on all the input documents to perform automatic classification.
[0062]
Note that, in the application examples to the above-described respective systems, an example using a thesaurus has been described. However, since data having a hierarchical structure is generally widely used, the present invention relates to a method for interchanging data other than a thesaurus, such as a file directory. It can also be applied to matching.
[0063]
【The invention's effect】
As is apparent from the above description, according to the present invention, comparison between data having a hierarchical structure such as a thesaurus can be performed at high speed only by simple processing based on simple bit logical operations such as AND and XOR. It is possible to execute. Further, based on the bit logical operation processing, not only realization by software but also hardware realization is easy.
[0064]
According to the present invention, comparison can be made even in the case of data between data straddling the middle of a hierarchy, which has been difficult with simple processing such as the number of layers up to a common upper node in the related art. In addition, by using the hierarchical level difference, it is possible to provide a detailed comparison criterion according to the type of data and to assign a priority.
[0065]
As described above, since the matching process including thesaurus expansion, which has been difficult in the past, can be executed easily and at high speed, for example, similar text search and automatic text classification using the matching process of the present invention can be performed at high speed and at low cost. This has the effect of being feasible.
[Brief description of the drawings]
FIG. 1 is a schematic configuration diagram showing one embodiment of a matching device of the present invention.
FIG. 2 is a block diagram showing an example of a case where one embodiment of the matching device of the present invention is applied to a similar word search system.
FIG. 3 is a flowchart illustrating an example of an overall operation in an example of a similar word search system.
FIG. 4 is a flowchart illustrating an outline of an example of a matching algorithm.
FIG. 5 is an explanatory diagram showing a specific example of the configuration of a thesaurus.
FIG. 6 is a conceptual diagram showing an example of conversion into a thesaurus code.
FIG. 7 is an explanatory diagram of a specific example of a matching result.
FIG. 8 is a conceptual diagram in which a part of the conceptual distance in a specific example is two-dimensionally mapped.
FIG. 9 is an explanatory diagram of a method of calculating a conceptual distance using an OR gate.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... Input part, 2 ... Hierarchical relation coding part, 3 ... Matching part, 4 ... Output part, 10 ... Search request input part, 11 ... Terminal, 12 ... OCR, 13 ... Telephone and voice recognition part, 14 ... Storage device , 15: infrared / wireless receiving unit, 20: thesaurus coding unit, 21: thesaurus, 30: matching unit, 31: distance calculating unit, 32: hierarchical difference calculating unit, 33: priority calculating unit, 34: output instruction unit , 40 ... output unit, 41 ... terminal, 42 ... facsimile / printer, 43 ... telephone / pager, 44 ... infrared / wireless transmission unit, 50 ... database unit, 51 ... database.

Claims (6)

In a matching device that performs matching based on a hierarchical relationship between data, each data of a data group having a hierarchical relationship is stored together with a value sequentially allocated to each data having the same parent of each hierarchical level, and the hierarchical level is maintained until input data. A hierarchical data encoding means for converting values assigned to the data of each hierarchy when tracing into a code arranged, and a conceptual distance indicating a closeness between data in a hierarchical relationship of at least two given data. The largest mask code in which the logical product of the exclusive OR of the codes obtained by the hierarchical data coding means from the two data and the mask code of the value of each layer is non-zero is associated with the mask code. A distance calculating means for calculating a value of the distance as the conceptual distance. Further, for the given at least two data, the hierarchical level of each data is obtained from the minimum mask code in which the logical product of the code obtained by the hierarchical data encoding means and the mask code of the value of each hierarchical level becomes non-zero. 2. The matching apparatus according to claim 1, further comprising a hierarchical difference calculating unit that calculates a hierarchical level difference between the data. 3. The matching device according to claim 2, further comprising a priority calculation unit that calculates a priority based on the results of the distance calculation unit and the hierarchy difference calculation unit. In a matching method for performing matching based on a hierarchical relationship between data, each data of a data group having a hierarchical relationship is stored in a hierarchical data encoding unit together with a value sequentially allocated to each data having the same parent of each hierarchy. The hierarchical data encoding means converts the values assigned to the data of the respective layers when the data is traced up to the data with respect to at least two given data, and converts the values into a code in which the obtained codes are mutually exclusive. The value associated with the mask code is calculated by the distance calculating means from the largest mask code in which the logical product of the logical OR and the mask code of the value of each layer is non-zero, and the value is calculated as the data in the layer relation. A matching method characterized by using a conceptual distance indicating closeness between the two. In a matching method for performing matching based on a hierarchical relationship between data, each data of a data group having a hierarchical relationship is stored in a hierarchical data encoding unit together with a value sequentially allocated to each data having the same parent of each hierarchy. The hierarchical data encoding means converts the values assigned to the data of the respective layers when the data is traced up to the data with respect to at least two given data, and converts the values into a code in which the obtained codes are mutually exclusive. The value associated with the mask code is calculated by the distance calculation means from the largest mask code in which the logical product of the logical OR and the mask code of the value of each layer is non-zero, and the distance between the data in the layer relation is calculated. And a value obtained by converting the code of the data and the value of each layer, Matching wherein the logical product of the Sukukodo calculates the difference between the hierarchical levels between the data seek hierarchical level of the data from the minimum of the mask code to be non-zero in a hierarchical difference calculating means. In a matching method for performing matching based on a hierarchical relationship between data, each data of a data group having a hierarchical relationship is stored in a hierarchical data encoding unit together with a value sequentially allocated to each data having the same parent of each hierarchy. The hierarchical data encoding means converts the values assigned to the data of the respective layers when the data is traced up to the data with respect to at least two given data, and converts the values into a code in which the obtained codes are mutually exclusive. The value associated with the mask code is calculated by the distance calculation means from the largest mask code in which the logical product of the logical OR and the mask code of the value of each layer is non-zero, and the distance between the data in the layer relation is calculated. And a value obtained by converting the code of the data and the value of each layer, The hierarchical level of each data is determined from the minimum mask code whose logical product with the code is non-zero, and the hierarchical level difference between the data is calculated by the hierarchical difference calculating means, and the conceptual distance and the hierarchical level difference are calculated. A matching method, wherein a priority is calculated by a priority calculation unit based on the priority.

Images