JP3495253B2 - Error elimination method for automatic combination of heterogeneous data having address information and its processing apparatus - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention uses a computer for two types of data including address-related information (town, street, address, number, dialect, resident name, etc.) and refers to the address-related information. The present invention relates to a method and a device for efficiently detecting and eliminating a coupling error due to automatic coupling, as opposed to a method for efficiently and automatically finding and coupling (combining) data of the same resident.

[0002]

2. Description of the Related Art For example, by combining customer data with detailed residential map data having resident information for each building, a building on a map can be specified and displayed from customer information such as a telephone number. To do this, it is necessary to combine different types of data based on address information.

However, the address information of different types of data created separately includes fluctuations and incompleteness of notation, and inaccuracies such as address information. For example, telephone number book database (DB) and commercial When combined with the computerized house map DB of above with perfect matching, only about 40% is automatically combined (Chapter 6 of Reference 1). If the remaining data are combined manually, for example, combining the customer information published in the telephone numbers by occupation (about 1 million cases) in Tokyo's 23 wards and the map would require a great deal of work.

[Reference 1]: Yasuda, Matsumura, Mizumachi, Karasawa "Map guidance system using telephone / FAX", Proc. Of the 3rd functional figure information system symposium, 1992.
/ 4, pp.81-86. What is the fluctuation of the notation in the address information?
For example, it may be written in Kanji or Katakana, written as "Co., Ltd." or "(shares) ~"
It may be abbreviated, such as
It means that it may be written in full-width characters or half-width characters. There are various other fluctuations.

For this reason, conventionally, as a method of increasing the accuracy of automatic connection in a computer, there are two types of DB address information (Machichome.
No.) is used only for narrowing down comparison data without strict matching, and by focusing on comparison of resident information, morpheme analysis is performed on Japanese names for resident names, and the unique A method of extracting information and determining the degree of coincidence has been proposed (reference document 1 above). By using this method, it has become possible to increase the automatic coupling rate by computer to more than 80% and significantly reduce human labor.

Further, as shown in Japanese Patent Application No. 10-21617, "Method and apparatus for combining dissimilar data by address information", character matching is evaluated and combined according to address matching. There are also methods. This method changes the judgment criteria of the degree of coincidence of the characters of the resident name information in consideration of the degree of coincidence of the addresses, and makes it possible to perform an automatic combination by fine evaluation, and moreover, the character string as a set of words. It is a simple set of processing because it is treated as a set of characters and is processed without paying attention to the existence of common characters. It is resistant to fluctuations in writing and does not require Japanese processing.

[0007]

However, in any of the methods that allow ambiguity to improve the coupling ratio, since automatic coupling is used, for example, "AA car rental car" and "AA car shop" are the same. It is inevitable that a mistaken connection will occur, such as being considered and combined. In particular, the user will notice an error only when he visits the site by seeing a map that is combined based on this information.
Such a situation should be eliminated as much as a service, but in order to eliminate this error, in the past,
There was a problem that the whole legitimacy of the combined result had to be manually reviewed, and an unrealistically huge man-hour was required depending on the number of data.

An object of the present invention is to provide means for detecting and eliminating an error in an automatic joining result and improving the joining reliability with the smallest number of steps.

[0009]

FIG. 1 is a block diagram showing the configuration of the main part of the present invention. The processing device 100 is a CPU
And a memory, and a plurality of coupling processing units 1
10, 120, result match classification unit 130, error result elimination unit 1
40, each processing unit of the combination result combining unit 150.

Two kinds of databases (DB) DBa10
1 and a plurality of types of combination processing units 110 and 120 having different methods of combining attribute data for the same building from the address information (address, resident name, etc.) of the DBb 102 and executing the respective combination processes, and the result match classification The processing is transferred to the unit 130, and the combined results are matched and classified into those in which the results do not match and those in which they match. For the unmatched ones, the display device 160 and the input device 170, which can be manually selected to send to the error result elimination unit 140 to determine which of the two results is to be used or not to use both results, In addition to eliminating erroneous merging, the results of both can be merged for those that can be judged as successful merging.

The combination result combining unit 150 combines this result and the result of which the result of the process of the combination matching classifying unit 130 matches to generate a final combination DB 200. The operation of the present invention is as follows.

In the processing apparatus 100 having the configuration shown in FIG. 1, two types of combination processing units 110 and 120 having different methods are used, such as one that focuses on a proper word and one that focuses only on the degree of coincidence of characters to combine them. Since the results are compared with each other and the results are compared, for records with large fluctuations in the name and fluctuations in the address, which are prone to automatic join failures, join errors peculiar to each method are likely to occur, and the results do not match. By manually checking only the objects, it is possible to detect errors in the automatic combination result extremely efficiently.

It should be noted that here, two types of combination processing units 11 are used.
Although the example of 0 and 120 is shown, of course, three or more types may be used.

[0014]

BEST MODE FOR CARRYING OUT THE INVENTION Next, embodiments of the present invention will be described with reference to the drawings. FIG. 2 is a diagram specifically showing the outline of the processing flow by the processing apparatus of the present invention shown in FIG.

DBa 101, DB which is the target of the combining process
As shown in FIG. 2, each b 102 has an address and other information for a certain name. Join processing unit 11
0 and the join processing unit 120 respectively access the DBa 101 and the DBb 102, read the respective records, perform the join processing by their own methods, and output the intermediate join result a 103 and the intermediate join result b 104. The intermediate join result holds the join result and the names and addresses of both the original DBa 101 and DBb 102 as shown in FIG. The connection flag in this data is "1" when the connection is successful and "0" when the connection is unsuccessful. In the case of unsuccessful joining, the intermediate joining result has only the information of DBa 101 and not the information of DBb 102.

The result matching classifying unit 130 includes a result comparing unit 13
1, a combination result mismatch data group 105, a combination result match data group 106, and a combination failure data group 107. DBb1 combined based on the information of DBa101 with respect to the intermediate combination results a103 and b104
The information 02 is compared in record units, and if the join results do not match, they are stored in the join result mismatch data group 105, and if they match, they are stored in the join result match data group 106. The data is stored in the combination failure data group 107.

The erroneous result elimination unit 140 is composed of a mismatch result selection unit 141 and a storage unit for the combination approval data group 108 and the combination rejection data group 109. The mismatch result selection unit 141 reads from the combination result mismatch data group 105. The displayed data group is displayed on a display device 160 such as a display, and a person judges the validity of the combination by looking at the address and name information corresponding to the result of the combination, and whether one is OK, whether both are NG. It is judged whether or not it is notified from the input device 170 such as a keyboard. The non-coincidence result selection unit 141 sorts and sorts the results, and if one is OK, it is stored in the combination approval data group 108, and if both are NG, it is stored in the combination rejection data group 109.

FIG. 3 shows an example in which the result is different between the processing A of the combination processing unit 110 and the processing B of the combination processing unit 120, and a person judges that the combination has failed because neither of them has adopted it. In contrast to [5-2 / Tanaka Clinic] in [(address) block number-lot number / nominal] of DBa101, in the process A of the combining processing unit 110, a matching record of DBb102 could not be found at the same address, so the surrounding area And found [6-10 / Ichiro Tanaka],
We identified this as a combination target by matching the important proper noun "Tanaka".

On the other hand, in the process B of the merging processing unit 120, since no matching record is found in the address matching as well, the periphery is searched to find [5-12 / Nakai Clinic],
Since 3 characters match 4 characters of Tanaka Clinic, the degree of matching is high, so it was certified as a combination target.

These results are intermediate join results a, respectively.
103 and the intermediate combination result b104. Both of the combination flags are "1", indicating that the combination is successful. The result matching classifying unit 130 compares the intermediate join result a103 and the intermediate join result b104 by the result comparing unit 131 and stores them in the join result mismatch data group 105 because they do not match.

Mismatch result selection unit 1 of error result elimination unit 140
Reference numeral 41 displays these combined results read from the combined result disagreement data group 105 on the display device 160 in order to determine whether they are erroneous results. In this case, since the person who judges the result determines that there is a high risk of combining both after seeing the combination result of both, and selects the combination failure, both results are not adopted. Therefore, the result is stored in the combination rejection data group 109.

The combination result coalescing unit 150 totals the results based on the result of the error result eliminating unit 140 and the data group of the combination matching classifying unit 130 to create a combination DB 200.

[0023]

[Example] Specifically, a telephone directory DB (about 100
For reference to the connection between the (1 million cases) and the housing map DB, reference 1 (Yasuda, Matsumura, Mizumachi, Karasawa “Map guidance system using telephone / FAX”, 3rd Functional Graphic Information System Symposium Proceedings, 1992/4, pp.81-86.) Japanese-language specific information comparison method (Process A)
The result of automatic joining is about 85% of successful joining,
The coupling failure was about 15%, and the estimated mis-coupling due to sampling was 5% (coupling reliability 95%).

On the other hand, in contrast to this, Japanese Patent Application No. 10-21617
The result of combining processing by the method (Process B) proposed in No. is about 90% of successful combination and about 10% of unsuccessful combination. Was 20% (about 5% for treatment A alone, about 10% for treatment B alone, and 5% for both bonds, resulting in disagreement) and 5% for unbound.

Further, as a result of manually checking only those inconsistent with each other in the results, 2% of both formulas can be judged to be correct, and the combination succeeds in selection, and 3% (60% of all errors) are combination errors. I was able to judge and eliminate it.

For example, in the case of using only the "Processing A" method, in order to improve the overall reliability by 3%, an error of 10
Assuming that the data evenly exists in the data of 0,000 records, it is considered necessary to manually check the results of the 600,000 records to eliminate erroneous combinations. Moreover, in order to judge whether or not the combined result is incorrect by looking at only one combined result, it is necessary to proceed while checking the surrounding information on the housing map etc. and judging whether or not there is more suitable one. There is, Tokyo 2
In the case of 3 wards, the efficiency was about 200 cases / person day.

On the other hand, in the example in which the method of the present invention is adopted and only the result mismatch (5%) of the two methods is manually checked,
It is only necessary to check 10,000 cases (5%), and to exclude only those that are not suitable for combining the results of the two methods (high risk of incorrect connection). , 3000 cases / man-day efficiency could be improved. As a result, the number of man-hours required to eliminate 60% of all misconnections was reduced to 1/180.

[0028]

As described above, according to the present invention,
By comparing only the disagreement parts of the results of the multiple types of join processing methods, it is possible to eliminate the automatic join error very efficiently and with a small number of man-hours, and improve the reliability of the entire join result.

[Brief description of drawings]

FIG. 1 is a block diagram showing a main configuration of the present invention.

FIG. 2 is a diagram specifically showing an outline of a processing flow by the processing apparatus of the present invention shown in FIG.

FIG. 3 is a diagram showing an example in which a combination result by a plurality of combination processing units (process A and process B) is different, and a person does not adopt either of them and the combination fails.

[Explanation of symbols]

101 Database to be combined (DBa) 102 Database to be combined (DBb) 110 Combine Processing Unit (Process A) 120 Combine Processing Unit (Process B) 130 Combined Butt Classifier 140 False result elimination section 150 Combined result unit 160 display device 170 input device 200 combined database (DB)

Front page continuation (72) Inventor Kensaku Fujii 3-19-2 Nishishinjuku, Shinjuku-ku, Tokyo Nihon Telegraph and Telephone Corporation (56) Reference JP-A-9-259141 (JP, A) JP-A-9 -62700 (JP, A) Hideki Kawabe et al., 104 number guidance operation search support method, NTT R & D, Japan, Japan Telecommunications Association, May 10, 1996, Vol. 45, No. 5, pp. 433- 440 pages Yuriko Yoshizawa et al., Multimedia map system (1) System overview and DB construction, Proc. Of the 41st national conference (2nd half) (4), Japan, Japan Society for Information Processing, 1990 Sept. 6, pp. 150-151 (58) Fields investigated (Int.Cl. ⁷ , DB name) G06F 17/30 JISST file (JOIS)

Claims (4)

(57) [Claims] The method according to claim 1] 2 types of data that is different with the address information
It is equipped with storage means for the database to store
Join processing part, result match classification part, and
The error result elimination unit that has an output interface and the combined result
In a system including a computer having a body part,
Processing method der two types of the computer the data linking
I, each of the plurality of coupling processing unit, said database
The process of combining the two types of data to be combined, which are to be combined, by a plurality of kinds of combining processing methods having different combining evaluation methods, and the result matching classifying unit match the combining results of the plurality of kinds of combining processing methods. Then, the process of extracting the ones in which the results do not match, and the error result excluding unit determines whether the result extracted by the result matching classifying unit is an incorrect combination or not, as the input / output interface.
An over extent you eliminate erroneous coupling inputs and outputs by interface, the coupling results coalescence portion, the combined binding results due to the binding processing method of the plurality of types which eliminated the false bond, and a process in which the binding results to determine An error elimination method for automatic combination of different types of data having address information. 2. The plurality of types of combination processing methods used by the plurality of combination processing units are to process elements such as a resident name and a building name in address information in Japanese and evaluate important words to combine them. The method of claim 1 and the second method of evaluating the degree of coincidence of characters according to the degree of coincidence of addresses. 3. A processing device for connecting two different kinds of data having address information by a computer, and a plurality of combining processes for combining the two kinds of data to be combined by different combining evaluation methods. Section and the combination results of the plurality of combination processing sections are compared,
A result match classification unit that extracts results that do not match, and an error result elimination unit that has an input / output interface for eliminating the wrong combination as a target for determining whether the result extracted by the result match classification unit is a wrong combination. An error elimination processing apparatus for automatic heterogeneous data combination having address information, comprising: a combination result combining unit that obtains a combination result obtained by combining the combination results of the plurality of combination processing units excluding erroneous combination. 4. A first combination processing unit that processes elements such as a resident name and a building name in address information in Japanese in the plurality of kinds of combination processing units, evaluates important words, and combines them. 4. An error elimination processing apparatus for automatic combination of different types of data with address information according to claim 3, further comprising a second combination processing unit that evaluates the degree of coincidence of characters according to the degree of coincidence of addresses.