JP3350594B2 - Search device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION The present invention calculates an effective similarity from a problem and presents a case with a high evaluation as a retrieval result, calculates the effectiveness of the case, and presents a case with a low evaluation as an object to be deleted. At the time of the search. Evaluate the effectiveness of cases that gradually decrease over time from the time of case registration in a search device that searches cases with similar contents to problems entered by users from a case database that stores past problem solving cases It is desired to do.

[0002]

2. Description of the Related Art Conventionally, in the business of providing a service responding to a problem (question) from a user, past QA (problem / response) cases are stored in a database.
For a newly received problem, a case search device that searches a database for QA cases having contents similar to that of the problem is used, and it can be expected that the efficiency of work will be greatly improved.

[0003] In the conventional case retrieval apparatus as described above, cases made into a database gradually become obsolete with the passage of time, and usually become useless information for users.

[0004]

When a case is made into a database, it gradually loses its effectiveness over time,
Since a response is made based on this case, an appropriate response cannot be returned, and unnecessary unreferenced cases accumulate in the database, causing a problem that the memory cannot be effectively used. In order to avoid this, data that has passed a certain period from the data registration point is extracted from the database, the data is displayed on the screen, and the administrator judges whether the data is necessary or unnecessary and deletes the unnecessary data. As a result, data that is no longer valid after a certain period of time may be deleted, but data that has high validity and is frequently referenced by users after a certain period of time should be left. Further, there is a problem that the administrator has to judge the data one by one to judge whether or not the data is necessary or unnecessary, and the operation is difficult.

[0005] The present invention solves these problems,
Similarity and validity period (or actual validity period calculated from reference frequency) of data retrieved by similar search of database
The validity calculated from the data and the effective similarity calculated from the data are displayed at a predetermined value or more, or the validity calculated from the validity period of the data in the database (or the actual validity period calculated from the reference frequency) is displayed at a predetermined value or less. The purpose of the present invention is to present only data having high validity of similar searched data and to delete data having low validity from the database as unnecessary data.

[0006]

Means for Solving the Problems [Note: Batch conversion from similarity search to similarity search has been completed] FIG. 1 is a diagram showing the principle configuration of the present invention. In FIG. 1, a database 2 stores data so that the data can be easily searched. Here, a registration date and time, a validity period, and a reference frequency are registered in association with the data.

The validity calculating means 3 calculates the validity from the validity period, the registration date and time, and the current date and time. The actual validity period calculating means 10 calculates the actual validity period based on the data reference frequency and the validity period.

The similarity calculating means 6 searches the database 2 based on a keyword extracted from a problem for which a search request has been made and retrieves similar data (for example, data containing a keyword). For example, it is to calculate a ratio including a plurality of keywords extracted from the question).

The effective similarity calculating means 8 calculates an effective similarity based on the validity and the similarity. The presenting means 22 displays, on the display device 23, data whose effective similarity is equal to or more than a predetermined threshold value at the time of retrieval, or displays data whose validity is equal to or less than the predetermined value on the display device 23 at the time of deletion. / Requesting unnecessary input.

The display device 23 displays data. The deletion means 24 deletes the data for which deletion has been instructed from the database 2.

[0011]

According to the present invention, as shown in FIG. 1, in response to a search instruction of the database 2 specifying a problem described in a natural language, the similarity calculating means 6 uses a database based on keywords extracted from the problem. 2. When similar data containing the keyword is retrieved by searching for the keyword 2, the similarity is calculated, and the validity calculating means 3 validates the retrieved data based on the validity period, registration date and time, and current date and time. The effective similarity calculating means 8 calculates the effective similarity based on the similarity and the validity, and the presenting means 22 presents data having the effective similarity equal to or more than a predetermined value as a search result. ing.

Also, in response to a search instruction of the database 2 specifying a problem described in a natural language, the similarity calculating means 6 searches the database 2 based on the keyword extracted from the problem and includes the keyword. The similarity data is extracted and the similarity is calculated.
Calculates the validity based on the actual validity period, the registration date and the current date and time obtained from the reference frequency and the validity period of the extracted data, and the effective similarity calculating means 8 calculates the validity based on the similarity and the validity. The similarity is calculated, and the presentation means 22 presents data whose effective similarity is equal to or greater than a predetermined value.

In response to the deletion request, the validity calculating means 3 calculates the validity based on the validity period of the data in the database 2, the registration date and time, and the current date and time. The validity is calculated based on the actual validity period obtained from the validity period, the registration date and time, and the current date and time, and the calculated validity is data having a predetermined value or less.
2 is presented as an unnecessary data candidate, and is deleted from the database 2 in response to the deletion means 24 being instructed to delete.

Therefore, if the effective similarity calculated from the similarity of the data extracted by performing a similar search on the database and the validity obtained from the validity period (or the actual validity period obtained from the reference frequency) is equal to or greater than a predetermined value. By displaying the validity of the database data from the validity period (or the actual validity period calculated from the reference frequency) and displaying it below a predetermined value and deleting it when necessary. It is possible to present only the data with high validity of the retrieved data or to delete data with low validity from the database.

[0015]

Next, the configuration and operation of an embodiment of the present invention will be sequentially described in detail with reference to FIGS.

FIG. 2 is a block diagram showing one embodiment of the present invention.
In FIG. 2, the current date and time 1 is the current date and time. The case database 15 is an example of the database 2 and is registered so that cases can be easily searched. Here, the case registered as data in the case database 15 includes a registration date and time 2a, a validity period 2b, a reference frequency 2c, and a case content 2d.

The registration date and time 2a is the date and time when the case content 2d was registered in the case database 15. The validity period 2b is
This is a valid period after the case contents 2d are registered in the case database 15.

The reference frequency 2c is obtained by registering a reference frequency of +1 for each reference to the case content 2d. Case content 2d
Is the content of the actual case.

The validity calculating means 3 calculates the validity 4, and here, for example, the current date and time at the time of search, the registration date and time 2a of the searched case, and the validity period 2 of the searched case
The validity is calculated based on b (or, if the reference frequency 2c of the searched case is registered, the actual validity period obtained from the curve in FIG. 7 described later).

The effectiveness is calculated by a function having the following properties. E (0) = 1 (Equation 1) limE (t) = 0 (Equation 2) t → ∞ E (t0) ≧ E (t1) (t0 ≦ t1) (Equation 3) where E: Effectiveness t: Elapsed time (= Current date / time−Registration date / time) (Equation 1) Indicates that the validity is 1 (completely valid).

(Equation 2) indicates that the validity is 0 (invalid) when a long time has passed since the registration date and time. (Equation 3) indicates that the effectiveness always decreases with time.

As long as the properties of (Equation 1), (Equation 2) and (Equation 3) are satisfied, the form of E (t) may be any. As an example of E (t), Here, R: validity period (or actual validity period) a: a parameter that determines the speed of reduction of the validity near t = R. This is schematically shown in FIG.

Problem 5 is a problem (question) in which the user makes an inquiry in a natural language. When this question 5 is input, although not shown, a keyword is extracted with reference to a dictionary, and a case database 15 is searched based on the keyword group extracted from the question 5 to search for a case in which the keyword is partially included. The content 2d is extracted as a search result.

The similarity calculating means 6 responds to the search request by retrieving the similarity 7 of the case searched from the case database 15.
Is calculated, and the similarity 7 is calculated based on the content of the problem and the case. For example, the similarity 7 is calculated based on the ratio of the keyword extracted from the problem to the content of the case (described later with reference to FIG. 4).

The effective similarity calculating means 8 calculates an effective similarity 9 based on the validity 4 and the similarity 7.
For example, it is calculated by the following equation (Equation 5). Effective similarity = effectiveness × similarity (Equation 5) The actual effective period calculating means 10 calculates the actual effective period 11 based on the effective period 2b and the reference frequency 2c (to be described later with reference to FIG. 7). ).

The unnecessary case discriminating means 13 is for discriminating a case whose validity 4 is equal to or less than the validity threshold 12 from the unnecessary case 14. The similar case identification means 18 identifies a case in which the effective similarity 9 is equal to or greater than the effective similarity threshold as a similar case 19.

The output case buffer 20 stores a case (unnecessary case 14 or similar case 1) which is a result identified by the unnecessary case identifying unit 13 or the similar case identifying unit 18.
9) is stored.

The presentation means 21 outputs the case (unnecessary case 14 or similar case 19) stored in the output case buffer 20.
Is displayed on the display device 22. For example, the unnecessary case 14 is presented as a deleted case candidate from the case database 15 to urge the manager to input a necessity / unnecessary instruction, or presented to the user as a similar search result candidate. It is.

The display device 22 displays the unnecessary case 14, the similar case 19, and the like. The case deletion means 23
The unnecessary case 14 is displayed on the display device 22, and when a designation of unnecessary is input from the administrator, the case database 1 is displayed.
5 (described later with reference to FIGS. 8 and 9).

The device control means controls the entire system. Next, the operation at the time of searching the configuration of FIG. 2 will be described in detail according to the order shown in the flowchart of FIG.

In FIG. 3, S1 issues a search request. In this case, the user inputs the question 5 in natural language, and requests the case database 15 to be searched. S2 performs a similarity search. This corresponds to the search request from the user in S1,
As described above, keywords are extracted by referring to a keyword dictionary from a problem described in a natural language, and a case database 15 is searched based on these extracted keyword sets to search for a case in which a keyword is partially included. All are taken out, and the similarity is calculated based on the ratio at which the keyword is included in the case.

In step S3, the effectiveness is calculated. This is S2
Based on the registration date / time, validity period, reference frequency, and current date / time of each case extracted in the above, the validity is calculated using the function of (Equation 4).

In step S4, an effective similarity is calculated based on the effectiveness and the similarity. For example, it is calculated as effective similarity = effectiveness × similarity.

In step S5, after sorting, a value less than a predetermined value is deleted. In this case, among the search result candidate cases searched from the case database 15 based on the effective similarity in S4, those that are equal to or less than a predetermined value are deleted as unnecessary from the search result candidates.

In step S6, a list is displayed. In this case, the case after the unnecessary case is deleted in S5 is displayed on the display device 22 as a search result, and is presented to the user. Upon seeing this, the user selects a desired search result from the search result, extracts the content, and displays the content.

As described above, only the user inputs a question in a natural language, the apparatus of the present invention automatically extracts a keyword, and among the cases searched for similarity from the case database 15, those having a high effective similarity. The user selects only one of them and presents the list, and the user selects a desired one from these lists and further displays the contents. As a result, it is possible to always present only search results having a high effective similarity from the case database 15 at the time of search.

FIG. 4 shows a flowchart for calculating the similarity according to the present invention. This is because the similarity search of S2 in FIG.
It is a detailed flowchart at the time of calculating the similarity of the case which carried out similarity search.

In FIG. 4, a step S11 inputs a question in a natural language. In this method, a user inputs a question (question sentence) in a natural language for a problem to be searched. In step S12, keywords are extracted with reference to a dictionary (keyword dictionary). This is to extract all matching keywords by referring to the keyword dictionary for the problem in the natural language input by the user in S11.

In step S13, the case database 15 is searched for similarity. In this case, the case database 15 is searched based on the keywords extracted in S12, and all cases including the matching keywords are extracted.

In step S14, a similarity is calculated. In this method, the similarity is calculated as the ratio of the keyword in question in the search case. As described above, in response to the user inputting a problem described in a natural language and requesting a similar search, all keywords are extracted from the problem by referring to the keyword dictionary, and based on these extracted keywords, A similar search is performed on the case database 15 to extract all cases including the keyword, and the similarity of the extracted cases is expressed by the ratio of the included keywords. Thus, it is possible to calculate the similarity of the searched case.

FIG. 5 shows a detailed flowchart of the search according to the present invention. This is a detailed flowchart of the search flowchart of FIG. In FIG. 5, S21 extracts a case. That is, in response to a search request specifying a problem in a natural language from a user, the case database 15 is searched based on the keyword extracted from the problem and a case including the keyword is extracted.

In S22, it is checked whether or not the reference frequency 2c of the case extracted in S21 is set. If there is a setting, the actual valid period (described later with reference to FIG. 7) is calculated by applying the curve in FIG. 7 based on the reference frequency 2c and the valid period 2b in S23, and then the process proceeds to S24. on the other hand,
If no setting has been made, the process proceeds directly to S24.

In step S24, the validity is calculated. This is S
When the reference frequency has not been set for the case extracted in 21, the registration date and time, the current date and time, and the validity period are substituted into (Equation 4) to calculate the validity.

On the other hand, regarding the case taken out in S21,
If the reference frequency has been set, the actual validity period is obtained by applying the reference frequency and the validity period to the following equation. R _eff (f) = R / (1−f) Here, R _eff represents an actual effective period, R represents an effective period, and f represents a reference frequency. The actual effective period can be obtained, for example, by applying the reference frequency and the effective period to FIG. Then, the obtained effective validity period, registration date and time, and current date and time are substituted into (Equation 4) to calculate the effectiveness.

In step S25, a similarity is calculated. This is calculated for the case extracted from the case database 15 based on the ratio of keywords included in the case, as shown in the flowchart of FIG. 4 described above.

In step S26, an effective similarity is calculated. That is, the effective similarity is calculated by the following equation based on the effectiveness calculated in S24 and the similarity calculated in S25. Effective similarity = effectiveness × similarity S27 identifies a similar case. This means that the effective similarity calculated in S26 is similar to a predetermined value or more, and the following is identified as dissimilar.

In step S28, a similar case is checked. O
In the case of N, that is, when the effective similarity of the case is equal to or more than the predetermined value, it is determined that the case is similar, so the case is stored in S29, and the process proceeds to S30. On the other hand, if the case is OFF, that is, if the effective similarity of the case is equal to or less than the predetermined value, it is determined that the case is not similar, so the case is deleted without being stored, and the process proceeds to S30.

In S30, it is checked whether there is any unprocessed case. If there is no processing, the next case is extracted in S21, and S22 and subsequent steps are repeated. On the other hand, if there is no unprocessed data, the case is output in S31, and the display device 2 is displayed to the user.
2 and display and present the search results.

As described above, it is possible to extract the case including the keyword extracted from the problem from the case database 15, obtain the effective similarity for the extracted case, and present only the important cases having a predetermined value or more.

FIG. 6 shows an example of the effectiveness curve of the present invention. This is an effectiveness curve for calculating the effectiveness based on the elapsed time (current date / time-registration date / time) described above. As can be seen from this validity curve, the elapsed time (current date and time-registered date and time)
Becomes larger, the validity decreases exponentially. Here, the horizontal axis represents elapsed time, and the vertical axis represents effectiveness.

FIG. 7 shows the actual effective period curve of the present invention.
Here, the horizontal axis represents the reference frequency of the case, and the vertical axis represents the actual effective period / effective period. The valid period is the valid period added to the case. The reference frequency is the frequency of referring to the case. Therefore, in order to calculate the actual effective period of the case, the value of the reference frequency of the case is plotted on the horizontal axis, and the value of the actual effective period / effective period on the vertical axis at the intersection with the curve at that time is read, and the effective value is read from this value. The actual effective period to be obtained is calculated by multiplying the periods.

When the reference frequency is set in the case as described above, the value of the actual effective period / effective period at that time increases as the reference frequency increases. Calculate the validity period. As a result, if the reference frequency increases, the actual effective period becomes longer even if the effective periods are the same. For example, when the validity is obtained by using (Equation 4), the validity increases as the validity period increases. Therefore, even if the similarity is the same, the effective similarity is large and is evaluated as important. Become.

Next, the operation of the configuration shown in FIG. 2 in the case where a case with a low degree of effectiveness is presented from the case database 15 and deleted in response to a deletion instruction will be described in detail with reference to FIGS. 8 and 9.

FIG. 8 shows a deletion flowchart of the present invention. In FIG. 8, S41 issues a deletion request. this is,
An administrator issues a request to delete the case database 15.

In step S42, the effectiveness is calculated. In this case, the validity of one case extracted from the case database 15 is calculated in S21 to S24 of FIG. That is, when the reference frequency is set in the case, the actual valid period is calculated by applying the curve of FIG. 7 based on the reference frequency and the valid period, and the validity is calculated using (Equation 4). I do. On the other hand, if the reference frequency has not been set for the case, the validity is calculated using (Equation 4).

In step S43, sorting is performed. In this case, the cases are sorted in descending order of the validity obtained in S42. S44
Presents a value equal to or less than a predetermined value. In this case, among the cases sorted in S43, a case having a low evaluation whose validity is equal to or less than a predetermined value is displayed on the display device 22 and presented to the administrator as a deletion target.

In step S45, the administrator confirms the deletion is OK for the presentation of the deletion target in step S44. S46 deletes. This is because the case database 15 is deleted at S45.
Delete K and the case confirmed by the administrator.

As described above, in response to a deletion request from the manager, the validity of the case in the case database 15 is calculated, and a case with a low evaluation below a predetermined value is presented to the manager. The deleted case is deleted from the case database 15. Thus, only by inputting a deletion request, the validity of a case is automatically calculated, and only a case with a low evaluation is presented as a case to be deleted, so that a case database can be easily deleted. It became possible.

FIG. 9 is a detailed flowchart of the deletion according to the present invention. This is a detail of the deletion flowchart of FIG. Here, S51 to S54 correspond to S21 to S in FIG.
24, and a description thereof will be omitted. Here, S5
From 1 to S54, the validity of the cases sequentially extracted one by one from the case database 15 for which the deletion request is made is calculated.

A step S55 identifies unnecessary cases. this is,
Cases whose validity calculated in S54 is equal to or less than a predetermined value are identified as unnecessary cases. In S56, the unnecessary case is checked. If the case is ON, that is, if the validity of the case is equal to or less than the predetermined value, it is determined that the case is unnecessary. Therefore, the case is stored in S57, and the process proceeds to S58. On the other hand, when the case is OFF, that is, when the validity of the case is equal to or more than the predetermined value, it is determined that the case is an important case, and the process directly proceeds to S58 without storing.

A step S58 checks whether there is any unprocessed case. If there is no processing, the next case is extracted in S51, and S52 and subsequent steps are repeated. On the other hand, if no processing has been performed, the case is output in S59, and the display device 2 is displayed to the administrator.
2. The search result is displayed and presented on the screen 2, and the deletion is OK for the case with a low evaluation of the deletion target presented by the administrator in S <b> 60.
Is input, the case is deleted from the case database 15 in S61.

As described above, the validity of the case extracted from the case database 15 is obtained, and a value less than a predetermined value is presented to the administrator as a low-evaluation deletion target, and the case designated as OK is deleted from the case database 15. As a result, only cases with low evaluations are presented, so that the administrator can easily and quickly delete obsolete cases in the case database 15.

[0063]

As described above, according to the present invention,
Similarity and validity period (or actual validity period calculated from reference frequency) of data retrieved by similar search of database
If the validity calculated from the data and the effective similarity calculated from the data are equal to or more than a predetermined value, the search result is presented, or the validity period of the data in the database (or the actual validity period calculated from the reference frequency)
Is displayed when the validity obtained from is less than or equal to a predetermined value and it is necessary to delete the data. It is now possible to easily and easily delete low-profile data from the database. By these, each data of the database (each case)
Evaluate the effectiveness of the search and present the search results, or remove the less effective cases from the database and present them for deletion.
Can be easily deleted in response to the instruction.

[Brief description of the drawings]

FIG. 1 is a principle configuration diagram of the present invention.

FIG. 2 is a configuration diagram of one embodiment of the present invention.

FIG. 3 is a search flowchart of the present invention.

FIG. 4 is a similarity calculation flowchart according to the present invention.

FIG. 5 is a detailed search flowchart of the present invention.

FIG. 6 is an example of an effectiveness curve according to the present invention.

FIG. 7 is an example of an actual effective period curve according to the present invention.

FIG. 8 is a deletion flowchart of the present invention.

FIG. 9 is a detailed flowchart of deletion according to the present invention.

[Explanation of symbols]

 1: Current date and time 2: Database 2a: Registration date and time 2b: Validity period 2c: Reference frequency 2d: Case content 3: Effectiveness calculating means 4: Effectiveness 5: Problem 6: Similarity calculating means 7: Similarity 8: Effective similarity Degree calculation means 9: Effective similarity 10: Actual effective period calculation means 11: Actual effective period 12: Effectiveness threshold value 13: Unnecessary case identification means 14: Unnecessary case 15: Case database 17: Effective similarity threshold value 18: Similar case identification Means 19: Similar case 20: Output case buffer 21: Presenting means 22: Display device 23: Case deleting means

Continued on the front page (72) Inventor Keiko Matsumoto 1015 Uedanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture Inside Fujitsu Limited (72) Inventor Kazuhiko Tazawa 1015 Uedanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture Fujitsu Limited (56) Reference Documents JP-A-3-172966 (JP, A) JP-A-3-228179 (JP, A) JP-A-58-3029 (JP, A) JP-A-5-181715 (JP, A) Hiroshi Matsuo, Kazuto Uchino , Text-based search method based on semantic attributes, Transactions of the Information Processing Society of Japan, September 15, 1991, Vol. 32, No. 9, p. 1172-1179 (58) Field surveyed (Int. Cl. ⁷ , DB name) G06F 17/30 G06F 12/00 JICST file (JOIS)

Claims (3)

(57) [Claims] When a database (2) is searched based on a keyword extracted from a problem and data containing the keyword is extracted in response to a search instruction designating the problem, the similarity of the data is determined. Similarity calculating means (6) for calculating; validity calculating means (3) for calculating validity based on the validity period, registration date and time, and current date and time of the extracted data; A retrieval apparatus comprising: an effective similarity calculating means (8) for calculating an effective similarity based on the information; and a presentation means (22) for presenting data whose effective similarity is equal to or greater than a predetermined value. 2. Searching a database (2) based on a keyword extracted from a problem and extracting data including the keyword in response to a search instruction designating the problem, the similarity of the data is determined. A similarity calculating means for calculating; a validity calculating means for calculating a validity based on an actual validity period, a registration date and a current date and time obtained from a reference frequency and a validity period of the extracted data; An effective similarity calculating means (8) for calculating an effective similarity based on the similarity and the validity; and a presentation means (22) for presenting data with the effective similarity being a predetermined value or more. A retrieval device characterized by the following. 3. An actual validity period calculated based on a validity period of data in the database (2), a registration date and a current date and time, or an actual validity period obtained from a reference frequency and a validity period of the data in the database (2). A validity calculating means (3) for calculating the validity based on the registration date and the current date and time; a presenting means (22) for presenting data whose calculated validity is equal to or less than a predetermined value as unnecessary data candidates; A deletion unit (24) for deleting from the database (2) in response to an instruction to delete the data from the database (2).