JPH05204980A - Information retrieving device - Google Patents

Abstract

PURPOSE:To obtain a retrieving result matched with the retrieving request as much as possible when a data base is retrieved based on a retrieving question sentence consisting of the ambiguous conditions. CONSTITUTION:An adaptability calculating part 505 calculates the adaptability between the ambiguous retrieving conditions included in a retrieving question sentence and the data. A variable connection calculating part 506 calculates the adaptability to an entire retrieving question sentence through a prescribed function arithmetic and with the adaptability calculated by the part 505 used as an input. A parameter control part 509 controls the function arithmetic parameter so as to minimize the error produced between the adaptability calculated by the part 506 and the teacher data inputted by a retriever. Thus it is possible to automatically change the function so that the retrieving result is matched with the retrieving request.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an information retrieval device for retrieving predetermined information from a database.

[0002]

2. Description of the Related Art Conventionally, as a method for searching a database using ambiguous search query sentences, for example, information
Procesing and Ant Management (Information Processing & M
anagement), No.13, pp.289-303. FIG. 6 shows a block diagram of an example of a conventional information retrieval apparatus using the membership function. In FIG. 6, a database 601 stores data to be searched, a membership function storage unit 602 stores membership functions, and a search question sentence input unit 603 receives a search question sentence including a fuzzy proposition. A means for calculating the goodness of fit, means for obtaining data from the database 601 and using the membership function of the membership function storage unit 602 to find the goodness of fit of the data to the fuzzy proposition input from the search question sentence input unit 603. The join operation unit 605 is means for performing min operation and max operation with the goodness of fit of the data for the fuzzy proposition as input, and the search result output unit 606 outputs data as a search result in descending order of operation result of the join operation unit 605. It is a means.

As an example of the data stored in the database 601, a person database is shown in (Table 1).

[0004]

[Table 1]

The person database 601 has "name", "age", and "annual income" as attributes. Further, FIG. 7 shows μ _{AGE = YOUNG} representing “age = young” and μ _{SALARY = HIGH} representing “annual income = high” as examples of membership functions stored in the membership function storage unit 602.

The operation of the conventional information retrieval apparatus configured as described above will be explained below using a retrieval question sentence consisting of a fuzzy proposition "Search for a person who is young and has a high annual income" as an example. To do. The search question sentence input unit 603 decomposes the entire search question sentence into two fuzzy propositions “age = young”, “annual income = high”, and “Katsu”, and the fitness calculation unit 604.
Send to. The goodness-of-fit calculation unit 604 extracts all person data D _i (i is a data number) from the database 601, and uses the membership function of the membership function storage unit 602 to obtain the fuzzy proposition “age = young” of the data D _i. And "annual income =
High) _{= AGE = YOUNG} (AGE (D _i )) (AGE
(D _i ): Age of data D _i ) and μ _{SALARY = HIGH} (SALARY (D _i )) (SA
LARY (D _i ): Annual income of data D _i ) and μ _{AGE = YOUNG} (AGE
(D _i )) and μ _{SALARY = HIGH} (SALARY (D _i )) are sent to the join operation unit 605 together with “Katsu”. The join operation unit 605 obtains the goodness of fit μ (D _i ) of the data D _i with respect to the entire search question sentence by performing the process on “Katsu”, and sends μ (D _i ) to the search result output unit 606. Here, the processing for “Katsu” means the matching degree μ of the data D _i with respect to the fuzzy propositions “age = young” and “annual income = high” as shown in (Equation 1).
_{AGE = YOUNG (AGE (D i} )) and _{μ SALARY = HIGH (SALARY (D} i)) is a process of performing min operation as inputs, fitness mu (D to the search query entire sentence resulting data D _{i i} ).

[0007]

[Equation 1]

Μ _{AGE = YOUNG} (AGE (D _i )) of each person data and μ
_{SALARY = HIGH} (SALARY (D _i )) and μ (D _i ) are shown in (Table 2).

[0009]

[Table 2]

The search result output unit 606 outputs data as a search result in the descending order of the goodness of fit μ (D _i ) for the entire search question sentence. The search results are shown in (Table 3).

[0011]

[Table 3]

The operation when a search is performed using the search query sentence "search for a person who is young or has a high annual income" will be described focusing on the difference from the operation of the above search example. In the combination operation unit 605, instead of the min operation which is the processing for "and", the processing for "or" is performed.
Calculate μ (D _i ) using max operation. The other processes are the same, so detailed description thereof will be omitted.

[0013]

However, in general, the meaning of "and" or "or" in a search question sentence consisting of two or more ambiguous search conditions depends on the searcher or the search question sentence. Different.
However, in the conventional information retrieval device having such a configuration,
Despite the different meanings of "and" and "or", the min and max operations are performed uniformly as processing for "and" and "or", respectively. Therefore, there is a problem that the search result that the searcher intends cannot be obtained.

The present invention solves the above-mentioned conventional problems. In the case of performing a search using a search question sentence composed of at least two or more fuzzy propositions, "and" or "or".
It is an object of the present invention to provide an information search device capable of obtaining an appropriate search result for a searcher's intention by setting an appropriate combination operation as a process for.

[0015]

According to the present invention, there is provided a database for storing data to be searched, a search question sentence input section for accepting an input of a search question sentence to the database, and a membership function storage for storing a membership function. Section, a fitness calculation unit that retrieves data from the database, and that uses the membership function to determine the fitness of the data with respect to the ambiguous search condition in the search question sentence, and at least a function that performs a calculation between the fitness levels. One or more join operation storage units, a join operation selection unit that selects a function of the join operation storage unit according to a searcher's input, and a match operation selection unit that selects the suitability as input And a detection unit that outputs the calculation result of at least the combination calculation unit as a search result. An information retrieval apparatus and a result output unit.

The present invention also includes a database for storing data to be searched, a search question sentence input unit for receiving an input of a search question sentence to the database, a membership function storage unit for storing a membership function, and Retrieval of data from a database and acceptance of the input of parameters of a fitness calculation unit that obtains the fitness of the data for the ambiguous search condition in the search query sentence using the membership function, and the function for performing the computation between the fitness A parameter input unit, a variable coupling operation unit that performs a coupling operation between the fitnesses by a function created by the parameters using the fitness of the fitness computing unit as an input, and at least a calculation result of the variable coupling calculation unit as a search result. And a search result output unit for outputting as.

Further, according to the present invention, there is provided a database for storing data to be searched, a search question sentence input unit for receiving input of a search question sentence to the database, a membership function storage unit for storing a membership function, and The data is retrieved from the database, and the membership function is used to perform a goodness-of-fit calculation unit that obtains a goodness-of-fit for the ambiguous search condition in the search query sentence of the data, and a function calculation that obtains a goodness-of-fit for the search condition indicating negation. A variable negation operation unit, a variable combination operation unit that performs a function operation between the adaptation degrees by using the adaptation degrees obtained by the fitness degree operation unit and the variable negation operation unit as an input, and in the database for the search question sentence. A teacher data input unit that receives an input of the fitness of data, a calculation result of the function of the variable coupling calculation unit, and An error calculation unit that obtains an error from the fitness input to the teacher data input unit, and a parameter that changes the functions of the variable negation calculation unit and the variable combination calculation unit by using the error obtained by the error calculation unit. It is an information search device having an adjustment unit and a search result output unit that outputs at least a calculation result of a function of the variable coupling calculation unit as a search result.

The present invention also includes a database for storing data to be searched, a search question sentence input section for receiving an input of a search question sentence to the database, a membership function storage section for storing a membership function, and The data is retrieved from the database, and the goodness-of-fit calculation unit for obtaining the goodness-of-fit for the ambiguous search condition in the search question sentence of the data using the membership function, and the goodness-of-fit calculated by the goodness-of-fit calculation unit as inputs A variable combination calculation unit that performs a function calculation between the goodnesses of fit, a teacher data input unit that receives an input of the goodness of fit of the data in the database with respect to the search question sentence, a calculation result of the function of the variable combination calculation unit, and the teacher. An error calculation unit that calculates an error from the conformity input to the data input unit, and an error calculated by the error calculation unit A teacher data determination unit that determines whether a predetermined criterion is satisfied, a parameter adjustment unit that changes the function using the error when the error satisfies the criterion, and at least a function of the variable coupling operation unit. An information search device having a search result output unit that outputs the calculation result of as a search result.

[0019]

According to the present invention, with the above configuration, a searcher can appropriately select a function used for processing "and" or "or" between ambiguous search conditions in a search question sentence. When a search question sentence consisting of at least two ambiguous search conditions is input to this information retrieval device, this search question sentence is divided into individual ambiguous search conditions and "and" or "between" these ambiguous search conditions. Or "decomposes into". Next, the degree of matching between all the data in the database and each of these ambiguous search conditions is obtained. Furthermore, by inputting these relevance values and performing a join operation between the relevance values using a function selected by the searcher, the relevance values of all the data with respect to the entire search question sentence are obtained. Finally, the goodness of fit for the entire search question is output as a search result. This allows the searcher to select a function to be used for the processing of "and" or "or" that he or she intends, so that the search result matching the search request can be obtained.

Further, according to the present invention, the searcher can appropriately set the parameters of the function used for the processing of "and" or "or" between ambiguous search conditions in the search query by the above configuration. When a search question sentence consisting of at least two ambiguous search conditions is input to this information retrieval device, this search question sentence is divided into individual ambiguous search conditions and "and" or "between" these ambiguous search conditions. Or "decomposes into". Next, the degree of matching between all the data in the database and each of these ambiguous search conditions is obtained. Further, by performing a join operation between the fitness values using a function created by these fitness values and the parameters set by the searcher, the fitness values of all the data with respect to the entire search question sentence are obtained. Finally, the goodness of fit for the entire search question is output as a search result. This allows the searcher to
Since the function used for the processing of "and" or "or" which is intended by the user can be steplessly changed by inputting the parameter, it is possible to obtain the search result that matches the search request.

Further, according to the present invention, when a search question text consisting of at least two ambiguous search conditions is input, the search question text is converted into an ambiguous search condition and an ambiguous search condition. Denial of "no"
And “and” or “or” between these ambiguous search conditions. Next, the degree of matching between all the data in the database and each of these ambiguous search conditions is obtained.
Further, the degree of conformity of the ambiguous search condition with "absent" is input, and a negative operation is performed using a function to change the degree of conformity. By using these functions as inputs and performing a join operation between the fitness values using a function, the fitness values of all the data with respect to the entire search question sentence are obtained. An error between the goodness of fit and the goodness of fit input by the searcher is obtained, and the error is used to change the functions of the NOT operation and the join operation. Finally, the goodness of fit for the entire search question is output as a search result. This allows the searcher to automatically adjust the function used for the processing of "and", "or", and negation "not" if he / she inputs the degree of conformity to the data as teacher data. Therefore, it is possible to obtain a search result that matches the search request.

Further, according to the present invention, when a search question text consisting of at least two ambiguous search conditions is input, the search question text is converted into individual ambiguous search conditions and these ambiguous search conditions. "Katsu" between search conditions
Or "or" to decompose. Next, the degree of matching between all the data in the database and each of these ambiguous search conditions is obtained. Further, by using these functions as inputs and performing a joint operation between the fitness values using a function, the fitness values of all the data with respect to the entire search question sentence are obtained. An error between the goodness of fit and the goodness of fit input by the searcher is obtained, and when the error satisfies a predetermined criterion, the function is changed using the error. Finally, the goodness of fit and the data for the entire search question sentence are output as a search result. With this, when the searcher inputs the degree of conformity to the data as the teacher data, even if the teacher data is input by mistake, the error can be identified and the wrong teacher data can be eliminated. It is possible to automatically adjust the function used for the processing of “” as intended, and obtain a search result that matches the search request.

[0023]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 shows a block diagram of an information retrieval apparatus which is an embodiment of the present invention according to claim 1. In FIG. 1, a database 101 is a means for storing data to be searched, a membership function storage unit 102 is a means for storing membership functions, and a join operation storage unit 103 is a function for obtaining one output from at least two or more inputs. Means for storing at least two or more, and the join operation selecting unit 104 includes the join operation storage unit 1
Means for accepting input for selecting one function from 03,
There are at least two search question text input units 105 m
The means for accepting the input of the search question sentence consisting of fuzzy propositions, the goodness-of-fit calculation unit 106 retrieves data from the database 101, and uses the membership function of the membership function storage unit 102 to make each of the m fuzzy data. A means for obtaining a goodness of fit with respect to a proposition, a join operation unit 10
Reference numeral 7 is means for performing a join operation between the fitness values by using the function selected by the input to the join operation selection unit 104 with the fitness value as an input, and the search result output unit 108 is a join operation unit 107.
Is a means for outputting the calculation result of the function and the data as the search result.

As an example of the data stored in the database 101, the same person database (Table 1) as the conventional example is used. As an example of the membership function stored in the membership function storage unit 102, the same membership function of FIG. 7 as in the conventional example is used. The functions stored in the combination operation storage unit 103 are shown in (Equation 2) (Equation 3) (Equation 4) (Equation 5) (Equation 6) (Equation 7).

[0026]

[Equation 2]

[0027]

[Equation 3]

[0028]

[Equation 4]

[0029]

[Equation 5]

[0030]

[Equation 6]

[0031]

[Equation 7]

Where μ _k (k = 1,2,3, ...) is the input of the function,
μ is the output. (Equation 2) (Equation 3) (Equation 4) is "Katsu"
Is a function used for the processing of (Equation 5) (Equation 6) (Equation 7)
Is a function used for processing “or”.

The operation of the embodiment of the present invention according to claim 1 will be described below.

The searcher inputs at least two or more m search questions consisting of fuzzy propositions into the search question text input unit 105, and selects "Katsu" or "or" in the search question texts.
The function * to be used for the processing of (2) is selected from (Equation 2) to (Equation 7) and input to the join operation selection unit 104. The search question sentence input unit 105 converts the entire search question sentence into individual fuzzy propositions F _j (2 ≦ 2
j ≦ m) and “fu” or “or” between these fuzzy propositions, and send them to the fitness calculation unit 106.
The goodness-of-fit calculation unit 106 extracts all person data D _i (i is a data number) from the database 101, and uses the membership function of the membership function storage unit 102 to obtain the data.
D _i calculated fuzzy proposition F fit to _j mu _Fj attribute A _j attribute value A _j of _{(D i) (A j (} D i)), combine these fit together "cutlet" and "or" It is sent to the calculation unit 107. On the other hand, the join operation selection unit 104 retrieves the function * selected by the searcher's input from the join operation storage unit 103 and sends it to the join operation unit 107. The join operation unit 107 performs a process for “and” or “or”. As shown in (Equation 8), the processing for “Katsu” and “or” means the goodness of fit μ _Fj for the m fuzzy propositions F _j of the attribute value A _j (D _i ) of the data D _i.
This is a process in which (A _j (D _i )) is used as an input and a join operation is performed by the function * selected by the join operation selection unit 104.

[0035]

[Equation 8]

The search result output unit 10 sets the calculation result of this function as the goodness of fit μ (D _i ) of the data D _i with respect to the entire search question sentence.
Send to 8. The search result output unit 108 outputs data as search results in the descending order of goodness of fit μ (D _i ) for the entire search question sentence.

The operation of this embodiment will be described below when a search is actually performed using a specific example of the search question sentence "search for a person who is young and has a high annual income."

The retrieval question sentence input unit 105 decomposes the retrieval question sentence into two fuzzy propositions F ₁ = “age = young”, F ₂ = “annual income = high”, and “Katsu”. Goodness-of-fit calculation unit 106
Retrieves all the person data D _i from the database 101 and uses the membership function μ _{AGE = YOUNG} of the membership function storage unit 102 to set the attribute value AGE of “age” of the data D _i.
Request (D _i) of the fuzzy proposition F ₁ = "Age = young" for adaptability _{μ AGE = YOUNG (AGE (D} i)). Similarly, the fitness μ _{SALARY = HIGH} (SALARY (D _i )) for the fuzzy proposition F ₂ = “annual income = high” of the attribute value SALARY (D _i ) of “annual income” is also obtained. This μ _{AGE = YOUNG} (AGE (D _i )) and μ _{SALARY = HIGH} (SALARY (D _i ))
And are sent to the join operation unit 107 together with “Katsu”. Μ _{AGE = YOUNG} (AGE (D _i )) and μ _{SALARY = HIGH} (SALARY (DAL
_i )) and are shown in (Table 4) and (Table 5), respectively.

[0039]

[Table 4]

[0040]

[Table 5]

On the other hand, the join operation selection unit 104 retrieves the function * selected by the input of the searcher from the join operation storage unit 103 and sends it to the join operation unit 107. Join operation unit 10
7 calculates the conformity μ (D _i ) of the data D _i with respect to the entire search question sentence by performing the operation of (Equation 9) as a process for “Katsu”.

[0042]

[Equation 9]

The search result output unit 108 outputs data as a search result in the descending order of the goodness of fit μ (D _i ) for the entire search question sentence. The search results when (Equation 2) (Equation 3) (Equation 4) are selected as the functions used for the processing of "Katsu" are shown in (Table 6) (Table 7) (Table 8), respectively.

[0044]

[Table 6]

[0045]

[Table 7]

[0046]

[Table 8]

As described above, according to the present invention, a join operation storage unit that stores at least two or more functions, a join operation selection unit that receives an input for selecting one function from the join operation storage unit, and a join operation By providing a join operation unit that performs a join operation between the fitness values using the function selected in the operation selection unit, the function used by the searcher to process "and" or "or" that he or she intended Since it is possible to select and search, it is possible to obtain a search result that matches the search request of the searcher.

In the embodiment of the invention of claim 1,
(Equation 2) (Equation 3) (Equation 4) (Equation 5) (Equation 6) (Equation 7) is used as the function used for the processing of "and" or "or", but other functions may be used. ..

Next, FIG. 2 shows a block diagram of an information retrieval apparatus which is an embodiment of the present invention of claim 2. In FIG. 2, the database 201 is a means for storing data to be searched,
The membership function storage unit 202 stores the membership function, the search question sentence input unit 203 receives the search question sentence composed of at least two fuzzy propositions m, and the fitness calculation unit 204 includes A means for extracting data from the database 201 and obtaining a goodness of fit of each of the m fuzzy propositions of the data by using the membership function of the membership function storage unit 202, and the variable combination calculation unit 205 has at least the goodness of fit of the m pieces of data. A means for performing a joint operation between the fitness values by using a function that receives one or more parameters and one output, and the parameter input unit 206 receives the input of the parameters of the function,
The parameter adjustment means and the search result output unit 207 are means for outputting the operation result of the function of the variable coupling operation unit 205 and the data as the search result.

As an example of the data stored in the database 201, the same person database (Table 1) as the conventional example is used. As an example of the membership function stored in the membership function storage unit 202, the same membership function as in the conventional example shown in FIG. 7 is used. The function * _p stored in the variable combination calculation unit 205 is shown in (Equation 10) and (Equation 11).

[0051]

[Equation 10]

[0052]

[Equation 11]

However, μ _k (k = 1, 2, 3, ...) Is an input of the function * _p , μ is an output, and p is a parameter for adjusting the function * _p . (Equation 10) is a function used for the processing of "Katsu"
(Equation 11) is a function used for the processing of “or”.
(Equation 10) and (Equation 11) are functions that perform a continuously changing operation by changing the value of the parameter p.

The operation of the second embodiment of the present invention will be described below.

The searcher inputs a search question text consisting of at least two or more fuzzy propositions into the search question text input section 203, and "kats" and "or" in the search question texts are input.
The value of the parameter p of the function * _p used for the processing of is input to the parameter input unit 206. M fuzzy propositions of the attribute value A _j (D _i ) of the attribute A _j of the data D _i from the input search question sentence
The process of _{obtaining the} fitness μ F _j (A _j (D _i )) with respect to F _{j and sending it} to the variable combination calculation unit 205 is similar to that of the embodiment of claim 1 described above, and thus detailed description thereof is omitted. .. On the other hand, the parameter input unit 206 displays the parameters input by the searcher.
Using the value of p, the variable combination operation unit 205 (Equation 10)
Adjust the function * _p of (Equation 11). Variable combination operation unit 20
5 is a process for “and” or “or”, as shown in (Equation 12), the goodness of fit μ _Fj (A of the attribute value A _j (D _i ) of the data D _i with respect to m fuzzy propositions F _j (A _j (D _i )) is input and a function * _p adjusted by the parameter p is performed, and the calculation result of this function is used to determine the goodness of fit μ of the data D _i with respect to the entire search question sentence.
(D _i ) to the search result output unit 207. The search result output unit 207 outputs data as search results in the descending order of goodness of fit μ (D _i ) for the entire search question sentence.

The operation of the present embodiment when a search is actually performed using the specific example of the search question sentence "search for a person who is young and has a high annual income" will be described below.

The search question text input unit 203 decomposes the search question text into two fuzzy propositions F ₁ = “age = young”, F ₂ = “annual income = high”, and “Katsu”. Fitness calculation unit 204
Retrieves all the person data D _i from the database 201 and uses the membership function μ _{AGE = YOUNG} of the membership function storage unit 202 to acquire the attribute value AGE of “age” of the data D _i.
Request (D _i) of the fuzzy proposition F ₁ = "Age = young" for adaptability _{μ AGE = YOUNG (AGE (D} i)). Similarly, the fitness μ _{SALARY = HIGH} (SALARY (D _i )) for the fuzzy proposition F ₂ = “annual income = high” of the attribute value SALARY (D _i ) of “annual income” is also obtained. This μ _{AGE = YOUNG} (AGE (D _i )) and μ _{SALARY = HIGH} (SALARY (D _i ))
And are sent to the variable combination calculation unit 205 together with “Katsu”. Μ _{AGE = YOUNG} (AGE (D _i )) and μ _{SALARY = HIGH} (SALA
RY (D _i )) is the same as (Table 3) and (Table 4) in the embodiment of claim 1, respectively. The variable combination calculation unit 205 uses "Katsu"
The calculation of (Equation 12) is performed as a process for (1) to obtain the goodness of fit μ (D _i ) of the data D _i with respect to the entire search question sentence.

[0058]

[Equation 12]

The search result output unit 207 outputs the data as the search result in the descending order of the goodness of fit μ (D _i ) for the entire search question sentence. As the value of the parameter p of the function * _p , p = 0.1,1,
The search results when 5 is input are shown in (Table 9), (Table 10) and (Table 11), respectively.

[0060]

[Table 9]

[0061]

[Table 10]

[0062]

[Table 11]

As described above, according to the present invention, "and"
A parameter input unit that accepts the input of the parameter of the function used for the processing of or “or” and adjusts the function, and a join operation unit that performs the join operation between the fitness using the function adjusted by the input parameter. By providing the searcher, the function used for processing "and" or "or" that the searcher intends can be set more finely than that in the embodiment of claim 1, and the searcher can search. You can get search results that match your requirements.

In the embodiment of claim 2, as a function used for the processing of "Katsu" and "or" (Equation 10)
Although (Equation 11) is used, another function may be used.

Next, FIG. 3 shows a block diagram of an information retrieval apparatus which is an embodiment of the present invention of claim 3. In FIG. 3, a database 301 is a means for storing data to be searched,
The membership function storage unit 302 stores the membership function, the search question sentence input unit 303 receives the search question sentence including m fuzzy propositions of at least two, and the teacher data input unit 304 includes A means for accepting the teacher data input of the fitness of the data in the database with respect to the input search question sentence, the fitness calculator 305 retrieves the data from the database 301 and uses the membership function of the membership function storage 302 to store the data. The variable negation operation unit 306, which is a means for obtaining the goodness of fit for the fuzzy proposition, and the goodness of fit obtained by the goodness of fit calculation unit 305 for the fuzzy proposition, when a negative qualifier is applied to the fuzzy proposition, Using a function that takes at least one or more parameters and one output The variable combination calculation unit 307 is a means for performing a constant calculation and changing the conformity, and the variable combination calculation unit 307 receives m conformity degrees and at least one parameter obtained by the conformity calculation unit 305 and the variable negation calculation unit 306 as inputs. The error calculation unit 308 is a means for performing a connection calculation between the fitness values using a function that obtains one output, and the error calculation unit 308 calculates the error between the calculation result calculated by the variable combination calculation unit 306 and the fitness level input to the teacher data input unit 304. The means for obtaining, the parameter adjusting unit 309 adjusts the value of the parameter of the function of the variable combination calculating unit 306 using the error obtained by the error calculating unit 307, and the search result output unit 310 is the function of the variable combining calculating unit 306. Is a means for outputting the calculation result and the data as the search result.

As an example of the data stored in the database 301, the same person database (Table 1) as the conventional example is used. As an example of the membership function stored in the membership function storage unit 302, FIG.
Use the membership function of. As the function * _p stored in the variable combination calculation unit, the same (Equation 10) and (Equation 11) as in the embodiment of claim 2 are used. The function stored in the variable negation operation unit 306 is shown in (Equation 12). However, μ is the input of the function, μ _N is the output, and p is the parameter for adjusting the function. (Equation 12) is a function that performs a continuously changing operation by changing the value of the parameter p.

The operation of the third embodiment of the present invention will be described below.

The searcher inputs into the search question text input unit 303 a search question text consisting of m fuzzy propositions of at least two. Data D _i from the entered search questions
The process of _{obtaining the} goodness of fit μ _Fj (A _j (D _i )) with respect to the fuzzy proposition F _j of the attribute value A _j (D _i ) of the attribute A _j of Since it is similar to the embodiment of
Detailed description is omitted. When the fuzzy proposition F _j of the search question sentence is negative “not”, the variable negation operation unit 306 _applies ( _Equation 13) to the goodness of fit μ F _j (A _j (D _i )) for the fuzzy proposition F _j . Negative operation using μ _Fj (A _j (D _i ))
To change.

[0069]

[Equation 13]

However, the parameter p of the negation operation function
, A certain initial value is set in advance. Variable combination performance
The arithmetic unit 307 performs processing for "Katsu" and "or".
Then, using (Equation 12) as in the embodiment of claim 2,
Data D_iAttribute value A_j(D_i) Fuzzy proposition F_jConformance to
μ_Fj(A_j(D_i)) For the function *_pAnd perform this function
The calculation result of is data D_iOf the search query
μ (D_i) Is sent to the search result output unit 310. However, Seki
number*_pParameter p is set to some initial value in advance
I'll do it. The search result output unit 310 displays the entire search question text.
Goodness of fit μ (D _i), In order of higher
Output as a search result.

If the searcher determines that the output search result does not match his own search request, the searcher determines that
With respect to n pieces of data D _k (1 ≦ k ≦ n) out of the output data, the fitness level μ ^T (D _k ) that matches one's sense is input to the teacher data input unit 304. The variable negation calculation unit 306, the variable combination calculation unit 307, the error calculation unit 308, and the parameter adjustment unit 309 are based on the steepest descent method (see Imano, Yamashita: Nonlinear Programming, Nikka Giren (1979)) described below. The processes from [Step 1] to [Step 6] are performed. [Step 1] The error calculation unit 308 is the variable combination calculation unit 3
The error (Equation 17) between the goodness-of-fit μ (D _k ) of the data D _k obtained in 06 and the goodness-of-fit μ ^T (D _k ) of the data D _k input to the teacher data input unit 304 is obtained and the parameter is calculated. It is sent to the adjustment unit 309.

[0072]

[Equation 17]

[Step 2] Parameter adjusting unit 309
Calculates (Equation 16), and (Equation 16) and the error calculator 30
Δp of (Equation 15) is calculated from (Equation 17) obtained in 7, and Δp is added to the parameter p to change p.

[0074]

[Equation 15]

[0075]

[Equation 16]

[Step 3] Parameter adjusting unit 309
Is the parameter p adjusted by the processing in [Step 3]
Using the function of the variable negation operation unit 306 and the variable combination operation
Function of part 307 *_pAdjust. [Step 4] Variable NOT operation unit 306 and variable combination operation unit
307 is the adjusted function *_pSearch by using
All data with degree input D_kSearch for whole question
Degree μ (D _k) Is calculated. [Step 5] Error calculator 308 and parameter adjuster 3
09 is the processing from [Step 1] to [Step 2]
Is the total data D for which the searcher has entered the goodness of fit._kAbout line
U [Step 6] Variable NOT operation unit 306 and variable combination operation unit
307, error calculation unit 308, and parameter adjustment unit 309
Performs the processing from [Step 1] to [Step 5],
This is performed until the absolute value of the error of (Equation 17) becomes the minimum.

After the processing from [step 1] to [step 6], the search result output unit 310 again searches the data together with the goodness of fit in the descending order of goodness of fit μ (D _i ) for the entire search question sentence. Output as.

The principle of the steepest descent method used in the above steps [Step 1] to [Step 6] will be described below with reference to FIG. 4 which is a schematic diagram.

The graphs (a) and (b) of FIG. 4 show the relationship between p and E, where the horizontal axis is the parameter p and the vertical axis is the squared error E. In the figure, p is the current value of the parameter p, and p ^* is the value of the parameter p that minimizes E. As can be seen from FIG. 4 (a), when the slope ∂E / ∂p of the graph is positive, E decreases if p is changed in the negative direction. Also, as can be seen from (b) in Fig. 4, when the slope of the graph ∂E / ∂p is negative, p
If E is changed in the positive direction, E decreases. That is, ∂E
If the parameter p is changed in the opposite direction of / ∂p, E will decrease. That is, the adjustment amount Δp of the parameter p that reduces E
Can be expressed by (Equation 15) using a constant α (> 0).
Therefore, E can be minimized by repeating the parameter adjustment of (Equation 15).

In this embodiment, ∂E / ∂p is the effect of μ (D _k ) on E ∂E / ∂μ (D _k ), and μ (D _k ) is a small change in p.
The effect on ∂μ (D _k ) / ∂p is expressed as (Equation 16). ∂E / ∂μ (D _k ) on the right side of (Equation 16) is (Equation 1
Since 4) is differentiated by μ (D _k ) and expressed as (Equation 17), ∂E / ∂p can be calculated using (Equation 16) and (Equation 17). Therefore, E can be minimized by repeatedly performing the parameter adjustment of (Equation 15) on the data D _k for which the searcher has input the goodness of fit.

[0081]

[Equation 14]

The operation of this embodiment will be described below when a search is actually performed using a specific example of the search question sentence "search for a person who is young and has a high annual income." Since the processing from the input of the search question text in the search question text input unit 303 to the output of the first search result by the search result output unit 309 is the same as that of the embodiment of claim 2, detailed description will be omitted. To do. However, in the present invention, a certain initial value is set in advance for the parameter p of the function * _p . In this example, the initial value of p is 1. The results of the first search are shown in (Table 12).

[0083]

[Table 12]

However, the searcher determines that the search result (Table 12) does not meet his own search request, and the teacher data input unit 304 is given a matching degree (Table 13) that matches his sense of the person data E and C. input.

[0085]

[Table 13]

Variable combination operation unit 306 and error operation unit 307
And the parameter adjusting unit 308 performs the above-mentioned [Step 1].
To [Step 6], the parameter p is adjusted. In this example, the value of parameter p is changed to 5. After adjusting the parameters, the search result output unit 309 again sets in order of the goodness of fit μ (D _i ) for the entire search query sentence,
The data is output as a search result together with the goodness of fit. The search results using the adjusted parameter p are shown in (Table 14).

[0087]

[Table 14]

As described above, according to the present invention, the teacher data input unit that receives the input of the fitness of the data in the database with respect to the search question sentence, the fitness calculated by the variable combination operation unit, and the teacher data input unit. The parameter that adjusts the parameter of the function used for the processing of "and", "or", and "not" by using the error calculator that calculates the error from the goodness of fit input to By providing the adjustment unit, the searcher can automatically set the function to be used for the processing of "Katsu", "or", and "No" that he or she intends, without setting the parameter by himself. Therefore, the search result matching the search request of the searcher can be obtained more easily than the invention of claim 2.

In the embodiment of the invention of claim 3,
Although the steepest descent method is used as the method of adjusting the parameter p of the parameter adjusting unit, other methods such as the conjugate direction method and the quasi-Newton method may be used. Also, as a function used for the processing of "Katsu", "or", and "not" (Equation 10)
(Equation 11) and (Equation 13) are used, but it goes without saying that other functions may be used.

Next, FIG. 5 shows a block diagram of an information retrieval apparatus which is an embodiment of the present invention of claim 4. In FIG. 5, a database 501 is means for storing data to be searched,
The membership function storage unit 502 is a means for storing a membership function, the search question sentence input unit 503 is a means for accepting an input of a search question sentence composed of at least two m fuzzy propositions, and the teacher data input unit 504 is A means for accepting the teacher data input of the fitness of the data in the database with respect to the input search query, the fitness calculator 505 retrieves the data from the database 501 and uses the membership function of the membership function storage 502 to store the data. The means for obtaining the goodness of fit for each of the m fuzzy propositions, the variable combination calculation unit 506 uses the function for obtaining one output with the m goodnesses of fit and at least one or more parameters as input. The error calculation unit 507, which performs the combination calculation, uses the fitness calculated by the variable combination calculation unit 506 and the teacher data. The means for obtaining an error from the conformity input to the input unit 504, the teacher data determination unit 508 determines whether or not the error obtained by the error calculation unit 507 satisfies a predetermined criterion, and the parameter adjustment unit 509 is the above-mentioned. A means for adjusting the parameter of the function of the variable combination calculation unit 506 by using the error when the error obtained by the error calculation unit satisfies the criterion, and the search result output unit 510 calculates the function of the variable combination calculation unit 506. It is a means for outputting a result and data as a search result.

As an example of the data stored in the database 501, the same person database (Table 1) as the conventional example is used. As an example of the membership function stored in the membership function storage unit 502, FIG.
Use the membership function of. As the function * _p stored in the variable combination calculation unit, the same (Equation 10) and (Equation 11) as in the embodiment of claim 2 are used.

The operation of the fourth embodiment of the present invention will be described below.

After the searcher inputs a search question sentence consisting of m fuzzy propositions of at least two into the search question sentence input unit 503, the search result output unit 510 outputs the first search result. Since the processing up to this is the same as that of the embodiment of claim 3, detailed description will be omitted.

If the searcher determines that the output search result does not match his own search request, the searcher determines that
With respect to n pieces of data D _k (1 ≦ k ≦ n) out of the output data, the fitness level μ ^T (D _k ) that matches one's sense is input to the teacher data input unit 504. The variable combination calculation unit 506, the error calculation unit 507, the teacher data determination unit 508, and the parameter adjustment unit 509 perform the following processes from [Step 1] to [Step 7] based on the steepest descent method described above. [Step 1] The error calculation unit 507 is the variable combination calculation unit 3
The error (Equation 17) between the fitness μ (D _k ) of the data D _k obtained in 06 and the fitness μ ^T (D _k ) of the data D _k input to the teacher data input unit 304 is obtained, and the teacher The data is sent to the data determination unit 508. [Step 2] The teacher data determination unit 508 sends the error to the parameter adjustment unit 509 only when the absolute value of the error calculated by the error calculation unit 507 is less than or equal to a certain threshold value. [Step 3] If an error is sent from the teacher data determination unit 508, the parameter adjustment unit 509 (Equation 16)
Is calculated, and Δp of (Equation 15) is calculated from (Equation 16) and (Equation 17) obtained by the error calculation unit 507, and Δp is added to the parameter p to change p. [Step 4] The parameter adjustment unit 509 changes the function * _p of the variable combination calculation unit 506 using the parameter p adjusted by the process of [Step 3]. [Step 5] The variable combination operation unit 506 uses the adjusted function * _p to obtain all the data D _{k for} which the searcher inputs the goodness of fit.
Calculate the goodness of fit μ (D _k ) for the entire search question sentence of. [Step 6] Error calculation unit 507 and teacher data determination unit 5
08 and the parameter adjusting unit 509 perform the processing from [Step 1] to [Step 5] for all data D _k for which the searcher has input the degree of matching. [Step 7] Variable combination calculation unit 506 and error calculation unit 50
7, a teacher data determination unit 508, and a parameter adjustment unit 509
Is the process from [Step 1] to [Step 6]
This is performed until the absolute value of the error of (Equation 17) becomes the minimum.

After the processing from [step 1] to [step 7], the search result output unit 510 again retrieves the data together with the goodness of fit in the descending order of goodness of fit μ (D _i ) for the entire search question sentence. Output as.

Actually, the operation of the present embodiment when a search is performed using a specific example of the search question sentence "search for a person who is young and has a high annual income" will be described below.

Since the processing from the input of the search question text to the search question text input unit 503 to the output of the first search result by the search result output unit 510 is the same as that of the third embodiment, detailed description will be given. The description is omitted. However, the initial value of p is 1 as in the third embodiment. The result of the first search is the same as that of the embodiment of claim 3 (Table 12). here,
The searcher determines that the search result (Table 12) does not meet his or her search request, and inputs the matching degree (Table 15) that matches his or her own feelings about the person data E and C to the teacher data input unit 504.

[0098]

[Table 15]

Variable combination calculation unit 306 and error calculation unit 307
And the parameter adjusting unit 308 performs the above-mentioned [Step 1].
To [Step 7], the parameter p is adjusted. Here, for the person data E, C, and A, the absolute values of the errors between the goodness of fit of the first search result (Table 12) and the goodness of fit entered by the searcher are 0.126, 0.17, and 0.4, respectively.
Is 15. The threshold in the teacher data determination unit 508 is 0.4
When the setting is made, the teacher data 508 determines that the matching degree given to the A by the searcher is an error and does not use it for adjusting the parameter p. Therefore, the value of the parameter p is adjusted to 5 as in the embodiment of the invention of claim 3. As the search result, the same as the embodiment of the invention of claim 3 (Table 14) is obtained.

As described above, according to the present invention, the teacher data input unit that receives the input of the fitness of the data in the database with respect to the search question sentence, the fitness calculated by the variable combination operation unit, and the teacher data input unit. The error calculated by the error calculation section, the teacher data determination section that determines whether the error calculated by the error calculation section satisfies a predetermined criterion, and the error calculated by the error calculation section. When the above criteria are satisfied, the searcher himself / herself can set the parameter by adjusting the parameter of the function used for the processing of “and” or “or” by using the error obtained by the error calculating section. Without setting, you can automatically set and search the function used for the processing of "and" and "or" that you intend, and the searcher can determine the degree of relevance to the data. When inputting as teacher data, incorrect teacher data can be mistakenly input and incorrect teacher data can be eliminated, so that the search that surely matches the search request of the searcher can be performed. The result can be obtained.

In the embodiment of claim 4, the error determining unit 508 determines that the absolute value of the error is less than or equal to the threshold value. However, the determination condition by another function of the error may be used. .. Also, the parameter p of the parameter adjustment unit
Although the steepest descent method is used as a method of adjusting the above, other methods may be used. In addition, as a function (Equation 10) (Equation 1
Although 1) is used, it goes without saying that other functions may be used.

Further, in the above-mentioned all inventions, each means is not limited to the parts as in the above-mentioned embodiment, but may be any means having those functions. For example, it may be realized by software using a computer or by using a dedicated hardware circuit having those functions.

[0103]

As described above, according to the present invention, a searcher can select and search a function intended by the searcher, and thus a search result matching a search request of the searcher can be obtained.

Further, according to the present invention, the searcher can finely set the function intended by the searcher to perform the search, so that the search result matching the search request of the searcher can be obtained.

Further, according to the present invention, the searcher can automatically create a function intended by the searcher without setting the parameter by himself or herself, so that the search that meets the search request of the searcher can be easily performed. The result can be obtained.

Further, according to the present invention, the searcher can automatically create and search a function intended by the searcher without setting the parameter by himself, and the searcher can teach the degree of conformity to the data. When inputting as data, even if incorrect teacher data is input by mistake, the error can be discriminated and the incorrect teacher data can be eliminated, so that the search result that surely matches the search request of the searcher can be obtained.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an embodiment of the present invention according to claim 1.

FIG. 2 is a block diagram showing a configuration of an embodiment of the present invention according to claim 2;

FIG. 3 is a block diagram showing the configuration of an embodiment of the present invention according to claim 3;

FIG. 4 is a schematic graph for explaining the principle of the steepest descent method in the embodiment of the present invention according to claim 3;

FIG. 5 is a block diagram showing the configuration of an embodiment of the present invention according to claim 4;

FIG. 6 is a block diagram showing a configuration of an example of a conventional information search device.

FIG. 7 is a graph showing a membership function used in a conventional information search device.

[Explanation of symbols]

101,201,301,501,601 Database 102,202,302,502,602 Membership function storage unit (means) 103 Join operation storage unit (means) 104 Join operation selecting unit (means) 105,203,303,503 , 603 Search question text input unit (means) 106,204,305,505,604 Fitness calculation unit (means) 107,605 Combined calculation unit (means) 108,207,310,510,606 Search result output unit (means) ) 205, 307, 506 Variable combination calculation unit (means) 206 Parameter input unit (means) 304, 504 Teacher data input unit (means) 306 Variable negation calculation unit (means) 308, 507 Error calculation unit (means) 309, 509 Parameter adjusting unit (means) 508 Teacher data judging unit (means)

Claims (5)

[Claims] 1. A database for storing data to be searched, a search question sentence input means for receiving an input of a search question sentence to the database, a membership function storing means for storing a membership function, and data from the database. Relevance calculation means for obtaining the goodness of fit of the data with respect to the search condition in the search question sentence by using the membership function, and combination operation storage means for storing at least two or more functions for performing the calculation between the goodnesses of fit. A combination operation selection means for selecting a function of the combination operation storage means by a searcher's input, and a function selected by the combination operation selection means with the compatibility degree of the compatibility calculation means as an input, And a search for outputting at least a calculation result of the combination calculation means as a search result. Information retrieval apparatus characterized by comprising a fruit output means. 2. A database for storing data to be searched, a search question sentence input means for receiving an input of a search question sentence to the database, a membership function storing means for storing a membership function, and data from the database. Relevance calculating means for obtaining the goodness of fit of the data with respect to the search condition in the search question sentence by using the membership function, and parameter input means for receiving the input of the parameter of the function for calculating the goodness of fit, A variable combination calculation means for performing a combination calculation between the matching degrees by a function created by the parameter with the goodness of fit of the compatibility calculation means as an input, and a search result output for outputting at least a calculation result of the variable combination calculation means as a search result. An information retrieval device comprising means. 3. A database for storing data to be searched, a search question sentence input means for receiving an input of a search question sentence to the database, a membership function storing means for storing a membership function, and data from the database. Retrieving, using the membership function, a fitness calculating means for obtaining the fitness of the data with respect to the search condition in the search question sentence; and a variable negating means for performing a function operation for obtaining the fitness with respect to the search condition representing negation, A variable combination calculation means for performing a function calculation between the goodnesses of fit by inputting the goodness of fit obtained by the goodness of fit calculation means and the variable negation calculation means, and an input of the goodness of fit of the data in the database with respect to the search question sentence The teacher data input means for receiving, the calculation result of the function of the variable coupling calculation means and the teacher data input means. Error calculation means for obtaining an error from the conformity input to the means, and parameter adjustment means for changing the functions of the variable negation calculation means and the variable combination calculation means using the error obtained by the error calculation means. And at least a search result output means for outputting a calculation result of the function of the variable coupling calculation means as a search result. 4. A database for storing data to be searched, a search question sentence input means for receiving an input of a search question sentence to the database, a membership function storing means for storing a membership function, and data from the database. Relevance calculation means for obtaining the goodness of fit of the data with respect to the search condition in the search question sentence of the data using the membership function, and between the goodness of fit with the goodness of fit obtained by the goodness of fit calculation means as an input. A variable combination calculation means for performing a function calculation, a teacher data input means for receiving an input of the fitness of the data in the database with respect to the search question sentence, a calculation result of a function of the variable combination calculation means, and the teacher data input means. The error calculation means for obtaining an error from the input conformance and the error calculated by the error calculation means Of teacher data determining means for determining whether or not satisfies a predetermined criterion, parameter adjusting means for changing the function using the error when the error satisfies the criterion, and at least the variable combination computing means. An information retrieval apparatus comprising: a retrieval result output means for outputting a calculation result of a function as a retrieval result. 5. The information search device according to claim 1, wherein the output of the search result means outputs the data as well as the calculation result of the function.