JPH03260769A - Fuzzy retrieving method - Google Patents

Abstract

PURPOSE:To obtain an answer to an ambiguous retrieving request by converting the attribute value included in a data base into a form corresponding to the ambiguous retrieving request by means of the general adaptable degree. CONSTITUTION:A simple fuzzy attribute, a composite fuzzy attribute, and a relative fuzzy attribute are generated based on the attribute values included in a data base. The adaptable degree is calculated for each attribute with selec tion of one of several conditions set between an OR and an AND. Then those attributes are combined with each other for calculation of the general adaptable degree, and the retrieving subjects are successively arrayed based on the calcu lated adaptable degree. Thus an answer is obtained to an ambiguous retrieving request with use of a conventional data base. Then the higher approximation is secured to the human judgement and the higher satisfaction is secured.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a fuzzy search method for obtaining answers to ambiguous search requests from conventionally used databases.

(Conventional technology) Conventionally known database management systems include:
Data models such as relational, tonal, hierarchical, functional, and object-oriented have been proposed. For example, when searching a relational database, data is expressed using several attributes, and set operations are performed according to conditions set by the user to obtain search results.

(Problem to be solved by the invention) However, since the conventional search described above is bound by the binary logic that is the basis of computers, such as 0 or 1, yes or no, it is difficult to enter ambiguous search requests. However, there was a problem that fuzzy search results could not be expected.

(Means for Solving the Problem) However, this invention allows attribute values in a conventional database to be
The above-mentioned conventional problems have been solved by making it possible to obtain an answer to an ambiguous search request by converting it to correspond to an ambiguous search request using the overall relevance.

That is, the present invention generates a simple fuzzy attribute, a compound fuzzy attribute, or a relational fuzzy attribute in order to respond to an ambiguous search request from attribute values in a database, and each of the attributes is set among several conditions between or and and. A fuzzy search method characterized by selecting one of the above, calculating a person suitability, then calculating a total suitability by combining each of the above-mentioned attributes, and sequentially arranging search objects based on the calculated value. It is. In addition, another invention generates a simple fuzzy attribute, a compound fuzzy attribute, or a relational fuzzy attribute, respectively, in order to respond to an ambiguous search request from attribute values in a database, and each of the attributes has several values between or and and. After selecting one of the conditions, calculating the degree of suitability for each, and assigning weights to the calculated values, the overall suitability is calculated by combining each of the attributes, and the search target is sequentially selected based on the calculated value. This is a fuzzy search method characterized by arranging.

Furthermore, another invention is characterized in that when calculating a simple fuzzy attribute, a complex fuzzy attribute, or a relational fuzzy attribute, a desired one is selected from a predetermined text or menu, and calculations are made based on this. This is what I did.

The simple fuzzy attribute mentioned above is an attribute that is the basis of a fuzzy search, and is used when the search request and the attribute value in the database correspond to each other on an l-to-l basis.

For example, when the condition [8GE=young J is required for the attribute value rAGE J in the database, the attribute value rAGE J is called a simple fuzzy attribute. In this case, let the fuzzy set ryoung J be the membership function μ,. By defining ung, the degree of conformity (grade) to the condition is calculated.

μ,. If u,,R are defined as shown in Fig. 1, then μ,. u
The goodness of fit is calculated as n((15)=0.7.

This method can be widely used for fuzzy searches.

The compound fuzzy attribute No. 1111 is an attribute with a higher level of abstraction that is formed by combining several attributes. This composite fuzzy attribute is derived by mapping from a plurality of attributes having a part-of relationship, and is mainly intended for numerical data.

For example, if we take a car as an example, the size of the car is the "overall length".
, "total width" and "total height", "total length"
, "overall width", and "overall height" are all the same as "the size of the car".
Since there is a part-of relationship, this is combined to generate a composite fuzzy attribute called "car size" (FIG. 2).

When deriving a composite fuzzy attribute, a reference attribute in a part-of relationship with numerical data,
From the relationship between the composite fuzzy attributes, the value of the composite fuzzy attribute increases as the value of the reference attribute (II'j) increases, and conversely, the value of the composite fuzzy attribute increases as the value of the reference attribute decreases. The former is called a plus factor, and the latter is called a minus factor, and reference attributes can be divided into these two types.

For example, according to Figure 3, when the attributes ``capacity,'' ``interior length,'' 1-interior width, and ``interior height'' are used as reference attributes for the complex fuzzy attribute ``car comfort,'' Among them, “indoor manager”;
``Indoor width J'' and ``indoor height'' are positive factors;
"capacity" can be considered a negative factor. Furthermore, each attribute in FIG. 2 is a plus factor.

Next, an algorithm for generating composite fuzzy attributes will be explained.

First, the generated composite fuzzy attribute is
n) The reference attributes are a, (j = l, 21., m). All data in the database can be expressed as point data in m-dimensional Euclidean space from a to a. The composite fuzzy attribute is calculated by the user-defined evaluation function F as follows:
i = 1. 2. , , n) on the evaluation axis. FIG. 4 shows an example of two attributes (al, a2).

For example, assuming that the weight between each attribute is constant, the reference attributes a+, a2..., a are the attribute values in the database that are proportionally converted between [0, 1], and the evaluation axis is al =az",,,"a+++ was used.

However, if al is a negative factor, al”
It was set as I at. Moreover, as the evaluation function F, the following one was used.

■ min (al, a2.,, -, a+++)■
]-m/(Σ (] / (I aJ )))■
maX (al , aa , , , , am ) The above middle ■ and ■ are a in the usual fuzzy theory, respectively.
It is equivalent to extending nd and or to m dimensions. In addition, the above ■, ■, ■, ■, ■ are mean operators, ■ is a harmonic mean, ■ is a geometric mean, ■ is an arithmetic mean, ■ is a dual geometric mean, and ■ is a dual harmonic mean with m dimensions. It has the same value as expanded to . A two-dimensional example of these evaluation functions is shown in FIG.

The magnitude relationship of the fliI evaluation functions is ■≦■≦■≦■≦■≦■≦■ It is necessary to use these evaluation functions properly depending on the relationship between each reference attribute.

Next, let's consider the evaluation function in a little more detail. Fifth
As you can see from the diagram, ■, ■, ■ are and or a.
Close to nd, ■ is an intermediate value between and and or, ■, ■, ■
takes a value close to 0 or or. Therefore, the operators are divided into three groups and these are called the and system, the arithmetic system, and the or system. Examples of these two-dimensional calculations are shown in Table 1.

Table 1 Example of Calculation of Evaluation Algebra When the difference in the values of the reference attributes is small, as in case 1 above, the values will be the same regardless of the evaluation function. Conversely, when the difference is large, it indicates a significant difference. The following can be said as the characteristics of operators.

(1) Because it is affected by small values in the and system reference attributes,
The calculated value is likely to be a small value.

(2) Arithmetic system It is likely to take a value near the center.

(3) Because it is affected by large values in the or system reference attributes,
It is likely to take a large value.

From this, each operator is affected by one distinct value among the reference attributes. The degree of this influence is an
In the d system, the geometric mean is the smallest, and min is the largest (the or system is its dual).

Therefore, the evaluation function for creating fuzzy attributes is selected depending on the relationship between reference attributes. Specifically, when the reference attributes Δ3 and A2 are AND systems, if the fuzzy attribute must strictly satisfy both A+ and A2, use min, and if it is okay to satisfy only one, use geometric mean. , and so on.

However, when the number of reference attributes increases, the values of the calculated fuzzy attributes are generally small (less than 0.5) in the and system, and near the center (0.5 or less) in the arithmetic system.
, 5), and the OR system is likely to take a large value (more than 0°5). This means that one of the values is smaller (
This is because the possibility of When increasing the number of reference attributes, you need to be careful about this point.

Next, to give an example of a composite fuzzy attribute,
Attributes in the database include “height” and “weight”
If you have information such as "chest circumference", "visual acuity", "blood pressure", "sitting height", etc. (such as a database in a school health room), if you want to search this database using the condition "Who is fat?", you can use the complex fuzzy attribute " Create a body type. There are three possible reference attributes: height, body weight, and chest circumference.Height is considered a negative factor because it is considered that the larger it is, the less fat it is.Here, the evaluation function is determined. Here, we intuitively assume that it is an and system, and create an attribute using the geometric mean of ■.

Also, the relational fuzzy attribute sets the muddy layer to 1s.
This attribute is generated by giving the degree of similarity between specialized hierarchies created when classified according to the -a relationship, and is mainly intended for non-numeric data. For example, when searching for a car, the attribute car type is set to [ko3 door-11B] based on the 1s-a relationship.
J, r 4 door-3EDANJ, "4 door-
It^RDT Kano”, r 2 door-COUPJ, r
lBOXJ, rRVJ, rVfAGON J (car name is trademark Teal) can be classified as a subset of several specialized classes such as Germany. By giving mutual similarities between these subsets, a relational fuzzy attribute called "car type" can be created (Figure 6).

The relational fuzzy attribute is given by the similarity of a specialized subset from the 1s-a relation.

Let the relational fuzzy attributes be y+, + (i, j = l, 2. , , ,
n) A subset specialized in the is-a relation is defined as bl(
+ = I, 2. ,,,,n).

Here, if the user-defined evaluation function is G, then the relational fuzzy attribute y+i is 3'1J=G (bl
, ba ) (i, j = 1.2. , , , n ). The evaluation function G here indicates the degree of similarity between b and [), and is expressed by a fuzzy relationship.

As an example of relational fuzzy attributes, Table 2 shows the fuzzy relationships for "car type" shown in Figure 6.
It was shown to. In this case, the search request [3door HIIJ
The goodness of fit for "a door song" is 1.0, r4
door SD J is 0.2, r IBOXJ is 0, etc.

Table 2 Relationship Fuzzy Attribute "Type" Above, we proposed a method for generating data from crisp sets to fuzzy sets. This will calculate the degree of suitability for each condition set by the user. However, human needs are extremely complex and are made up of multiple conditions. Therefore, a method for processing more complicated condition settings will be described.

In conventional fuzzy databases, when multiple conditions are set, and indicates the minimum fitness value for each attribute, and
The maximum value of or was taken as the overall evaluation. but,
In this method, the final result for the set condition is highly dependent on the degree of suitability for one attribute. for example,
The goodness of fit for the two conditions (R3°R2) is DI (+
, 0.0.2) 1) When 2 (0, 2, 0, 3), the degree of conformance to the condition (111 and R2) becomes the same value 0.2 for both Dl and 02. This is clearly counter to human intuition. Therefore, an
It is necessary to bring it closer to human thinking by giving d or Or a more average meaning.

In place of the conventional and and or, the present invention uses an average operator, and by using the average operator, conditions can be set more flexibly.

For example, the replacement with a n cl, or is as follows.

and -1 ■min (normal and) ■harmonic mean ■geometric mean and, or intermediate value → ■arithmetic mean or -■
Dual geometric mean ■ Dual harmonic mean ■ max (ordinary or) For example, for conditions R,, R2, the goodness of fit of Ai (aa2) is evaluated for conditions (R) and Rz). Shown in 3.

Table 3 Calculation of fitness (and) Looking at the above-mentioned Example 3, Δ is always in first place, but A1 has two ranks due to the geometric mean. This is because the value of suitability becomes larger by using the average operator, but this tendency is stronger when the difference in suitability for each condition is large. Furthermore, when there are many conditions, there is a very high possibility that the degree of suitability will be influenced by one. To take an extreme example, the fitness for each of the five conditions connected by a n (] is (+, 0.1.0.1.0.1.
0.0.1), the overall fitness is 0.1 for min, 0.36 for harmonic mean, and 0.63 for geometric mean, which shows a large difference. As we will see, even if two conditions are not met, if the other fitness values are high and you want to give priority to them, you can use the geometric mean or the harmonic mean depending on the degree of suitability. Similarly, o
The same can be said for r.

In this way, by using the average operator, it becomes possible to create more ambiguous conditions. Users can freely
By selecting n, d or or, it helps to effectively eliminate individual differences in human sensitivity.

In conventional fuzzy databases, it is not possible to weight certain attributes when setting conditions.

However, this invention is a method that allows weighting between attributes for setting conditions. This method can be easily extended by using the average operator described above.

Let the generated overall fitness be X, and the fitness of each referenced attribute be a, (i = 1.2., ,
.

n), and the weight for a is W+ (+ = 1.
2°, , n).

Table 4 Example of goodness of fit when weighting is taken into account.

As an example, FIG. 7 shows an example in which the fitness value a I + 82 is set for and using the geometric mean of ■, and a weight is attached to it.

As an example, for the case where each degree of fitness is (0, 1, 0, 5, 1, 0), the table below shows the case where the geometric mean is used and weights are added.
4.

From Cloth-4, it is confirmed that by applying weight, that attribute is given importance.

By assigning weights between attributes, it is possible to clarify which attributes are important in the search conditions, and more complex conditions can be set, so the answer is likely to be more satisfying.

(Example) Next, an example in which the present invention is applied to a specific article will be described.

First, the attributes of the database were set as follows.

Car name 11) Engine specs 2) Overall length
12) Displacement 3) Overall height 13) Maximum power 4) Overall width +41 Maximum torque 5) Wheelbase 15) 10 Dry fuel consumption 6) Weight +6
)ΔT/M T7) Interior length 17) Car type 8) Interior width 18) Price 9) Interior height 10) Capacity These are all crisp data, and it is necessary to create fuzzy attributes to perform fuzzy searches. .

Therefore, we will specifically explain the case of the prototype system ``Car Selection Database J'' as an example.

First, as fuzzy attributes for search, "type",
"Size", "Power", "Livability", "Economy", "
Select 7 options: ``Price'' and ``Purpose of Use.'' The types of these fuzzy attributes and reference attributes were determined as follows.

■ Type (related fuzzy attributes)...Type of car■ Size (compound fuzzy attributes)...Overall length, width, overall height■ Power (compound fuzzy attributes)...Maximum output/maximum torque■ Comfort (compound fuzzy attributes)... Attributes)...interior length, interior width, interior height, number of occupants■ Economy (compound fuzzy attributes)...displacement, fuel efficiency■ Price (simple fuzzy attributes)...new vehicle price■ Purpose of use (compound fuzzy attributes) Among the selected reference attributes, numerical attributes are [
0,1] conversion was performed. This is to operate multiple attributes with different units and ranges in one space. The method used in the actual system is to give the minimum value Δmin and maximum value 1aΔmax to the reference attribute 8, and
5. Conversion was performed as in the fuzzy relationship example of "type".

Next, a method for specifically creating seven fuzzy attributes will be described.

■Shape When classifying the types of automobiles stored in the database, “
l1atch-back”, 5edan “1l
ard-top”, “Coupe, Recr
ational-viicle “Wagon,”
"1Box" can be classified into roughly 7 types (
(Car name is a trademark). These were classified according to the 1s-a relationship, and were considered to be relational fuzzy attributes, and the fuzzy relationships were given subjectively. The created fuzzy relationships are shown in Table 5. For example, if you select "Coupe" as a condition, you may be willing to buy a "1la Lch-Back" car if it's good. However, “Co
It is unlikely that a person who chooses ``UPE'' will be interested in buying ``IBOX.'' From this perspective, the fuzzy relationship is created.

From this table, for example, for “Coupe”, “l1ar
dtop” has a fitness of 0.7, “l1aLch-bac
The goodness of fit is determined such that the goodness of fit of k'' is 0.5.

■ Size (Size) Reference attributes, total length 1□enl, width Wi
Convert d+, total height 1lei+ to [0, 1] respectively, and LW
, ,I+, are created. To do this, the maximum and minimum values of each attribute are multiplied. This may be the maximum and minimum values in the actual database, but we made it possible to determine them subjectively. l, W,, I+, were derived from the following equation.

Considering that if any one of these is extremely small, the impression of the car is quite small, we decided to use the harmonic mean as the evaluation function and set "5ize" S as shown in the following formula.

As described above, there are seven evaluation functions F for creating a composite fuzzy attribute: the a and d system, the arithmetic mean, and the or system. Used when it cannot be said that the conditions are completely met.

OR system: Used when it can be said that if any one of the values is satisfied, it is completely satisfied regardless of the other values.

Arithmetic mean: Take a value between a and d and or, and
0 is used when it cannot be clearly stated.

In this system, the maximum output and maximum torque, which are the reference attributes of the total length, width, and height, are
Similar to "size", it was converted to [0, 1] and the arithmetic average was taken.It is thought that there is a significant correlation between maximum output and maximum torque, and in this case, it is generally true that both the and system and the or system Since this is not possible, we used the arithmetic mean to save time.

The value of the fuzzy attribute "Power" P rl is derived from the maximum output Pl and the maximum torque 'I' converted to [0, 1] as follows.

P, +1 P rl2 ■ Comfort This depends on the interior space and the number of passengers, so it is necessary to take these into consideration (in reality, it is also affected by the equipment and seat type, etc., but this time (I haven't created a database that detailed, so I haven't taken that into account.) Although it is easy to express the size of the room, it is impossible to predict the number of passengers, so I decided to calculate the number of passengers for the time being. The larger the room, the better. (Plus factor)
The concept of livability was derived from the fact that the more people are on board, the worse the livability becomes (minus factor). Indoor length (R1, en), indoor width (Wid), indoor height (
R11ei) are converted to [0,]] respectively, and R162, R
111,, R1+, closed. Also, since the capacity (Meml) is a negative factor, it is converted to [0°1] and O and 1 are inverted. At this time, the value M.

In this case, livability is considered to be a correlation between indoor volume and capacity. Specifically, no matter how long or wide the room is, if the ceiling is low, it will not be comfortable to live in, and no matter how small the capacity, if it is too small, it will not be comfortable to live in, so it is a strict AND system. The harmonic mean was used to derive the livability. Habitability"
Comfort"Cut is expressed as follows.

CL” l/M, + l/RL, + I/RW, +
I/R11 ■Economy It is a problem of maintenance costs. A car with good fuel efficiency is of course more economical, but the larger the engine displacement, the more taxes you have to pay, so we thought of it from the perspective of the negative factor. The [0,1] converted value of the 10-day road fuel consumption was Md, and the value obtained by converting the displacement to [0°] and inverting O and 1 was C7. Economy
E y+ was considered to be a weak and, and was derived as follows.

Ey+= (c+ *Md,) I/2■ Price (Pr
ice ) is a simple fuzzy attribute with a price of [0,]]
The converted product was used as is.

■ Purpose of use (Purpose) Make a selection according to the purpose of use. “5portsuse
, -0utdoor use, -Fami
ly-use.

We have prepared four “Town uses.” Each of these derives its fitness from a complex condition. In other words, from 0 to 0
Instead of inputting compound conditions of terms, you can make selections that suit all four purposes at once by manually inputting this one item (however,
There is a greater degree of subjectivity at play here). These are: 5ports-use: (type = (Coupe o
r 1lard-top) ) and (output = large) OuLdoor-use: (type = (R
ecreationalviicle or tlat
ch-back)) and (Family Iy-use): (type = (Sedan)
or 1lard-top or IBox or Wa
gon) ) and (livability = good) Town-use: (size = small) an
d (economic efficiency = good). In order to provide ambiguity for a and d, the geometric mean was used.

We have established the rules for creating fuzzy attributes as described above. In reality, the rules are set at the system configuration stage 31, and it is impossible to easily set the rules while the system is running, but by changing these, you can create more personal attributes. You can also do that.

Next, a method for inputting user requests will be described. All conditions are handled manually by selecting menu items.

[Car Selection Database] When the system is started, the main menu screen as shown in FIG. 8 is displayed first, so select the attribute for which you want to input the condition by clicking on the icon (or button).

When you select and click an attribute from the main menu, a window will open prompting you to enter the respective conditions ((submenu). For example, if you select [ltl J, the 9th
A window like the one shown will open, and you can click on the model of car you want to search for. If you click "Absolute" at the bottom left, the fitness of types other than the selected one becomes O. Also, “Size (S
ize) J conditions are entered by selecting one from "large", "medium", and "small" (first
Figure 0).

In Figure 10, "large", "medium", and "small" mean the first
The membership function is set as shown in diagram I. 10th
Regarding the functions of the “considerably” and “somewhat” buttons at the bottom left of the diagram,
Search for details.

The above-mentioned “Power J”, “Livability (com)
forL)” and “Economy”.
J, "Price" Inputting conditions such as J is done through three-step selection similar to the size, and the membership function is also the same as above.However, like price, it is generally recognized as a numerical value. The attributes that are
Rather than expressions such as “high” or “low,” the actual numerical value “l”
It is easier to understand if it is ``less than 00,000 yen''.

However, the current system places emphasis on vague searches, and uses a display like the one shown in Figure 12 to make it easy to see how much is considered "expensive."

Such an expression is effective for a simple fuzzy attribute such as "price", but is not suitable for a complex fuzzy attribute such as "livability".

Finally, the Purpose J is entered, as in the case of the Shape J, by selecting the icon shown in Figure 13.

After entering these conditions, click “RE'rRI” in Figure 8.
When the user clicks on the "EVING" button, the weight input screen shown in FIG. 14 is displayed, and weights between attributes can be input in three stages.

The weight is expressed as a ratio, and in Figure 14, “type”: “size”: “output”: “livability”: “U resistance”: “price”: “purpose of use” = I:
2:I:3:]:3:2 has been set.

The method of weighting has been described above, but in the above example, when the goodness of fit is calculated using the geometric mean, the goodness of fit of each attribute is a−(k=], 2°, , 7). Then, in the case of the weights shown in Figure 14, the overall fitness X is (a + * a 22 * a 3 * a' * a6
*ae3*a72) ”+3.

The conditions input in this invention are all AND conditions. The overall suitability is determined by calculating the suitability of each fuzzy attribute for the conditions and using the geometric mean, which is a weak AND. When the degree of suitability of each attribute to the condition is a, if there are 5 input conditions, then
The overall suitability X is “” (a l* a2 * ,,,*
a, ) Izk. Here, when "considerably" was added to the condition, the degree of fitness of the fuzzy trivico, -to was squared, and when "somewhat" was added to the condition, it was multiplied by square. The former means stricter conditions, and the latter means looser conditions. Also, when a is multiplied by the weight Wl, − (a, “I * a 2”2 books・,・*akw
k) l′1wl *wll+ − “wkl.

The search results were determined to have an overall degree of suitability for the conditions greater than 0, and were sorted and output in descending order.

Here, a certain search example is shown. The data prepared includes 51 randomly selected types of automobiles, and the model and grade of 1J were also selected arbitrarily.

[Search example] Which car has a hard top, is medium in size, and has a lot of power? Search request Type = hard top size = medium output = high weight is I: I: l When searching, the retrieved data is 7 data, and its compatibility with the car model is 1st Toyota Subra 2600 0. 8881st Toyota Seriki 2000 GTRO, 8883rd Nissan Fairlady 2
3000 0.8164th Mazda RX -70T-
1."T'D O, 8125th place Toyota MR2200
00,6936th Toyota Soarer 2000 0
．． 6307th place Nissan Skyline 2000 0.5
It became like 88.

Table 6 shows the degree of conformance to each fuzzy attribute.

Table 6: Conformity of each attribute to conditions Next, let's make a request that emphasizes power. Weight I: ]: 3,
The ranking is shown below.

1st place Toyota Subra 2600 0.9311st place Toyota Seriki 2000 GTRO, 9313rd place 1st production Fairlady 23000 0.88504th place Toyota M
R220000, 80305th place Mazda RX-7GT-
L, TI) 0.7926th place Toyota Soarer 2000
0.7217th place 1st generation Skyline 2000
By focusing on 0.682 power, the 4th place RX7 and the 5th place Mn2 are reversed. Of these two cars,
Considering the goodness of fit for each condition, r'? X
-7 Size = medium → 1.000, output = high → 0.7
64 MF< 2 Size = Medium - ○, 475, Output = High → 1. It is called ODD. In other words, M R2 is N]o
Since the degree of conformity to the condition wcr =llighJ is large, the overall degree of conformity increases significantly when weight is applied to power.

(Effects of the Invention) According to the present invention, it is possible to obtain answers to ambiguous search requests using conventionally used databases, so it is possible to get closer to human judgment and improve satisfaction. It has the effect of

Also, by adding weights between attributes,
This has the effect of making it possible to more closely match the searcher's intentions according to the search purpose.

[Brief explanation of drawings]

Figure 1 is an example graph of a simple fuzzy attribute of the present invention, Figure 2 is a diagram of a composite fuzzy attribute, Figure 3 is a diagram of plus and minus factors, and Figure 4 is a diagram of the generation of a composite fuzzy attribute. Figure 5 is a graph showing the average operator, Figure 6 is a three-dimensional diagram of relational fuzzy attributes, Figure 7 is a graph of fitness considering smallness, and Figure 8 is a graph showing the average operator.
The figures are the same (the main menu screen of the embodiment, Figure 9 is the same) the submenu diagram of the embodiment, Figure 1Q is the same (the block diagram of other submenus of the embodiment, Figure 11 is the same) The graph of the fuzzy attribute representing the size of , Figure 12 is the same.
The figure shows the same <type selection menu of the embodiment, and FIG. 14 shows the weight determination diagram of the same <embodiment. Great′

Claims (1)

[Scope of Claims] 1. In order to respond to ambiguous search requests from attribute values in a database, simple fuzzy attributes, compound fuzzy attributes, or relational fuzzy attributes are respectively generated, and each of the attributes has several values between or and and. A fuzzy system characterized by selecting one of the conditions, calculating the degree of suitability for each, then combining each of the above-mentioned attributes to calculate a total suitability, and sequentially arranging search objects based on the calculated value. Search Method 2 In order to respond to ambiguous search requests from attribute values in the database, a simple fuzzy attribute, a compound fuzzy attribute, or a relational fuzzy attribute is generated, and each of the attributes is one of several conditions between or and and. After selecting one of the attributes, calculating the degree of relevance for each, assigning weights to the calculated values, calculating the overall degree of relevance by combining each of the attributes, and sequentially arranging the search objects based on the calculated values. Featured fuzzy search method 3: When calculating simple fuzzy attributes, complex fuzzy attributes, or relational fuzzy attributes, a desired one is selected from a predetermined text or menu, and calculations are made based on this. Fuzzy search method described in Section 1 or 2