JPH086795A - Grouping method and grouping device - Google Patents

Abstract

(57) [Abstract] [Purpose] A wide variety of grouping is possible without using a large storage capacity. The learning function makes use of past know-how to easily modify the grouping template. [Structure] An input device 300 for inputting a certain number of individuals and the number of people in a group, creating a contingency table by paying attention to two individual attributes, and setting the groups of interest sequentially to set a target group and others Processing device 100 that calculates a discriminant attribute value related to the number of assigned units of the group and the combination of the discriminant attribute values is used as a condition part, and the discriminant result of whether to allocate the unit to the target group or move the target group is displayed. Knowledge base 20 that stores rules to be executed
0 and an output device 400. The processing device 100 is
Based on the calculated discriminant attribute value, the knowledge base 200 is searched and inference processing is performed to recalculate the discriminant attribute value.
Units are distributed according to the inference result to create a grouping template, and individual data is assigned to this template for grouping.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a group of a certain number of individuals to be grouped (the data of each individual is a plurality of attributes (hereinafter referred to as individual attributes) indicating the characteristics of the individual. A grouping method in which a grouping template is created based on the number of groups with a specified number of members, and a grouping method that performs grouping based on this template, and The present invention relates to a grouping device that executes this.

Specifically, for example, when considering a group division problem in in-house training, etc., when dividing a large number of employees (individuals) into appropriate groups, the members of the created groups have equal individual attributes. There is
It is desirable for conducting an equal evaluation as much as possible. That is, when all groups are viewed based on the individual attributes (affiliation department, years of service, place of work, age, etc.) of the database relating to employees, it is desirable to perform grouping so that the individual attributes are not biased among the groups.

On the other hand, in some cases, grouping is required so that individuals having a specific individual attribute are concentrated in a specific group. This is a case where individuals having similar individual attributes are divided into the same group, and it is fully conceivable to realize this by cluster analysis of a conventional method. The present invention particularly relates to the former problem in the grouping problem as described above, and also relates to a grouping method and a grouping device applicable to the latter problem.

[0004]

2. Description of the Related Art Conventionally, this type of grouping is performed manually, but at this time, consideration is given so that bias and concentration of individual attributes do not occur between groups. For example, when a group consisting of a certain number of individuals is divided into a plurality of groups each having a specified number of people, grouping is performed according to the following procedure.

Individual objects to be grouped are made into cards one by one. Based on the individual attributes to be most considered and the individual attributes with the highest priority (for example, the department to which they belong), all the cards are assigned to the individual attribute groups. If necessary, subdivide the lumps created by using the individual attributes to be considered next. Cards are sequentially picked up from a group formed by allocating the cards to form a predetermined number of groups, and the bias of individual attributes among the groups is adjusted.

[0006]

In the case of such manual work, there is a high possibility that human error such as oversight or misunderstanding will occur. A cluster analysis method is used as a method of collecting individuals with similar individual attributes and dividing them into several groups (clusters). It can be said that it is not suitable for planning problems such as grouping in which groups are divided so that no bias and concentration occur.

This kind of problem is generally called an assignment problem or a class organization problem, and it is desirable that the user can freely and easily change which individual attribute is selected and in what order as a priority. However, there is a problem that it is difficult to formulate all complicated constraint conditions in advance. In other words, this kind of problem is simplified to the i-th person (i
It can be expressed as a problem of whether or not the (th individual) is assigned to the j-th group, and can be formulated by introducing the following i × j variables.

Maximization: Σ _{i = 1, m} Σ _{j = 1, n} c _ij x _ij Constraint condition: Σ _{i = 1, m} x _ij = 1 (i = 1, 2, ...).
......, m), Σ _{j = 1, n} x _ij = b _j (j = 1,2, ...
..., n) x _ij ε {0, 1} (0: do not assign the i-th person to the j-th group, 1: assign the i-th person to the j-th group, b _j : the maximum number of the j-th group)

This formulation is such that, when c _ij is the degree of suitability or preference when the i-th person is assigned to the j-th group,
The sum total is maximized under the above constraint condition. However, it is up to the user to decide what kind of evaluation function c _ij is, and according to a change in conditions such as averaging the goodness of fit of the groups or giving the individual attributes a bias between the groups, Each time, the trouble of having to recreate the evaluation function and the constraint conditions occurs. In addition, since the individual is a human being, it cannot be divided, and the number of individuals does not have a negative value, so it is a discrete type problem, and the simplex method (simplex method) etc. cannot be used as it is. Problems also occur.

Further, as another method, when a group of a certain number of individuals to be grouped and a number of groups of a designated capacity are input, group candidates are generated from all combinations, and from among them, A method of selecting the optimum grouping result is also possible. However, this method has a practical problem because the amount of data of the group candidates generated becomes enormous. For example, 6 individuals with 30 individuals, 7 individuals,
The number of combinations divided into 5, 8, and 4 people is ₃₀ C ₆ × ₂₄ C ₇
× ₁₇ C ₅ × ₁₂ C ₈ ≈ 6 × 10 ¹⁶ patterns, which is practically impossible to calculate.

A method is conceivable in which various grouping templates are prepared in advance according to the total number of individuals that will form a group, and a grouping template having the same number of individuals is searched and used as necessary. However, in addition to the huge amount of data to be stored, there is concern about whether to deal with various practical grouping problems. Furthermore, in the above-mentioned various methods, know-how at the time of past grouping, for example, priority of individual attributes at the time of performing grouping, distribution of individual attributes of each group, etc. are used,
It is considered difficult to automatically perform group division similar to the grouping that users create on a daily basis.

[0012] In addition, in the above-mentioned conventional technique, it is difficult to formulate a concrete execution method and procedure of grouping and complicated constraint conditions in advance by using mathematical formulas, etc. It seems that it is extremely difficult for the user to correct and change the conditions. Furthermore, in order to reuse the experience and know-how gained through trial and error at the time of past grouping at the time of another grouping, it is necessary to record it in some form such as a document, but such recording work requires a lot of labor. is necessary.

Even if discriminant conditions based on experience and know-how are created as a knowledge base, it is necessary to organize the relationship between complicated constraint conditions and their grouping methods in all cases. It is a heavy burden. In particular, the burden on an expert to sort out complicated constraint conditions and grouping methods first before others was extremely heavy.

The present invention has been made to solve the above-mentioned various problems, and an object thereof is to be able to automatically execute a wide variety of groupings without requiring a large amount of arithmetic processing and storage capacity. Another object of the present invention is to provide a grouping method and a grouping device for executing the grouping method, in which the know-how at the time of past grouping can be effectively utilized by the learning function and the grouping template can be easily corrected.

Incidentally, the inventor previously mentioned that the Japanese Patent Application Laid-Open No. 6-6806.
As No. 6, we proposed a discriminant condition generation method that can be applied to plant failure diagnosis systems. This application is, in various equipment, when some observation data is given, whether the observation data is in a specific range, or whether the discrete data belongs to a specific set ,
The judgment condition for judging whether or not the equipment is suitable for a certain purpose is defined by a case of the storage device in which case data including the attribute value and the judgment result based on the attribute value and the combination of the attribute value are set. It is a method of generating based on the contents. In the grouping device of the present invention, the learning process disclosed in the discrimination condition generating method is used as necessary.

[0016]

In order to solve the above-mentioned object, the grouping method of the first invention divides a group of a certain number of individuals to be grouped into a plurality of groups with a specified number of persons. In the grouping method, attention is paid to two individual attributes out of a large number of individual attributes showing the characteristics of individuals, and the frequency of 1 unit of the individuals corresponding to the combination of the contents of these individual attributes is regarded as the unit which is the minimum unit of grouping. Create the expressed contingency table, set the groups to be focused on in sequence for each group with the specified number of people, and consider the discrimination attributes regarding the number of allocated units and the number of remaining groups in the focused group and other groups. , Assign the unit to the attention group,
Alternatively, it is determined whether the target group should be moved to the next group, and by allocating the units based on this determination result, a grouping template expressing the combination of individual attributes forming each group is created. Using the combination of conditions relating to the discriminant attribute considered at the time of creating and the discrimination result, a discrimination process is executed under a newly given certain number of individuals, the capacity of each group and the number of groups, A grouping model is formed by allocating a plurality of groups of a specified number of people, and each individual is assigned to each group based on this grouping model to perform grouping.

As a result, when the number of groups in which the number of individuals and the number of persons are designated is presented, the number of individuals and the number of groups can be reduced without storing a huge amount of grouping templates considering all combinations of individuals. It is flexible enough to handle all combinations of numbers, and it is possible to create an optimum grouping template and perform grouping under any conditions.

The grouping device according to the second aspect of the present invention includes an input device for inputting a group of a certain number of individuals to be grouped and a number of groups of an arbitrary number of persons, and a large number of individual attributes showing characteristics of the individual. Focusing on two individual attributes from the inside, a contingency table was created in which one unit of the frequency of individuals corresponding to the combination of the contents of these individual attributes was expressed as a unit that is the minimum unit of grouping, and the number of people was specified. For each group, set the group of interest sequentially, the processing device that calculates the discriminant attribute value regarding the number of allocated units and the number of remaining groups in the target group and other groups, and the combination of the discriminant attribute value as the condition part, Assign the unit in the attention group, or
A knowledge base that stores a rule whose execution unit is a result of determining whether or not to move a target group to the next group, wherein the processing device uses a knowledge base that has a condition unit that satisfies the calculated determination attribute value. And perform the inference process with the discrimination result described in the execution part of the rule as the inference result, recalculate the discrimination attribute value each time this process is executed, and allocate units according to the inference result. A grouping device for creating a grouping template by allocating individual data to the grouping model, and having an output device for displaying the grouping model created by the processing device and the grouping result.

According to the present invention, when the group of a certain number of individuals to be grouped and the number of groups of an arbitrary number of persons are input, the discrimination attribute values necessary for the processing device to evaluate the discrimination condition are successively obtained. Calculate and infer the discrimination result by searching the knowledge base based on this discrimination attribute value. Also, the discrimination attribute value is recalculated each time this inference is executed. A grouping template is automatically created by assigning a corresponding unit to the attention group or moving the attention group to the next group according to the inference result, and the grouping template is displayed on the output device. Further, the grouping results are actually created by assigning individual data to the grouping template and displayed on the output device.

In the grouping device of the third invention, in the second invention, the processing device assigns a group in this grouping template as a target group based on the completed grouping template, and assigns all units to any group. Allocation in the target group and other groups in each state is simulated by simulating changes in the unit allocation state from the starting state to the final state, with the grouping template as the final state and the grouping template as the final state. The case data is created by associating the discriminant attribute values regarding the number of units, the number of remaining groups, etc. with the discrimination result applied to match the final state in each state.

In the present invention, a grouping template created by the second invention and modified and completed by the user through the input device, or directly created and completed by the user without the second invention. For the grouping template, the processing device simulates changes in the allocation status of the units from the allocation start status to the final status, so that the case data that can be used to update the knowledge base, that is, used for learning processing Possible case data is generated.

The group creating apparatus of the fourth invention executes the learning process using the case data created by the third invention, and updates the rules in the knowledge base used when creating the grouping template. .

In the present invention, the case data created by the third invention is converted into discrete discriminant attribute values, and the combination of these discriminant attribute values is stored in the storage device. Then, a combination of the discriminant attribute values is used as the condition part of the rule, a rule is generated that uses the determination result of whether or not a certain unit is assigned to the target group as the execution part, and these are stored in the knowledge base, whereby the knowledge base Update the content.

In the grouping method of the fifth invention, in the first invention, necessity for creating a contingency table for each individual attribute, priority, and satisfaction degree (fitness) for the user are shown. The weighting factor for calculating the evaluation value can be arbitrarily set.

In the present invention, in the grouping problem, it is not necessary to use all of the plurality of individual attributes that characterize individuals for grouping, but whether or not to use in creating the contingency table is set for each individual attribute, and the grouping is performed. The user arbitrarily adds priorities among the individual attributes to be targeted, and the weighting coefficient used to calculate the evaluation value (score) that indicates the degree of satisfaction for the user is arbitrarily set, and these priorities are set. Records and manages weighting factors.

A grouping method of a sixth invention is the grouping method of the fifth invention, wherein for all group candidates satisfying the combination of the individual attributes of the grouping template, a judgment value and a weighting coefficient arbitrarily set for each individual attribute are set. Then, the evaluation value of each group candidate is calculated.

According to the present invention, a plurality of group candidates having a possibility of a grouping result are created based on the grouping template, and the evaluation value of each group candidate is calculated. It is possible to judge which group candidate most reflects the intention of the user.

The grouping method of the seventh invention is such that, with respect to a plurality of group candidates created by the sixth invention and having the evaluation values added, the individuals do not overlap among the group candidates, and the evaluation values are The integer grouping is used to obtain the grouping result so that is optimal.

In the present invention, the cost (evaluation value) and the coefficient matrix to be optimized are created based on the group candidates, the optimum solution is obtained as a set partitioning problem for these, and the optimum grouping result is obtained. Output as.
That is, in the first invention, a tentative grouping result that simply fits the grouping template is obtained, whereas in the present invention, an optimum grouping result that fits the user's intention can be obtained.

The grouping method according to the eighth invention is the grouping method according to the first invention, wherein the individual attribute values are composed of discrete individual attribute values using character strings or symbols and continuous individual attribute values using numerical values. Representing, for discrete individual attribute values, give detailed discrete individual attribute values for classifying similar ones into the same classification, and the types of data classified in these attribute values, For continuous individual attribute values, it provides detailed discrete individual attribute values for classifying adjacent numerical values into the same classification and a threshold value that serves as a boundary for identifying these individual attribute values. .

According to the present invention, for both discrete and continuous individual attribute values, the discretization defining unit newly defines discrete individual attribute values. When setting, the inconvenience of setting combinations for all actual data and the inconvenience of increased storage capacity is eliminated, and the actual data is attributed to a new discrete individual attribute value within a certain range Set combination conditions between values.

In the grouping method of the ninth invention, in the eighth invention, the compatibility between the detailed contents of the discrete individual attributes is numerically set and managed as a combination condition of the attribute values, and this numerical value is This is used for calculating the evaluation value of the group candidate. As a result, the compatibility between the contents of the individual attributes can be reflected in the evaluation value of the group candidate.

The grouping method of the tenth invention is the fifth invention.
In the invention, the top-level contingency table is created by paying attention to the two highest individual attributes among the priorities designated by the user, and if the frequency in this contingency table is 2 or more, the priority is given priority. Creating a contingency table by paying attention to the two individual attributes located at the next level in the ranking, and hierarchically managing the combinations of individual attribute values forming the group candidates in multiple stages to create a grouping template. It is a thing. As a result, it is possible to perform grouping capable of meeting a more detailed request regarding the combination of individual attribute values.

[0034]

Embodiments of the present invention will be described below with reference to the drawings. First, an embodiment of the first invention will be described.
It is a database showing a group of individuals to be grouped, and one individual having a plurality of individual attributes is recorded in each row. For example, "No. 1"
The employee name of “Wada” is “management” for the “affiliation department” that is an individual attribute, “3 (years)” for “years of service”, “Kyoto” for “work location”, and “division” for “classification”. The leader. "

[0035]

[Table 1]

Table 2 shows a division table for dividing a group by paying attention to any two of the individual attributes shown in Table 1. In this example, the individual attributes "belonging department" and "individual attribute" are shown. The nine individuals in Table 1 are divided on the basis of "division". Here, the individual attribute “belonging department” has discrete sales attribute values “sales”, “design”, and “management”, and the individual attribute “classification” has the same individual attribute value “leader”, It is assumed that there is "general". The number in the contingency table is the number of units indicating one unit of frequency of individuals belonging to a combination of individual attributes (for example, (sales, leader)).

[0037]

[Table 2]

Table 3 shows each group indicating what kind of combination of individual attributes each group should have when the group shown in Table 1 is divided into groups each having a capacity of 3 persons. The combination of individual attributes of is shown as a grouping template. For example, the first group is
As a combination of discrete individual attributes related to “Department” and “Division”, (sales, general), (design, leader),
It is shown that it is composed of three units (management, general).

[0039]

[Table 3]

In this embodiment, when the number of individuals, the number of groups, and the number of people constituting the group are designated, the first group,
The groups of interest are sequentially set as in the second group, and so on, and the unit that is the minimum unit of grouping (frequency unit of 1 in the contingency table shown in Table 2) is assigned to each group, or the group of interest is set. The grouping template as shown in Table 3 is created only by the operation of moving to the next group. Here, the rule determination items necessary when creating the grouping template, that is, the determination results described in the rule execution unit and the determination attributes to be considered are listed below.

(1) Discrimination Item (Discrimination Result) Allocation: Allocate a unit to a target group. Do not assign: Move the attention group to the next group without assigning a unit to the attention group.

(2) Discrimination attribute relating to the group to be evaluated This is a discrimination attribute indicating the number of the group currently being evaluated among all the groups. Discrimination attribute 1: Indicates the number of groups after the evaluation is started. Discrimination attribute 2: Indicates how many groups that have not been evaluated (not evaluated) remain.

(3) Discrimination attribute related to the status of units already assigned to the group to be evaluated This is a discrimination attribute related to the units already assigned to the group to be evaluated. Discrimination attribute 3: Discrimination attribute relating to the number of units of the same type as the evaluation target unit Discrimination attribute 4: Discrimination attribute relating to the ratio of the number of units assigned to the evaluation target group regardless of the type of unit

(4) Discrimination attribute relating to the number of units in the previous group to be evaluated This is a discrimination attribute relating to the number of units already assigned to the group immediately preceding the group currently being evaluated. Discrimination attribute 5: Discrimination attribute regarding the number of units of the same type as the evaluation target unit Discrimination attribute 6: Discrimination attribute regarding the ratio of the number of units already assigned to the preceding group regardless of the type of unit

(5) Discrimination attribute relating to the number of units in the next group to be evaluated This is a discrimination attribute relating to the number of units already assigned to the group next to the group currently being evaluated. Discrimination attribute 7: Discrimination attribute regarding the number of units of the same type as the evaluation target unit Discrimination attribute 8: Discrimination attribute regarding the ratio of the number of units already assigned to the next group regardless of the type of unit

(6) Discrimination attribute relating to the deviation from the average value of units of the type currently assigned. Discrimination attribute 9: The number of units that are currently assigned to the type of units that are already assigned to the group to be evaluated, and its average value (when assuming that the units of the target type are equally divided into all groups) Value) discriminant attribute related to deviation

(7) Discrimination attribute relating to the margin for assigning units to the group currently being evaluated. Discrimination attribute 10: Discrimination attribute relating to the ratio of the number of units that can be assigned to the group to be evaluated regardless of the unit type

(8) Discrimination attribute relating to the unassigned degree of the unit of the type currently assigned. Discrimination attribute 11: Discrimination attribute relating to the number of units in which the group to be assigned is not fixed among the types of units currently assigned (the unit currently assigned is also unassigned). )

(9) Discrimination attribute relating to the number of units that can be assigned to the group to be evaluated. Discrimination attribute 12: Discrimination attribute relating to the number of units, such as how many more units can be assigned to the group to be evaluated regardless of the type of unit

(10) Discrimination attribute relating to the ratio of the types of units to be allocated. Judgment attribute 13: Judgment attribute relating to the ratio of the type of unit currently allocated to the total number of units

(11) Other discrimination attributes. Discrimination attribute 14 and later: These are arbitrarily set as necessary.

As described above, Table 3 shows a state in which all the units indicated by the frequency 1 unit in the contingency table of Table 2 are assigned to all the groups. Table 4 shows combinations of individual attributes of each unit in the contingency table of Table 2, where a ₁₂ , a ₂₁ , a ₂₂ , a ₃₁ , a ₃₂ are combinations of the same individual attributes. For convenience, the same unit number is given to the unit group to have. As is clear from Table 4, the unit group a ₁₂ has three units
₂₁ and a ₃₂ are composed of two units, and the same a ₂₂ and a ₃₁ are composed of one unit.

[0053]

[Table 4]

In this embodiment, an initial state is a state in which no unit is assigned to all groups,
First, the first group is a target group at the start of grouping template creation.

First, of all 9 units shown in Table 4, a unit having a combination (sales, general) of the first individual attribute in the unit group a ₁₂ (hereinafter, a unit (sales, general) as required). The same applies to other items) is assigned to the first group. Next, the second unit (sales, general) in the unit group a ₁₂ is
The target group is moved without being assigned to the first group, and is assigned to the second group. Further, the third unit (sales, general) in the unit group a ₁₂ moves the point of interest without allocating it to the first group, and moves it to the third group without allocating it to the second group. wear.

As a result of the series of operations described above, the three units of the unit group a ₁₂ are
One is assigned to each of the third groups.

[0057]

[Table 5]

This means that among the above-mentioned discrimination items, "allocation" is applied three times and "no allocation" is applied three times. Since the allocation of the unit group a _12, that is, the three units (business, general) has been completed, the target group is returned to the first group again.

Next, the fourth unit from the beginning, that is, the first unit (design, leader) in the unit group a ₂₁ is assigned to the first group. Further, the fifth unit from the beginning, that is the second unit in the unit group a ₂₁ (design, reader) moves the target group without assigned to the first group, allocates to the second group. Since the allocation of the unit group a _21, that is, the two units (design, leader) is completed in this way, the target group is returned to the first group again.

Similarly, the sixth unit from the beginning,
That is, the units (design, general) of the unit group a ₂₂ move the attention group without assigning them to the first group, and further move the attention group without assigning them to the second group,
Assign to the third group. In this way, since the allocation of one unit (design, general) of the unit group a ₂₂ is completed, the target group is returned to the first group again.

Similarly, the seventh unit from the beginning,
That is, the unit (management, leader) of the unit group a ₃₁ moves the attention group without allocating it to the first group, further moves the attention group without allocating it to the second group, and allocates it to the third group. Thus unit group a
_Since the allocation of ₃₁ units (management, leader) is completed, the target group is returned to the first group again.

Further, the eighth unit from the beginning, that is, the first unit (management, general) of the unit group a ₃₂ is assigned to the first group. Then, the last unit, that is, the second unit (management, general) of the unit group a ₃₂ is moved to the target group without being assigned to the first group, and is moved to the second group without being assigned to the second group. Assign to 3 groups. This completes the allocation of the two units (management, general) of the unit group a ₃₂ . By the above operation, the grouping template as shown in Table 3 is created.

In order to obtain the above-mentioned discrimination item (discrimination result) "allocation" or "non-allocation",
The plurality of discrimination attributes 1 to 14 described above are considered. For example, as a result of considering these discrimination attributes, the first to third
The number of units in each group is three, and the discrimination attributes 1
As a result of considering 14 and the like, units having the same combination of individual attributes, for example, units (sales, general) are evenly allocated.

That is, in the present embodiment, given a group of a certain number of individuals to be grouped, the number of groups to be divided, and the quorum of each group, first,
The number of individuals is two individual attributes A (discrete individual attribute value A ₁ ,
A _2, with a .......... ) And the individual attribute B (having discrete individual attribute values B ₁ , B ₂ , ...), the frequency of the contingency table, that is, the number of units. Then, for each group from the first group to the last group, set the group to be focused on sequentially and assign the unit to the group currently being focused on, or pay attention to the case when no unit is assigned to that group. A grouping template is created while sequentially determining whether to move the group to the next group in consideration of the determination attributes 1 to 14 regarding the number of allocated units and the number of remaining groups in the current target group and other groups.

By appropriately storing the combination of the conditions relating to the discrimination attributes considered when creating the grouping template and the discrimination result, a new group of individuals to be grouped, the number of groups to be divided, and The combination of these discriminant attributes and the discriminant result can be used for the discriminating process when the capacity of each group is given. Therefore, even if the number of individuals, the number of groups, and the maximum number of people in each group that are completely different from the original grouping template are given, the operation of assigning each unit to the attention group and setting the attention group to the next group A grouping template can be created by only two operations of moving, and actual grouping can be performed based on this template.

Next, an embodiment of the second invention will be described. The present invention relates to a grouping device using a knowledge base and an inference engine. In this embodiment, a case where a group of 12 people is divided into 4 groups of 3 people will be described.

FIG. 1 shows the structure of a grouping device. Reference numeral 100 denotes a processing device composed of a personal computer or the like, which is provided with an inference program. Reference numeral 200 denotes a knowledge base, which has a combination of conditions related to the discrimination attribute values, which are values of various discrimination attributes as described above, as a condition part, and a rule in which a determination result of “assign” and “not assign” is an execution part. Many are stored.

When the input device 300 inputs a group of a certain number of individuals to be grouped and a number of groups of an arbitrary number of persons, the processing apparatus 100 is expressed by arbitrary two individual attributes. One unit of the frequency of the contingency table is expressed as the minimum unit, and a contingency table as shown in Table 2 is created. Next, the processing device 100 makes an inference according to the contents of the knowledge base 200, and outputs the result as a grouping template and a grouping result to the output device 40.
Output to 0.

Details of the processing in the processing apparatus 100 will be described below. Now, suppose that the rules stored in the knowledge base 200 are as shown in Tables 6 and 7, for example. Table 6 shows rules relating to “assignment” (7 rules in total from rule 01 to rule 07) and Table 7 shows rules relating to “not assigning” (total rules 8 to 15 in total 8).
Individual).

[0070]

[Table 6]

[0071]

[Table 7]

In these rules, for example, "discriminant attribute 1-1" means that "discriminant attribute 1: indicates which number group from the start of the discriminant", It is a discrimination attribute indicating that it is the beginning of the group. In addition, "discrimination attribute 2-2" means that in "discrimination attribute 2: indicating how many un-evaluated groups remain," about half of the groups remain after the evaluation is started. It is a discrimination attribute indicating that.

FIG. 2 shows the processing apparatus 10 in this embodiment.
It is a flowchart of the process performed by 0. First, the processing apparatus 100 calculates all the discriminant attribute values (attribute values relating to the discriminant attributes 1 to 14) necessary for creating the grouping template for the unit to be evaluated (S1), and these discriminant attribute values are calculated. The knowledge base 20 shown in Tables 6 and 7 has a rule having a conditional part that satisfies the conditions.
Searching from the content of 0, the determination result described in the execution part of the rule "assign" or "not assign"
Is set as the inference result (S2).

Then, the above inference result is judged (S3
1) If the inference result is "do not allocate", move the focused group to the next group (S41, S42),
Recalculation, inference and judgment of attribute value (S1, S2, S31)
Is repeatedly executed. Then, when the inference result in S31 is “not assigned” and the group of interest reaches the last group in S41, or when there is no unit to be assigned, the inference is terminated and the grouping template Creation is complete.

If the result of inference in S31 is "assign", one unit is assigned to the currently focused group (S32), and the processes after S1 are re-executed. In this way, the processing device 100 evaluates the rule in the knowledge base 200 when the discriminant attribute value is calculated, makes a conclusion of “allocate” or “not allocate”, and finishes one inference cycle. Is taking.

When the inference is continued using the rules shown in Tables 6 and 7, when the group of 12 people shown in Table 8 is divided into 4 groups of 3 people, finally, Table 10 A grouping template as shown in is created. It should be noted that Table 9 is a division table formed by focusing on the “affiliation department” and the “division”.

[0077]

[Table 8]

[0078]

[Table 9]

[0079]

[Table 10]

The output device 400 displays the grouping template and the grouping result as shown in Table 11 formed by fitting the actual individual data to the grouping template.

[0081]

[Table 11]

As described above, in this embodiment, the processing device 10
When the inference is completed once by 0, the discriminant attribute value is recalculated, the condition part of each rule is examined based on this discriminant attribute value, and the execution part of the rule in which the condition part is satisfied is set as the inference result. When is not assigned, the target group is moved to the next group and the same inference is repeatedly executed. As a result, a grouping template as shown in Table 10 is finally created, and by applying actual data to this grouping template, the grouping result as shown in Table 11 can be obtained.

Next, an embodiment of the third invention will be described with reference to FIG. In this embodiment, the grouping template of Table 10 created by the embodiment of the second invention (created by the knowledge base and the inference program) does not match the user's intention (for example, in the grouping template of Table 10 , Leaders are not assigned to each group, etc.). In this case, a case that can be used by the user to create and input a new grouping template as shown in Table 12 using the input device 300 and update the contents of the knowledge base 200 (rules when performing inference) The data, that is, the case data 500 that can be used for the learning process is created by the processing device 100 and stored in the case data file.

[0084]

[Table 12]

First, as shown in Table 12, a model grouping template is created using the input device 300. The grouping template shown in Table 12 is divided into four groups, each of which has 12 individuals as a grouping target and has a maximum capacity of 3 as described above. In Table 12, the underlined portions are the portions changed from Table 10.

When a model grouping template is created in this way, the processing apparatus 100 creates the case data 500 by using this as a completed grouping template. That is,
The initial state is a state in which no unit is assigned to all groups, and the final state is the completed grouping template as shown in Table 12 above. The state change in the process of assigning each unit to each group is processed. Device 10
Simulate within 0 and create each case data with each discrimination attribute value in each state and the discrimination result of "allocation" or "not allocation" applied to match the final state in each state. , Add to case data fill.

Table 13 shows a part of the case data 500 created from the grouping template of Table 12. The case data 500 includes discrimination attributes 1 to 1 for each target concept (discrimination result).
The 13 determination attribute values are associated with each other. As a result of accumulating such case data 500, this case data 5
00 can be used to update the rules that are the contents of the knowledge base 200 as described below.

[0088]

[Table 13]

Next, an embodiment of the fourth invention is shown in FIGS.
Will be described with reference to. In FIG. 4, the processing device 100
Stores case data 500 consisting of accumulated numerical data as shown in Table 13 with the conditions for defining the magnitude relation between the numerical data and the threshold value and new discrete discriminant attribute values such as words. The attribute value definition unit 600 (see FIG.
Referring to (a), for each discriminant attribute, discrete discriminant attribute values (a1 and a in FIG. 5A and 5B are shown for convenience).
2, b1, ... In the present embodiment, “discrimination attribute 10-3” and “discrimination attribute 1” in Tables 6 and 7 are used.
2-4 ”, ..., And so on), and a combination of these discrimination attribute values is generated as the contents of the storage device 700.

Here, the storage device 700 is substantially the same as the storage device according to the first invention described in Japanese Patent Laid-Open No. 6-68066 of the same applicant, and an example is shown in FIG. 5 (b). As shown in the table, the combination of discrimination attribute values, the number of positive examples, the number of negative examples, and the column of necessity are included. In the storage device 700 of FIG. 5B, discrete discriminant attribute values a1, a2,
b1, b2, c1 and c2 are combined so that the same discrimination attribute value does not overlap. In addition, the number of positive examples is a number as a value for determining whether it is the positive example itself described below or a combination of the determination attribute values of the positive example. It is an integrated value of the number of negative examples mentioned. Further, the necessity means the necessity of the combination of the discrimination attribute values.

In the present embodiment, first, the storage device 70 is used as an example of the determination result that "allocation" is "positive" (positive example) and as an example of the determination result that "not assigned" is "negative" (negative example).
A combination of discriminant attribute values is generated in 0, and a rule having “assign” is generated as an execution part of the rule. Then, contrary to the above, a combination of discriminant attribute values is generated in the storage device 700, with “not allocated” as a positive example and “allocated” as a negative example,
Generate a rule with "not assigned" as the rule execution part.

In other words, in the knowledge base 200, when “assignment” is taken as a positive example, the necessity column is “necessary” in the combination of the discrimination attribute values in the storage device 700 at that time. A rule is stored in which a combination of discrimination attribute values in a row is used as a condition part of a rule, and a discrimination result “assign” is used as a rule execution part. Similarly, when “not assigned” is taken as a positive example, among the combinations of the determination attribute values in the storage device 700 at that time, the combination of the determination attribute values of the row whose necessity column is “necessary” is selected. The rule is stored as the condition part of the rule, and the rule execution part is “not assigned” as the determination result. As a result, the contents of the knowledge base 200 will be updated to new ones.

Therefore, the person in charge of grouping does not need to directly give the knowledge, and the example of the grouping template according to the user's intention as shown in Table 12 is used as a model according to the embodiment of the third invention. By accumulating data 500 and learning using the case data 500, the rules in the knowledge base 200 can be improved. That is, since the user only needs to indirectly give the knowledge, the burden on the user is significantly reduced.

Case data 50 accumulated as shown in Table 13
As a result of learning using 0, the contents of the knowledge base 200 are
As shown in Table 14, there are 11 rules (rule 01 to rule 11) related to “allocation”. Further, although only a part is shown in Table 15, there are 34 rules (rule 12 to rule 45) relating to "not allocated". In addition,
Of the case data in Table 13, the 18th (*) data from the top is peculiar data, so it was excluded from the case data for learning and learned.

[0095]

[Table 14]

[0096]

[Table 15]

It is assumed that the discriminant attribute value in the condition part of the rule is a new discriminant attribute value (discrete value such as a word) defined by the attribute value defining part 600. Further, in this embodiment, in order to facilitate understanding, learning is performed only with the case data created by the new grouping template (see Table 12).

Next, an embodiment of the fifth invention will be described with reference to FIG. In this embodiment, as shown in FIG. 6, an attribute condition definition unit 800 is added to the processing device 100, and
A predetermined process is added to the processing device 100 to set and display the necessity of individual attributes, the priority order, and the weighting coefficient when creating a contingency table. here,
The weighting coefficient is a coefficient used when calculating an evaluation value (score) indicating the degree of satisfaction of each group candidate for the user. That is, giving a large weighting coefficient to a certain individual attribute means that the content of the individual attribute is largely reflected in the evaluation value when the group candidate is evaluated.

The processing device 100 executes the following processing. (1) Database to be grouped (may be input via the input device 300)
The individual attribute name is read from the attribute condition definition unit 800
In addition to writing in the column of individual attribute names, the list of individual attribute names is displayed on the output device 400 in the format shown in Table 16.

[0100]

[Table 16]

For example, when the database shown in Table 1 above is input through the input device 300, the names of individual attributes are “affiliated department”, “years of service”, “work location”,
The “category” is set in the individual attribute name column of the attribute condition definition unit 800 and is displayed on the output device 400. The “No.” and “employee name” in Table 1 are used to identify each record (each row) in the database.

(2) With respect to the necessity of individual attributes, the priority order, and the weighting factor, the contents input from the input device 300 are read, and the attribute condition defining unit 80 having the format shown in Table 16 is read.
Fill in each column of 0. For example, in the database of Table 1, if the contingency table is created and grouped by paying attention only to the “affiliated department” and the “division” of the individual attribute name, the user can arbitrarily set values as shown in Table 16. Can be set. That is, since the contingency table is created by paying attention to the “affiliation department” and the “classification”, “necessity” is set in the “necessity” column for these. In the “priority order” column, the user arbitrarily assigns a number from the ones that are considered to have the highest priority order among the individual attributes. Furthermore, a weighting factor for calculating the evaluation value of each group candidate described later is given as appropriate. Here, if only considering creating a contingency table, it is possible to create using the top two individual attributes with high priorities.
The “necessity” column may not be provided.

As described above, according to this embodiment, it is not necessary to use all individual attributes that characterize individual elements for grouping, and the user can arbitrarily specify whether or not to create a contingency table for each individual attribute. Moreover, the priority order of individual attributes can be arbitrarily set. Further, when calculating the evaluation value of each group candidate at the time of grouping, the user can freely set the weighting coefficient α _i (i = 1, 2, 3, ...) In the following Equation 1 at the site. There is an advantage that you can. The determination value (score) A _i for each individual attribute in Expression 1 is also arbitrarily set by the user.

[0104]

## EQU1 ## Evaluation value = α ₁ × A ₁ + α ₂ × A ₂ + ... α _i (i = 1,2,3, ...): Weighting coefficient A _i (i = 1,2,3, ...) …): Judgment value (score) for each individual attribute

Next, an embodiment of the sixth invention will be described with reference to FIG. In this embodiment, as shown in FIG. 7, a group candidate part 900 is added to the processing device 100, and
A predetermined process is added to the processing device 100, and a large number of group candidates are created by using individuals (data) having various individual attributes based on the grouping template shown in Table 3.
The evaluation value indicating the satisfaction level of these group candidates for the user is calculated.

The processing apparatus 100 executes the following processing (method 1). (1) Read one line from the grouping template. (2) A large number of group candidates satisfying the combination of individual attributes of the grouping template are created from the data to be grouped so that their contents do not overlap. For example, when the first group is read from the grouping template of Table 3, data satisfying the combination of individual attributes of this first group is shown in Table 1.
Searching the database of "No. 3" for "Suzuki", "No. 5" for "Kanda", "No. 6" for "Goto", "No. 3" for "Suzuki", " No.
There are combinations such as "Kanda" of "5" and "Kawasaki" of "No. 7".

(3) If the same candidate as the group candidate created in (2) above does not exist in the group candidate section 900, it is recorded in the group candidate section 900. (4) If another line exists in the grouping template, the above-mentioned processes (1) to (3) are repeated.

(5) With respect to all the group candidates of the group candidate section 900, discrete individual attributes (“sales”, “sales”, “sections” and individual attributes as shown in Table 17 are included. The largest value is set in the judgment value column for each corresponding individual attribute of the group candidate part 900 based on the combination condition (set by numerical value) of “design”, “management”, “leader”, and “general”).

Table 17 shows that, for each individual attribute, the user arbitrarily sets the judgment value A _i of a combination of discrete individual attributes indicating the contents thereof. It can be said that it defines the compatibility between the two.

[0110]

[Table 17]

In the above example, “No.
"3", "Suzuki", "No. 5", "Kanda", "No.
In the combination of “Goto” of “6”, from Table 1, combinations of (sales, design), (sales, management), (design, management) can be considered for “affiliated department”, and these combinations are shown in Table 17. According to, the judgment values are “3”, “3”, and “3”, respectively.
Is. Therefore, “3” as the largest value is set in the judgment value column of “belonging department” in the group candidate section 900 as shown in Table 18 described later.

Here, "1 to 5 years", "6 to 10 years", "11 to 15" set for "years of service" in Table 17 are set.
"Kanto" set for "year" and "work location",
"Kansai" and "west of China" are discrete individual attribute values newly introduced by the embodiment of the eighth invention described later.
However, in the sixth invention, even if the eighth invention is not premised, for example, regarding “belonging department” and “classification”, the discrete individual attributes (“management”, “ Design ”, ……,“ Reader ”,“ General ”
Etc.) can be used to create group candidates.
Further, as for “years of service” and “working place”, the group candidate as shown in Table 18 is set by setting the judgment value corresponding to the combination of the contents by another appropriate method without depending on the eighth invention. It is possible to create and obtain the evaluation value.

(6) In the group candidate section 900, when values have been set in the judgment value columns of all group candidates, the evaluation value of each group candidate is calculated by the above-mentioned mathematical expression 1 and Table 1
Set in the evaluation value column of 8.

[0114]

[Table 18]

Further, in order to make the processing device 100 create a group candidate, the following processing (method 2) can be executed. This method 2 is a modification of the method 1 in order to prevent duplicate group candidates from being created. (1) Read one line from the grouping template. (2) One group candidate satisfying the combination of individual attributes of the grouping template is created from the grouping target data. At the same time, the data used to create the group candidates are excluded from the database. (3) If the same candidate as the group candidate created in (2) above does not exist in the group candidate section 900, it is recorded in the group candidate section 900. (4) If another line exists in the grouping template, the above (1) to (3) are repeated.

(5) The above processes (1) to (4) are repeated a plurality of times. In (2), the use of random numbers or the like prevents duplicate group candidates. (6) Based on the combination of discrete individual attribute values that are the contents of the individual attributes “belonging department” and “classification” as shown in Table 17, the largest among all the group candidates of the group candidate part 900 The value is set in the judgment value column for each corresponding individual attribute of the group candidate part 900. (7) When values are set in the judgment value columns of all group candidates, the evaluation value of each group candidate is calculated and set in the “evaluation value” column of Table 18. In addition, the calculation formula of the evaluation value is Formula 1 as described above.

In this embodiment, for example, based on the database as shown in Table 1, using the contents of the attribute condition definition unit 800 shown in Table 16 and the judgment value for each individual attribute shown in Table 17, group candidates By creating the above, it is possible to create a table 18 (one row per one candidate in this table) in the group candidate section 900. Here, in the column of "group constituent element" of Table 18, the individuals forming each group candidate are shown using "No." of the database shown in Table 1.

As described above, according to the present embodiment, it is possible to create a plurality of group candidates conforming to this based on the grouping template, and by referring to the evaluation values of these group candidates, which group indicates the user's intention. You can easily and objectively judge whether it is the one that reflects the best and is satisfactory.

Next, an embodiment of the seventh invention will be described with reference to FIG. In this embodiment, as shown in FIG. 8, the processing device 100 includes a group candidate unit 900 and an optimization processing unit 10.
00 and a predetermined process are added to the processing device 100, so that a large number of group candidates have an integer plan so that individual elements do not overlap between groups and the evaluation value is optimized. This method is used to obtain the optimum grouping result. That is, according to the first aspect of the present invention, a tentative grouping result that simply conforms to the grouping template is obtained, whereas the present invention is made to obtain an optimum grouping result that matches the user's intention. is there.

The processing apparatus 100 executes the following processing. (1) Based on the group candidates as shown in Table 18, the coefficient matrix A and the cost C to be optimized are created as in Equation 2 and Equation 3, respectively, and these are generated by the optimization processing unit 1
Enter in 000.

[0121]

[Equation 2]

[0122]

(Equation 3)

Each row of the coefficient matrix A shown in Equation 2 is
Each individual in the group represents a condition that must be included in any group, and each column represents one row of group candidates. For example, in the first column of the coefficient matrix A, the third, fifth, and sixth from the top are “1”, which means that the first (first row) group candidate (No. It consists of 3, 5 and 6 individuals).
Furthermore, the evaluation value shown in Table 18 expresses the cost C of Expression 3.

(2) The optimization processing section 1000 applies the data created in (1) to the following equation and solves it as a set division problem.

Minimization: Σ _{j = 1, n} c _j x _j , Constraint condition: Σ _{j = 1, n} a _ij x _j = 1 (i = 1, 2, 3, ... M, j = 1, 2, 3, ... n) x _j ε {0,1}

(3) The optimum solution is edited and presented to the user via the output device 400 in a format in which the contents and evaluation values of each group candidate are displayed as shown in Table 19.

[0127]

[Table 19]

As described above, according to this embodiment, grouping results can be determined such that individual elements do not overlap between groups and the evaluation value is optimum, and the grouping template is simply adapted. It is possible to obtain an optimum grouping result that matches the user's wishes regarding the detailed contents of individual attributes, not the grouping result.

Next, an eighth embodiment of the invention will be described with reference to FIG. In this embodiment, as shown in FIG. 9, the discretization definition unit 1100 is added to the processing device 100, and a predetermined process is added to the processing device 100, so that the “work place” is set.
In the case of individual attributes represented by discrete and qualitative data such as, a new discrete individual attribute (eg "Kanto", "Kansai") for classifying similar items into the same classification
And the type of data classified into each of the attributes (for example, in the case of "Kanto", "Tokyo, Yokohama, Chiba").
On the other hand, unlike qualitative data, in the case of individual attributes represented by continuous numerical values such as "years of service", new discrete individual attributes for classifying similar values (For example, "1-5 years", "6-10 years", etc.)
A threshold value (for example, “6-1
In the case of "0 years", the individual attributes are converted by giving "6", "10").

The processing apparatus 100 executes the following processing. (1) The individual attribute name is read from the database as shown in Table 1 and set in the individual attribute column of the discretization defining unit 1100. (2) The contents of the discretization defining unit 1100 are presented to the user via the output device 400. (3) The user uses the input device 3
When a new discrete individual attribute and the type and threshold value of data classified therein are given via 00, these are recorded in the discretization defining unit 1100 as shown in Table 20.

[0131]

[Table 20]

Here, the new individual attribute (new discrete individual attribute) shown in Table 20 is a discrete attribute that constitutes each individual attribute (“years of service”, “work location”) shown in Table 17 above. It is the same as the typical individual attribute.

As described above, according to this embodiment, when setting the combination condition of individual attributes as shown in the following ninth invention, waste and storage of setting the combination for all the actual data The combination condition of individual attributes can be set in a certain unit without causing the disadvantage of increasing the capacity.

Next, an embodiment of the ninth invention will be described with reference to FIG. In this embodiment, as shown in FIG.
A combination condition definition unit 1200 is added to the processing device 100, and predetermined processing is added to the processing device 100.
Numerical values are used as the combination conditions of individual attributes (for example, the compatibility between discrete individual attributes “sales” and “design” in individual attribute “belonging department” in Table 17) as the individual attribute combination conditions. Settings are managed and these numerical values can be used when calculating the evaluation value of each group candidate at the time of grouping.

The processing device 100 executes the following processing. (1) The name of the individual attribute is read from the database as shown in Table 1, the content of the discretization defining unit 1100 is referred to, and the new discrete individual attribute name is entered in the name column of the combination condition defining unit 1200. Set. (2) Output the contents of the combination condition definition unit 1200 to the output device 40
Present to the user via 0. (3) When the user gives the combination conditions of the individual attributes as numerical values via the input device 300, these are recorded in the combination condition definition unit 1200.

As described above, according to this embodiment, the compatibility between the discrete individual attributes, which are the contents of the individual attributes, can be arbitrarily changed and managed by a numerical value as a combination condition, and these The numerical value can be used as the judgment value A _i (i = 1, 2, 3, ...) In the above mathematical expression 1 when calculating the evaluation value of each group candidate at the time of grouping.

Next, an embodiment of the tenth invention will be described with reference to FIGS. 11 to 13. This embodiment is shown in FIG.
As shown in FIG. 1, the hierarchy management unit 1300 of the group candidate is added to the processing device 100, and a predetermined process is added to the processing device 100, so that the top two ranks in the priority order shown in Table 16 specified by the user are added. After creating a contingency table as shown in Table 2 at the highest level by paying attention to individual attributes, when there are combinations with frequencies of 2 or more in the contingency table, for example, "belonging department" is " If “business” and “classification” are “general”, pay attention to the two individual attributes “years of service” and “working place” that are located in the next priority, and divide the table as shown in Table 21. Is further created and the combination of individual attributes forming the group candidates is managed in the hierarchy management unit 1300 in a multi-tiered and hierarchical manner as shown in FIG. 12, thereby creating a detailed grouping template.

[0138]

[Table 21]

The processing device 100 executes the following processing. (1) In the contingency table of Table 2, when a user specifies a portion to be detailed (for example, a combination of (sales, general)),
As shown in Table 21, a new contingency table focusing on “years of service” and “work location” is displayed to the user via the output device 400. (2) When the user gives a second level grouping template (see FIG. 12) created based on the contingency table of Table 21 from the input device 300, this is recorded in the hierarchy management unit 1300.

As described above, according to this embodiment, there is an advantage that the combination of individual attributes forming a group candidate can be managed in a multi-tiered and hierarchical manner, and the characteristics of the group candidate can be changed in detail. Further, the processing (2) of the processing device 100 in the embodiment of the sixth aspect of the invention is expanded and executed as described below, thereby making it possible to create a group candidate using a grouping template managed hierarchically.

(1) One line is read from the grouping template. For example, {(sales, general), (1-5 years, Kansai)}, (design, leader), (management, general) are read from the grouping template of FIG. (2) The constituent elements of the group candidates are sequentially determined from the data satisfying the combination condition of the most detailed individual attributes of the grouping template, and a large number of them are created from the grouping target data so that the contents do not overlap.

That is, the data for which the detailed condition of combination of individual attributes satisfies (sales, general) and (1-5 years, Kansai) is the data of "Kitano" of "No. 4" in Table 1. Only. Therefore, the data of the combination condition (design, leader) of individual attributes and (management, general) are added to this data.
By adding the data of, the group candidate as shown in FIG. 13 can be created. 13, for example, (4 5 6) in the first row represents one group candidate, and “Kitano” and “N” in “No. 4” in Table 1
o. 1 by "Kanda" of "5" and "Goto" of "No. 6"
One group candidate is configured. Note that these group candidates are also included in the group candidates in Table 18.

[0143]

As described above, according to the first aspect of the invention, 2
Create a contingency table focusing on individual individual attributes, create a grouping model by assigning each unit to each group by performing the discrimination process while considering the predetermined discrimination attributes, and set a certain number of individuals and the number of people. When the specified number of groups is presented, the discrimination process is performed using the combination of the conditions relating to the discrimination attributes considered at the time of creating the grouping template and the discrimination result, and the units are distributed to create the desired grouping template. be able to. As a result, it is possible to create a grouping template that can accommodate any combination of the number of individuals and the number of people in a group without having to store an enormous amount of grouping templates that take all combinations of individuals into consideration. , Actual grouping can be performed based on this template.

In the second invention, the processing device searches the knowledge base for a rule having a conditional part satisfying the calculated discrimination attribute value, and infers the discrimination result described in the execution part of the rule as an inference result. Inference processing is performed, the attribute value is recalculated each time this processing is executed, units are distributed according to the inference results, grouping templates are created, and actual grouping is performed based on this grouping template. is there. As a result, since the inference processing is sequentially executed based on the attribute value that changes with each inference, grouping can be performed without defining complicated constraint conditions in advance.

In the third invention, the completed grouping template is set as the final state, and the state in which no unit is assigned to each group is set as the start state, and the unit assignment state from the start state to the final state is set. The processing device simulates a change to create case data having a discrimination attribute value in each state and a discrimination result applied to match the final state in each state. A fourth aspect of the invention is to update a rule, which is the content of the knowledge base, by executing learning processing using case data. According to these inventions, the learning function enables the effective use of the know-how at the time of grouping in the past, and when the grouping template does not match the intention, it is corrected and further used for creating the grouping template. You can easily modify the existing rules.

In the fifth invention, necessity for creating a contingency table, priority order, and weighting coefficient for calculating an evaluation value indicating a satisfaction level for a user are arbitrarily set for each individual attribute. It can be set. With this, it is possible to use appropriately selected ones among multiple individual attributes for grouping, and to determine the user's intention or purpose depending on the necessity of use in creating a contingency table, the priority order, and the setting content of the weighting factor. Flexible grouping can be performed.

A sixth aspect of the present invention uses all the group candidates satisfying the combination of the individual attributes of the grouping template, using the judgment value and the weighting coefficient arbitrarily set for each individual attribute, and the evaluation value of each group candidate. Is to calculate. Thus, it is possible to determine which group candidate reflects the user's intention most by using the evaluation value.

A seventh aspect of the present invention uses an integer programming method for a plurality of group candidates to which the evaluation value is added so that the individuals do not overlap among the group candidates and the evaluation value is optimized. To obtain grouping results,
The cost and coefficient matrix to be optimized are created, the optimum solution is obtained as a set division problem for these, and the optimum grouping result is obtained. That is, the first
In the invention described above, a tentative grouping result that simply conforms to the grouping template is obtained, whereas in the present invention, an optimum grouping result that matches the user's intention can be obtained.

In the eighth invention, individual attribute values are represented by discrete and continuous individual attribute values, and the discrete individual attribute values are detailed for classifying similar ones into the same classification. Given discrete individual attribute values and the types of data classified into these attribute values, for continuous individual attribute values, the detailed discrete It provides individual attribute values and a threshold value that serves as a boundary for identifying these individual attribute values.

Thus, for both discrete and continuous individual attribute values, by defining new discrete individual attribute values by the discretization defining unit, the combination condition of individual attribute values in the ninth invention can be defined. At the time of setting, it is possible to eliminate the waste of setting a combination for all actual data and the inconvenience of increasing the storage capacity.

In a ninth aspect of the present invention, the compatibility between the contents of the detailed discrete individual attributes is set and managed by a numerical value as a combination condition of the attribute values, and this numerical value is used for calculating the evaluation value of each group candidate. To do. As a result, the compatibility between the contents of the individual attributes can be reflected in the evaluation value of the group candidate.

The tenth aspect of the invention is to create a contingency table at the highest level while paying attention to the upper two individual attributes of the priority designated by the user, and the frequency in the contingency table is 2 or more. In that case, pay attention to the two individual attributes located at the next level in the priority order, create a contingency table, and manage the combinations of individual attribute values forming the group candidates in a multi-tiered manner. A grouping template is created. As a result, it is possible to perform grouping capable of meeting a more detailed request regarding the combination of individual attribute values.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing an embodiment of a second invention.

FIG. 2 is a flowchart of a process in the embodiment of the second invention.

FIG. 3 is a configuration diagram showing an embodiment of a third invention.

FIG. 4 is a configuration diagram showing an embodiment of the fourth invention.

FIG. 5 is an explanatory diagram of an attribute value defining unit and a storage device according to an embodiment of the fourth invention.

FIG. 6 is a configuration diagram showing an embodiment of the fifth invention.

FIG. 7 is a configuration diagram showing an embodiment of the sixth invention.

FIG. 8 is a configuration diagram showing an embodiment of the seventh invention.

FIG. 9 is a configuration diagram showing an embodiment of the eighth invention.

FIG. 10 is a configuration diagram showing an embodiment of the ninth invention.

FIG. 11 is a configuration diagram showing an embodiment of a tenth invention.

FIG. 12 is an explanatory diagram of the detailing of the grouping template in the embodiment of the tenth invention.

FIG. 13 is an explanatory diagram of group candidates in the example of the tenth invention.

[Explanation of Codes] 100 Processing Device 200 Knowledge Base 300 Input Device 400 Output Device 500 Case Data 600 Attribute Value Definition Section 700 Storage Device 800 Attribute Condition Definition Section 900 Group Candidate Section 1000 Optimization Processing Section 1100 Discretization Definition Section 1200 Combination Conditions Definition section 1300 Tier management section

Claims (10)

[Claims] 1. A grouping method in which a group of a certain number of individuals to be grouped is divided into a plurality of groups with a specified number of persons, and two individual attributes out of a large number of individual attributes showing the characteristics of the individual. Paying attention to, create a contingency table that expresses one unit of the frequency of individuals corresponding to the combination of the contents of these individual attributes as a unit that is the minimum unit of grouping, and sequentially pay attention to each group for which the capacity is specified. Set a group and assign the unit to the attention group or move the attention group to the next group, considering the discrimination attributes regarding the number of assigned units and the number of remaining groups in the attention group and other groups. By deciding whether the individual attributes of each group are Create a grouping template expressing the match, using the combination of the conditions regarding the discrimination attributes considered when creating the grouping template and the discrimination result, a newly given group of a certain number of individuals, the number of persons in each group, and By performing the discrimination process based on the number of groups and allocating the units, a grouping template consisting of multiple groups with a specified number of people is created, and based on this grouping template, each individual is assigned to each group for grouping. A grouping method characterized by the following. 2. An input device for inputting a group of a certain number of individuals to be grouped and a number of groups of an arbitrary number of persons, and two individual attributes out of a large number of individual attributes showing the characteristics of the individual. Paying attention to, create a contingency table that expresses one unit of the frequency of individuals corresponding to the combination of the contents of these individual attributes as a unit that is the minimum unit of grouping, and sequentially pay attention to each group for which the capacity is specified. A group is set, and the processing unit that calculates the discriminant attribute value related to the number of assigned units and the number of remaining groups in the target group and other groups and the combination of the discriminant attribute values are used as the condition part, and the relevant unit is allocated in the target group. A knowledge base that stores rules that have the result of determining whether to attach or move the focused group to the next group. The processing device searches the knowledge base for a rule having a conditional part that satisfies the calculated discrimination attribute value, and performs an inference process using the discrimination result described in the execution part of the rule as the inference result. A grouping device that recalculates the discriminant attribute value each time processing is performed, distributes units according to the inference result, creates a grouping template, and allocates individual data to this grouping template to perform grouping. A grouping device having an output device for displaying the grouping template and the grouping result created by. 3. The grouping device according to claim 2, wherein the processing device is based on the completed grouping template, sets a group in the grouping template as a target group, and all units are not assigned to any group. As a starting state, the grouping template is used as a final state, and by simulating changes in the allocation state of the units from the starting state to the final state, the number of allocated units and the remaining number in the target group and other groups in each state are simulated. A grouping device, which creates case data in which a discrimination attribute value relating to the number of groups and the like and a discrimination result applied to match the final state in each state are associated with each other. 4. The grouping device according to claim 3, wherein the processing device executes a learning process using the case data to update the rule in the knowledge base. 5. The grouping method according to claim 1, wherein the necessity for creating a contingency table for each individual attribute, the priority, and the weight for calculating the evaluation value indicating the degree of satisfaction for the user. A grouping method characterized in that coefficients and can be set arbitrarily. 6. The grouping method according to claim 5, wherein for all group candidates that satisfy the combination of individual attributes of the grouping template, a determination value and a weighting coefficient that are arbitrarily set for each individual attribute are used for each group candidate. A grouping method characterized by calculating an evaluation value of a group candidate. 7. The grouping method according to claim 6, wherein for a plurality of group candidates to which evaluation values are added, individuals do not overlap among the group candidates and the evaluation values are optimized. A grouping method characterized by obtaining a grouping result using integer programming. 8. The grouping method according to claim 1, wherein the individual attribute value is expressed by a discrete individual attribute value using a character string or a symbol and a continuous individual attribute value using a numerical value, For specific individual attribute values, the detailed discrete individual attribute values for classifying similar ones into the same classification and the types of data classified in these attribute values are given, and continuous individual attribute values are given. Regarding the attribute value, a grouping method characterized by giving a detailed discrete individual attribute value for classifying adjacent numerical values to the same classification and a threshold value serving as a boundary for distinguishing these individual attribute values . 9. The grouping method according to claim 8, wherein compatibility between the contents of the detailed discrete individual attributes is set and managed as a combination condition of attribute values, and this numerical value is evaluated for each group candidate. A grouping method characterized by being used for calculating values. 10. The grouping method according to claim 5, wherein the top-level contingency table is created by paying attention to the top two individual attributes among the priorities designated by the user, and If the frequency is 2 or more, pay attention to the two individual attributes located at the next level in the priority order, create a contingency table, and create a hierarchical combination of the individual attribute values that make up the group candidate in multiple stages. A grouping method characterized by managing groupings and creating grouping templates.