JP6852785B2 - Deployment knowledge generation system, deployment knowledge generation method and deployment knowledge generation program - Google Patents

Description

The present invention relates to an expansion knowledge generation system, an expansion knowledge generation method, and an expansion knowledge generation program that generate expansion knowledge developed based on knowledge.

Markov Logic Network, Bayesian Network, Answer Set Programming, etc. are known as methodologies for evaluating propositional logic. In the above-mentioned methodology, knowledge may be expressed by a logical expression in which observable predicates are combined by logical symbols. Since this predicate corresponds to each term when it is expressed by a logical expression, this predicate may be referred to as an item in the following description.

For example, Patent Document 1 describes a system for discovering database rules. The system described in Patent Document 1 is a free item set consisting of attribute / value pairs in a database, and generates a set of free item sets whose frequency in the database is equal to or higher than a predetermined threshold value. Then, in the system described in Patent Document 1, as rule candidates, the free item set α is used as the condition part of the rule, the item x that does not share the attribute with the free item set is used as the consequent part of the rule, and the attribute of the premise part of the rule. A rule is generated in which the set is obtained by a depth priority search and the attributes are not included in α or x.

Further, in the above methodology, a process (grounding) of creating information by applying specific data to knowledge may be performed. In the following explanation, the information obtained by applying a specific value to knowledge is referred to as expanded knowledge. The expanded knowledge is used, for example, as information when an inference engine executes inference.

International Publication No. 2013/172310

Hereinafter, a specific method for generating expansion knowledge will be described. Here, for the sake of simplification of the explanation, the following two pieces of knowledge will be illustrated.
Knowledge 1: mail_address_sender (x1, m1) ^ mail_address_receiver (x2, m1)
=> competitor (x1, x2)
Knowledge 2: mailild (e1, m1) ^ competitor (x1, x2) ^ meet (e1, x1, x2) ^ change (e2, x1, ”price”)
=> query (x1, e1, ”cartel”)

Further, FIG. 11 is an explanatory diagram showing an example of observation data for sending and receiving e-mails. In the example shown in FIG. 11, it is assumed that there are from company A to company E, and mail is transmitted in the direction indicated by the arrow (company A to company B, company A to company C, company C to company B, company D to company B). Is observed.

On the left side of knowledge 1, the predicate "mail_address_sender (x1, m1)" indicates that x1 sent mail m1, and the predicate "mail_address_receiver (x2, m1)" indicates that x2 received mail m1. Shown. Further, on the right side of knowledge 1, "competitor (x1, x2)" indicates that x1 and x2 are in a competitive relationship.

Similarly, on the left side of knowledge 2, the predicate "mailild (e1, m1)" indicates that the content of the mail m1 is e1, and the predicate "meet (e1, x1, x2)" is x1 in e1. The predicate "change (e2, x1," price ")" indicates that x1 has changed the price in the content e2. The predicate "competitor (x1, x2)" on the left side of knowledge 2 has the same meaning as the predicate on the right side of knowledge 1. Further, on the right side of knowledge 2, the predicate "query (x1, e1," cartel ")" indicates that x1 may be a cartel by e1.

As the first expansion method, in each knowledge, there is a method of generating expansion knowledge from observation data satisfying the left side, that is, a method of generating expansion knowledge using the data observed in all the terms on the left side. The number of expanded knowledge generated in this case is a combination of the number observed for knowledge 1 (combination of observed (mail_address_sender × mail_address_receiver)) and a combination of the number observed for knowledge 2 (observed (mailild)). It can be calculated by the combination of × competitor × meet × change). The competitor in knowledge 2 can be specified from the result of knowledge 1.

For example, in the example shown in FIG. 11, the number combination observed for knowledge 1 is 4 (company A to company B, company A to company C, company C to company B, company D to company B).

When the expansion knowledge is generated by the first expansion method, since the generated expansion knowledge is information created based on the actually observed data, it is suppressed that the amount becomes enormous. On the other hand, it can be said that the development knowledge generated in this way alone cannot generate a creative hypothesis, and is insufficient as information when the inference engine executes inference.

Therefore, as a second development method, there is a method of narrowing down to the closed-world hypothesis (that is, narrowing down to the observation data of each item). The number of expanded knowledge generated in this case is the combination of the number observed in each item for knowledge 1 (the number of observed mail_address_sender x the number of observed mail_address_receiver) and the number of knowledge 2 observed in each item. It can be calculated by the combination of the numbers (the number of observed maililds x the number of observed competitors x the number of observed meets x the number of observed changes).

For example, in the example shown in FIG. 11, the number observed for mail_address_sender of knowledge 1 is 3 (company A, company C and company D), and the number observed for mail_address_receiver of knowledge 2 is 2 (company B and company C). Become. Therefore, the number of expanded knowledge is 6 (= 3 × 2). If it is preferable to exclude the information from the C company to the C company, the number of developed knowledge is 5 (= 6-1).

By generating the expansion knowledge by the second expansion method, as compared with the first expansion method, not only the data actually observed but also the data that can be used as the expansion knowledge (for example, from company D to company C). Can be generated. However, even when the second expansion method is used, it is not possible to generate a creative hypothesis for an unobserved range for each item. For example, in the example shown in FIG. 11, since there is no observation data for company E, it is not possible to generate development knowledge about company E even when the second expansion method is used.

In order to solve the above problem, as a third expansion method, it is conceivable to generate expansion knowledge including all unobserved data. The number of expanded knowledge generated in this case is a combination of the number expected for each item for knowledge 1 (the number of all mail_address_senders x the number of all mail_address_receivers) and the combination of the number expected for each item for knowledge 2. It can be calculated by (the number of all maililds x the number of all competitors x the number of all meets x the number of all changes).

For example, in the example shown in FIG. 11, the number assumed for the mail_address_sender of knowledge 1 is 5, and the number assumed for the mail_address_receiver of knowledge 2 is also 5. Therefore, the number of expanded knowledge is 25 (= 5 × 5). If it is preferable to exclude transmission to the same company, the number of development knowledge is 20 (= 25-5).

However, in the case of the method of grounding all possible data such as the third expansion method, the number of combinations increases rapidly as the number of combinations of data candidates and predicates increases, which is not realistic. ..

By using the method described in Patent Document 1, if the rules (knowledge) can be reduced, the amount of developed knowledge generated can also be suppressed. However, even if the rules (knowledge) can be reduced, there is still a problem that a combinatorial explosion occurs in the method of performing full expansion such as the third expansion method.

Therefore, an object of the present invention is to provide an expansion knowledge generation system, an expansion knowledge generation method, and an expansion knowledge generation program capable of generating expansion knowledge capable of generating creative hypotheses while suppressing combinatorial explosion. ..

The expansion knowledge generation system according to the present invention includes an expansion knowledge generation unit that generates expansion knowledge by substituting observed data or unobserved data into knowledge, and the expansion knowledge generation unit is based on the unobserved data assigned to knowledge. The expanded knowledge is generated so that the calculated index satisfies the evaluation criteria, and the unobserved data is substituted into the knowledge to generate the expanded knowledge. The feature is that it is generated so as to be equal to or less than the threshold value.

The expansion knowledge generation method according to the present invention is based on the unobserved data assigned to the knowledge when the information processing apparatus substitutes the observed data or the unobserved data into the knowledge to generate the expansion knowledge and generates the expansion knowledge. The calculated index generates expansion knowledge so as to satisfy the evaluation criteria, and the expansion knowledge generated by substituting the unobserved data into the knowledge is the expansion number, which is the number of unobserved data included in the expansion knowledge. It is characterized in that it is generated so as to be equal to or less than the threshold value.

The expanded knowledge generation program according to the present invention causes a computer to execute an expanded knowledge generation process of substituting observed data or unobserved data into knowledge to generate expanded knowledge, and the unobserved data assigned to the knowledge in the expanded knowledge generation process. The index calculated based on the data generates the expansion knowledge so as to satisfy the evaluation criteria, and the expansion knowledge generated by substituting the unobserved data into the knowledge is calculated by the number of unobserved data included in the expansion knowledge. It is generated so that the number of expansion is less than the threshold value, characterized in Rukoto.

According to the present invention, it is possible to generate developmental knowledge capable of generating creative hypotheses while suppressing combinatorial explosions.

It is a block diagram which shows the structural example of the 1st Embodiment of the development knowledge generation system by this invention. It is a flowchart which shows the operation example of the development knowledge generation system 100 of 1st Embodiment. It is a block diagram which shows the structural example of the 2nd Embodiment of the development knowledge generation system by this invention. It is a flowchart which shows the operation example of the development knowledge generation system 200 of 2nd Embodiment. It is a block diagram which shows the structural example of the 3rd Embodiment of the development knowledge generation system by this invention. It is a flowchart which shows the operation example of the development knowledge generation system 300 of 3rd Embodiment. It is a block diagram which shows the structural example of the 4th Embodiment of the development knowledge generation system by this invention. It is a flowchart which shows the operation example of the development knowledge generation system 400 of 4th Embodiment. It is a block diagram which shows the outline of the development knowledge generation system by this invention. It is a schematic block diagram which shows the structure of the computer which concerns on at least one Embodiment. It is explanatory drawing which shows the example of the observation data of mail transmission / reception.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the following explanation, the goal is to perform so-called pruning of unfolded knowledge within a range that does not significantly lose the optimum (range of approximation with a certain degree of accuracy reduced). In the following description, knowledge means a formula that combines observable predicates (or items) with logical symbols. The predicate represents each event and contains one or more elements to which a specific value is assigned. The value assigned here is either the actually observed data (hereinafter, observed data) or the data that can be taken as a value but is not actually observed (hereinafter, unobserved data).

In the case of the above example, the observation data is, for example, all mail data between an employee of a certain company and an employee of another company. The mail data includes, for example, the recipient's mail address (To, Cc, Bcc), the date and time of transmission, the text, and the like.

Further, the list for each item (predicate) including unobserved data is simply referred to as an item list or an item-by-item list. In addition, the data included in the item list may be referred to as item list data or items. For example, in the case of sender and recipient item lists, the item list data includes, for example, the email addresses of all employees and all other companies known to interact with each employee.

Embodiment 1.
FIG. 1 is a block diagram showing a configuration example of a first embodiment of the deployment knowledge generation system according to the present invention. The expansion knowledge generation system 100 of the present embodiment includes an expansion knowledge generation unit 10a, a knowledge data storage unit 21, an item-by-item list data storage unit 22, an observation data storage unit 23, and an inference unit 30. The expansion knowledge generation unit 10a includes an input unit 11, a combination pattern generation unit 12, an item-by-item closed world list generation unit 13, and a knowledge development unit 14.

The knowledge data storage unit 21 stores the knowledge used when generating the expanded knowledge.

The item-by-item list data storage unit 22 stores an item list of each item included in the knowledge.

The observation data storage unit 23 stores the observation data. The observation data storage unit 23 may store observation data of items not included in the knowledge.

The input unit 11 inputs the evaluation criteria that the generated development knowledge should satisfy. In the present embodiment, the input unit 11 inputs the number of unobserved data included in the development knowledge as an evaluation standard. In the following description, the number of unobserved data included in the expansion knowledge is referred to as the expansion number or the flip number. For example, the fact that the expansion knowledge includes N unobserved data is referred to as N flip.

The number of flips to be input is determined in consideration of the number of development knowledge generated, the calculation time, and the like. If the number of flips becomes too large, a combinatorial explosion will occur. Therefore, the number of flips is preferably about 1 to 3.

The combination pattern generation unit 12 generates a combination pattern of unobserved data and observed data to be substituted for the knowledge when generating the development knowledge.

For example, suppose that a combination pattern of knowledge 1 and knowledge 2 shown above is generated. When N = 1 (that is, in the case of 1 flip), the combination pattern generation unit 12 generates two combination patterns (knowledge 1a and knowledge 1b) illustrated below from knowledge 1.

In addition, the item in which "[Up to one item without observation is OK]" indicates that the item is expanded by substituting the variable (data) included in the list data for each item. In other words, this item is an item to be expanded by substituting either unobserved data or observed data. On the other hand, the items not described indicate that the items are expanded by substituting only the variables (data) existing in the observation data.

Knowledge 1a: [Up to 1 without observation is OK] mail_address_sender (x1, m1) ^
mail_address_receiver (x2, m1)
=> competitor (x1, x2)
Knowledge 1b: mail_address_sender (x1, m1) ^
[Up to 1 without observation is OK] mail_address_receiver (x2, m1)
=> competitor (x1, x2)

Similarly, the combination pattern generation unit 12 generates four combination patterns (knowledge 2a, knowledge 2b, knowledge 2c, and knowledge 2d) illustrated below from knowledge 2.

Knowledge 2a: [Up to 1 without observation] mailild (e1, m1) ^ competitor (x1, x2) ^
meet (e1, x1, x2) ^ change (e2, x1, ”price”)
=> query (x1, e1, ”cartel”)
Knowledge 2b: mailild (e1, m1) ^ [Up to 1 without observation is OK] competitor (x1, x2) ^
meet (e1, x1, x2) ^ change (e2, x1, ”price”)
=> query (x1, e1, ”cartel”)
Knowledge 2c: mailild (e1, m1) ^ competitor (x1, x2) ^
[Up to 1 without observation] meet (e1, x1, x2) ^ change (e2, x1, "price")
=> query (x1, e1, ”cartel”)
Knowledge 2d: mailild (e1, m1) ^ competitor (x1, x2) ^
meet (e1, x1, x2) ^ [up to 1 without observation] change (e2, x1, ”price”)
=> query (x1, e1, ”cartel”)

The same applies to the case of one flip or more. Hereinafter, in order to simplify the explanation, it is assumed that the knowledge X: knowledge X: A ^ B ^ C ^ D → E exists. For example, a two-flip combination pattern is generated as follows. The same applies to the third and subsequent flips.

Knowledge Xa: [Up to 1 without observation is OK] A ^ [Up to 1 without observation is OK] B ^ C ^ D → E
Knowledge Xb: [Up to 1 without observation is OK] A ^ B ^ [Up to 1 without observation is OK] C ^ D → E
Knowledge Xc: [Up to 1 without observation is OK] A ^ B ^ C ^ [Up to 1 without observation is OK] D → E
Knowledge Xd: A ^ [No observation up to 1 OK] B ^ [No observation up to 1 OK] C ^ D → E
Knowledge Xe: A ^ [No observation up to 1 OK] B ^ C ^ [No observation up to 1 OK] D → E
Knowledge Xf: A ^ B ^ [No observation up to 1 OK] C ^ [No observation up to 1 OK] D → E

The item-by-item closed-world list generation unit 13 extracts and lists the variables existing in the observation data for each item. For example, when the observation data is mail data, the item-by-item closed-world list generation unit 13 extracts the sender and the receiver and generates a list. Hereinafter, the list generated by the item-by-item closed-world list generation unit 13 may be referred to as an item-by-item closed-world list.

The knowledge development unit 14 generates development knowledge so that the index calculated based on the unobserved data substituted for the knowledge satisfies the evaluation criteria provided in advance. In the present embodiment, the knowledge expansion unit 14 generates the expansion knowledge so that the number of unobserved data (that is, the expansion number) included in the expansion knowledge is equal to or less than the threshold value. This threshold value is an evaluation standard input by the input unit 11.

Specifically, the knowledge development unit 14 substitutes the observed data or the unobserved data into the combination pattern generated by the combination pattern generation unit 12. In the case of the above example, the knowledge development unit 14 expands the item in which [up to one item without observation is OK] with the variable of the corresponding item stored in the item-by-item list data storage unit 22. Further, the knowledge development unit 14 expands the items for which [up to one item without observation is OK] is expanded by the variables included in the list generated by the item-by-item closed-world list generation unit 13.

For example, when N = 1, when generating the expanded knowledge for the above knowledge 1, (the number of observed mail_address_senders x the number of all mail_address_receivers) and (the number of all mail_address_senders x the number of observed mail_address_receivers) were totaled. Number expansion knowledge is generated.

The reasoning unit 30 makes inferences based on the generated development knowledge and generates a creative hypothesis. At that time, the inference unit 30 outputs the probability of occurrence of the generated expansion knowledge. Since the method of executing the inference based on the inference generated by the inference unit 30 and the method of calculating the probability of occurrence of the inference knowledge are widely known, detailed description thereof will be omitted here.

The input unit 11, the combination pattern generation unit 12, the item-by-item closed-world list generation unit 13, the knowledge development unit 14, and the inference unit 30 are realized by a computer CPU that operates according to a program (expansion knowledge generation program). To. For example, the program is stored in a storage unit (not shown) included in the expansion knowledge generation system 100, and the CPU reads the program and generates an input unit 11, a combination pattern generation unit 12, and a closed world list for each item according to the program. It may operate as a unit 13, a knowledge development unit 14, and an inference unit 30.

Further, the input unit 11, the combination pattern generation unit 12, the item-by-item closed-world list generation unit 13, the knowledge development unit 14, and the inference unit 30 may be realized by dedicated hardware, respectively. Further, the knowledge data storage unit 21, the item-by-item list data storage unit 22, and the observation data storage unit 23 are realized by, for example, a magnetic disk or the like. Each data may be input by, for example, the input unit 11, and the combination pattern generation unit 12 and the item-by-item closed world list generation unit 13 may, as necessary, the knowledge data storage unit 21, the item-by-item list data storage unit 22, and the item-by-item list data storage unit 22. Each may be acquired from the observation data storage unit 23.

Next, the operation of the deployment knowledge generation system 100 of this embodiment will be described. FIG. 2 is a flowchart showing an operation example of the deployment knowledge generation system 100 of the present embodiment.

First, the input unit 11 inputs the number of flips N, knowledge data, item-by-item list data, and observation data (step S11). The combination pattern generation unit 12 generates a knowledge combination pattern (step S12). The item-by-item closed-world list generation unit 13 generates an item-by-item closed-world list (step S13). The order of steps S12 and S13 may be reversed.

The knowledge development unit 14 generates development knowledge (step S14) and outputs it to the inference unit 30 (step S15). After that, the inference unit 30 makes an inference using the expansion knowledge.

As described above, according to the present embodiment, the knowledge expansion unit 14 uses the expanded knowledge generated by substituting the unobserved data into the knowledge, and the number of unobserved data (expanded number) included in the expanded knowledge. Generate so that it is equal to or less than the threshold N. Therefore, it is possible to generate developmental knowledge that enables creative hypothesis generation while suppressing combinatorial explosion.

Embodiment 2.
Next, a second embodiment of the development knowledge generation system according to the present invention will be described. In the first embodiment, expansion knowledge based on the number of expansions is generated for each knowledge. In addition, there may be a relationship in which the first knowledge uses the second knowledge (hereinafter referred to as a parent-child relationship of knowledge) among a plurality of knowledges. When N ≧ 1, rather than calculating the number of expansions for each knowledge, the second knowledge is expanded to the first knowledge, and complex new knowledge (hereinafter referred to as compound knowledge) is developed. It is preferable to generate and develop based on the combined knowledge. Therefore, in the present embodiment, a method of generating development knowledge using knowledge having a parent-child relationship will be described.

FIG. 3 is a block diagram showing a configuration example of a second embodiment of the deployment knowledge generation system according to the present invention. The same configurations as those in the first embodiment are designated by the same reference numerals as those in FIG. 1, and detailed description thereof will be omitted. The expanded knowledge generation system 200 of the present embodiment includes an expanded knowledge generation unit 10b, a knowledge data storage unit 21, an item-by-item list data storage unit 22, an observation data storage unit 23, and an inference unit 30. The expansion knowledge generation unit 10b includes an input unit 11, a combination pattern generation unit 15, an item-by-item closed world list generation unit 13, and a knowledge development unit 14.

Similar to the first embodiment, the combination pattern generation unit 15 generates a combination pattern of unobserved data and observed data to be substituted for the knowledge when generating the development knowledge. Further, the combination pattern generation unit 15 of the present embodiment extracts the parent-child relationship of the above-mentioned knowledge. That is, the combination pattern generation unit 15 operates as a parent-child relationship extraction unit that extracts the parent-child relationship of knowledge.

Then, the combination pattern generation unit 15 develops the second knowledge into the first knowledge to generate the compound knowledge based on the extracted parent-child relationship, and generates the combination pattern of the generated compound knowledge.

For example, in the case of knowledge 1 and knowledge 2, the competitors (x1, x2) included in knowledge 1 are used in knowledge 2. Therefore, it can be said that Knowledge 1 and Knowledge 2 have a parent-child relationship. In this case, the combination pattern generation unit 15 can expand the knowledge 1 into the knowledge 2 and generate the composite knowledge shown below. In some cases, the ability to acquire the contents directly from the observation data is referred to as the primary observation, and the ability to acquire the contents using knowledge is referred to as the secondary observation.

Combined knowledge: mailild (e1, m1) ^ {mail_address_sender (x1, m1) ^
mail_address_receiver (x2, m1)} ^ meet (e1, x1, x2) ^ change (e2, x1, ”price”)
=> query (x1, e1, ”cartel”)

For example, when the number of flips is N, at least 5-N items will be expanded in the observation data. In this way, it can be said that the combination pattern generation unit 15 recursively generates compound knowledge based on the parent-child relationship and calculates the number of expansions.

The input unit 11, the combination pattern generation unit 15, the item-by-item closed-world list generation unit 13, the knowledge development unit 14, and the inference unit 30 are realized by a computer CPU that operates according to a program (expansion knowledge generation program). To.

Next, the operation of the deployment knowledge generation system 200 of this embodiment will be described. FIG. 4 is a flowchart showing an operation example of the deployment knowledge generation system 200 of the present embodiment. First, the input unit 11 inputs the number of flips N, a plurality of knowledge data, item-by-item list data, and observation data (step S21). The combination pattern generation unit 15 extracts the parent-child relationship of knowledge and generates complex knowledge (step S22). Then, the combination pattern generation unit 15 generates a combination pattern of the generated composite knowledge (step S23).

After that, the processes from step S13 to step S15 from generating the closed world list for each item to generating the expansion knowledge are the same as the processes shown in FIG.

As described above, according to the present embodiment, the combination pattern generation unit 15 extracts the parent-child relationship, and the knowledge development unit 14 recursively calculates the number of expansions based on the parent-child relationship. Therefore, in addition to the effect of the first embodiment, the development knowledge can be generated more appropriately.

Next, a modification of the first embodiment and the second embodiment will be described. In the first embodiment and the second embodiment, a method of generating expansion knowledge by fixing the number of flips N input to the input unit 11 has been described. In general, the amount of calculation required for inference by the inference unit 30 is larger than the generation of expansion knowledge. Therefore, it is preferable that the number of expanded knowledge can be generated based on the number of expanded knowledge used by the inference unit 30 for inference and the calculation time of the expanded knowledge.

Therefore, the input unit 11 may input a desired expansion knowledge number M instead of the expansion number (flip number N). Then, the combination pattern generation unit 12 or the combination pattern generation unit 15 sequentially creates a combination pattern in which the number of flips is increased by 1 from 0, and the knowledge development unit 14 generates expansion knowledge using the generated combination pattern. Then, the generation of the expansion knowledge may be repeated until the number of expansion knowledge generated exceeds the input expansion knowledge number M.

Specifically, the combination pattern generation unit 12 or the combination pattern generation unit 15 first generates a combination pattern with a flip number of 0 (no expansion based on unobserved data). The knowledge development unit 14 generates development knowledge using the generated combination pattern. At that time, when the generated expansion knowledge exceeds the input expansion knowledge number M, the knowledge expansion unit 14 ends the process there. On the other hand, when the generated expansion knowledge does not exceed the input expansion knowledge number M, the combination pattern generation unit 12 or the combination pattern generation unit 15 first increases the number of flips by 1 (that is, the number of flips is 1). Generate a combination pattern. The knowledge development unit 14 generates development knowledge using a combination pattern in which the number of flips is increased by 1. After that, the above process is repeated until the number of expanded knowledge cases M is exceeded.

Further, the input unit 11 may input a desired calculation time instead of the desired number of development knowledge cases M. In this case, the combination pattern generation unit 12, the combination pattern generation unit 15, and the knowledge development unit 14 may generate the development knowledge until a desired calculation time elapses.

That is, in this modification, the combination pattern generation unit 12, the combination pattern generation unit 15, and the knowledge expansion unit 14 change the threshold value of the expansion number until the specified expansion knowledge number or calculation time (specifically, the combination pattern generation unit 12 or the combination pattern generation unit 15 and the knowledge expansion unit 14). (Increase the threshold of the number of expansions by 1 from 0), and repeat the generation of expansion knowledge. With such a configuration, it is possible to generate a desired number of development knowledge. That is, according to this modification, a so-called anytime algorithm (an algorithm that can stop at an arbitrary time and output a solution with a certain accuracy at the time of stopping) can be realized, so that grounding is performed according to a predetermined number of cases and calculation time. You can manipulate the amount.

Embodiment 3.
Next, a third embodiment of the development knowledge generation system according to the present invention will be described. FIG. 5 is a block diagram showing a configuration example of a third embodiment of the deployment knowledge generation system according to the present invention. The same configurations as those in the first embodiment are designated by the same reference numerals as those in FIG. 1, and detailed description thereof will be omitted. The expansion knowledge generation system 300 of the present embodiment includes an expansion knowledge generation unit 10c, a knowledge data storage unit 21, an item-by-item list data storage unit 22, an observation data storage unit 23, and an inference unit 30. The expansion knowledge generation unit 10c includes an input unit 11, an occurrence probability information generation unit 16, an expansion knowledge establishment probability calculation unit 17, and an expansion knowledge selection unit 18.

In the present embodiment, the input unit 11 inputs the knowledge establishment probability calculated based on the establishment probability of each item (hereinafter, referred to as the knowledge establishment probability) as an evaluation standard. The calculation method of the establishment probability will be described later.

The generation probability information generation unit 16 generates the establishment probability of each item from the observation data and the item-by-item list data. Specifically, the occurrence probability information generation unit 16 calculates the occurrence probability of the data of each item based on the frequency with which the data included in the item list is included in the observation data, and the establishment of each item based on the occurrence probability. Calculate the probability.

The probability of establishment of each item is generated based on the probability of occurrence of the primary observation. For example, for the item of primary observation, the occurrence probability information generation unit 16 sets the occurrence probability of the data existing in the observation data to 1, and sets the occurrence probability of the data existing in the list data for each item but not in the observation data. It may be set to m (for example, 0.05). However, the value of m does not have to be set uniformly and depends on the content of the observation data. From this, it can be said that the probability of establishment is an index calculated at least based on unobserved data.

For example, when the observation data covers 90% of all the target data, the occurrence probability information generation unit 16 can make the value of m smaller than the case where the observation data covers 50%. Moreover, when the data of a certain department is small, m cannot be reduced even if there is no observation data about the department. Further, the occurrence probability information generation unit 16 may statistically determine the value of m.

Regarding the items of the secondary observation, the occurrence probability information generation unit 16 may manually set the establishment probability of each knowledge. For example, when the knowledge 1 and the knowledge 2 exist, the occurrence probability information generation unit 16 may set the knowledge establishment probability of the knowledge 1 to 0.5, the knowledge establishment probability of the knowledge 2 to 0.95, and the like. Further, for example, when the probability that the relationship between the sender and the receiver becomes a conflict is low, the probability of the expanded knowledge expanded by substituting a variable for knowledge 1 is a smaller value than 1 (for example, 0.4). It may be.

In addition, the probability of establishment may be statistically calculated in advance. For example, if there is teacher data indicating the relationship between the mail data and the presence or absence of conflict, the probability of establishment may be calculated statistically from the data. For example, when the observation data of the mail data and the competition relationship data exists for the knowledge 1, the occurrence probability information generation unit 16 calculates the establishment probability by calculating the ratio of the mails of the senders and receivers in the competition relationship. You may. The calculation method of the probability of establishment is the same for the third and subsequent observations.

The expansion knowledge establishment probability calculation unit 17 calculates the expansion knowledge establishment probability based on the occurrence probability of each item. The probability of occurrence of each item includes the probability of occurrence of primary observation and the probability of establishment of each knowledge. The expansion knowledge establishment probability calculation unit 17 may calculate the knowledge establishment probability by, for example, multiplying the term of the secondary observation on the right side by the occurrence probability of the primary observation on the left side and the establishment probability of the knowledge. Specifically, the expansion knowledge establishment probability calculation unit 17 may calculate the knowledge establishment probability based on the following equation 1.

Π _i p _i (Equation 1)

For example, regarding the above knowledge 2, the probability of establishment of each item is mailild (e1, m1): probability 1, competitor (x1, x2): probability 0.9, meet (e1, x1, x2): probability 0.7, respectively. , And change (e2, x1, "price"): Suppose the probability is 0.6. In this case, the expansion knowledge establishment probability calculation unit 17 may calculate the establishment probability as 1 × 0.9 × 0.7 × 0.6 = 0.378 by multiplying the probabilities of each knowledge.

Further, the expansion knowledge establishment probability calculation unit 17 may calculate the knowledge establishment probability based on the average of the establishment probabilities. Specifically, the expansion knowledge establishment probability calculation unit 17 may calculate the knowledge establishment probability based on the following equation 2. In Equation 2, I is the number of items on the left side.

^{_{{Π i p i} 1 /}} I ( Equation 2)

The probability of each item may differ depending on the value of the variable to be specifically assigned. For example, the competitor included in the above knowledge 2 is included in the right side of the knowledge 1. Then, for example, it is assumed that the probability of the competitor (A, B) is 0.95 and the probability of the competitor (C, D) is 0.5. In this case, the probability of establishment of the development knowledge of knowledge 2 may be calculated differently depending on the variables.

The expansion knowledge selection unit 18 selects expansion knowledge whose knowledge establishment probability calculated by the expansion knowledge establishment probability calculation unit 17 is equal to or higher than this threshold value, using the establishment probability input to the input unit 11 as an evaluation criterion as a threshold value. In other words, since the establishment probability based on the observed data and the unobserved data is set for each item constituting the knowledge, the expansion knowledge selection unit 18 generates the expansion knowledge so that the establishment probability becomes equal to or more than the threshold value. To do.

For example, it is assumed that the probability P = 0.4 is input as the threshold value of the establishment probability. As in the above example, when the probability of establishment = 0.378 is calculated, it falls below the threshold value, so the expansion knowledge selection unit 18 determines that this knowledge is not expanded.

On the other hand, when it is determined that the expansion knowledge is generated, the expansion knowledge selection unit 18 expands the knowledge with unobserved data of a predetermined number of flips (the number of flips is a natural number). For example, when the number of flips is set to 1 for the knowledge 2 determined to be expanded, the expansion knowledge selection unit 18 acquires the expansion knowledge based on the patterns shown by the knowledge 2a to 2d described in the first embodiment. Just generate it.

Further, for example, as in the above example, it is assumed that the probability of the competitor (A, B) is 0.98 and the probability of the competitor (C, D) is 0.5, depending on the variables. At this time, since the probability of establishment of the former is calculated to be 0.4116, the expansion knowledge selection unit 18 may determine that this expansion knowledge is selected. On the other hand, since the probability of establishment of the latter is calculated to be 0.21, the expansion knowledge selection unit 18 may determine that this expansion knowledge is not selected.

The input unit 11, the generation probability information generation unit 16, the expansion knowledge establishment probability calculation unit 17, and the expansion knowledge selection unit 18 are realized by a computer CPU that operates according to a program (expansion knowledge generation program).

Next, the operation of the deployment knowledge generation system of this embodiment will be described. FIG. 6 is a flowchart showing an operation example of the deployment knowledge generation system 300 of the present embodiment.

First, the input unit 11 inputs the establishment probability P, knowledge data, item-by-item list data, and observation data (step S31). The generation probability information generation unit 16 generates the establishment probability for each item (step S32). The expansion knowledge establishment probability calculation unit 17 calculates the expansion knowledge establishment probability (knowledge establishment probability) (step S33). The expansion knowledge selection unit 18 selects the expansion knowledge whose knowledge establishment probability is equal to or higher than the establishment probability P (step S34), and outputs the selected expansion knowledge to the inference unit 30 (step S35). After that, the inference unit 30 makes an inference using the expansion knowledge.

As described above, according to the present embodiment, the establishment probability based on the observed data and the unobserved data is set for each item constituting the knowledge, and the expansion knowledge selection unit 18 is based on the establishment probability. Generate a knowledge development probability that the calculated probability is equal to or greater than the threshold P. Therefore, as in the first embodiment, it is possible to generate developmental knowledge capable of generating creative hypotheses while suppressing combinatorial explosions.

Next, a modified example of this embodiment will be described. In the above embodiment, a method of generating expanded knowledge with a predetermined number of flips when the knowledge establishment probability satisfies a threshold value has been described. The expansion knowledge selection unit 18 may determine the number of flips according to the knowledge establishment probability. In other words, the establishment probability P may be determined according to the number of flips, and the expansion knowledge selection unit 18 may generate expansion knowledge so that the number of flips is equal to or less than the threshold value and the establishment probability P is equal to or greater than the threshold value.

For example, the expansion knowledge selection unit 18 does not expand when the probability of establishment of knowledge X: A ^ B ^ C ^ D is _{less than P 0} , and expands with one flip when it is _{P 0} or more and _{less than P 1.} In _{the case of P 1} or more, it may be decided to develop with 2 flips. The same applies to more cases.

Further, the expansion knowledge selection unit 18 may expand the knowledge one flip at a time and decide whether to further increase the number of flips according to the probability of establishment of the expanded knowledge. For example, the expansion knowledge selection unit 18 determines that the knowledge is not expanded when the probability of establishment of the knowledge X: A ^ B ^ C ^ D is less than P, and determines that the knowledge is expanded in one flip when it is P or more. In this case, it is expanded to the following four knowledges.

Knowledge Xg: [Up to 1 without observation] A ^ B ^ C ^ D → E
Knowledge Xh: A ^ [Up to 1 is OK without observation] B ^ C ^ D → E
Knowledge Xi: A ^ B ^ [Up to 1 is OK without observation] C ^ D → E
Knowledge Xj: A ^ B ^ C ^ [Up to 1 is OK without observation] D → E

Further, the occurrence probability information generation unit 16 is [OK up to 1 without observation] A, [OK up to 1 without observation] B, [OK up to 1 without observation] C and [OK up to 1 without observation]. ] D, the probability of establishment is generated. Then, the expansion knowledge establishment probability calculation unit 17 calculates the establishment probability of each expansion knowledge. Of the establishment probabilities, for example, when a predetermined number of establishment probabilities are less than the establishment probability P, the expansion knowledge selection unit 18 determines that no further expansion knowledge is generated. On the other hand, when the probability of establishment of a predetermined number is equal to or greater than the probability of establishment P, the expansion knowledge selection unit 18 determines to generate expansion knowledge in which the number of flips is further increased.

Next, another modification of the present embodiment will be described. The expansion knowledge generation system of the present embodiment may also generate expansion knowledge in consideration of the parent-child relationship of knowledge, as in the second embodiment. Specifically, the expansion knowledge establishment probability calculation unit 17 may extract the parent-child relationship of knowledge in the same manner as the combination pattern generation unit 15 of the second embodiment. That is, the expansion knowledge establishment probability calculation unit 17 may operate as a parent-child relationship extraction unit that extracts the parent-child relationship of knowledge.

Then, the expansion knowledge establishment probability calculation unit 17 may reflect the establishment probability calculated on the knowledge side of the child on the knowledge side of the parent. For example, in the case of the above knowledge 1 and knowledge 2, the knowledge establishment probability of knowledge 2 may be calculated by reflecting the knowledge establishment probability calculated by knowledge 1 in the establishment probability of knowledge 2. In this way, it can be said that the expansion knowledge establishment probability calculation unit 17 recursively calculates the establishment probability based on the parent-child relationship. With such a configuration, in addition to the effect of the third embodiment, it becomes possible to make a more precise comparison with the threshold value of the establishment probability.

Embodiment 4.
Next, a fourth embodiment of the development knowledge generation system according to the present invention will be described. In the present embodiment, a method of updating the probability of establishment of each item by feeding back the result of the inference unit 30 will be described.

FIG. 7 is a block diagram showing a configuration example of a fourth embodiment of the deployment knowledge generation system according to the present invention. The same components as those in the third embodiment are designated by the same reference numerals as those in FIG. 5, and detailed description thereof will be omitted. The expanded knowledge generation system 400 of the present embodiment includes an expanded knowledge generation unit 10d, a knowledge data storage unit 21, an item-by-item list data storage unit 22, an observation data storage unit 23, and an inference unit 30. The expansion knowledge generation unit 10d includes an input unit 11, an occurrence probability information generation unit 19, an expansion knowledge establishment probability calculation unit 17, and an expansion knowledge selection unit 18.

The inference unit 30 of the present embodiment inputs the occurrence probability of the expansion knowledge to the occurrence probability information generation unit 19. The inference unit 30 outputs, for example, a score (establishment probability) for the following development knowledge.
mail_address_sender (person A, mail1) ^ mail_address_receiver (person B, mail1)
=> competitor (Person A, Person B): Score 0.9

The occurrence probability information generation unit 19 calculates the establishment probability based on the occurrence probability output as a result of the inference, and updates the establishment probability set for each item constituting the knowledge. The method of calculating the probability of establishment of knowledge based on the probability of occurrence of expanded knowledge is the same as the method of calculating the probability of establishment of knowledge by the occurrence probability information generation unit 16 in the third embodiment.

The input unit 11, the generation probability information generation unit 19, the expansion knowledge establishment probability calculation unit 17, and the expansion knowledge selection unit 18 are realized by a computer CPU that operates according to a program (expansion knowledge generation program).

Next, the operation of the deployment knowledge generation system of this embodiment will be described. FIG. 8 is a flowchart showing an operation example of the deployment knowledge generation system 400 of the present embodiment. The process of calculating the establishment probability for each item, selecting the expansion knowledge, and outputting it to the inference unit 30 is the same as the process from step S31 to step S35 illustrated in FIG.

The inference unit 30 executes inference (step S36), and inputs the inference result to the occurrence probability information generation unit 19. The generation probability information generation unit 19 updates the establishment probability based on the inference result, and repeats the processing after step S32.

As described above, in the present embodiment, the inference unit 30 executes inference based on the generated expansion knowledge, and the occurrence probability information generation unit 19 for each item constituting the knowledge based on the inference. Update the set probability of establishment. Therefore, in addition to the effect of the third embodiment, it is possible to generate development knowledge that reflects the inference result.

Next, a modification of the third embodiment and the fourth embodiment will be described. In the third embodiment and the fourth embodiment, a method of generating expansion knowledge by fixing the establishment probability P input to the input unit 11 has been described. As described in the first embodiment and the modified examples of the second embodiment, it is preferable that the number of expanded knowledge can be generated based on the number of expanded knowledge used by the inference unit 30 for inference and the calculation time of the expanded knowledge.

Therefore, the input unit 11 may input the desired number of development knowledge cases M instead of the establishment probability P. Then, the expansion knowledge selection unit 18 gradually decreases the probability from the establishment probability P = 1 to increase the expansion knowledge to be selected, until the number of expansion knowledge selected exceeds the input expansion knowledge number M. , The selection of development knowledge may be repeated.

Specifically, the expansion knowledge selection unit 18 first selects expansion knowledge that satisfies the establishment probability P = 1. At that time, when the selected expansion knowledge exceeds the input expansion knowledge number M, the expansion knowledge selection unit 18 ends the process there. On the other hand, when the selected expansion knowledge does not exceed the input expansion knowledge number M, the expansion knowledge selection unit 18 reduces the establishment probability P by a predetermined value or rate. The expansion knowledge selection unit 18 selects expansion knowledge that satisfies the reduced probability of establishment P. After that, the above process is repeated until the number of expanded knowledge cases M is exceeded.

Further, the input unit 11 may input a desired calculation time instead of the desired number of development knowledge cases M. In this case, the expansion knowledge selection unit 18 may select the expansion knowledge until the desired calculation time elapses.

That is, in this modification, the expansion knowledge selection unit 18 changes the threshold of the establishment probability until the specified number of expansion knowledge or the calculation time (specifically, the threshold of the establishment probability is reduced from 1). , Repeat the generation of development knowledge. With such a configuration, it is possible to generate a desired number of development knowledge. That is, according to this modification, the grounding amount can be manipulated according to a predetermined number of cases and calculation time, as in the modification of the first embodiment and the second embodiment.

Next, the outline of the present invention will be described. FIG. 9 is a block diagram showing an outline of the development knowledge generation system according to the present invention. The expansion knowledge generation system 80 according to the present invention includes an expansion knowledge generation unit 81 that generates expansion knowledge by substituting observed data or unobserved data into knowledge. In the expansion knowledge generation unit 81, the index (for example, the number of unobserved data, the probability of establishment) calculated based on the unobserved data assigned to the knowledge satisfies the evaluation criteria (for example, the number of flips N, the probability of establishment P). To generate expansion knowledge.

The expansion knowledge generation unit 81 is the expansion knowledge generation unit 10a (combination pattern generation unit 12, item-by-item closed world list generation unit 13 and knowledge expansion unit 14) in the first embodiment, and the development in the second embodiment. Knowledge generation unit 10b (combination pattern generation unit 15, closed world list generation unit 13 for each item and knowledge expansion unit 14), expansion knowledge generation unit 10c in the third embodiment (occurrence probability information generation unit 16, expansion knowledge establishment probability calculation) It corresponds to the expansion knowledge generation unit 10d (generation probability information generation unit 19, expansion knowledge establishment probability calculation unit 17 and expansion knowledge selection unit 18) in the fourth embodiment, as well as the expansion knowledge generation unit 17 and the expansion knowledge selection unit 18).

With such a configuration, it is possible to generate developmental knowledge capable of generating creative hypotheses while suppressing combinatorial explosions. That is, grounding can be performed according to the level of creative hypothesis generation. Further, according to the present invention, it is possible to estimate the grounding amount and the feasible creative hypothesis generation for the possessed data (that is, it is possible to judge whether or not it can be solved as a problem).

Specifically, the expansion knowledge generation unit 81 (for example, the expansion knowledge generation unit 10a and the expansion knowledge generation unit 10b) includes the expansion knowledge generated by substituting unobserved data into the knowledge in the expansion knowledge. It may be generated so that the number of expansions, which is the number of unobserved data, is equal to or less than the threshold value.

In addition, the probability of establishment based on observed data and unobserved data may be set for each item constituting the knowledge. In this case, the expansion knowledge generation unit 81 (for example, the expansion knowledge generation unit 10c, the expansion knowledge generation unit 10d) may generate the expansion probability of the knowledge whose probability calculated based on the establishment probability is equal to or more than the threshold value.

Further, the probability of establishment set for each item constituting the knowledge may be determined according to the number of unobserved data included in the expanded knowledge. In this case, the expansion knowledge generation unit 81 may generate expansion knowledge so that the number of expansions is equal to or less than the threshold value and the probability of establishment is equal to or greater than the threshold value.

The expansion knowledge generation unit 81 may repeat the generation of expansion knowledge while changing at least one of the threshold value of the expansion number and the establishment probability until the specified expansion knowledge number or calculation time.

Specifically, the expansion knowledge generation unit 81 may repeat the generation of expansion knowledge by increasing the threshold value of the expansion number from 0 to 1 until the designated expansion knowledge number or calculation time.

Further, the expansion knowledge generation unit 81 may repeat the generation of expansion knowledge by reducing the threshold value of the establishment probability from 1 until the designated number of expansion knowledge or the calculation time.

Further, the expansion knowledge generation system 80 is a parent-child relationship extraction unit that extracts a parent-child relationship of knowledge representing a relationship in which one knowledge uses another knowledge (for example, a combination pattern generation unit 15 and an expansion knowledge establishment probability calculation unit 17). May be provided. Then, the expansion knowledge generation unit 81 may recursively calculate at least the expansion number or the establishment probability based on the parent-child relationship.

Further, the expansion knowledge generation system 80 may include an inference unit (for example, an inference unit 30) that executes inference based on the generated expansion knowledge. Then, the expansion knowledge generation unit 81 may update the establishment probability set for each item constituting the knowledge based on the inference.

FIG. 10 is a schematic block diagram showing the configuration of a computer according to at least one embodiment. The computer 1000 includes a CPU 1001, a main storage device 1002, an auxiliary storage device 1003, and an interface 1004.

Each of the above model generation systems is implemented in the computer 1000. The operation of each processing unit described above is stored in the auxiliary storage device 1003 in the form of a program (expansion knowledge generation program). The CPU 1001 reads a program from the auxiliary storage device 1003, expands it into the main storage device 1002, and executes the above processing according to the program.

In at least one embodiment, the auxiliary storage device 1003 is an example of a non-temporary tangible medium. Other examples of non-temporary tangible media include magnetic disks, magneto-optical disks, CD-ROMs, DVD-ROMs, semiconductor memories, etc. connected via interface 1004. When this program is distributed to the computer 1000 via a communication line, the distributed computer 1000 may expand the program to the main storage device 1002 and execute the above processing.

Further, the program may be for realizing a part of the above-mentioned functions. Further, the program may be a so-called difference file (difference program) that realizes the above-mentioned function in combination with another program already stored in the auxiliary storage device 1003.

Some or all of the above embodiments may also be described, but not limited to:

(Appendix 1) An expansion knowledge generation unit that generates expansion knowledge by substituting observed data or unobserved data into knowledge is provided, and the expansion knowledge generation unit is calculated based on the unobserved data assigned to the knowledge. A deployment knowledge generation system characterized in that an index generates deployment knowledge so as to meet evaluation criteria.

(Appendix 2) The expansion knowledge generation unit generates expansion knowledge generated by substituting unobserved data into the knowledge so that the expansion number, which is the number of unobserved data included in the expansion knowledge, is equal to or less than the threshold value. The development knowledge generation system described in Appendix 1.

(Appendix 3) Knowledge that a probability of establishment based on observed data and unobserved data is set for each item constituting the knowledge, and the development knowledge generation unit has a knowledge that the probability calculated based on the probability of establishment is equal to or higher than the threshold value. The expansion knowledge generation system according to Appendix 1 or Appendix 2 that generates the expansion probability of.

(Appendix 4) The establishment probability set for each item constituting the knowledge is determined according to the number of unobserved data included in the expanded knowledge, and the expanded knowledge generation unit determines the expanded number. The expansion knowledge generation system according to any one of Appendix 1 to Appendix 3, which generates expansion knowledge so that the establishment probability is equal to or less than the threshold value and the establishment probability is equal to or greater than the threshold value.

(Appendix 5) The expansion knowledge generation unit repeats the generation of expansion knowledge while changing at least one of the threshold value of the expansion number and the establishment probability until the specified expansion knowledge number or calculation time. system.

(Appendix 6) The expansion knowledge generation system according to Appendix 2, wherein the expansion knowledge generation unit repeats the generation of expansion knowledge by increasing the threshold value of the expansion number from 0 to 1 until the specified expansion knowledge number or calculation time.

(Appendix 7) The expansion knowledge generation system according to Appendix 3, wherein the expansion knowledge generation unit reduces the threshold value of the establishment probability from 1 and repeats the generation of expansion knowledge until the specified number of expansion knowledge or calculation time.

(Appendix 8) A parent-child relationship extraction unit for extracting a parent-child relationship of knowledge representing a relationship in which one knowledge uses another knowledge is provided, and the expansion knowledge generation unit determines at least the number of expansions or the establishment probability based on the parent-child relationship. The expanded knowledge generation system according to Appendix 4 or Appendix 5 that calculates recursively.

(Appendix 9) An inference unit that executes inference based on the generated expansion knowledge is provided, and the expansion knowledge generation unit updates the establishment probability set for each item constituting the knowledge based on the inference. The development knowledge generation system according to any one of Supplementary note 3 to Supplementary note 6.

(Appendix 10) When the observed data or unobserved data is substituted into the knowledge to generate the expanded knowledge and the expanded knowledge is generated, the index calculated based on the unobserved data substituted into the knowledge is the evaluation standard. A method of generating developmental knowledge, which comprises generating developmental knowledge so as to satisfy.

(Appendix 11) The expansion knowledge described in Appendix 10 for generating the expansion knowledge generated by substituting the unobserved data into the knowledge so that the expansion number, which is the number of unobserved data included in the expansion knowledge, is equal to or less than the threshold value. Generation method.

(Appendix 12) A computer is made to execute an expansion knowledge generation process for generating expansion knowledge by substituting observed data or unobserved data into knowledge, and the expansion knowledge generation process is based on the unobserved data substituted for the knowledge. A development knowledge generation program for generating development knowledge so that the index calculated by the above meets the evaluation criteria.

(Appendix 13) The expansion knowledge generated by substituting unobserved data into the knowledge in the expansion knowledge generation process on the computer is set so that the expansion number, which is the number of unobserved data included in the expansion knowledge, is equal to or less than the threshold value. The development knowledge generation program according to Appendix 12, which is generated in.

10a, 10b, 10c, 10d Expansion knowledge generation unit 11 Input unit 12, 15 Combination pattern generation unit 13 Item-by-item closed world list generation unit 14 Knowledge expansion unit 16, 19 Occurrence probability information generation unit 17 Expansion knowledge establishment probability calculation unit 18 Expansion Knowledge selection unit 21 Knowledge data storage unit 22 Item-by-item list data storage unit 23 Observation data storage unit 100, 200, 300, 400 Expanded knowledge generation system

Claims (9)

Equipped with an expansion knowledge generation unit that generates expansion knowledge by substituting observed data or unobserved data into knowledge.
The development knowledge generation unit
The index calculated based on the unobserved data assigned to the knowledge generates the expanded knowledge so as to satisfy the evaluation criteria .
An expansion knowledge generation system characterized in that expansion knowledge generated by substituting unobserved data into the knowledge is generated so that the number of expansions, which is the number of unobserved data included in the expansion knowledge, is equal to or less than a threshold value. .. For each item that composes knowledge, the probability of establishment based on observed data and unobserved data is set.
Expand knowledge generating unit, development knowledge generation system of claim 1, wherein generating a deployment probability knowledge probability is calculated based on the establishment probability is above a threshold value. The probability of establishment set for each item that composes the knowledge is determined according to the number of unobserved data included in the expanded knowledge.
The expansion knowledge generation system according to claim 1 or 2 , wherein the expansion knowledge generation unit generates expansion knowledge so that the expansion number is equal to or less than a threshold value and the establishment probability is equal to or more than a threshold value. The expansion knowledge generation system according to claim 3 , wherein the expansion knowledge generation unit repeats generation of expansion knowledge while changing at least one of the threshold value of the expansion number and the establishment probability until the specified expansion knowledge number or calculation time. The expansion knowledge generation system according to claim 1 , wherein the expansion knowledge generation unit increases the threshold value of the expansion number by 1 from 0 to 1 until the specified expansion knowledge number or calculation time, and repeats the generation of the expansion knowledge. The expansion knowledge generation system according to claim 2 , wherein the expansion knowledge generation unit reduces the threshold value of the establishment probability from 1 until the specified number of expansion knowledge or the calculation time, and repeats the generation of the expansion knowledge. It is equipped with a parent-child relationship extraction unit that extracts the parent-child relationship of knowledge that represents the relationship in which one knowledge uses the other knowledge.
The expansion knowledge generation system according to claim 3 or 4 , wherein the expansion knowledge generation unit recursively calculates at least the expansion number or the establishment probability based on the parent-child relationship. Information processing device
Substituting observed or unobserved data into knowledge to generate expanded knowledge,
When generating the expansion knowledge, the expansion knowledge is generated so that the index calculated based on the unobserved data substituted for the knowledge satisfies the evaluation criteria .
A method for generating expansion knowledge, characterized in that expansion knowledge generated by substituting unobserved data into the knowledge is generated so that the number of expansions, which is the number of unobserved data included in the expansion knowledge, is equal to or less than a threshold value. .. On the computer
Execute the expansion knowledge generation process to generate expansion knowledge by substituting the observed data or unobserved data into the knowledge.
In the expansion knowledge generation process,
The index calculated based on the unobserved data assigned to the knowledge generates the expanded knowledge so as to satisfy the evaluation criteria .
An expansion knowledge generation program for generating expansion knowledge generated by substituting unobserved data into the knowledge so that the expansion number, which is the number of unobserved data included in the expansion knowledge, is equal to or less than a threshold value.

Images