Learning/thinking machine and learning/thinking method based on structured knowledge, computer system, and information generation method

A learning/thinking machine based on a structured knowledge comprising:

• means for extracting a semantic relation from collected information in accordance with a plurality of rules;
• means for analyzing information based on the extracted semantic relation and in which a link indicates a semantic relation between a node and a node for indicating meaning so as to sufficiently express a semantic means, said node and said link have structure so as to exchange their roles and a knowledge expressed by said node and said link is stored;
• means for generating new information according to predetermined inference by verifying a knowledge based in which knowledge structured based on said semantic relation is stored in response to an inquiry or request with a unit in which a relating node is stored;
• means for evaluating generated new information;
• means for judging sequencing of evaluated results;
• means for deciding an optimal solution based on a judged result;
• an input unit for receiving an inquiry or request from the outside;
• and
• verifying means for verifying said inquiry or request from the outside with said knowledge base, wherein if it is determined by the verified result of said verifying means that said inquiry or request are completely coincident with said knowledge base, then understanding of such coincidence is transmitted to a central unit, if it is determined by the verified result of said verifying means that said inquiry or request are partly coincident with said knowledge base, then said knowledge based and said unit in which said relating node is stored are verified and said information generating means supplements insufficient information and generates information so that knowledge structured by said node and said link based on said semantic relation has new semantic content and new semantic relation.

A learning/thinking method based on a structured knowledge comprising:

• a knowledge input step for inputting inclusively collected data, information and knowledge;
• a knowledge structuring step in which a semantic relation is extracted from said inputted data, information and knowledge in accordance with a plurality of rules, meaning of information is analyzed based on said extracted semantic relation, a link indicates a semantic relation between a node and a node for indicating a meaning, said node and said link have structures so as to exchange their roles and structured knowledge expressed by said node and said link is stored;
• an information generating step for generating new information by predetermined inference such that a knowledge structured by said node and said link based on said semantic relation has new semantic content and semantic relation;
• a value judging step for evaluating and judging a new knowledge of generated information by verifying said information generated result with said knowledge base;
• a knowledge increasing step for accumulating said evaluated/judged result and new information generated knowledge in said knowledge base to increase knowledge;
• and
• an optimal solution deciding step for deciding and outputting an optimal solution in response to an inquiry or request from then outside, wherein said information generating step comprises:
• a relating node retrieving step for retrieving only a unit in which a relating node is stored;
• a relating link retrieving step for retrieving only a unit in which a relating link is stored;
• and
• a step for executing inference by using at least any one of analogical reasoning, inductive inference, abduction and association based on a retrieved result of said relating node retrieving step or said relating link retrieving step.

In a computer system in which a central control computer and a plurality of cell computer are coupled so as to communicate with each other, a computer system characterized in that:

• said central control computer stores therein only addresses of said plurality of cell computer before being operated, said central control computer can input a question to said plurality of cell computers and can output an answer to said question;
• said plurality of cell computers have nodes each indicating meaning and a link indicating a semantic relation between the nodes, said node and said link have structures so as to exchange their roles and each cell computer stores therein information of one node or link and information of other cell computer storing therein a node or link having a connection of a semantic relation to said node or link in order to hold structured knowledge expressed by said node and said link;
• when a question is inputted, said question is transmitted from said central control computer to all of said plurality of cell computer, as presupposed operation, said plurality of cell computers are changed to a processing target state or unsuitable state based on connection information of a semantic relation of a knowledge which each of said plurality of cell computers has relative to said question;
• a cell computer having no semantic relation to said question is placed in the unsuitable state to return information indicative of the unsuitable state to said central control computer and only a cell computer having a connection of a semantic relation is placed in the processing target state to return an answer to said central control computer;
• and
• said central control computer continues to output a question to only said cell computer placed in said processing target state, said central control computer analyze an answer from said cell computer in said processing target state and understands a meaning of said answer to generate new information, converts said generated new information into an output form corresponding to said question and outputs said converted information as an answer.

In an information generating method using a computer system in which a central control computer and a plurality of cell computers are coupled so as to communicate with each other, an information generating method characterized in that:

• said central control computer stores therein only addresses of said plurality of cell computer before being operated, said central control computer can input a question to said plurality of cell computers and can output an answer to said question;
• and
• said plurality of cell computers have nodes each indicating meaning and a link indicating a semantic relation between the nodes, said node and said link have structures so as to exchange their roles and each cell computer stores therein information of one node or link and information of other cell computer storing therein a node or link having a connection of a semantic relation to said node or link in order to hold structured knowledge expressed by said node and said link;
• said information generating method comprises:
• a step in which when a question is inputted, said question is transmitted from said central control computer to all of said plurality of cell computer;
• a step in which as presupposing operation, said plurality of cell computers are changed to a processing target state or unsuitable state based on connection information of a semantic relation of a knowledge which each of said plurality of cell computers has relative to said question;
• a step in which a cell computer having no semantic relation to said question is placed in the unsuitable state to return information indicative of the unsuitable state to said central control computer and only a cell computer having a connection of a semantic relation is placed in the processing target state to return an answer to said central control computer;
• a step in which said central control computer continues to output a question to only said cell computer placed in said processing target state;
• and
• a step in which said central control computer analyze an answer from said cell computer in said processing target state and understands a meaning of said answer to generate new information, converts said generated new information into an output form corresponding to said question and outputs said converted information as an answer.