Design And Implementation Of Problem-solving System Based On Self-learning

At present,the rapid development of artificial intelligence has made great breakthroughs in many aspects.With the development of natural language understanding,people hope that artificial intelligence can reach the level of cognitive intelligence,and cognitive intelligence is also a standard for measuring intelligence to a certain extent.Logical reasoning ability,as a comprehensive reflection of human memory ability,learning ability and judgment ability,is also an urgent problem to be solved by cognitive intelligence.The content of this thesis is the design and implementation of an automatic problem-solving system based on self-learning.The theoretical basis of the problem-solving system is permutation and unity,and the realization method is a rule-based production system,which mainly involves the construction of a problem-solving system and self-learning.The specific research contents are as follows:（1）Representation of elementary mathematics knowledgeKnowledge representation is the first step in the automatic solution of the problemsolving system.Only by converting the mathematical text into the data structure of the computer can reasoning calculation be performed.The final system uses a knowledge graph to represent the relationships between abstract entities.In the implementation process,the class structure of Java is used to instantiate the entity relationship in the topic,and the inheritance idea of Java class also realizes the extensibility of knowledge representation in the knowledge graph.（2）Design and construction of inference engineAs the core of the problem-solving system,the reasoning engine uses the traditional forward inference method in the reasoning method,and integrates with the calculation in the reasoning process,and iteratively generates new knowledge.In terms of implementation,a modular design scheme is used to ensure that each module is relatively independent.The core matching algorithm adopts the mixed matching mode,which combines various matching schemes to speed up the matching of rule entities,form a mapping table,and finally complete the update of knowledge.This thesis has completed the design and construction of the inference engine,and conducted batch tests on different types of questions.The pass rate of the problem solving is 73.4%,and the solving time does not exceed 5 minutes.（3）Construction of self-learning moduleIn the rule-based production reasoning system,the rule base is an important driving force and the basis for generating new knowledge.The construction method studied in this thesis is based on data-driven bottom-up automatic construction.The source of the data is the standard answer problem solving process in the teaching supplement.The standard answer is combined with the problem-solving system to convert the answer into a rule and add it to the rule base according to a certain algorithm.The standard answer generates a short sentence tag library through self-learning,and uses the language model to complete the automatic selection function of the rules.