Knowledge Representation And Realization Of Sequence Questions Based On Knowledge Graph

The study of automated solving systems for elementary mathematics problems is a touchstone in the field of automated mathematical reasoning.It attempts to practice automated reasoning in the field of elementary mathematics,and sequence problems are a particularly difficult and challenging type of problem in elementary mathematics.The difficulties lie mainly in the representation and implementation of knowledge,which are the foundation and prerequisite for automated problem solving,and need to be addressed urgently.The main research contents of the thesis are as follows:1.Guided by problem-solving,the thesis analyzes and summarizes various typical sequence problems in elementary mathematics,using a knowledge graph as the underlying structure for knowledge representation.It attempts to construct entity relationship triplets for sequence problems in elementary mathematics using a combination of Chinese and English mathematical natural language text as input.With regards to knowledge representation,the focus is on modeling sequence-related problems and semantic expression in the knowledge graph.In terms of implementation,the emphasis is on entity recognition and relationship extraction of sequence problem triplets.2.For entity recognition,the thesis combines the Harbin Institute of Technology LTP model,which is more suitable for Chinese in natural language processing models,for preprocessing.Due to the particularity of sequence problems,the thesis focuses on studying a combination of rules and dependency syntax analysis.3.In the process of relationship extraction for sequence problems,the thesis further optimizes the relationship extraction based on the BERT model and the classic distance model Trans H on the knowledge graph.4.The specification for writing instantiation rules for sequence problem solving,and constructing the corresponding instantiated knowledge base.5.Ultimately,the thesis achieves a human-like solution to sequence problems,capable of accepting inputs in both text and mathematical formula La Te X formats.For a mixed test set of multiple-choice,fill-in-the-blank,and open-ended sequence problems in elementary mathematics,the developed and optimized sequence module using a combination of Chinese and English mathematical natural language processing for upstream tasks achieves a knowledge representation completeness of 82.9% and an overall score of 74.3% for solving sequence problems when combined with downstream reasoning tasks.