Elementary Mathematical Relation Extraction And Application Based On Knowledge Graph

In recent years,as the field of artificial intelligence has advanced by leaps and bounds,deep learning technology has brought great developments in all directions of natural lan-guage processing.With the help of natural language processing technology and deep learn-ing algorithms,people can extract structured information from text.Extracting knowledge from mathematical texts and achieving the goal of human-like answering through knowl-edge reasoning has become the research direction of some scholars at home and abroad,among which elementary mathematical relationship extraction has become one of the im-portant research directions.Relying on the work and research results of predecessors,We apply the knowledge graph to the study of elementary mathematical relationship extrac-tion.The main research contents are as follows:1.We propose a relation extraction algorithm based on text feature vector and math-ematical concept knowledge graph.Based on this algorithm,an elementary mathematical relationship extraction system is designed.The idea of this algorithm is that if two texts are similar,the corresponding entities contained in the text also have similar entity re-lationships.Further use BERT to obtain text features,use the features of the text to be matched and the text features in the relational database to calculate the similarity,and then selectively retain them from The triples contained in the relational database text are then pruned to remove the unreasonable matching triples from the knowledge graph rela-tionship,finally the result of the relationship extraction is obtained.2.We have built an elementary mathematical relationship labeling platform.Using this relationship annotation platform can allow more people to participate in the anno-tation of relational data,which can increase the relational database data faster and more conveniently,which provides great help to improve the accuracy of relation extraction?at the same time,the annotated data is constructed A mathematical text relation library,which stores mathematical text and its contained entity relation triples,which is structured data.Therefore,it is convenient to load the triple data of the relation extraction system and quickly store the data of the relation labeling platform.3.We propose a solution to the long-distance entity dependence problem in the field of elementary mathematics.In order to reduce the complexity of the relationship extrac-tion system,the mathematical text is divided into multiple short texts according to certain principles,and the entity relationship in the short text is relatively complete as far as possi-ble,but the problem of entity relationship extraction across multiple short texts is encoun-tered.A solution to this problem is proposed to use entity placeholders when labeling data and the shortest distance matching principle when extracting cross-text relationships,and this solution also solves the problem of reference in the field of elementary mathematics.Based on the above scheme,we designed and implemented the elementary mathemat-ics relation extraction system,and applied the system to the project of ”Key Techniques and Systems for Answering Questions for Elementary Mathematics”.There are 5,623 different mathematical short sentences and 10,875 entity triples contained in the elemen-tary mathematical relation library we constructed.In order to conduct a comprehensive test on the system,we randomly selected 1,000 math questions from the elementary math question bank,and the accuracy of the system's extraction of triples was 95.6%.