Identification Of Named Entities For Elementary Mathematical Problems Based On CRF Model

In recent years the development of machine learning and deep learning technology in artificial intelligence has made great achievements in speech recognition and image recognition.Therefore,artificial intelligence technology has been paid more and more attention by experts and scholars at home and abroad.And knowledge reasoning is one of the most important problems in machine learning and deeplearning.So autosolve math problem based on knowledge reasoning have been proposed.Knowledge reasoning basis on the correct acquisition of mathematical knowledge.So in this articlewe focus on how to correctly and efficiently extract the field of knowledge of mathematics.So we research from the fellow aspects.Firstly we have studied the problem of elementary mathematics based on automatic problem solving.In this paper by analyzing the characteristics of elementary mathematics language and the automatic solution of elementary mathematics,we define the categories of mathematical named entities based on the problem of geometry and algebra.And the research content provides a direction for entity identification.Secondly we propose a new method of naming entity based on problem solving.In this paper we analyse the different characteristics of algebra and geometric parts.Because the algebraic part of the entity length is longer,the physical boundary judgment is prone to error,so we propose a new 6-word entity annotation method.The experiment show that the algebraic part of the 6-word annotation method is better than the 4-bit and 2-word annotationmethod.Thirdly A new automatic generation dictionary is proposed for entity recognition post-processing.Because the recognition of entities based on statistics does not reach 100% correct rate.But solve the problem need to be completely correct knowledge to ensure that the reasoning is correct.So in this paper,after the statistical model recognition,we add the post-processing algorithm of automatic generation dictionary,which greatly improves the effect of entity recognition.Fourth we propose the pruning strategy for Viterbi.In this paper we enumerates the transfer probability between the state of the mathematical entity.We found that some search paths of the Vterbi algorithm in the decoding process are redundant and affect the efficiency of decoding.So we can prune the algorithm with the rule.Finally,based on a series of methods,we constructs a named entity recognition system based on CRF for elementary mathematical problems.The test results show that the system is very effective in extracting knowledge.