A Study On Automatic Generation Of Arithmetic Applied Problems Text

With the rapid development of artificial intelligence,there are more and more demands on natural language comprehension,interpretation and generation.However,there are few researches on the natural language description methods or models of the four operational word problems.Based on such background conditions,this thesis studies the characteristics and attributes of the text of the four operational word problems,and summarizes the methods or models from the number relationship to the natural language description of the text of the word problems,which provides significance for this study.mainly starts from the expressions of the four operational problems,and concludes that the expressions are composed of the smallest units.Then,I studied the word problem text and found that it was composed of a single sentence,analyzed the part of speech of each single sentence,and extracted the donor,verb,acceptor,number and quantifier.The content of corpus construction is determined by the representation of concept map and its part of speech.Then the relationship between the minimum unit and the single sentence is studied.It can be seen that each minimum unit corresponds to two single sentences and a relationship model.With such a relationship,the expression can be decomposed to the smallest unit,generating a single sentence for each smallest unit,and finally regenerated into a word problem.In this study,the corpus,relational model and the model represented by concept diagram for the four operational word problems were summarized.In addition,the most innovative point of this study is to create the algorithm rules that must be followed in the process of automatic generation of word problems.On the basis of word problem generation,one arithmetic expression can generate multiple word problem texts,which brings great convenience to automatic word problem generation.The evaluation methods automatically generated by natural language mainly include internal evaluation and external evaluation.The internal evaluation method was used in this study.From the perspective of objective evaluation,the ROUGE evaluation model was used.From the perspective of subjective evaluation,the generated test cases are scored manually.According to the evaluation results,the flexibility,correctness,readability and consistency of the word problems generated by this method have reached the level of artificial word problem design.In general,a method or model from number relation to natural language description of word problem text is established through this study,which provides a direction for further research on automatic text generation of four operational word problems.