A Model Based On Deep Learning For Plane Automatic Problem Solving

How to solve geometric problems automatically has always been a a highly challenging task.The current challenge in the field of solving plane geometry problems is that the data sets on plane geometry problems are either small in scale or cannot be made public;there are still deficiencies in the recognition of graphs and texts;using expression trees,when the amount of data increases,the search space will rise exponentially,these methods are not yet as robust as expected..In the past one or two years,deep learning related technologies have shown great potential in the field of solving plane geometry problems.This paper uses deep learning technology to construct an algorithm that can automatically solve multiple-choice problems in plane geometry.This method can automatically deduce the solution of plane geometry problems from text and graphic information.The main work and contributions of this paper are as follows:(1)A geometric problem solving method is proposed that jointly models textual and graphical information and generates a feasible,correct and grammatical rule-compliant sequence of programs through a program decoder to finally obtain the answer to the problem.A neural network-based geometric problem solver with the advantage of automatically learning features and patterns from the data without the need to manually design rules or templates.(2)Using pre-training can improve the model’s understanding of structural and semantic content,and enhance its performance in solving geometry problems.(3)In response to the deficiency of data sets,data augmentation techniques are employed to broaden the training set.