Research On Key Technologies And System Implementation Ofautomatic Solving Algorithms For Math Word Problems

The automatic problem-solving algorithm for math word problems aims to parse human readable problem text into machine understandable logical codes,and generate objective equations through quantitative reasoning to ultimately obtain the answer to the problem.However,the existing models mainly focus on the mapping relationship between the problem text and the target equation,ignoring the logical semantic similarity between the problem text,resulting in the decoding of the same type of questions into different equations.In addition,existing models lack precise prediction and explicit processing of implicit constants in problem text,which are important components of the objective expression.In order to solve the above problems,this paper improves the accuracy of target equation generation and final answer by aggregating training the logical semantics of the problem text and statistical prediction of implicit constants.The main research work of this article is as follows:(1)An ICP(Implicit Constants Predictor)implicit constant prediction algorithm is designed and proposed,which can accurately predict the implicit constant knowledge in problem text.And three fusion methods were proposed,incorporating implicit constants as prompt information into the problem solving algorithm,enriching the representation information of the problem text and enhancing the decoding ability of the model.The experimental results on publicly available datasets indicate that the ICP algorithm can improve the accuracy of generating objective equations.Besides,the ICP algorithm has good portability and can be embedded into other models for joint training,improving the model’s problem-solving performance.(2)An LSA(Logical Semantic Aggregator)logical semantic aggregation algorithm is designed and proposed.The logical semantic similarity of the problem text is constructed according to the structural similarity of the target equation,and high-level logical Semantic information is extracted from the problem text,which solves the problem that the existing model is difficult to learn logical representation and easy to be interfered by topic description.The experimental results show that the LSA algorithm enhances the accuracy and adaptability of the target equation generation algorithm,and can accurately extract high-level logical semantics even when the subject description changes.In addition,the LSA algorithm can collaborate with the ICP algorithm to improve the overall model’s problem-solving accuracy.(3)An automatic problem-solving system for mathematical application problems is designed and implemented,which includes three major functional modules: an automatic problem-solving module based on rule library template matching;Automatic problem-solving module based on knowledge graph for logical reasoning;The problem text similarity retrieval module,which meets the practical needs of repeated practice and drawing inferences from one instance on wrong questions and difficult problems in the actual educational scene.