代数机器解答是智能教育领域中的核心技术。本项目针对代数机器解答研究中的主要难点问题提出创新解决方法，即研究提升题目理解能力的新方法，增强自动求解能力的新方法，以及集成题目理解和自动求解的新方法提升代数机器解答算法性能的方法。为此，本项目拟从四个方面展开研究：第一，针对申请人在预研中建立的模型池缺乏泛化能力的不足，建立更好地体现数学含义的词语分类体系，在此基础上构建泛化能力更强的模型池；第二，针对没有优选直陈数学关系方法的不足，研究既抽取又优选直陈数学关系的方法，进而研究构建嵌入模型池的深度关系网络提升该方法的能力的方法；第三，针对发现隐含数学关系能力不强的不足，研究结合隐含数学关系库构建深度关系网络提升发现隐含数学关系能力的方法；第四，集成用深度关系网络获取直陈和隐含数学关系和增强自动求解能力等新方法，构建高性能代数机器解答新算法。这些研究具有重要的学术价值，同时也具有十分广泛的应用前景。

Algorithms for solving algebraic problems are the crucial technology in developing educational intelligent systems. This proposal suggests a batch of new methods against the issues in developing algorithms for solving algebraic problems, including the methods to enhance the performance of understanding problem and the performance of solving the group of acquired math relations, and the method for fusing the proposed new methods to establish high performance algorithms for solving algebraic problems. Hence, this project aims to establish new theories and methods for improving the performance of solving algebraic problems in the following four aspects. (1) Since the set of syntax-semantic models established while preparing this project lacks generalization ability, this proposing project builds a new parts-of-speech classification system for natural languages and establishes a new set of syntax-semantic models leveraged on the system for a natural language. (2) Since no method chooses the qualified group from the extracted stated math relations, it proposes an extracting-and-choosing method that extracts stated math relations by using all possible models and then that chooses a qualified group from the extracted relations. Further, it designs a deep relational network for lifting the performance of the extracting-and-choosing method. (3) To improve the ability to discover hidden math relations, this proposing project suggests of designing a deep relational network for discovering hidden relations by adopting the techniques of encoding, decoding, deep matching network, and knowledge database. (4) This project also proposes an improved algorithm for solving algebraic problems by integrating the new methods for understanding problems and the enforced methods for solving groups of math relations. The methods that enforces the method for solving groups of math relations include the method for adding auxiliary lines and the method for transforming algebraic solution into arithmetic solution. Therefore, the proposed methods and algorithms developed in this project have their high value for both the academia and the industry as well as a great potential for application.