Research And Application Of The Approach To Understanding Elementary Mathematical Problems Based On Sentence Framework

In recent years, issues concerning artificial intelligence based problem solving, automatic proof and human-like test-taking have become a hot spot of research, which achieved new breakthroughs constantly. Natural language processing is an important content of artificial intelligence research, and its methods are classified into mainly two types: rule-based method and statistical-based method. With the gradual maturity of large-scale corpus-based machine learning algorithms, statistical-based natural language processing method have become a mainstream. However, as the problem that needs to be solved is getting complicated, it is difficult to obtain high efficiency for single natural language processing method. In this paper, the sentence framework theory which was put forward in the 1990 s, combined with machine learning methods, is used for natural language processing of elementary mathematical problems in text form, thus to realize understanding of the meaning of mathematical problems by machine. This paper concentrates on the following researches:Firstly, under the guidance of sentence framework theory, expression characteristics of elementary mathematical problems in Chinese form are combined to classify and elaborate verbs in mathematical problems. And on this basis, common sentence framework formed of verb-centered construction is studied, including simple form and complicated form. Sentence framework research of mathematical problems provides a solid foundation for semantic processing of mathematical problems in this paper.Secondly, according to the hierarchy and complexity of concepts of mathematical problems, an object-oriented approach is used to represent knowledge of mathematical problems. Meanwhile, a object-oriented knowledge representation framework with hierarchical construction is proposed. Object-oriented approach has abstraction, encapsulation, inheritance and other basic features, and with these features, representation of mathematical concepts and relationships between concepts can be solved very well. Traditional knowledge object model is improved in this paper by means of separating rule properties inside from it, which simplifies the structure of the knowledge object.The next step is the research of method on meaning understanding of mathematical problems, the core of which is the semantic processing algorithms of sentence based on sentence framework. This kind of algorithm will realize the semantic normalization of sentences which belong to different sentence framework but have equivalent meaning, and then extract the knowledge contained in the sentences. Prior to this, representation of sentence framework is a critical issue to clarify which mainly includes characteristics of sentence framework, semantic compositional data and normalization function and so on. In addition, mathematical entities recognition based on conditional random fields has also been studied in this paper in order to extract the tagging features of sentences. Through experimental verification, recognition accuracy and F1-value of this method is up to 97.13% and 94.98% respectively.At last, with the method proposed above, a question understanding system is established for question analysis of elementary mathematical problems. The performance of system and the effect of question understanding have been tested and analyzed and the effect shows that the question understanding system established in this paper can basically achieve the purpose of analyzing mathematical problems.