Research On Causal Reasoning Geography Examination Problem Solving Method

With the rapid development of computer technology, the way of information resources acquisition has been changed for human beings, from which original yellow pages to search engines, and then from traditional question answering system to intelligent personal assistant. However, all of these systems mainly handle factoid problems. Human beings are not satisfied with finding answers simply, but expect that computer system can understand their purposes and do more things by causal reasoning. In recent years, domestic and international academic institutions have begun to focus on the study of intelligent computing reasoning technology, such as the Japanese started a robot project aiming to pass college entrance examinations. As we known that, the automatic solving problems technology for examination of human has become a hot research topic wherever domestic or abroad, which can strongly improve whole automatic question answering research domain level. The purpose of this research is to construct a system which can automatically solve the causal reasoning problems of high school geography.The main contents of this paper include the analysis of the problems of senior geography, the method of solving the problem based on the classification model, and the method of solving the problem based on knowledge reasoning. At present, the research of automatic solving problems technology for senior examination is not mature. This paper first analyzed the problems’ category of senior geography, and then gave a formal definition and analyzed the general reasoning method with the real casual reasoning problems. Then, this paper used classification model based solving method to make original problem into four sub yes/no problems and judge these sub-problems with the help of classification technology. Because word embedding model can express some latent semantic information, this paper used it to represent those features of problems’ text, and combined support vector machine model to computing problem’s answer. Finally, it is found that the geographical problems can be divided into different categories according to their knowledge points background. For each category of problems, its description usually contains some key information, which corresponding its right answer. At last, this paper drove a general method to computing answer, in which language model and entity recognition technology were employed orderly to extract the key information of problems, and reasoning out answer by computing the matching similarity with knowledge reasoning rules base. The latter model proved that it is performs better when adopting extracting key information accurately with the help of entity recognize by sequencing labeling.The experimental data of this paper mainly consist of the college entrance test problems, the simulation test problems, the senior geography teaching materials, the geography vocabulary entries from Baidu encyclopedia and the whole Wikipedia entries. When using the classification model based method to solve problems, it got 0.344 and 0.329 average accuracy values on college entrance corpora and simulation test corpora respectively. However, the knowledge reasoning based method achieved a higher average accuracy 0.652 when testing it on geography industrial location factor problems. Actually, these two methods have a better accuracy than the search strategy based method.