Research On Data Analysis And Evaluation Methods Of Examinees’ Ability In College Entrance Examination

With the rapid development of Internet technology and information technology, learning data in the educational field increase day by day. With the aid of the computer and other tools, a large number of related data can be collected. Therefore, how to transform the massive educational data into knowledge and information and how to serve the evaluation of learners, teaching decision-making and learning promotion, have become a hotspot in the field of educational data research.The existing researches are mainly based on the open and intelligent online learning system, revealing students’ learning characteristics and learning states through the use of student modeling. But, the researches based on answering data are comparatively fewer and more of them starts from the perspective of psychology. In the field of psychology, Cognitive Diagnostic Theory could understand the processing procedures of individuals’ internal micro psychology, completing the diagnosis of individual cognition states. However, Cognitive Diagnosis generally use Q-matrix to guide the preparation of test before doing diagnosis analysis, which is much different from the actual education examination. Moreover, Cognitive Diagnostic Models generally have high complexity, and most of them apply only to 0-1 score data, which makes the existing diagnosis models have some limitations in practical application.For the above issues, the paper is to put forward a kind of data analysis and evaluation method of examinees’ ability in Entrance Examination suitable for multi-value scoring data based on existing researches.Firstly, based on the examinees’ score information, analyze the relationship between students’ reaction behavior and their potential characteristics(also known as ability) by means of the Item Response Theory to evaluate individual examinees’ ability. Specifically use Rasch model and maximum likelihood estimation method to estimate the parameter to calculate each examinee’s total ability value θ. This ability value reflects the consistent intrinsic characteristics of examinee in a particular test, which is important for examinees’ deeper ability mining.Secondly, because IRT adopt single dimension assumption, meaning it is only suitable for analyzing single dimension ability attribute. Therefore, the paper use Q-matrix theory to connect examinees’ ability state which is not directly observable with their response reaction which can be observed on certain items in order to specify students’ ability to various levels of ability attribute, and then finally provide theoretical evidence for students’ ability analysis and evaluation in a deeper sense. Specific processes are as follows:(1) detailedly analyze the unobservable ability properties and their hierarchy of the test items;(2) convert ability states into observable item response pattern by Q-matrix theory and improved Q-matrix theory to make it suitable for the multi-value score in the actual test to obtain the expected response pattern.Finally, consider examinees’ ability analysis and evaluation as a multi-classified problem, combine examinee’s ability value and item response pattern to construct examinees’ feature vector based on IRT and Q-matrix, use Probabilistic Neural Network algorithm for classification, and then mine examinees’ ability states to the deeper level. Simulation experiments show that the proposed Probabilistic Neural Network algorithm based on IRT and Q-matrix can improve the accuracy of examinees’ ability. In addition, the empirical research based on college entrance examination data shows that the proposed method in this paper could effectively realize the analysis and evaluation of examinees’ ability. It can also find out obvious ability difference of different group’s examinees through comparative analysis between different regions, schools and student categories, which is of a certain value for actual applications.