The Feature Selection Method Of The Scoring Information System And Its Application

Examination is an important method of assessing the quality of teaching.Through analysis of test papers,we can not only understand the learning effect of students,but also find the problems of the question-setting and test papers,which are of practical guiding significance to evaluate teaching and standardize the questions.However,the existing evaluation indicators of test papers can only reflect the statistical characteristics of the papers,the final evaluation results are merely concluded by a simple "good" or "bad".The feature selection method can effectively make up the shortcomings of the existing test paper evaluation methods,find out the hidden rules in the test paper data,and provide decision support for teachers.Therefore,this paper presents a test paper data mining scheme which takes the test paper scoring information system as the foundation and the heuristic feature selection algorithm as the core.First of all,the scheme defines a data model for the paper scoring information system.Each column represents an evaluation item,for example,a question or a knowledge point.And each row represents a test object,such as a candidate or a survey object.The model establishes the mapping from the score of the test object to the ranking of the object,thus further defining the data model of the order relation information system.This model stores the ranking of all the test objects under any combination of the evaluation items.Secondly,this paper proposes a strict attribute reduction problem and two loose feature selection sub-problems.The attribute reduction problem takes the order relation information system as the input and the attribute subset as the output.The attribute subset is consistent with the order relation of the total score,and they are the constraint conditions.The minimum of the base of the attribute subset is the optimal target.However,in practical applications,it is very strict to maintain a consistent order relation,which can cause the problem to be unsolved,that is,an attribute subset cannot be found or no element can be removed from the complete set of evaluation items.Therefore,it is necessary to loosen the consistency constraint in the attribute reduction problem and change it to similarity constraint to obtain a more compact and meaningful character subset.In the feature selection problem,the constraint condition is that the similarity between the character subset and the order relation of the total score satisfies a threshold given by the expert in advance.The rest of the parameters are consistent with the problem of attribute reduction.In addition,due to the loosening of constraint condition,there may be multiple solutions to the feature selection problem.Thus,the feature selection problem is divided into two sub-problems:finding an optimal character subset and finding all the optimal character subsets.Then,by calculating tasks aimed at the similarity in the feature selection problem,a dominance relation similarity is proposed,and the classic Manhattan similarity and cosine similarity are improved.In order to make the three similarities comparable,the general properties of similarity are given first.Secondly,the author argues theoretically the satisfaction of these three indicators to the general properties,and illustrates the situation when various indicators acquire the maximum value(0 and 1).Finally,since the problem of feature selection defined by the preceding article does not satisfy the monotonicity,it is very difficult to solve the optimal solution.Therefore,a greedy algorithm is proposed to quickly obtain an optimal solution.In addition,the backtracking algorithm is improved to verify the results of the heuristic algorithm.Finally,this paper carries out a wealth of experiments on the data of real test papers.The data set consists of data of eight classes,and a standardized data set is obtained.The experimental results show that:1)in the current scene(the data structure test paper analysis),it is impossible to obtain a meaningful solution to the problem of attribute reduction to keep order relation consistent;2)in most cases,greedy algorithm can find the optimal solution,and its accuracy rate of the original data set is higher than that of the standardized data set;3)Manhattan similarity is the most reasonable,followed by the dominance relation similarity,and the cosine similarity is the worst in reasonability;4)the result of the feature selection method is meaningful;5)results obtained on the basis of feature selection can provide meaningful suggestions for teachers.