Study And Implementation Of Natural Language Supported Intelligent Scoring Algorithms

English teaching becomes one of the most important and hardest domestic educational tasks.Its importance is obvious to exist in various critical tests like NMT(National Matriculation Test),GEE(Graduate Entrance Exam),DEE(Doctorial Entrance Exam),and overseas study exams.Its hardness lies on the lack of an English environment and the insufficiency of English teachers.Taking the English writing for example,there needs the careful evaluation to large-scaled compositions of students.The internal demand of the teaching process requires not only an intensive work of English teachers but also the ability of English writing of the teaching staff.At present,domestic English teachers are basically Chinese with a non-English native language.They can not undertake such a job,which needs both energetic and talented English teaching quality.In the process of practical writing teaching,the number of students is very large while the number of teachers is relatively small,it leads to low efficiency of working and long ?test-scoring-feedback? cycle,greatly decreased the training opportunity of writing the essay.It results that the errors can't be revised because of the late feedback.And in the process of scoring,it's susceptible to the influence of teachers' subjective feeling and other troubles.With the rapid development of computational intelligence,"automatic scoring" approaches in recent years into public view.It has overcome the shortcomings of low efficiency through artificial scoring,and compensated the defects of teachers' feelings mixed within the paper.In addition,the high accurate and efficiency as well as the objective and consistent evaluation free the teachers from the heavy labors,which helps them contributing to meaningful works.The research on intelligent writing review is relatively early overseas and the mature systems emerged show excellent reliability in actual applications.However,these systems are designed to review for English as the mother tongue.For domestic English examinations,taking into account that neither students nor scoring teachers are native English speakers who aspire fewer skills than English indigene,there could exist "incompatible" during the manual review according to such systems.Based on the culture demands coming from Chinese students,some improvements have been made upon index selections and feedbacks on the scoring algorithm.In this paper,we use the intelligence scoring algorithm to explore the background,collecting the samples of real exams,summing up the marking indexes that may affect the scoring results,summarizing the algorithmic indexes that have significant influence on the composition scores,and extract the general rules of the examinations.To create a hierarchical scoring model.This paper constructs a set of index system,taking into account the basic elements of the composition of the words,sentences and text structure of the three major indicators,through the indicators with the indicators of the four elements of the composition of the standard,The scores of the composition are analyzed,and the influencing factors are extracted,and the indexes are analyzed.According to the linear fitting of the index and the score of the composition,the optimal threshold is obtained,and a hierarchical evaluation mode is constructed.However,The score is not the ultimate goal.The intelligent evaluation system proposed in this study includes not only the evaluation score,but also the natural language feedback of the analysis information and the individual learning suggestion.About 312 writings from college students with three different topics are tested and analyzed with the help of three parsing tools.The algorithm developed to score all essays in the data set reports an overall 79.66% exact accuracy and 94% adjacent accuracy between the predicted essay scores and the human scores.The maximum error rate is less than 20% and there is no singular value error in the intelligent evaluation system.The result suggests that the Adaboost/CT algorithm could be applied to intelligent evaluation efficiently.