Research On Automatic Scoring Algorithm Of Subjective Questions Based On Ensemble Learning

In recent years,with the rapid development of online education on the Internet,people's learning mode has gradually shifted from traditional offline education to online learning.On the one hand,this has brought about large-scale online scoring tasks;on the other hand,the demand for automatic computer scoring has soared.Standardized tests,including written ones,are becoming more expensive as each subjective question is graded by two teachers at high cost.At present,the existing intelligent marking system for subjective questions mainly completes marking by regular matching the text of key word.There are some problems in this method,such as low accuracy,less dimension in marking and unrecognizable synonym.Therefore,this thesis proposes a machine learning algorithm based on multi-model feature integration,which overcomes the problems in the traditional scoring model above and effectively completes the tasks related to automatic scoring of subjective questions.Multi-model feature integration machine learning algorithm is an algorithm that uses different models to extract different dimensional features from the same data samples,and then using the relevant machine learning algorithm to complete the relevant prediction.This thesis studies the primary grading methods and further defines the subjective grading problem.By studying the logic of grading subjective questions,data processing of corresponding answering texts was completed.Meanwhile,TF-IDF,Word2 vec,LDA and other quantitative features representing semantics were extracted from multiple dimensions.Finally,the predictive effects of different machine learning classification models on current features were compared.Experimental results show that on the XGBoost model,the accuracy rate of predicting the score points of the response text is over 82%.Through the machine learning algorithm of multi-model feature integration extracted in this thesis,the single dimension problem of traditional scoring model is solved.At the same time,high accuracy is guaranteed in the scoring process.By using the character representation of digital vector,the problem of unrecognized synonym is solved.On the whole,the subjective grading algorithm based on multi-model integration proposed in this thesis has obvious grading effect,which implements the automatic grading function,improves the accuracy of grading,and provides a feasible reference value for solving large-scale grading tasks and other related grading problems.