Research On Semantic Classification Model Of Teaching Evaluation Based On Feature Weighted Stacking Algorithm

College students as participants in teaching activities,their evaluation of teaching activities can objectively reflect the quality of teaching in schools,and have an important feedback effect on the development and reform of colleges,but there is little research on text mining of teaching evaluation,resulting in its value not fully reflected.By doing text mining on teaching evaluation,we can understand students' attitudes towards teaching activities,which can provide suggestions for college teaching arrangements and promote the improvement of teaching quality.It is of practical significance to study the teaching evaluation analysis model.This paper mainly studies on the sentiment classification model about teaching evaluation through machine learning theories and technologies.The main research work are as follows:1.We propose an improved Stacking algorithm based on feature weighting method.For the traditional Stacking algorithm,the K-fold cross-validation method ignores the difference between the primary classifiers and does not reflect the true prediction of the primary classifiers,we design a method for assigning weights to prediction based on the training performance of the primary classifiers.This method can provide more predictive information of the primary classifiers for the meta-classifier,and the final experiment also confirms the effectiveness of our method.2.We preprocess the teaching evaluation data set and implement the primary classifiers.The pre-processing process includes data cleaning,Chinese word segmentation,filtering stop words,using Word2 vec technology to train feature vectors and feature extraction,and then implementing primary classifiers based on SVM,KNN,NB and LSTM.We determine the word segmentation tools through experiments and compare the performance of the primary classifiers on the test data set.The experimental results show that on the teaching evaluation data set,the four primary classifiers in the order of classification performance from high to low are: LSTM> SVM> KNN> NB.3.We build a sentiment classification model based on the improved Stacking algorithm.The improved Stacking algorithm is used as the framework of the integrated model.The first layer of the framework is the four primary classifiers that have been implemented.The primary classifiers are trained according to the improved Stacking algorithm,and then the algorithm of meta-classifier will be selected through experiments.Finally,the performance of the Stacking algorithm before and after improvement is compared.Experimental results show that the SVM algorithm can be used to improve the performance of the model,and the improved algorithm is indeed effective and feasible.The research results show that the classification model based on the feature weighted Stacking algorithm is suitable for teaching evaluation sentiment classification tasks and shows good performance.