Research On The Test-questions Similarity Detection And Classflcation Based On RNN

With the rapid development of information technology,all kinds of schools at different levels are vigorously developing online-based education,and established large-scale test-questions system for online evaluation.However,as time goes by,a large number of similar questions will be included in the database,and when the automatic test paper examination system for test-item system questions need to be avoided in the same test paper,duplicate test points will appear in similar test-questions.The assessment and evaluation of knowledge acquisition and the evaluation tasks of the automated examination system all have adverse effects.In addition,many of the questions in the existing test-question database do not have associated classifications corresponding to the knowledge points,which seriously hampers the number of test questions and the proportion of scores assigned to questions in the system automation.Therefore,the necessary techniques need to be taken for the similarity test-questions measurement.And the similarity of the questions and the automated relevance of the knowledge to the questions.The main contents of this paper are as follows:(1)According to the large number of massive test-questions,the task of the similarity of test-questions dection need to be taken,a similarity measurement algorithm based on the combination of TF-IDF and word2 vec was introduced.Experiments show that this method of calculating the similarity using word2 vec dection algorithm is superior to traditional TF-IDF method.On this baseline,a related study of the similarity calculation model of test-questions based on the recurrent neural network is make a further step developed.Using this model reach a reasonable solution for the semantic similarity questions in the test-questions dataset.A series of comparative experiments were carried out to optimize the model,and then verify that the optimized model can complete the test-questions similarity evaluation task and reach a much better result.(2)In order to achieve the classification of the test-questions based on the knowledge points,a pre-training word2 vec was introduced to design the test-questions classification model.A calculation model combining TF-IDF and word vector was proposed to classify the test-questions based on the knowledge points.Experiments showed that the method can complete the classification task.On this baseline,and then using the test-questions dataset of labeled knowledge points,a further step study on the classification model of test questions based on the recurrent neural network was introduced,and the recurrent neural network was used to semantically code the test-questions,and a multi-group relevant adjustment experiments were performed on the classification model.The optimized model completes the multi-classification tasks of the questions.