The Urgency Post Classification And Topic Mining For MOOC Discussion Forums

The forum of Massive Open Online Courses(MOOC)is an important tool for teachers to understand and intervene learners' learning activities.There are a large number of learners' learning processes and experience data in the forums.Because of the small number of teachers and managers,the effective information in a large number of comments is easily ignored.The classification of the urgency of MOOC comments can help teachers timely follow up the comments on the forum,find out the learner's posts that need to be replied to,and respond accordingly,so as to help learners solve problems in a timely manner.The topic mining of students' "urgent" posts can help teachers understand the reasons for the problems caused by learners in the learning process and the feelings of learners in the course,which is of great research significance for improving the MOOC teaching platform and improving the course quality.This paper takes MOOC posts as the research object,and the main research contents are as follows:(1)Aiming at the inaccurate semantic information expression of the existing word embedding technology,which cannot solve the problem of "polysemy" of words,a classification model based on recurrent convolutional neural network(BLNN)is proposed.The model uses Bi LSTM to obtain contextual information of each word from the two grammatical dependency directions.The contextual information is combined with the information of the word itself to obtain a word representation with real semantic information.The model has achieved 1.9%,2.1% and 2.7% better than the existing classical distributed vector representation methods in three sets of experiments on the Stanford MOOC review data set,respectively,which proves the advancedness of the model.(2)Aiming at the problems that the BLNN model cannot learn information such as spelling errors and special symbols in MOOC posts,and the existing models have limited ability to learn the semantic and structural information of sentences,this paper proposes the Attention-Based Character-Word Hybrid Neural Networks(ATHNN).The model extracts information such as misspellings and special symbols in the review text through character embedding and Convolutional Neural Networks(CNN);at the same time,the proposed CNN-GRU module can simultaneously learn semantic and structural information of sentences.The model achieved F1 values of 92.2%,91.6%,and 89.2% in the MOOC review classification task,which is superior to all existing advanced models..(3)Aiming at the problem that probabilistic topic models lack text semantic information and separate word vector-based deep learning models ignore potential topic information of text,a MOOC review topic mining model(DCSM)combining a topic model and a neural network is proposed.LDA model was used to extract the subject keywords in the text,and then CNN was used to extract the subject features in the subject keyword matrix,which was integrated with the text semantic features learned from the LSTM model to obtain the text representation containing both the subject features and semantic information.The model achieved 71.7% F1 value,better than the existing advanced model 2.5%.This paper researches the classification algorithm of MOOC posts urgency and topic mining algorithm,improves the accuracy of MOOC review classification and topic mining,which helps teachers better manage MOOC forums.