Research On Online Course Evaluation Sentiment Analysis Algorithm Based On Deep Learning

With the rapid development of Internet technology,Internet-based e-commerce platforms have sprung up.It has become a habit for users to express opinions and express their personal emotions on these platforms.Massive comment data has been generated and users’ emotional tendencies have been discovered.It has become an important way for product after-sales information feedback.At present,the main research objects in text sentiment analysis tasks are Weibo reviews,movie reviews,and e-commerce reviews.Relatively few sentiment analysis is conducted on online course reviews.However,online teaching has already occupied a place in my country’s primary and secondary education and the teaching of ordinary colleges and universities.Emotion analysis of online course reviews can understand students’ evaluations of courses and analyze student needs,so as to better serve teachers.Provide personalized guidance,but also can help students choose the most suitable course for them.Therefore,using online course reviews as the research object of sentiment analysis tasks can play a certain role in promoting the development of education.Deep learning is currently the mainstream method in sentiment analysis tasks.Commonly used are convolutional neural network,recurrent neural network,and bidirectional long short-term memory network algorithms,but these algorithms have the following shortcomings:1.The problem of ambiguity in word vector conversion cannot be solved,resulting in The final accuracy of the algorithm is not high;2.The structure adopted in the design of the feature learning network is relatively simple,which leads to incomplete features;3.Lack of distinguishing the importance of different features.In this paper,a new sentiment analysis model is designed in response to the above problems.The new model is based on deep learning technology and studies the currently popular sentiment analysis algorithms,and then improves the related algorithms for the specific research object of online course reviews.The main performance of the improved algorithm when performing sentiment analysis tasks is improved.The main work of this paper is as follows:(1)This article has conducted research on convolutional neural network(CNN)and bidirectional long short-term memory network(Bi-LSTM),and found that CNN network has outstanding performance in feature learning,but because the algorithm cannot combine context in text analysis The final result is not ideal;the Bi-LSTM network can link the contextual information when analyzing the target word,but the network processes the sentences in the sequence of the text,and obtains the global semantic information.Mining more The ability of advanced features is far less than that of the CNN network.This paper hopes to design a feature extraction network that can combine the advantages of the above two networks to increase the breadth of features.For this reason,the CBi-LSTM model is first proposed,in which CNN and Bi-LSTM are connected in series to mine more advanced features.The experiment proved that the model did not achieve the expected effect.After analyzing and summarizing the reasons,the CNN-Bi-LSTM model was proposed.The CNN-Bi-LSTM model adopts a parallel method,allowing the two networks to perform feature learning separately,and then combine the output results to increase the breadth of feature learning.Experiments have verified that the F1 value of the model is significantly higher than that of the CNN network and Bi-LSTM network,which verifies the effectiveness of the model.(2)The data set used in this article is obtained from the online learning platform of the Chinese University MOOC national quality course.Through research,it is found that these comments are generally relatively short,and the verbal expression is more colloquial and youthful,and the words used are often There is a phenomenon of polysemous words.In ordinary sentiment analysis tasks,word vectors are often obtained through Word2Vec training and learning,but the static representation of words obtained through the Word2Vec model cannot solve the problem of polysemous words.In order to make the data set of this article more accurate in the representation of word vectors,this article uses the Bert language pre-training model to train the word vectors to realize the vector representation of words.(3)In order to improve the weight of features with great emotional inclination in sentences,this paper introduces an attention mechanism based on the CNN-Bi-LSTM model proposed earlier,and puts it on the CNN network and Bi-LSTM The specific location of the network allows the feature extraction network to readjust the weights according to its importance while learning features.The experimental results show that the improved Bi-LSTMA-CNNA model has a better classification effect in the sentiment analysis task of online course reviews.The Bi-LSTMA-CNNA algorithm model proposed in this paper has a certain improvement in performance compared with traditional algorithms when performing sentiment analysis tasks on online course reviews.It provides an effective way for online education platforms to collect user feedback and improve course quality.Technical Support.