Research And Application Of Sentiment Analysis For Course Reviews

The spatial separation of teachers and students has led to the rapid growth of online education,with more and more students joining the online learning team.Online education platforms are leading the way in the digital,remote and highly competitive world of education.Course review information in online learning is an effective feedback that can effectively reflect the quality of the course and provide valuable insights into the teaching and learning process.With the significant growth of online education platform users,course review data is increasing day by day,and it is very challenging to handle the huge amount of reviews by manual means.Therefore,sentiment analysis of a large number of course review texts has become a hot topic for research.The current course review sentiment tendency recognition relies on a large amount of annotated data,and it is extremely costly to annotate the huge amount of data,so semi-supervised course review-based sentiment analysis has become a hot research topic,and this method can perform sentiment analysis in small samples of data with very little annotated data.In addition,in the sentence-level and text-level coarse-grained sentiment analysis,a text has only one kind of sentiment,which ignores the aspect-level sentiment polarity of course-specific attributes in course reviews,making it difficult to fully reflect the detailed and comprehensive sentiment of course review users,which is not conducive to online education platforms to improve platform services in a targeted manner.To address the above problems,this paper conducts the following research using natural language processing-related techniques:(1)A prompt-based sentiment classification algorithm is proposed.Because of the reliance on large amounts of labeled data in traditional sentiment classification,this paper uses prompted learning to combine task descriptions with standard supervised learning in small sample situations to pre-train language models on raw text for scenarios with little or no labeled data.At the same time,the combination of automatic construction of prompt templates,which allows labeled data to learn the templates,can achieve better classification results with only a small number of parameters to be optimized.(2)From the aspect-level fine-grained perspective,to address the problem that existing sentiment analysis methods are insensitive to the location of category words and not well connected,this paper constructs an aspect feature localization model based on the BERT model,incorporating the multi-headed attention mechanism and location-related full connectivity,and constructs a new loss function in order to fully utilize the attention rights of different aspects for each word in a given context.Through experimental validation,the aspect-level sentiment analysis method proposed in this paper can obtain a more comprehensive sentiment classification by mining the sentiment tendencies of different aspects in the course review text,which has better performance than other aspect-level sentiment analysis models.(3)Designed and implemented a course review sentiment analysis system for visual presentation of the above work.The functions of the system include: course review data acquisition and course review sentiment analysis;Calling prompt-based sentiment classification algorithm and aspect-level sentiment analysis algorithm model to identify the sentiment tendency of course review text;In addition,users can find out students’ real feelings about the course and the shortcomings of the platform based on the results of the emotional polarity of the comments,which helps to adjust the teaching plan and improve the platform functions in a timely and accurate manner.