Research On Course-Comment-Oriented Fine-Grained Sentiment Analysis Method

Sentiment analysis is an important research direction of natural language processing.It is becoming more and more important to process and analyze a large number of unstructured data on the network using sentiment technique analyzing.Most of existing sentiment analysis results are about tourism,consumer products,public opinion and stock market prediction.Because there is no publicly collated educational data,the sentiment analysis of curriculum reviews in the field of education is relatively less.With the rise of large-scale online courses(MOOCs),many learners begin to learn in MOOCs.The commentary area of the platform stores many course reviews published by learners.Through sentiment analysis of course reviews,learners' learning feelings,motivation and suggestions for the platform and courses can be understood.Therefore,it is important to conduct sentiment analysis of course reviews with pragmatic value.Through the sentiment analysis in other fields,this dissertation explores the method of evaluating MOOCs curriculum through curriculum reviews,and scoring the curriculum from the perspectives of educators,learners and platforms to provide objective and reasonable curriculum evaluation reference for educators,learners and platform managers.Firstly,curriculum reviews from multiple disciplines on MOOCs were collected invalid and non-natural language texts were eliminated,and more than 18,000 effective reviews constitute a corpus of curriculum reviews.Secondly,based on the conditional random field,we extracted the opinion target and selected the features of words,parts of speech,sentiment words and ontology as feature selection.Because the implicit opinion target can not be extracted directly,the implicit opinion target was recognized based on document classification method,and the training set was constructed by using the relevance of opinion words and explicit opinion target,so that the implicit opinion target corresponding to opinion words can be inferred from the opinion words.Finally,the local perception of convolutional neural network was used to classify the attributes of reviews,and the polarity was calculated based on sentiment dictionary.In this dissertation,a method of attribute weight-based scoring was proposed.By retrieving the proportion of attributes and course reviews in the Web pages,it regards the proportion of common hits and all attributes hits as the weight value of scoring from the perspectives of educators,learners and platforms,and takes the sum of the product of weights and attribute scores as the total score of course reviews.The sentiment analysis method proposed in this dissertation was applied to the curriculum reviews,and the results were analyzed by the teaching authority of a university.It is considered that the results of the sentiment analysis are objective and reasonable.The ontology features proposed in this dissertation can effectively improve the accuracy and recall rate of opinion target extraction.Compared with other literatures,the classification accuracy of implicit opinion target recognition based on text classification is improved 4%.The attribute weight based on the number of hits and the polarity computation method based on sentiment dictionary are put forward.The course can be graded from three perspectives of educators,learners and platforms.