Research On Sentiment Analysis Of MOOC Curriculum Comment Text Combining Part Of Speech And Knowledge Graph

In the era of the digital economy,information dissemination is accelerating,and people’s personalized demands for learning and living continue to expand.Online course platforms have emerged rapidly in response to this trend.Among them,the Chinese University Massive Open Online Course(MOOC)platform has become a popular choice for many online learners,offering a diverse range of open online courses and after-class review systems.The substantial volume of after-class review data generated encompasses learners’ opinions,attitudes,and suggestions regarding course content,serving as a valuable reference for other learners.Analyzing this data and extracting the embedded sentiment information can help teachers better understand students’ needs and enhance the course and teaching direction accordingly.Traditional text sentiment analysis only determines the emotional tendencies expressed in the text.In fact,the emotions contained in the text are multifaceted,and a course review may involve multiple emotional tendencies.Simple sentiment polarity analysis cannot accurately extract the reviewer’s views on various aspects.Fine grained sentiment analysis can refine research into learners’ emotions and attitudes towards a specific aspect.This article will explore aspect-level sentiment analysis based on MOOC online course review texts.This study delves into aspect-level sentiment analysis,comprising two subtasks:aspect term extraction and sentiment polarity classification.In existing research on aspect term extraction,most studies encounter errors in segmentation,negatively impacting extraction results,and seldom utilize part-of-speech information.In sentiment polarity classification research,there is insufficient investigation or unsatisfactory progress in dependency trees and adjacency matrices,with a lack of consideration for integrating external knowledge.Hence,this paper explores and refines methods for aspect-level sentiment analysis based on MOOC course review text,making three primary contributions:(1)We assemble and organize experimental datasets,design several comparative experiments,and evaluate the effectiveness of the proposed models.We develop a MOOC course review dataset,which encompasses data collection,preprocessing,annotation,and statistical analysis of pertinent dataset information.Ultimately,we obtain a standardized dataset that addresses the gap in Chinese online course domain datasets to a certain degree.(2)We propose a Multi-granularity Semantic Feature Model(MGSFM)for the aspect term extraction task,featuring two significant enhancements over existing neural network models: first,we introduce part-of-speech as a feature to be converted into a vector input for the model,fully leveraging the information embedded in the text and addressing issues such as polysemy;second,we integrate character embeddings into the word vector representation.This process enriches the semantic feature information and reduces the impact of inaccurate segmentation.(3)We put forward a Simplified Graph Convolutional Network-TransE(SGC-TransE)model for the sentiment polarity classification task,incorporating the "knowledge + data" research paradigm,which allows the model to integrate external knowledge.At the "data" level,we employ a simplified adjacency matrix algorithm to refine the adjacency matrix.This emphasizes their connection and guarantees comprehensive extraction of sentiment information.At the "knowledge" level,we introduce prior knowledge through the sentiment knowledge graph,empowering the model to integrate external knowledge and address the problem of polysemy in various contexts,thus bolstering the model’s learning capability.