Study On Online Learning Text Feature Extraction Based On Sequence Model And Graph Convolution Model

Learning Analytics(LA)refers to the measurement,collection,analysis,and reporting of data about learners and their contexts,with the goal of understanding and optimizing learning and the environments in which it occurs.With the rise of data-driven technologies in recent years,researchers have used a variety of data mining methods to analyze learning from data in Centralized Educational Systems(CES),which has resulted in good feedback on the educational system.However,few researchers have applied learning analytics to online learning sites with massive amounts of data to improve user experience.Moreover,in the traditional learning analytics process,researchers usually need to construct features for course data guided by expert knowledge,and when the analysis task changes,the features often need to be redesigned,which is a tedious and time-consuming process.Therefore developing an end-to-end feature representation method for online learning materials can make learning analysis more efficient and accurate.On the Internet,course information is often available in various forms such as text,video,and audio,among which text form is the main source material for course features due to its compact and easy to handle information.It is of great importance to extract rich feature information from short texts containing a large number of specialized vocabulary.First,based on the correlation between text and text in online learning text data,we propose a structural feature-based online learning text feature extraction process by introducing a structural feature extraction module into the traditional text embedding model,in which the structural feature extraction module encodes the relationship between texts,which complements the semantic embedding obtained from the traditional text embedding model,and obtains a more feature-rich text embedding.The text embedding model in the proposed process can be any advanced text embedding model,and the quality of the features in this part can be tested with common metrics,while the corresponding metric Mean Rank is designed to evaluate the features obtained from the structural embedding extraction module,thus forming a closed-loop for process optimization and process evaluation.Besides,based on a large amount of data in online learning platforms and the small number of open-source online learning datasets,we build a more large-scale dataset Coursera Corpus to support the related research.By refining and extending the text feature extraction process proposed in this paper,this paper proposes a framework for extracting online learning text features based on structural features and provides multiple implementations for the structural embedding extraction module based on sequence models and graph convolution models according to the different structures that may be formed between classes within a course.Through experiments with multiple metrics on multiple datasets,the results show that the feature vectors obtained through the proposed framework can obtain better Mean Rank values without degrading text classification accuracy and can be used as a credible feature source for online education learning analysis.Finally,based on the structural feature-based online learning text feature extraction framework proposed in this paper,we implement an online learning text embedding extraction system by integrating common text embedding models and provide a detailed description of the positioning and application scenarios of this system in learning analytics-related research.We use this system to perform the analysis of the properties of the feature vectors obtained from the Quick thoughts model and the S-BERT model,and the comparison experiments between them and the proposed method in this paper.The experimental results verify the practicality of the online learning text embedding extraction system,which can provide highquality online text embedding vectors for relevant researchers.