Research And Implementation Of Tag Generation Technology For Educational Resources Based On Deep Learning

In the intelligent teaching system,recommending personalized learning content according to students’ learning needs and interests is a critical technology in the current system.Its realization requires the structured organization of learning resources first.Most teaching resource libraries are still organized by file storage methods with low retrieval efficiency and slow query speed.The key to the whole work is to complete the structural transformation of organization method.To this end,the structured organization of teaching content with Knowledge-Graph technology has become one of the optional technical routes.The essential work for reconstructing the teaching resource database through the Knowledge-Graph is to mark the digital teaching resources and represent the resource content information through the label to provide the data foundation for constructing the knowledge graph.However,most of the current resource labeling is implemented by manual crowdsourcing,which has high requirements for the professionalism of labeling personnel.It is difficult to control the labeling behavior,so the labeling quality is not qualified,the accuracy rate is low,and it is time-consuming and labor-intensive,which increases the project realization cost and the labeling behavior is challenging to control.Therefore,realizing the automatic resource labeling method is vital in constructing a knowledge map of teaching resources and realizing an intelligent teaching system.It is an effective and feasible way to use Deep Learning method to generate labels for educational resources.Through analysis,it is generally believed that the key to annotating resources is the annotation of video and audio resources.In this thesis,the online teaching resource platform of a university in Xi’an is selected as the research object to carry out experiments.A large number of high-quality audio and video teaching resources in the platform are labeled according to subject classification and knowledge points,aiming to seek effective and reliable general methods through experiments.During the experiment,the audio containing the teacher’s voice information is first extracted and converted using by speech recognition tools to obtain the content text data of the teaching resources,and a content text data set is constructed.Then,the text data is processed and analyzed using by the text classification and keyword extraction technology in the natural language processing technology to form a method for generating labels for educational resources based on Deep Learning.This method classifies the subject categories of audio and video resources by designing an educational text classification model based on BERT-CNN and designs a TextRank keyword extraction model that integrates external knowledge and semantic feature weights to realize the extraction of knowledge points contained in audio and video resources.Furthermore,the classification label and the keyword extraction result are taken together as the label of original audio and video resources.Through the design and implementation of the prototype system and experimental verification,the method of generating labels for educational resources based on machine learning is feasible.The constructed BERT-CNN classification model performs better than other Chinese baseline classification models on a university in Xi’an educational resource content text dataset.The accuracy of the TextRank keyword extraction model that integrates external knowledge and semantic feature weights is higher than other Chinese baseline keyword extraction models.The designed prototype system lays the technical foundation for developing the whole system.