基于深度学习的在线视频与习题匹配计算研究  

Research on Online Video and Exercises Matching Degree Calculation of Based on Deep Learning

作　　者：李浩君[1] 钟依莲 吕韵 LI Haojun;ZHONG Yilian;LU Yun(College of Education Science and Technology,Zhejiang University of Technology,Hangzhou 310023,China)

机构地区：[1]浙江工业大学教育科学与技术学院,杭州310023

出　　处：《小型微型计算机系统》2025年第1期81-89,共9页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(62077043)资助;浙江省哲学社会科学规划交叉学科重点支持项目(22JCXK05Z)资助。

摘　　要：在线学习多模态资源匹配的精准性是自适应学习服务效率提升的关键问题,而目前在线学习服务存在着不同模态资源关联特征挖掘浅层化、模态资源表征形式缺乏规范化以及模态资源间智能匹配计算低效化等问题.针对以上问题,本文聚焦在线视频与习题资源匹配研究问题,提出了一种基于深度学习的在线视频与习题匹配计算模型DL-VEMC(Online video and exercise matching calculation based on deep learning).首先,通过关键帧提取算法KEA、语音识别技术以及jieba分词技术深度挖掘在线资源多维度特征,实现在线视频与习题预处理;其次,使用CNN、注意力机制以及LSTM等深度学习技术协同开展视频关键帧表征,利用BERT技术对在线视频音频转录文本以及习题文本进行表征,获得在线视频与习题统一化语义表示;最后,融合在线视频与习题的语义信息,利用三层MLP拟合在线视频与习题匹配度值计算函数.实验结果表明,该模型的性能优于现有基线模型,消融实验和实际应用案例也验证了模型的有效性及可行性,为在线视频与习题匹配计算提供了理论依据.The accuracy of online learning multimodal resource matching is a key issue in improving the efficiency of adaptive learning services,at present,there are some problems in online learning services,such as shallow mining of association features of different modal resources,lack of standardization of modal resource representation,and inefficient intelligent matching calculation between modal resources.In response to the above problems,this paper focuses on the research of online video and exercise resource matching,and proposes an online video and exercise matching calculation model based on deep learning.Firstly,through the key frame extraction algorithm KEA,speech recognition technology and jieba word segmentation technology,the multi-dimensional features of online resources are deeply mined to realize online video and exercise preprocessing.Secondly,deep learning technologies such as CNN,attention mechanism and LSTM are used to collaboratively perform video key frame representation,and BERT technology is used to characterize online video audio transcription text and exercise text to obtain a unified semantic representation of online video and exercise.Finally,the semantic information of online video and exercises is fused,and the three-layer MLP is used to fit the calculation function of the matching degree between online video and exercises.The experimental results show that the performance of the model is better than the existing baseline model,and the ablation experiment and practical application cases also verify the effectiveness and feasibility of the model,which provides a theoretical basis for online video and exercise matching calculation.

关 键 词：在线视频 在线习题 匹配计算 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]