【作者】 刘阳；

【导师】 陈旺虎； 李扬；

【作者基本信息】 西北师范大学 ， 工程硕士（专业学位）， 2022， 硕士

【摘要】 大学生的校园在线行为分析对于其过程管理至关重要。鉴于大学生的校园行为具有个性化、多元化等特点,采用传统的方法对其进行跟踪监测极具挑战。本文基于大学生的校园网访问日志、Wi-Fi接入点日志以及校园一卡通记录等在线数据,探讨一种学生校园在线行为的特征表征模型以及基于此模型的行为分析方法,以促进学生的个性化和过程化管理。主要研究工作包括如下三个方面:(1)基于心理学注意力的学生校园在线行为分析方法。将学生的行为看作是注意力在其关注对象上的选择、分配与流动,学生行为模式的挖掘则可被转化为其注意力模式的挖掘,可为学生校园多元行为提供一种统一的表征方式。同时,为支持学生行为中的注意力计算,探讨了校园在线行为数据的融合。(2)基于心理学注意力的学生校园在线行为表征模型。在学生在线行为数据的驱动下,构建了心理学注意力的计算模型(Psychological Attention,PA)。将在线行为看作是学生对各种对象的关注,并由注意力的稳定性、广度、转移性和分配性四种品质来区分潜在的行为模式。所提出的PA模型为校园在线行为的分析提供了一个统一的行为特征基础。实验表明,与基于传统统计行为指标相比,基于PA模型提取的特征在K-means、DBSCAN、GMM、FCM以及FCM等传统的聚类算法上的SC、CHI和DBI三个聚类评价指标均表现最优。这表明,PA模型对学生行为的特征具有更好的区分能力。(3)基于心理学注意力模型的学生群体划分及学业表现分析。基于PA模型,探讨了一种基于一维卷积的学生注意力品质时间序列聚类方法(CNN-PA),并将其应用于学生的群体划分,探索学生的注意力品质与其学业表现之间的潜在关联。实验表明,CNN-PA应用于学生行为的聚类时,SC、CHI和DBI指标均有一定的提高。从实验结果来看,不同群体中学生的注意力品质存在一定的差异,注意力的稳定性与分配性对学生的学业表现有很大的积极影响,而广度和转移性的影响略显消极。该发现将来可用于学生的异常行为预测,并给予早期干预。更多还原

【Abstract】 The analysis of college students’ Campus Online Behaviors is crucial to their process management.Considering that students’ campus behaviors are much personalized and multivariate,it is extremely challenging to trace and monitor them by using traditional methods.Based on college students’ access logs of campus network,Wi-Fi access point logs and campus card records,this thesis intends to explore a feature representation model of students’ campus online behaviors and a behavior analysis method based on this model to support students’ personalization and proceduralization management.The main research work includes the following three aspects:(1)A campus online behavior analysis method of students based on psychological attention is proposed.Considering students’ behaviors as their attention distribution,selection and flowing on objects focused,their behavior pattern mining can be transformed into that of attention,which can provide an uniformly characterization way for students’ campus online behaviors.Meanwhile,the fusion of campus online behavior data is also explored to support the Psychological Attention calculation in students’ behaviors.(2)A campus online behavior representation model of students based on psychological attention is proposed.Driven by campus online behaviors data,a Psychological Attention(PA)computation model is constructed.Campus online behaviors are regarded as paying attentions to various objects,and potential behavioral patterns are then distinguished by attention qualities including Stability,Span,Shifting and Distribution.The proposed PA model provides a unique characterization for campus online behavior analysis.Compared with traditional statistical behavior features,those based on PA models show great advantages when it is used in student clustering.When such features are used into K-means,DBSCAN,GMM,FCM and FCM,the SCs,DBIs and CHIs all surpass those of traditional features.(3)Student clustering as well as its associations with academic performances are discussed based on PA model.A one-dimensional Convolutional Neural Networks based student clustering method is explored depending on attention quality time series.The clustering results are applied to investigate the association between students’ attention qualities and their academic performances.Experiments show that,when using CNN-PA to student behavior clustering,the indices SC,CHI and DBI have been improved to a certain extent.Experiments also represent that there are certain differences in attention qualities of students of different clusters.What’s more,it is found that both Stability and Distribution of attention have great active impacts on academic performance,but the impacts of Span and Shifting seem to be slightly negative.These discoveries can be used to predict students’ abnormal behaviors and give them early interventions.更多还原