Article Preview
TopIntroduction
The smart classroom provides more diversified, massive, real-time, and valuable data for educational big data in the smart teaching scene, as well as data for research of the in-class social network composed of students seats, interactive behaviors, etc. (Chen & Zhang, 2020). Studies related to in-class social networks are trying to go beyond the traditional methods of obtaining data through questionnaires and psychometric tests, which are not easily accessible and have subjective biases (Grunspan et al., 2014; Li & Stone, 2018; Van Rijsewijk et al., 2018). Wei and Yang (2012) used OpenCV and a skin color detector to identify students in the classroom, used linear regression to identify student seats, and constructed an in-class social network based on the co-occurrence of the corresponding students sitting in neighboring seats. Pei et al. (2018) modeled an in-class social network by capturing student photos before class using the AdaBoost algorithm for face detection and recognition, and then utilized the center projection principle and linear fitting algorithms to locate the position of students in the classroom. Beardsley et al. (2019) designed an online classroom orchestration tool, ClassMood App, to collect student data. However, the networks built by existing construction methods are mostly static networks, which cannot reflect changes in multiple areas, such as students’ seats and status, for intelligent teaching process-based education evaluation.
Based on the in-class social network, many studies have found correlations between the network features and student academic achievement, motivation, and student friendships. Pulgar (2021) noted that students with high network centrality not only have higher levels of class prestige and popularity, but also have many social connections with classmates and thus enjoy the advantages of information and collaboration. Buchenroth-Martin et al. (2017) studied the interactions of students in an evolutionary biology class and used social network analysis to find that factors such as student network centrality, gender, and attendance influenced student classroom performance.
However, most of the existing network analyses are static and holistic, and there is no study of individual students’ processes. At the same time, students are highly susceptible to the influence of peers around them in the classroom, and this subtle influence is often reflected in students’ classroom learning status and academic performance (Raca & Dillenbourg, 2013; Raca et al., 2013). Therefore, the researchers analyze the structural characteristics of the classroom network and discover the key nodes. The results of such analysis will reshape the student learning process in terms of feedback, personalization, and probabilistic prediction, which will allow educators to understand the student learning process, improve the effectiveness of classroom instruction, and predict student learning trends.
Based on the above analyses, this paper proposes a dynamic construction method of an in-class social network for real classroom environment scenarios. The authors adopt the Entropy-CRITIC combined weighted GRA-TOPSIS (EC-GTOPSIS) algorithm based on the combination of network characteristics to rank the student nodes of the in-class social network, and comprehensively analyze the relationship and evolution trend of ranking results with student friend nomination, negative friend nomination, and academic performance. The authors then mine student nodes with negative impacts. These are defined as student nodes with poor learning effects for the students themselves in the classroom setting, and a strong negative impact on those around them. Thus, this paper provides a novel perspective for teachers to evaluate students’ learning processes. It offers the following contributions:
- 1.
Construction of an in-class social network using the dynamics of student classroom seating similarity to better reveal what underlies student seating relationships and reflect subjective trends in student seating choices.
- 2.
Proposal of a multi-decision fusion algorithm for negative node mining within the in-class social network, which can not only select nodes with a negative impact in classroom, but also reflect the future trends of students.