Daily Peer Relationships and Academic Achievement among College Students: A Social Network Analysis Based on Behavioral Big Data

Abstract

This study aimed to detect college students’ daily peer networks through a behavioral big-data-driven social network analysis and to explore the relationship between college students’ daily peer relationships and academic achievement. We collected data on the class attendance, eating, and bathing records of 4738 undergraduate students who entered a university in 2018 to infer the daily peer relationship networks of students. The Louvain algorithm and some network indicators such as density and average clustering coefficient were used to investigate social network characteristics of peer relationship networks. The findings show that initially, students in the same dormitory tended to form daily peer relationships, gradually shifting toward relationships centered on classmates as time progressed. These peer networks often aligned with geographical location and living areas. Moreover, the peers of college students who received high-level scholarships were more likely to receive scholarships. The number of peers was positively correlated with the likelihood of receiving a scholarship. The research findings contribute to the application of information technology to promote the sustainable development of higher education and individual students.

1. Introduction

Peer relationships are interpersonal relationships that are established and developed in the course of interactions between individuals at comparable levels of psychological development, which are different from the vertical relationships that exist among individuals interacting with their parents or with older individuals [1,2,3]. For many students, the transition to college marks the first time they are separated from close family and friends [4], and they are required to spend time with strangers and adapt to a new social environment. By developing strong friendships with peers, college students can share each other’s experiences, learn from each other, and provide emotional support that can help with adjustment to college life and academic development. In addition, separation from family and friends may lead to some degree of loneliness, but by establishing connections with peers, college students are able to alleviate loneliness and reduce anxiety and depression [5,6,7]. However, not all peer relationships are positive. Some peer relationships may have a negative impact on college students. For example, some peer groups may encourage unhealthy behaviors, such as excessive drinking [8], vaping [9], and truancy [10]. When college students are in such peer relationships, they may be tempted to engage in undesirable behaviors. Therefore, it is of great significance to study the formation and influence of college students’ peer relationships on their personal development.

Prior research has confirmed that academic achievement is strongly linked to peer relationships [11,12]. For example, peer relationships can promote individuals’ cooperation with others [3], enhance achievement motivation [13], influence course preferences [14], reduce cell phone addiction [15], and dissipate the negative effects of online education on academic engagement [16]. Having a trustworthy and loyal friend during the first semester of college was found to be associated with a higher grade point average (GPA) [17]. Students who have good relationships with their peers are less likely to drop out [18]. Academic achievement also plays a role in peer relationship choices. The higher a student’s academic performance, the more likely that student is to be chosen as a friend, and the more often this better-performing student establishes friendships and academically beneficial relationships [19]. Students with similar academic achievements are more likely to be friends and help each other [20]. Most of the current research on peer relationships is based on self-reports or conventional classmate and group relationships, with small data samples and research contexts limited to the classroom or extracurricular activities.

However, college campuses not only provide a learning environment but are also important life centers for college students. Various activities in life such as class attendance, eating, and bathing may provide opportunities for peers to exert social influence, and only a few studies [21,22] have investigated the association between peer activities in college students’ daily lives and academic achievement. Due to the influence of Confucian culture in East Asian countries, Chinese universities emphasize the collective concept and collectivist values, and students have a strong collective identity and are willing to participate in collective life [23]. In the daily life of Chinese university students, eating, bathing, and attending classes are three common behaviors in which students have a high degree of autonomy and can choose their peers independently. Specifically, eating with peers is considered an important way to socialize [24]. A survey of 2000 adults over the age of 18 showed that 93% eat with family and friends at least sometimes, and eating with others was considered by respondents to be an important way to engage in community and strengthen friendships, facilitating social and emotional support [25]. The orderliness of eating was also found to be associated with the GPA of students [26]. Peers may also be inclined to choose similar courses and to travel to classrooms together; for example, previous research found that exposure to female peers who excelled in mathematics increased the likelihood that women would choose science courses [27]. Regarding bathing, due to the large number of students in Chinese higher education institutions, many schools in China do not have individual bathrooms in dormitories due to operational and maintenance costs, and students bathe mainly in public bathrooms, which creates conditions for students to travel to the bathrooms in pairs. For example, students usually go to the shower in pairs after playing sports together. In addition, although students have a variety of peer activities, such as going to the gym and bars, most of them are not as prevalent and necessary as eating, bathing, and attending classes. Therefore, eating, bathing, and attending classes reflect students’ everyday peer socialization and provide the premise for the analysis in this paper.

The development and application of behavioral big data have made it possible to observe the daily peer relationships of college students. The lives of college students living on modern campuses involve a combination of studying, eating, exercising, and socializing. The rapid development of modern information technology allows students’ activities (e.g., going to the library) to be recorded in databases so that students’ behaviors can be tracked. We can also collect records of students’ living and learning activities unobtrusively through MOOC, campus WIFI, and smart cards [26]. Recent studies have proposed utilizing these multi-source data in predicting academic performance to promote positive student–university interactions and improve the quality of higher education [22,28]. For example, by collecting longitudinal behavioral data on 6597 students’ smart cards, previous studies have found that students’ diligence, orderliness, and sleep patterns were strongly associated with academic performance. Students with similar behaviors were found to have correlated grades [29]. Some studies used online learning and teaching systems to collect data on time on tasks, class attendance, seating location, and group contact. They found that student performance was positively correlated with attendance, social stability in terms of peer grouping, and time spent on tasks and negatively correlated with the distance between the student’s seat and the lecturer’s location [30]. Data collected through questionnaires or self-reports are small in scale and may be affected by respondent bias, whereas behavioral data are large in scale and provide a relatively objective and nuanced picture of students’ daily activities.

Although the abovementioned studies initially used behavioral big data to explore the characteristics of college students’ behavioral patterns, they focused mainly on the association between individual behaviors and factors such as grades by profiling college students’ behavioral patterns, whereas less attention has been paid to the peer relationships among individual college students. As mentioned earlier, college students’ peer relationships constitute an important environment for the development of individuals’ academics, interests, and values. Therefore, this study was conducted to examine the process of forming peer relationship networks and their associations with academic achievement among college students in everyday situations by analyzing their static data and the data recorded on terminal devices for attending classes, eating, and bathing. To achieve the abovementioned objectives, students’ daily peer relationships were identified according to the terminal records of three activities of students, i.e., attending classes, eating, and bathing. Subsequently, we constructed peer relationship networks, identified peer communities within these networks using the Louvain algorithm, and explored the relationship between daily peer relationships and scholarship attainment. Specifically, the following research questions were addressed:

• Research Question 1: What factors are associated with the formation of daily peer relationships (reflected by attending classes, eating, and bathing) among college students after enrollment?
• Research Question 2: What are the characteristics of college students’ daily peer relationship networks?
• Research Question 3: What are the associations between college students’ daily peer relationships and their academic achievement?

The contribution of this study is to quantitatively describe students’ peer relationships and their association with academic achievement through behavioral data and a social network analysis. In particular, (1) this study provides a way of calculating peer relationships and helping to advance data analytics in the field of education based on the terminal data analysis of a university class of 2018 for 17 months after enrollment. This can provide empirical data support for the education theory and confirm the strong potential of behavioral data combined with archival data to reveal social regularities. (2) From a practical perspective, this study shows that more peer relationships could help students connect to college life and improve adaptability, which is also beneficial for their academic performance. This can help educational administrators guide students to perform better, as well as help identify undesirable and abnormal behaviors so that scientifically effective interventions can be implemented promptly. In addition, this study can provide a reference for universities to organize activities to promote education sustainability and a good learning atmosphere.

The remainder of this paper is structured as follows: Section 2 is the literature review of peer relationships and academic achievement, Section 3 presents the process of the data collection and analysis, Section 4 is the analysis of the empirical results, and Section 5 is the discussion and potential implications.

2. Literature Review

2.1. Peer Relationships of College Students

Peers play an integral role in higher education, shaping various aspects of college students’ lives, including their behaviors such as eating habits, bathing routines, and class attendance [1,31]. Peers have a strong influence on the behavior, cognition, and emotional engagement of college students. Increasing students’ exposure to highly engaged peers and avoiding the clustering of deviant peers promote greater academic success for youth [10].

Homogeneity, i.e., having characteristics that are similar in some way, can explain the formation of peer relationships, and college students tend to form friendships with classmates of the same age, gender, and race [32]. For example, female students are more likely to form peer relationships based on gender and age similarity, and both genders equally tend to form relationships based on informedness similarity regarding the content and study conditions of their study program [33]. In a study of email interaction data from students and recent graduates, it was found that two randomly selected White students interacted three times as often as one Black student and one White student [34]. Findings from a sample of Black college students suggest that students with strong racial identities tend to feel less connected to their college campuses and that Black students are more socially connected to students with similar backgrounds than to the campus as a whole [35]. Racial segregation is largely driven by preference rather than by the school system [36]. In addition, linguistic, cultural, and value differences within the same race can hinder the formation of peer relationships. For example, in a Hong Kong university with predominantly local students, only 18% of mainland students’ best friends were local [37].

Individual factors such as personality, academic performance, and experiences can also explain the establishment of peer relationships. It has been found that students who stayed in close contact with former friends were less likely to make new friends in college, and students who lived on-campus were more likely to separate from former friends and make new friends within the college community than those who lived off-campus [4]. Personality is an important factor in the formation of peer relationships. Prior research revealed that individuals high in extroversion tend to choose more friends than individuals low in extroversion, and individuals high in affinity tend to be chosen as friends by more people. In addition, individuals tend to choose friends with similar levels of affinity, extraversion, and openness [38]. Students with higher pro-social attitudes tend to nominate more of their peers as friends but are less likely to be nominated as friends by their peers [20]. College students also adjust peer relationships based on their academic performance. For example, by analyzing the social networks of 6000 Russian students from public high schools and universities over 42 months, it was found that the students’ social networks were academically homogeneous and that the students were more likely to gradually reorganize their social networks based on their level of performance rather than adjust their performance to reach the level of their social networks [2].

Peer relationships play a unique and irreplaceable role in an individual’s psychology, attitude, and behavior [39,40]. Specifically, on an attitudinal and psychological level, friendships among college students’ peers can provide social and emotional support, reduce loneliness, and help individuals adjust to college life [41,42]. By interacting with partners from different backgrounds as well as like-minded individuals, individuals can build confidence in their identity and increase their understanding of social diversity. College students often judge whether they are making progress by comparing and competing with their friends [43]. On a behavioral level, good peer relationships can reduce the negative effects of cell phone addiction on physical activity among college students, whereas poor peer relationships exacerbate cell phone dependency frequency and limit physical activity [44]. Furthermore, the impact of peer relationships extends to various everyday behaviors. For instance, the phenomenon of social eating is evident, where students often engage in meal consumption simply because their friends are doing so, even when not hungry or having recently eaten [45]. Similarly, students’ hygiene practices are shaped by peer influence, with instances of girls emulating their peers by carrying hand sanitizer or toilet paper in their bags [46]. This similarity in peer behavior once again proves that records of daily activities can reveal peer relationships.

2.2. Peer Relationships and Academic Achievement

Academic achievement measures allow educators to assess students’ abilities against specific learning goals. They are also used as criteria for various educational selection processes [47]. Research generally recognizes that peer relationships can have an impact on academic achievement. On the one hand, previous studies have compared the relationship between different types of peer relationships, such as friends, lovers, and study groups, and academic achievement. For example, the influence of study partners and friends was found to be differential, with student achievement increasing when the peers were friends-cum-study partners and study partners but not friends, whereas friends-cum-non-study partners did not have such an effect, and researchers suggest that peer-influenced knowledge-sharing channels have a greater impact than friendship-based role modeling channels [48]. Research has also examined the impact of academic support from the closest same-sex friends and lovers on academic success, finding that academic support from friends rather than lovers was positively associated with student engagement, which subsequently predicted the GPA [49]. Roommate peer effects have been found to potentially exist in a small group of students, but they were not found to be a key determinant of students’ GPA [50]. A negative impact of underachieving peers was found. Peers with grade rankings in the middle and lower rankings of the major were found to have a significant negative impact on individual students’ academic performance, whereas peers with grade rankings in the upper reaches of the major had a significant positive impact on individual academic performance. Male students were more likely to be negatively impacted by peers with lower academic performance, whereas females were more likely to be impacted by those with better academic performance [51]. Additionally, studies comparing the effects of classroom groups and randomized groupings during orientation week on the subsequent academic outcomes of college students have shown that socializing in an informal setting fails to produce significant peer effects [52]. On the other hand, the network structure of peer relationships also affects academic achievement. Study group cohesion is positively associated with student achievement, while achievement heterogeneity in study groups can have a negative effect [11]. In recent years, with the development of digital campuses, a study extracted student friendship relationships through co-occurring data from smart cards and found an inverted U relationship between the GPA ranking and the number of friends, with the richness of structural holes and homogeneity of friends positively affecting the GPA ranking [22].

Social capital theory can explain the influence of peers on academic achievement. Lin defines social capital as resources embedded in social relationships that are used for purposeful action [53]. Students access the social capital available in their networks through interactions with change agents. Social capital can be categorized as either expressive, which can provide emotional support (e.g., encouragement) and may help to maintain and enhance existing resources as well as emotional or physical well-being, or instrumental, which provides an individual with new or additional resources (e.g., information) to assist in the achievement of specific goals. From an expressive social capital perspective, previous research has shown that high-quality relationships promote higher academic achievement by increasing school well-being and a sense of belonging, enabling students to successfully adapt to the norms and attitudes of the academic culture [54,55,56]. Receiving feedback from peers helps students identify and understand their mistakes before taking exams and can improve academic self-concept and promote higher academic achievement [57]. It has also been found that students exposed to more persistent peers achieve higher grades. Students who encounter persistent peers at the beginning of their studies may develop better study habits and a different social network, which may yield academic rewards [58]. From an instrumental social capital perspective, students who refer to other students’ learning strategies during the learning process have been found to have more positive online learning behaviors, such as reading supplementary materials about incorrect answers [59]. In addition, it has also been found that peers can form norms about the importance of academic achievement, which in turn creates peer pressure on individuals to influence their academic achievement [60].

In conclusion, the literature suggests that college students’ peer relationships are the result of a combination of social, cultural, and individual personality and experience factors and that high-quality peer relationships can help individuals adapt to college life, enhance their sense of belonging, and, in turn, contribute to their academic achievement. Previous studies have presented limited contexts in which peer relationships can have an effect on classroom activities, and less consideration has been given to peer relationships in daily activities. Peer relationships in daily life are chosen by students on their initiative and are an important part of university life and an important scene for the formation and development of individual values. Furthermore, eating, bathing, and attending classes are common daily activities of college students and can reflect the social and peer relationships of college students, including those in China. Therefore, by focusing on daily behaviors such as eating, bathing, and class attendance, we aim to shed light on how peer relationships are formed, how they evolve, and how they are interrelated with academic achievement.

3. Methodology

3.1. Data Collection

Behavioral big data (BBD) refers to giant and rich multidimensional datasets containing information regarding human and social behaviors, which can be used by companies, governments, and researchers. Many researchers in the field of social sciences and management acquire and analyze BBD to extract knowledge and scientific findings [61]. Terminal device data are widely used by researchers in a behavioral big data analysis in fields such as higher education. An intelligent campus terminal system is an important part of digital campus construction, and its functions include authentication, financial transactions, comprehensive consumption, information services, and data management. For college students, who spend most of their time at school, most of their consumption behaviors can be accomplished and recorded through smart campus terminal systems [62]. Recent studies have begun to employ location data to profile students and use smart terminal logs from campus buildings and class attendance record data to investigate the relationship between students’ behavioral trajectory patterns and academic performance [63].

The student data in this article were obtained from and authorized by the university’s Network Information Centre, which is responsible for the overall planning and comprehensive coordination of the university’s network security and informatization work. To protect the students’ privacy, the data were desensitized to ensure that they cannot be re-identified, thus safeguarding the data security of student information. At the level of data collection, students’ information data were anonymized, and privacy was desensitized using non-sensitized collection. At the level of data use, the authority to apply for and approve data use was clarified, and effective process monitoring was carried out. Various levels of privilege control on data use were carried out to prevent illegal intrusion by unauthorized users and unauthorized use by authorized users. At the data storage level, the encrypted storage of student users’ private information was implemented. In addition, we analyzed data from a global perspective rather than an individual perspective, making individual contributions invisible to avoid the leakage of students’ privacy. Additionally, these data cannot be shared with third parties under privacy requirements. The above student data were used for research purposes only to better serve the management of student affairs.

Finally, we acquired terminal device-recorded data of 4738 undergraduate students enrolled in a Chinese university in 2018; the data were collected from September 2018 to January 2021 from two campuses. The data cover the first three semesters after enrollment and mainly include records of attending classes, eating meals, and bathing, which could reflect the main daily activities of students during their school years. In addition to behavioral data, we also obtained students’ information such as class, dormitory, and province, as well as scholarship and group activity data.

3.2. Construction of Daily Peer Relationship Networks and Community Detection

Social network analysis methods can help discover the characteristics of relationships among entities and provide insights into social phenomena [64]. They were used to analyze peer relationships among students. In the networks established, students are represented as nodes, and the peer relationship between them is represented as the edges between nodes and calculated through their records on terminal devices. According to previous studies [22], if two students are peers, they are more likely to use terminal devices at the same location much more frequently than a pair of strangers in a short time interval. Based on this idea, the daily peer relationship networks of students were established based on spatiotemporal co-occurrence relationships. Specifically, when a user checks in at a location, the server records the user ID and location information l with time information t, indicating the check-in access record of user u at location l at moment t. Accordingly, we considered that if two individuals are present at the same time within the time interval τ before and after the same swipe location l, there exists a spatiotemporal co-occurrence relationship that creates a connecting edge between the two user nodes. τ is a subjective parameter related to the actual application situation. The larger the value of τ, the more likely that students who are not peers will be treated as peers. Previous studies have typically set τ values ranging from 20 s to 1 min [22,65,66]. Accordingly, we set the value at 20 s to make it far less likely to identify students who did not know each other and would be inaccurately counted as friends. We considered that there was a peer relationship between two students if they checked in within 20 s on the same terminal device and there were less than or equal to four students checking in between them (excluding themselves).

We identified peer relationships from the terminal records of the three behaviors of attending classes, eating, and bathing separately using Networkx 3.1, a Python package for a network analysis. The peer relationships of all students were represented as weighted undirected networks where the edges between nodes were mutual and had a numerical weight representing the closeness of peer relationships. The weight of an edge was increased by 1 when two individuals co-occurred on the same day to indicate a stronger peer relationship between individuals. The threshold measures the closeness of the peer relationship. The higher the threshold, the closer the peer relationship. By adjusting the threshold, we can explore the relationship between peer relationships and academic achievement in the peer relationship network with different closeness. For example, when the threshold was 2, then the networks contained only peer relationships that co-occurred more than or equal to 2 times. In daily life, students may sometimes not use terminal devices with their peers but go to class, eat, and shower on their own. Therefore, they may use terminal devices before or after strangers. We believe that the number of co-occurrences between two students equal to 1 may represent coincidence and cannot accurately reveal the existence of a peer relationship between them.

However, for different behaviors, co-occurrence networks varied dramatically. For example, for the bathing data, since peers may go to a public bathroom together, we consider that students who use the same public bathroom within certain intervals are more likely to have a peer relationship. For the class networks, students who were selected to attend the same class had a high co-occurrence number, i.e., many students signed in at short intervals during the upcoming class, which led to a high clustering of students in the same class in the networks. In contrast, the eating networks preserved more spatiotemporal information and were less likely to be disturbed by other factors, better capturing the relationships among students. Therefore, to fully utilize the eating, classes, and bathing data to build student peer relationship networks, the networks formed for bathing and classes were again merged into the eating networks to form the final networks by adjusting the threshold upwards to incorporate the connecting edge information. Accordingly, we set up multiple networks for each month and the first year.

Furthermore, some students’ behaviors may not have a direct peer relationship, but their information may be transmitted with the help of others in the community to have an impact on individuals. For example, previous research has found that characteristics of both friends and friends of friends independently predict students’ college aspirations and their risk of dropping out of high school [67]. In a social network analysis, a community is defined as some sort of cohesive substructure with a high degree of connectivity [68]. Some members of the community may have an important impact on the community. The influence of the members of the community is not the same, and some members may have a significant impact on the community [64]. Therefore, we believe that students with different levels of academic performance may be associated with the overall academic performance of the community. To examine the association between student communities and students’ academic achievement, with the help of the well-known Louvain algorithm [69], we identified communities in the peer relationship networks. The algorithm has two steps. In the first step, it assigns each node to one community, and then for each node, it tries to find the maximum positive modularity gain by moving each node to all the communities of its neighbors. If no positive gain is found, the node remains in its original community. By adjusting the resolution parameter of Louvain’s algorithm, the size of the discovered communities can be adjusted.

3.3. The Analysis of Academic Achievement

Academic achievement can be measured in many ways. In this study, we chose scholarships to quantify academic achievement. Although previous studies have primarily used the GPA to measure students’ academic achievement, scholarships can provide information about a student’s overall performance. Scholarships are judged not only based on a student’s course grades but also based on the student’s participation in social services, clubs, group activities, and academic competitions. In addition, the GPA levels of different majors tend to have large differences due to different courses and different assessment methods. While scholarships are often distributed among majors based on the number of students, receiving a scholarship can prove that a student’s performance in their major is outstanding. Therefore, quantifying individual students’ academic achievement through scholarships can avoid the distractions associated with major attributes. This study focused on the awarding of four types of scholarships (A, B, C, and D) at the end of the first academic year and examined whether having peers who won scholarships affected the likelihood that an individual would win a scholarship. Scholarship A is a scholarship funded by the central government to reward outstanding students. It is the highest level of national scholarship that students of higher education can obtain at present. Its evaluation is the most standardized, and the standards are the most strict. Scholarship B is jointly funded and established by the central and local governments to reward and support students from poor families who have excellent moral and academic performance. Scholarship C is a type of scholarship set up by the school board, alumni, and caring people from all walks of life to donate money to the school foundation, and its difficulty in obtaining and evaluation criteria are second only to those of scholarship A. Scholarship D is a type of scholarship funded by the school that is less difficult to apply for.

4. Results

4.1. The Descriptive Analysis

Table 1. Data description of student characteristics (N = 4738).

Characteristics	Category	Count (%)	Characteristics	Category	Count (%)
Gender	Male	3223 (68.0)	Hometown	Northwest	1548 (32.7)
	Female	1515 (32.0)		Eastern	1038 (21.9)
Campus	a	3856 (81.4)		Central	669 (14.1)
	b	882 (18.6)		Northern	545 (11.5)
Living area	A	484 (10.2)		Southwest	502 (10.6)
	B	398 (8.4)		Northeast	199 (4.2)
	C	420 (8.8)		Southern	193 (4.1)
	D	391 (8.2)		Hong Kong, Macao, and Taiwan	44 (0.9)
	E	743 (15.6)	Scholarship	A	74 (1.6)
	F	381 (8.0)		B	196 (4.1)
	G	721 (15.2)		C	123 (2.6)
	H	557 (11.7)		D	1163 (24.5)
	I	643 (13.5)		None	3182 (67.2)

summarizes the characteristic distribution of the students. Of all students, 68% were male and 32% were female, which may be due to the school’s focus on science and engineering. According to statistics, the proportion of science and engineering students in the school is 66%, while the students majoring in literature and history and art account for 7% and 1%, respectively. The school has two campuses, with the majority of students (81.4%) concentrated on Campus a and the remaining students (18.6%) on Campus b. In terms of living areas, the school is divided into nine residential areas (A–I), and the number of people in each living area is roughly similar, among which living areas A and B are located on Campus b and the rest of the living areas are located on Campus a. In terms of origin, students are mainly from the northwestern (32.7%), eastern (21.9%), and central (14.1%) provinces of China. In the first academic year, the majority of students did not receive scholarships (67.2%); 24.5% received scholarship D, followed by B (4.1%) and C (2.6%), whereas only 1.6% received scholarship A.

4.2. Formation and Evolution of College Students’ Daily Peer Relationship Networks

We analyzed the effects of dormitory, class, and province on peer relationships (see

Table 2. The proportion of three types of peer relationships under different thresholds in the peer relationship network.

Month	Threshold	Province (%)	Class (%)	Dormitory (%)
1	2	10.02	7.95	7.95
	5	15.39	32.11	73.69
	10	13.95	42.63	87.63
2	2	10.31	8.23	8.39
	5	15.69	32.91	73.28
	10	15.45	35.19	86.70
3	2	10.23	7.68	7.46
	5	16.62	31.23	67.96
	10	15.94	40.58	84.93
7	2	9.30	8.75	6.48
	5	15.27	34.15	59.80
	10	16.71	47.56	76.61
12	2	14.33	19.03	11.03
	5	22.41	54.82	56.27
	10	26.19	71.43	66.67

). As the threshold increased, the daily peer relationship in the network became more intimate. In the first 3 months after enrollment, when the threshold was 10, the share of peer relationships in the same dormitory was much higher than that in the class and province. It can be argued that the dormitory was most important for the formation of peer relationships at the beginning of enrollment. However, as time elapsed, the proportion of peers in the same dormitory gradually decreased, and peer relationships were based more on classmates. In the 12th month, when the threshold value was 10, the peer relationship proportion of classmates was 71.43%, which was higher than that of peers in the same dormitory (66.67%).

4.3. Network Characteristics of Daily Peer Relationships of College Students

To characterize the networks of daily peer relationships, the networks in the first academic year were visualized using the Force Altas layout algorithm, which is designed to arrange the nodes and edges of a network in a way that reduces edge crossings, and groups nodes with stronger connections closer together [70]. As shown in Figure 1, each node represents a student, where the node-connecting edges indicate peer relationships and the node size indicates the degree of the node, which reflects the number of peers of an individual. The peer relationship communities present in the networks were discovered using Louvain’s algorithm, and the node color indicates the peer relationship community to which it belongs. As can be seen from the figure, the daily peer relationship networks presented two closely connected groups. We found that the left group mainly consisted of the students of the university’s Campus a, and the right group mainly consisted of the students of the university’s Campus b. The two campuses are located in two different areas of the city, so the peer relationships between the students of the two campuses were found to be relatively sparse. Some students in the networks were found to have peer relationships across the two campuses, which may be because they took classes across campuses or had friends across campuses. There was also a portion of students who were identified to be located on the fringes of the networks who were less likely to relate to other students, which may be because their homes were near the campus or they lived off-campus.

Furthermore, multiple peer relationship communities (indicated by different colors) were identified within both groups. It was found that students belonging to the same community were more likely to be from the same living area or to have the same major, which increased the likelihood that they would interact with each other. For example, for the community in green in the upper right corner, 86.78% of the members were identified to be from living area B and they were mainly medical students. For the larger blue neighborhood in the lower left corner, 44.63% of the members were from living area G and 38.55% of the members were from living area H, and the two living areas are the closest to each other in terms of location in the campus. The above analysis further confirms that the geographic location in which students live and that the similarity of disciplines affect their peer relationships in their daily activities. In addition, the relatively even distribution of node sizes in the figure suggests that many students established stable peer relationships with other students.

Figure 2 presents the changes in the network density and average clustering coefficient of college students’ daily peer relationships. Network density is the ratio of the number of edges that exist in the network to the upper limit of the number of edges that can be accommodated, which reflects the density of peer relationships. Figure 2 shows that the network density of college students’ peer relationships was relatively high at the beginning of the first semester and then gradually decreased, which may be due to the orientation and group activities at the beginning of the semester that promoted peer relationships among students, and with the gradual development of courses, the peer relationships of students tended to stabilize, and the density of the networks decreased. February and August 2019 were winter and summer vacations, which resulted in a lack of contact between students, leading to a decrease in networking density. Starting from September 2019, the density of the networks increased, which may also be due to the subdivision of majors that allowed students to develop new peer relationships. During this period, the students need to make a further choice of sub-specialty from the broad category of majors to achieve a professional education.

The orange line in Figure 2 demonstrates the change in the average clustering coefficient. The clustering coefficient reflects the probability that a student’s daily peers are peers with each other. During the first semester, the average clustering coefficient was found to gradually increase, reflecting the transmissive nature of peer relationships and the tendency for students to form cliques among themselves. The average clustering coefficient peaked during the winter break in February 2019, possibly because there were fewer students on campus during the winter break when school classes ended. Because the calculation of the average clustering coefficient may be underestimated due to the influence of misidentified peer relationships, while the recognition of peer relationships in winter vacation is less affected by the noise generated by a large number of students gathering to use terminal devices, the average clustering coefficient is higher and can better reflect the real peer relationships. In addition, the mean clustering coefficients showed an increase with the subdivision of majors in the third semester, which suggests that the establishment of new peer relationships may be based on common peers among students. The changes in the above-given network indicators correspond to real events and reflect the characteristics of peer relationships among college students.

4.4. Association between Daily Peer Relationships and Academic Achievement

Based on the peer relationship networks from September 2018 to August 2019 (before the scholarships were awarded in the first academic year), we analyzed the probability of winning scholarships for the direct peers of the scholarship winners and other students in the community. It can be seen in Figure 3 that as the threshold of the networks increased, the proportion of winning students in neighboring nodes increased. It was found that for different scholarships, peers of students who won scholarship A had the largest possibility of winning scholarships. When the threshold was 19, the probability that peers of students who won scholarship A would win a scholarship was 57.35%. The proportion of peers winning scholarships was close for students who won scholarships B and C. Students who won scholarship D provided less of a boost in the probability of their peers winning a scholarship, with peers of students who won scholarship D having a 32.14% probability of winning a scholarship when the threshold was 19. In addition, peers of students who did not win a scholarship had a lower probability of winning a scholarship. This reflects the homogeneity of the academic achievement levels among peers.

Figure 4 shows the association between a student receiving different types of scholarships and the proportion of peers in their community receiving scholarships. As the resolution parameter increased, the communities’ Louvain’s algorithm finds are smaller and more connected. On the one hand, similar to the results at different thresholds, the proportion of winning a scholarship was found to be greater for classmates in the same community as students who won scholarship A, followed by scholarships C, B, and D. The association between students who received scholarship D and the proportion of students in the community who received scholarships was similar to that of students who did not receive scholarships. This reveals that the students with better performance had a stronger positive link with the overall academic performance of the community. In addition, we found evidence that peer community influence was weaker than direct peer relationship influence. For example, when the resolution was 9, the networks were divided into 260 communities, and the probability that a peer of a community member with scholarship A would win a scholarship was calculated to be 25.59%. In contrast, at a threshold of 2, the probability that a direct peer of a scholarship A winner would win a scholarship was calculated to exceed 30%. It was established that different individuals had more influence on those closer to them while the influence on others in the community gradually decreased as the community increased.

To obtain more detailed conclusions to assist school administrators, the proportion of students winning scholarships based on the degree of centrality (number of peers) was further analyzed. Figure 5 shows the proportion of students winning scholarships at different quartile levels of degree centrality. While it is important to acknowledge the potential presence of measurement errors in the data, we observed a consistent trend. As the number of peers associated with a student increased, there was a corresponding increase in the probability of winning a scholarship. This probability tended to stabilize at approximately 35%. For students with fewer than or equal to three peers (representing the 10% quartile), the probability of receiving a scholarship was lowest, at 22%. Conversely, for students with a higher number of peers, falling within the 90% to 100% quartile (with each student having between 30 to 166 peers), the probability of scholarship attainment was highest, reaching 40.7%. To further validate this observation, our correlation analysis yielded consistent results. Spearman’s correlation coefficient between the probability of winning a scholarship and the number of peers was 0.18 (p < 0.001) for individuals with fewer than or equal to five peers and only 0.05 (p < 0.001) for individuals with more than five peers. Additionally, it is noteworthy that the quantity of peers demonstrated the most substantial correlation with receiving scholarship D (Spearman’s rho = 0.08, p < 0.001), followed by scholarship B (Spearman’s rho = 0.04, p < 0.01) and scholarship C (Spearman’s rho = 0.03, p < 0.05). Notably, no significant correlation emerged between scholarship A and the number of friends (Spearman’s rho = 0.02, p = 0.19).

Group activities and honors achievements can reflect educational success, so we analyzed students’ group living areas and group activities. The university in this study has nine living areas. Living areas have a certain degree of autonomy and are the units that organize and manage the extracurricular life of the students. Figure 6 presents the average clustering coefficients of the students within the different living areas in the university. It can be observed that the clustering coefficients of living areas A and B were higher and their students were more closely related to each other in terms of peer relationships. This may be because living areas A and B are both located on Campus b of the university, which has fewer people, making the students’ peers more concentrated within the living areas. In contrast, the relatively lower clustering coefficients for the other living areas may be because the main campus, Campus a, has a large number of students and a high diversity of majors. This makes students’ peers not only confined within the living areas but also between the living areas.

In addition, universities and their student unions organize a variety of activities after the start of the academic year, especially for new students, to help students integrate and build relationships on-campus [71]. The “Best Class” is a decades-old student event at the university. The unit of participation is not individual but an administrative class. The event aims to promote class cohesion, improve the overall quality of students, and spread social values through a variety of activities and presentations on a specific theme. The “Best Class” event is first organized in different living areas, and then some representative classes are selected to compete at the university level, so the degree of internal cohesion in living areas may be related to the participation in the “Best Class” activity. We compared the number of classes in different living areas that were awarded Best Class. Six classes in living area D were awarded Best Class, followed by four classes in living area G, and compared to other living areas on the same campus, their average clustering coefficients at thresholds of 12 and 22 were relatively high. Living areas F and H had only one class that earned Best Class, and their average clustering coefficients at thresholds of 12 and 22 were lower than those of the other classes. The average clustering coefficient at threshold 2 did not seem to be directly related to the number of Best Classes. This implies that there is a stronger correlation between the acquisition of collective honors and closer peer relationships. Increasing group activities and promoting closer peer relationships to enhance team cohesion may be considered in living areas with lower clustering coefficients.

5. Discussion and Conclusions

5.1. Main Findings

This study analyzed the formation and evolution of college students’ daily peer relationship networks based on the terminal device recordings of their daily classes, meals, and bathing in the first three semesters after their school enrollment and analyzed the relationship between college students’ daily peer relationship networks and their scholarship awards and group activity awards. It was found that the formation of college students’ peer relationship networks after enrollment was most closely related to the dormitory and that college students’ peer relationships gradually stabilized over time but were affected by external factors such as the subdivision of majors. Moreover, we observed changes in the density and clustering coefficients of college students’ peer relationship networks. By combining the scholarship data, it was found that being peers with classmates who won higher levels of scholarships increased the probability of winning scholarships for the students, and this peer influence existed not only in direct relationships but also in communities. Collective honor was also found to have a relationship with the average clustering coefficients of the networks. The results of the study are discussed below.

First, the peer relationship networks of enrolled college students were found to have associations with dormitories, classes, and provinces, while more intensive peer relationships occurred primarily in dormitories after enrollment. This is mainly due to the similarity of members within university dormitories. Despite the randomness of college dormitory assignments, dormitory members tend to enroll in the same majors and classes, which gives them similar schedules and trajectories. Another explanation is that at the beginning of the freshmen’s enrolment, peer relationships are not very obvious, relevant courses and activities have not yet been fully developed, and students’ social scope and social relationships are relatively limited, so their social circle is restricted to the dormitory field. We might argue that dormitory relationships have strong links with individuals’ sense of belonging to the school. Prior research has also revealed that dormitory roommates have a strong influence on individual decision making and that the density of roommates with the same major discourages students from changing majors [72]. In addition, as time progressed, the proportion of classmates among their peers increased and the proportion of dorm roommates among their peers declined, which indicates that the classroom becomes an important place for peer formation and influence. Classroom teaching can adopt some strategies such as group cooperation to promote the formation of peer relationships.

Furthermore, we analyzed the network characteristics of students’ daily peer relationships. Communities in the networks of peer relationships were found to be strongly linked to geographic location and major, with close peer relationships forming within different living areas. This is due to geographic proximity increasing the probability of students in the same living area meeting each other and due to the organizational management of the living areas. Living areas organize and manage students’ study and life based on dormitories, which enhances internal connections within the same living area through academic tutoring, extracurricular activities, and the establishment of learning communities. On the other hand, the findings on network density and average clustering coefficients were similar to the findings on the evolution of peer relationships. The densities of peer relationship networks were higher after enrollment and then began to decrease, reflecting the gradual stabilization of peer relationships. In contrast, the increase in network density and clustering coefficients after the major division indicates the expansion of peer relationships as well as the transmissibility of peer relationships. It was found that after the further division of majors, if two students had common peers, they were prone to establish new peer relationships. This reflects the fact that students’ existing peer relationships can help them acquire new peer relationships and help them integrate into the group in the future, but it may also exacerbate the solidification of the peer relationship structure and isolate students with fewer peer relationships. Therefore, university institutions should pay attention to important points in the periods when peer relationships are formed and help students on the periphery of social groups to facilitate their full integration into the community.

Moreover, academic achievement is strongly associated with college students’ daily peer networks, as demonstrated with the fact that the peers of students who won high-level scholarships were more likely to win scholarships. Prior research has explained the association between peer relationships and academic achievement through both influence and selection mechanisms. On the one hand, upon entering college, students tend to cluster with friends who perform similarly under the effect of homogeneity [60], which can be seen as a manifestation of social capital, as students receive support, encouragement, and information through connections with similar peers, which help improve their academic achievement. On the other hand, students’ academic achievement is linked with the academic achievement of their peers. From the perspective of social capital, the academic achievements of peers can be viewed as a resource that enables knowledge transfer, learning cooperation, and competition with each other through social networks. However, the task of clarifying the roles of these two is complex, and the association between students’ academic achievement and their peers often results from the competition and mixing of the two mechanisms [8]. Our results based on behavioral big data demonstrate that there was a positive correlation between the number of peers and scholarship attainment, and this link was stronger in groups with fewer peers, which is consistent with previous research that found a law of diminishing marginal effects of the number of friends on students’ GPA [73]. This may be because peers can provide individuals with social support and help them develop a sense of identity with the social norms in school, which in turn promotes individual motivation to win scholarships. However, as the number of peers increases, students need to balance their time and energy between befriending peers and studying, and some of the peers may be recreationally oriented peers that can distract students from their academic commitments, which makes the gains from the number of peers diminish and may even have a negative impact on academics. Therefore, college education should focus on marginalized groups in peer relationship networks to help them increase their recognition of the importance of academic achievement.

In addition, residential areas with closer ties are more likely to obtain collective honors. On the one hand, the presence of close social networks and friendships among students fosters a supportive environment. This environment, in turn, facilitates cooperation and the exchange of information among students. Consequently, it becomes easier for them to reach a consensus on the planning and objectives of group activities, ultimately contributing to the successful completion of projects. On the other hand, higher clustering coefficients in a residential area indicate the existence of numerous small groups with robust interpersonal connections. These groups can often be found within settings such as classes and dormitories. This structure enhances the organization and mobilization of group activities, as smaller, tightly knit groups are more effective in coordinating their efforts. School administrators should consider utilizing a variety of activities to enhance the closeness of collective ties.

5.2. Theoretical Implications

This study has several theoretical implications. Foremost, this study explored the relationship between everyday peer relationships and academic achievement based on behavioral big data and found that students with high academic achievement have peer relationships in everyday activities. This finding advances the understanding of the sites where peer relationships have an impact. While previous studies have focused on the relationship between learning help-seeking relationships and academic achievement, our study reveals the importance of peer relationships for academic achievement from the perspective of co-occurring networks in everyday behaviors, which provides new perspectives for understanding the influences on students’ academic achievement.

In addition, this study enriches the application of behavioral big data in the field of higher education research. Unlike previous studies that considered only individual behavioral data, we extracted network relationships through spatiotemporal co-occurrence data, thus reflecting peer relationships among students more objectively. This approach not only accurately captures the structure and evolution of peer relationships but also provides insights for future research based on behavioral data and broadens the space for choosing research methods.

Moreover, we found that the formation and evolution of daily peer relationships are related to external environments such as college student living areas, dormitories, provinces, classes, and majors. This finding emphasizes the important role of the external environment on peer relationships and implies the feasibility of encouraging students to form positive relationships with peers through external interventions. This study further reveals the complexity of peer relationship formation and evolution by analyzing the evolution of peer relationship networks. This is important for understanding the formation mechanism of students’ peer relationship networks and the shaping of students’ peer relationships in the higher education environment.

5.3. Practical Implications

This behavioral big data study helps to determine how a university can work to sustain the endurance of systems and processes, thus achieving the goal of sustainability in the long term. The practical implications are evident at two levels, i.e., university administration and higher education policy making.

This study has multiple practical implications. First, this study reveals the characteristics of college students’ peer relationship networks through behavioral big data, a method that can effectively identify the community structure of college students and offer the possibility of achieving precise and effective interventions to improve the quality of higher education. For example, we found that students who lacked peers had the lowest probability of winning scholarships, whereas peers winning scholarships were positively associated with an individual’s probability of winning scholarships, which suggests that school administrators should implement appropriate interventions, such as peer mentoring [74], to enable students who excel in academics to build relationships with those who have difficulty in learning and to help them set academic goals. By leveraging big data analytics, teaching administrations can better understand which students may need additional support to improve their academic achievement. This helps optimize education resource allocation, ensuring that resources are used efficiently while reducing waste.

Second, the daily peer relationships identified in the study can be linked to classroom instruction. Teachers can intervene in the formation of students’ daily peer relationships by grouping and guiding them to help them integrate into campus life, and they can also organize the more active students in their daily peer relationships to enhance the effectiveness of classroom discussions and to promote the role of the team in the practice sessions and the effectiveness of classroom teaching. Thus, schools should introduce the function of behavior big data to the learning management system and assist teachers in organizing their teaching. It is also suggested to implement a reform policy in higher education teaching to strengthen the development of big-data-based teaching.

Third, this study shows that living areas and geographic locations have a significant impact on students’ peer relationships, which suggests that management departments of living areas should improve organizational skills and help students integrate into campus life. Previous research also concluded that creating an informal, social environment that facilitates connections with like-minded peers is crucial for peer relationships and access to social support [75]. Therefore, schools should promote communication and interaction among students from different majors, classes, and dormitories through student government and other student self-governance organizations by conducting various activities, such as lectures, book clubs, and volunteer services. This is aligned with an increasing policy emphasis on the construction of the residential college system and “one-stop” student community. This can also provide insight for universities to help plan campuses and dormitories to better meet students’ social needs and reduce unnecessary traffic and resource consumption and thus meet environmental and urban planning goals in the principles of sustainability.

Last but not least, this study found that the acquisition of honors for collective activities was associated with a closer network of peer relationships, which means that instructional administrators should pay attention to the organization and construction of basic units such as peers, dormitories, and classes. When organizing campus activities, administrators should coordinate the strengths of multiple small groups and leverage trust and support within the small groups to increase student participation and overall team cohesion.

5.4. Limitations and Future Research

There are some limitations to this study. On the one hand, although we applied objective behavioral data, we only considered students’ daily behaviors of eating, bathing, and attending classes, while students’ peer relationships may exist in a variety of other scenarios, such as going to the library or the gym. On the other hand, although this study described the connection between daily peer relationships and academic achievement mainly from the perspective of a social network analysis, it did not reveal the influence strength and statistical significance between individual networks and academic achievement in a statistical sense. In addition, parameters chosen to determine the co-occurrence relationships may have affected the results that were found. However, this study demonstrates that behavioral big data have strong potential for analyzing college students’ peer relationships and providing inspiration for instructional management in practice. Future research may use a variety of statistical inference methods to examine the influence of the daily peer network structure on individual achievement and test the validity of this study’s findings. Future research may also aim to use richer behavioral data and combine questionnaire data, student profile data, etc., to verify and investigate the accuracy of peer relationships based on behavioral data.