The Relationship between the Facial Expression of People in University Campus and Host-City Variables

Featured Application

This work supplies a theoretical approach to evaluate public attitude towards university campuses and to detect the relationship with host-city variables using data about facial expression scores on social network services at the national scale.

Abstract

Public attitudes towards local university matters for the resource investment to sustainable science and technology. The application of machine learning techniques enables the evaluation of resource investments more precisely even at the national scale. In this study, a total number of 4327 selfies were collected from the social network services (SNS) platform of Sina Micro-Blog for check-in records of 92 211-Project university campuses from 82 cities of 31 Provinces across mainland China. Photos were analyzed by the FireFACE^TM-V1.0 software to obtain scores of happy and sad facial expressions and a positive response index (PRI) was calculated (happy-sad). One-way analysis of variance indicated that both happy and PRI scores were highest in Shandong University and lowest in Harbin Engineering University. The national distribution of positive expression scores was highest in Changchun, Jinan, and Guangzhou cities. The maximum likelihood estimates from general linear regression indicated that the city-variable of the number of regular institutions of higher learning had the positive contribution to the happy score. The number of internet accesses and area of residential housing contributed to the negative expression scores. Therefore, people tend to show positive expression at campuses in cities with more education infrastructures but fewer residences and internet users. The geospatial analysis of facial expression data can be one approach to supply theoretical evidence for the resource arrangement of sustainable science and technology from universities.

1. Introduction

Since 2008 more than half of the world population lived in cities and this is expected to reach 70% by 2050 [1]. Cities are widely regarded as important areas in the pursuit of global sustainability [2]. A sustainable society comprises five distinct elements for every human-being such as the proper education, a clean environment, a well-balanced safety, abundant resources for future generation, and contribution to a sustainable world [3]. Easy to collect metrics that rate environmental, economic, educational, and social variables are important to evaluate the global strategies for urban transformation towards sustainability that builds upon national and local scales [4]. A sustainable city should respond to residents’ needs through sustainable solutions for social and economic aspects [1]. The tradeoff between resource consumption and citizens’ demand can be sustainably solved by the use of information and communication technologies (ICT) and the internet of things (IoT), which can be accessed in a smart local university campus [5,6]. Therefore, a future-like relationship emerges between city variables and public attitude towards local university campuses.

Universities have a central role to create knowledge and tools to transfer information for societal transformations towards sustainability [2,7]. Universities can be a driving force to provide urban sustainable development by embedding knowledge to the local social and economic networks [8,9,10]. Besides the responsibilities of teaching and research, universities are increasingly expected to turn knowledge into innovation [11]. Universities can also be a partner with their host-city to develop the transformation to a smart community [1,5,6,7]. To test design principles, a university campus can be taken as a socioeconomic organization like a mini-city and the management and demands for resources therein can be acquired by the using data through the IoT. In China a model is being implemented to construct cities that are famous due to universities therein, but many of these programs are not successful as expected because of the poor educational outcomes and economic productivity [12]. The shortage of objective evaluation on local universities was at least partly responsible for the failure. People around the campus are frontiers that can give a precise evaluation on universities hence their attitude is the key to evaluate the university in its host-city.

City variables are known to affect the satisfaction of residents towards local universities [13,14]. An investigation using self-reported scores revealed that most undergraduates indicated satisfaction with settings of the city where their campus was located [13]. The group of variables out of these settings includes socializing [2,14], resident environment [2,13,14,15,16], socio-economic status [8,13,17], and industrialization [8] that have all been detected to have some relationship with perceived satisfaction towards campus although the magnitude was ever either positive or negative. The model of multiple city variables was proposed to measure the performance in the creativity of universities in cities [18]. This means that the multivariable model may also have contributions to the perceived satisfaction of people in university campuses. In addition, the current development of ICT can enable new methods and metrics to assess perceived satisfaction instead of questionnaire methodology. However, results about public attitude towards university campus were limited by the testing method and information that have been published on this matter.

Traditionally, questionnaires provided a common method to evaluate people perception. Evaluation through self-reported scores has several apparent biases from subjective emotion of respondents, real-time mood, problematic questions, and social-role restricted results [19]. Facial expression represents an emotional response to a stimulus and/or a communicative behavior in a social situation, which can be termed as Duchenne (a felt expression with an emotion cue) and non- Duchenne ways (an unfelt expression with a communicative cue), respectively [20]. The facial expression of a visitor’s photo at a place provides a novel way to show performative emotional satisfaction in the location. A selfie taken and shared by a person through social media is one way to collect the information of emotional expressions that would like to be exposed to the public. Facial expression scores with a check-in-recorded location enable geographical analysis of posed emotion towards environments in a visited location. Highly popularized social networking service (SNS) results in millions of facial-expression images uploaded to the data-cloud [21,22]. Therefore, to collect and analyze facial images from SNS with check-in locations supplies a new approach to assess satisfaction of people with a wide range of geographical locations. Regarding that people expose their selfies with check-in records to show posed facial expression at the location, all variables about the city where visitors is located can be used as the explanatory independent factors in a regression model together to build the relationship with expressional scores. To utilize data from SNS enables the precise evaluation of public attitude towards universities at the large geographical scale. However, to the best of our knowledge, the use of this methodology has not been tested.

In mainland China, 116 universities are classified in the ‘211-Project’. These universities were authorized by Ministry of Education of the People’s Republic of China (MEPRC) and are being distributed in 82 cities from 31 Provinces [23]. All universities within the 211-Project derive more financial and political support than other universities with an expectation of greater corresponding outcomes of science and technology. Therefore, cities with 211-Project universities are generally promoted by an enhanced scholar population, public services, daily livelihood, and social infrastructures. In this study, campuses of key universities in the 211-Project from mainland China were chosen as the research plots wherein selfies at check-in locations at campuses were collected and analyzed for intended facial expressions. We aimed to assess scores of happy and sad expressions of people in 211-Project universities and map them at the national scale. It was hypothesized that (i) people would pose more positive facial expressions at university campuses in cities with more development in economy and technology, and (ii) city variables about socializing and socio-economy had contrasting contributions to intended positive and negative expressions in university campuses.

2. Materials and Methods

2.1. Universities

In this study, key universities in the 211-Project were chosen as study locations. A total of 96 campuses of key universities were included in this study (Table S1). Campuses of some universities were also excluded in this study because the sample size of suitable images (as described below) was too small to support the data collection and therefore 96 universities in the 211-Project were included in this study. Overall, a total of 92 campuses of 211-Project universities were selected as the check-in-recorded locations exited for these from 82 cities in 31 provinces across mainland China. These 92 universities are also top scholar institutions in mainland China.

2.2. Photo Collection

The Sina Micro-Blog (SMB; Sina Inc., Beijing, China) was chosen as the platform of SNS to collect facial photos using the method of Wei et al. [19]. SMB is widely used by Chinese web-users and functions similar to Twitter (NYSE: TWTR, San Francisco, CA, USA) that supplies a platform to pose real-time mini-blogs to the internet with comments and images (photos). Both personal computer and mobile device terminals can enable users to pose blogs, but only mobile terminals can attach check-in location records. According to the policy of SMB, the use of text and photos in blogs published in SMB are open to the public if no private limit was claimed by the user. As of August 2018, SMB had 0.43 billion active users with 0.19 billion daily submissions of micro-blogs by active users [24]. SMB supplied an open platform to upload photos with check-in data through microblogs and to check and download images. All university campuses chosen in this study had been listed in the location records of SMB.

Photos were collected in the process following three phases:

(i)

All microblogs with uploaded photos with check-in data about geography of target university campuses were blocked by time between 15 February and 15 June 2019. This time was chosen because it covered the spring semester period for most universities we studied.

(ii)

Blogs with selfies were screened and only photos with young adult portraits were collected. Age of subjects in selfies was visually evaluated through the photo. Many users’ age information was hidden by a classified statement in SMB. This resulted in only a few of users’ age information published, but the number of revealed ages was too small to support a meaningful statistical analysis. Nearly all (95%) of initially collected selfies were uploaded by young adults. Therefore, we chose to use this part of photos for further analysis to keep the uniformity of age among users. Significant differences of perception and habit between international and Chinese students exist [25], thus, selfies with western-style faces were excluded. It was hard to distinguish nationalities among Eastern Asia countries in photos, hence it was assumed that all Asian faces were Chinese.

(iii)

Only photos clearly showing facial features were selected for analysis. Photos were excluded when makeup obscured facial expression (e.g., excessively beautified, over whitened, additionally decorated) or the face feature was twisted.

2.3. Face Expression Analysis

Collected photos were modified by cropping and rotating before analysis. Since some photos included more than one face, each face was separated from the original photo by cropping to generate a new photo file with one unique face. It was necessary to rotate photos to make the nose axis vertical to the horizon. Rotating was necessary for analysis precision of facial expressions. Photos were analyzed by the FireFACE^TM-V1.0 software (Zhilunpudao Agric. Sci. Inc., Changchun, China). This software was calibrated by training the computer program to recognize faces using the independent variables of oriental faces with posed facial expressions. Initially, about 30,000 photos were documented for training the machine to recognize basic expressions (happy, sad, and neutral) with about 10,000 photos for each of them. Photos were recognized and classified into files of known expressions manually then engineers wrote codes to enable the communication between computer and these files. The training was terminated until the machine has been tested to pass the aimed accuracy of 85% for happy and sad expressions and 80% for neutral expression (Figures S1 and S2). To train the software with high accuracy for recognizing faces of Chinese people, posed photos were recognized and classified into perceived expressions by Chinese experts. A total number of 4327 selfies (one person per photo) were analyzed for facial expressional scores.

2.4. Data about City Variables

Data depicting the variables of all 82 cities were obtained from the latest database of National Bureau of Statistics of China [26]. We employed city variables that were documented officially for all cities since the time of July 2019. As a result, four parameters (e.g., economy, public facility, habitation, and environments) were used in this study (

Table 1. Specific parameters used in city variables about economy, public facility, habitation, and environment.

Variables	Unit	Parameter
Economy	Yuan	GDP per capita
	×104 Yuan	Public financial expenditure
	×104 Yuan	Corporate profit of enterprises above designated scale 1
	Individuals	Population of employees of C&T work-units 2
	Yuan	Average wage of enrolled employees of work-units
Public facility	×104 Households	Number of internet wideband accesses
	Individuals	Real number of end-of-year taxicabs
	Individuals	Numbers of hospitals and health-centers
	Individuals	Number of regular institutions of higher education 3
	×103 Individuals	Collection of books in public libraries
Habitation	km2	Total area of residential lands
	×104 Individuals	End-of-year population of registered residents
	ha	Total area of green space in C&T parks
	×104 Tons	Domestic water consumption by residents
	×104 Kilowatt-hour	Electricity consumption of C&T residents
Environment	×104 Tons	Discharge of industrial sewage
	Tons	Emission of industrial SO2
	Tons	Emission of industrial dust
	Percent	Centralized disposal of sewage
	Percent	Disposal of household garbage

Note: ¹ Enterprises above designated scale, annual revenue over 20 million Yuan from the primary business of industrial enterprises; ² C&T, city and town; ³ institutions of higher education, educational institutions that can grant degrees that are higher than high-school education, including university, college, vocational technique university/college, etc.

). Economic parameters included government expenditure and resident income, which may affect financial status of local universities through tax investment at the national and local scales, respectively [27,28]. Public facilities covered aspects of communication, transportation, health care, education, and regional culture, which were all related to university students [29,30,31]. Parameters of habitation included the aspects that reflected the socializing and life utilities, which were found to be closely related to perceived satisfaction at campus [2,14]. Environment parameters (e.g., air pollution, water pollution, and garbage disposal) that may impact satisfaction under some given industrialization were extracted from databases of each local bureau [2,13,14,15,16].

2.5. Statistical Analysis

In addition to happy and sad expressions, we involved the positive response index (PRI) as a scoring metric to evaluate the net positive emotion [32]. PRI is defined as the difference between scores for happy and sad expressions.

SAS (ver. 9.4 64-bit, SAS Institute, Cary, NC, USA) software was used for data analysis. Data about happy and sad scores and PRI values were tested for the normal distribution, which was not found across universities campuses. Therefore, data were ranked to obtain a new set of distribution-free scores [33]. Ranked data were further analyzed by one-way analysis of variance (ANOVA) with the variation of universities as a source of variance. When a significant effect was found, means were arranged and compared by the Duncan Multiple Comparison test at α = 0.05 level. To detect the relationship between multiple city-variables and facial expressions, a model of maximum likelihood estimate on the general linear regression was made using the GENMOD procedure with multiple city-variables as the independent variables and facial expression scores as the dependent variables. Data used for the regression were pooled using averaged means for cities (n = 32); thereafter dependent variables were found to be normally distributed across cities and independent variables were log-transformed. The probability of a chi-square test was determined to be significant at α = 0.05 level for every estimated parameter. We did not distinguish observations from subjects therefore the negative effect of pseudoreplication on our results might happen [34,35]. To assess the possible impact of collinearity on regression, variance inflation (VIF) was detected in SAS by the REG procedure in advance.

3. Results

3.1. Facial Expressoin Scores among Universities

Results about analysis of one-way ANOVA on facial expression scores across university campuses are shown in

Table 2. Results of one-way analysis of variance (ANOVA) on ranked scores of happy and sad expressions, and positive response index (PRI) in visitors at university campuses across mainland China.

Variable	Degree of Freedom			Sum of Squares	F Value	p Value
	Model	Error	Correction
Happy	95	4231	4326	354,720,095	2.49	<0.0001
Sad				276,605,667	1.94	<0.0001
PRI				328,290,197	2.28	<0.0001

. Happy expression scores were higher in Shandong University, Xi’an Jiaotong University, and South China University than those in most of other universities (Figure 1). Harbin Engineering University had the lowest happy score. On the other hand, Ocean University of China, Inner Mongolia University, and Harbin Engineering University tended to have the highest sad expression scores, while sad scores appeared to be lower in Shandong University, Zhejiang University, and Xi’an Jiaotong University (Figure 1). The PRI generally showed the similar trend of happy score but had some distinctive inverses (Figure 1). For example, PRI in China Pharmaceutical University, Xinjiang University, and Ocean University of China were reduced by the relatively higher level of sad scores. In contrast, the sudden increase of PRI, such as that in Northwestern Polytechnical University, was caused by the relatively lower sad scores.

3.2. Geographical Distribution of Expression Scores among Host-Cities

Happy scores were distributed divisionally with higher values in central-northeast, northern East-China, and Southern China (Figure 2A). Cities of Changchun, Jinan, Wulumuqi, Xi’an, and Hangzhou tended to have higher happy scores, while those of Harbin, Huhetaote, and Chongqing tended to have low happy scores. Cities of Huhetaote, Zhengzhou, and Kunming tended to have high sad scores (Figure 2B). Cities of Hangzhou, Suzhou, and Xi’an tended to have low sad scores. As a result of interplay between happy and sad scores, PRI was highest in Cities of Jinan, and moderately higher in Changchun, Xi’an, Hangzhou, Shanghai, and Guangzhou.

3.3. The Bias of Collinearity

Analysis on VIF shows that the happy expression scores had the high risk of collinearity from variables of public financial expenditure, corporate profit of enterprises above designated scale, and average wage of enrolled employees of work-units (

Table 3. The analysis of collinearity among city variables on ranked scores of happy and sad expressions, and positive response index (PRI) in visitors at university campuses across mainland China.

Variable	DF 1	Happy		Sad		PRI
		Pr > |t|	VIF 2	Pr > |t|	VIF	Pr > |t|	VIF
Intercept	1	0.3208	0	0.7825	0	0.0192	0
GDP per capita	1	0.1992	8.5949	0.5681	8.5949	0.134	8.5949
Public financial expenditure	1	0.00243	264.95976	0.1355	264.95976	0.0033	264.95976
Corporate profit of enterprises above designated scale 4	1	0.0361	59.3944	0.0344	59.3944	0.0719	59.3944
Population of employees of C&T work-units 5	1	0.0511	221.303	0.0547	221.303	0.0945	221.303
Average wage of enrolled employees of work-units	1	0.0395	107.13049	0.9147	107.13049	0.0262	107.13049
Number of internet wideband accesses	1	0.1082	45.01435	0.953	45.01435	0.0897	45.01435
Real number of end-of-year taxicabs	1	0.826	174.32891	0.8198	174.32891	0.7701	174.32891
Numbers of hospitals and health-centers	1	0.1418	58.34608	0.3605	58.34608	0.1692	58.34608
Number of regular institutions of higher education 6	1	0.1438	14.73109	0.1323	14.73109	0.2172	14.73109
Collection of books in public libraries	1	0.5681	187.98365	0.316	187.98365	0.4092	187.98365
Total area of resident lands	1	0.6997	112.74508	0.0511	112.74508	0.9902	112.74508
End-of-year population of registered residents	1	0.3431	79.37711	0.4805	79.37711	0.3929	79.37711
Total area of green space in C&T parks	1	0.4886	69.92869	0.0783	69.92869	0.7245	69.92869
Domestic water consumption by residents	1	0.7263	63.86034	0.8968	63.86034	0.7279	63.86034
Electricity consumption of C&T residents	1	0.5377	178.41355	0.8228	178.41355	0.5412	178.41355
Discharge of industrial sewage	1	0.3671	22.32038	0.8054	22.32038	0.3135	22.32038
Emission of industrial SO2	1	0.777	14.73447	0.3799	14.73447	0.9083	14.73447
Emission of industrial dust	1	0.7675	10.03224	0.2676	10.03224	0.5754	10.03224
Centralized disposal of sewage	1	0.0942	5.49206	0.5721	5.49206	0.0562	5.49206
Disposal of household garbage	1	0.9116	1.58069	0.0156	1.58069	0.5084	1.58069

Note: ¹ DF, degree of freedom; ² VIF, variance inflation; ³ Values in bold font indicate significant VIF; ⁴ Enterprises above designated scale, annual revenue over 20 million Yuan from the primary business of industrial enterprises; ⁵ C&T, city and town; ⁶ institutions of higher education, educational institutions that can grant degrees that are higher than high-school education, including university, college, vocational technique university/college, etc.

). Sad expression scores showed possible collinearity from variables of corporate profit of enterprises above designated scale and disposal of household garbage. PRI exhibited potential collinearity from public financial expenditure and average wage of enrolled employees of work-units.

3.4. The Regression Analysis

Nearly all parameters about the economy had significant contribution to the happy score but their estimated coefficients were too small to be detectable (

Table 4. Maximum likelihood estimation of multiple city variables on happy expression scores of students from key universities of cities across mainland of China.

Parameter	Degree of Freedom	Estimate	Standard Error	Wald 95% Confidence		Wald Chi-Square	Pr > Chi-Square
Intercept1	1	−2.986	0.9664	−4.8802	−1.0918	9.55	0.002
GDP per capita	1	0	0	0	0	2.04	0.1536
Public financial expenditure	1	0	0	0	0	12.06	0.0005
Corporate profit of enterprises above designated scale2	1	0	0	0	0	5.76	0.0164
Population of employees of C&T work-units3	1	0	0	0	0	5.88	0.0153
Average wage of enrolled employees of work-units	1	0	0	0	0	5.72	0.0168
Number of internet wideband accesses	1	−0.0016	0.0008	−0.0031	0	4.08	0.0435
Real number of end-of-year taxicabs	1	0	0	0	0	0	0.9914
Numbers of hospitals and health-centers	1	−0.001	0.0005	−0.002	0.0001	3.33	0.068
Number of regular institutions of higher education4	1	0.0066	0.003	0.0006	0.0125	4.72	0.0299
Collection of books in public libraries	1	0	0	0	0	0.61	0.4358
Total area of resident lands	1	−0.0004	0.0013	−0.003	0.0022	0.07	0.7845
End-of-year population of registered residents	1	0.0004	0.0004	−0.0003	0.0011	1.55	0.2127
Total area of green space in C&T parks	1	0	0	0	0	0.99	0.3188
Domestic water consumption by residents	1	0	0	0	0	0.08	0.7836
Electricity consumption of C&T residents	1	0	0	0	0	0.32	0.5745
Discharge of industrial sewage	1	0	0	0	0	1.32	0.251
Emission of industrial SO2	1	0	0	0	0	0.12	0.7286
Emission of industrial dust	1	0	0	0	0	0.07	0.7971
Centralized disposal of sewage	1	0.0095	0.005	−0.0002	0.0192	3.66	0.0556
Disposal of household garbage	1	−0.0001	0.0037	−0.0074	0.0072	0	0.9788
Scale	1	0.0756	0.0053	0.0659	0.0867

Note: ¹ Values in bold font indicate significant contribution; ² Enterprises above designated scale, annual revenue over 20 million Yuan from the primary business of industrial enterprises; ³ C&T, city and town; ⁴ institutions of higher education, educational institutions that can grant degrees that are higher than high-school education, including university, college, vocational technique university/college, etc.

). Among variables of public facilities, number of internet wideband access had a negative contribution to the happy score while number of regular institutions of higher learning had a positive contribution. Neither of these two variables exhibited issues with collinearity. None of the rest of parameters was indicated to have any significant contributions.

In the economic field, parameters of the corporate profit of enterprises above designated scale and number of employees of city and town (C&T) work-units were indicated to have significant contributions to the sad scores but their estimated coefficients were too small to be detected (

Table 5. Maximum likelihood estimation of multiple city variables on sad expression scores of students from key universities of cities across mainland of China.

Parameter	Degree of Freedom	Estimate	Standard Error	Wald 95% Confidence		Wald Chi-Square	Pr > Chi-Square
Intercept1	1	−3.6837	1.2682	−6.1693	−1.1982	8.44	0.0037
GDP per capita	1	0	0	0	0	0.24	0.6208
Public financial expenditure	1	0	0	0	0	2.92	0.0873
Corporate profit of enterprises above designated scale2	1	0	0	0	0	6.3	0.0121
Population of employees of C&T work-units3	1	0	0	0	0	5.4	0.0201
Average wage of enrolled employees of work-units	1	0	0	0	0	0.01	0.9366
Number of internet wideband accesses	1	0	0.0009	−0.0017	0.0018	0	0.9621
Real number of end-of-year taxicabs	1	0	0	0	0	0.08	0.7727
Numbers of hospitals and health-centers	1	0.0007	0.0007	−0.0006	0.0019	1	0.3182
Number of regular institutions of higher education 4	1	−0.0066	0.004	−0.0145	0.0013	2.68	0.1018
Collection of books in public libraries	1	0	0	0	0	1.95	0.1624
Total area of resident lands	1	0.0041	0.0016	0.0009	0.0073	6.46	0.0110
End-of-year population of registered residents	1	−0.0003	0.0004	−0.0011	0.0005	0.59	0.4429
Total area of green space in C&T parks	1	0	0	0	0.0001	4.13	0.0421
Domestic water consumption by residents	1	0	0	0	0	0.01	0.9139
Electricity consumption of C&T residents	1	0	0	0	0	0.12	0.7326
Discharge of industrial sewage	1	0	0	0	0	0.22	0.6418
Emission of industrial SO2	1	0	0	0	0	1.32	0.2498
Emission of industrial dust	1	0	0	0	0	1.41	0.2345
Centralized disposal of sewage	1	0.0045	0.006	−0.0073	0.0162	0.55	0.4565
Disposal of household garbage	1	0.0155	0.0079	−0.0001	0.031	3.79	0.0515
Scale	1	0.0307	0.0021	0.0268	0.0352

Note: ¹ Values in bold font indicate significant contribution; ² Enterprises above designated scale, annual revenue over 20 million Yuan from the primary business of industrial enterprises; ³ C&T, city and town; ⁴ institutions of higher education, educational institutions that can grant degrees that are higher than high-school education, including university, college, vocational technique university/college, etc.

). The variable of corporate profit of enterprises above designated scale also suffered showed collinearity. In the field of habitation, the parameter of area of resident lands had a positive contribution to sad scores.

The variable of public financial expenditure showed collinearity (

Table 6. Maximum likelihood estimation of multiple city variables on PRI expression scores of students from key universities of cities across mainland of China.

Parameter	Degree of Freedom	Estimate	Standard Error	Wald 95% Confidence		Wald Chi-Square	Pr > Chi-Square
Intercept	1	0.5661	1.9418	−3.2396	4.3719	0.09	0.7706
GDP per capita	1	0	0	0	0	0.87	0.3512
Public financial expenditure1	1	0	0	0	0	9.99	0.0016
Corporate profit of enterprises above designated scale2	1	0	0	0	0	7.23	0.0072
Population of employees of C&T work-units3	1	0	0	0	0	7.86	0.0051
Average wage of enrolled employees of work-units	1	0	0	0	0.0001	3.36	0.067
Number of internet wideband accesses	1	−0.0028	0.0015	−0.0057	0.0001	3.54	0.0599
Real number of end-of-year taxicabs	1	0	0	0	0	0.14	0.7117
Numbers of hospitals and health-centers	1	−0.0022	0.0025	−0.0072	0.0027	0.77	0.3798
Number of regular institutions of higher education4	1	0.0147	0.0056	0.0037	0.0256	6.93	0.0085
Collection of books in public libraries	1	0	0	0	0	0.24	0.6262
Total area of resident lands	1	−0.0021	0.0011	−0.0042	0	4.03	0.0448
End-of-year population of registered residents	1	0.0011	0.0007	−0.0003	0.0024	2.26	0.1326
Total area of green space in C&T parks	1	0	0	−0.0001	0	3	0.0831
Domestic water consumption by residents	1	0	0	0	0	0.08	0.7752
Electricity consumption of C&T residents	1	0	0	0	0	0.02	0.8823
Discharge of industrial sewage	1	0	0	0	0.0001	0.78	0.3771
Emission of industrial SO2	1	0	0	0	0	0.5	0.4802
Emission of industrial dust	1	0	0	0	0	0.11	0.7397
Centralized disposal of sewage	1	0.0129	0.0096	−0.0059	0.0317	1.81	0.1781
Disposal of household garbage	1	−0.0067	0.0062	−0.0189	0.0054	1.19	0.2751
Scale	1	9.2193	0.6455	8.0371	10.5753

Note: ¹ Values in bold font indicate significant contribution; ² Enterprises above designated scale, annual revenue over 20 million Yuan from the primary business of industrial enterprises; ³ C&T, city and town; ⁴ institutions of higher education, educational institutions that can grant degrees that are higher than high-school education, including university, college, vocational technique university/college, etc.

). Area of resident lands and numbers of regular institutions of higher learning from the public facility had negative and positive contributions to PRI, respectively (Table 6). The absolute value of estimated number of regular institutions of higher learning was about 10-time higher than that of residential area.

4. Discussion

The most significant result in our study is that the happy score in Shandong university (Shandong Province) was highest among universities while geographical distribution also revealed that positive scores tended to be higher in Jinan City (Shandong Province). We surmise that the high happy expression score in a city was the result of the happy expression score in the campus therein. Another example is Changchun City, which obtained a medium-high happy score while the happy scores in the two universities in Jilin Province were medium or high. However, the facial expression score for the city may be null to be indicated by that for the universities therein. For example, overall universities in Beijing City showed a moderate score of positive expressions although some university campuses therein obtained higher happy expression scores (e.g., China University of Political Science and Law). In contrast, the positive expression score in some other universities at Beijing City was lower than the average level (e.g., China Mining University and Beijing Sport University). Changchun, Jinan, and Guangzhou were three cities with high scores for both happy expression and PRI and hence university campuses in these three cities result can be taken to have the highest perceived satisfaction.

Student satisfaction with a university can affect student enrollment and retention. Hajrasouliha [36] investigated quality scores of university campuses in the United States and the scores were associated with freshman retention and graduation rates. The authors also revealed the geographical distribution of scores in selected campuses across the United States and found higher scores in the northeastern cities. Thus, both results from our study and Hajrasouliha [36] revealed no response of distribution to a geographical gradient.

Several public facility parameters were tested for a relationship with facial expression scores. The number of regular institutions of higher education within a city was one parameter that had positive contribution to values of both a happy score and PRI. From the SNS platform of SMB we aimed to collect selfies about young people who can mainly come from students or a new teacher enrolled in the university. Undergraduates and most mater candidates look young, but some PhD candidates may look old as they may have spent several years to earn their degree. For students, it was found that faculty, advising staff, and the class itself all had significant impact on their perceived satisfaction in higher learning institutions [37]. Some lecturers and even associate and full-time professors can also look young if they achieved high scholar scores at young age. Therefore, all young adults who would like to pose their selfies in a campus are likely to feel the emotional cue that originated from campus-life. Our results can be interpreted that young adults as either students or tutors would enjoy the city with more educational institutions where they may show Duchenne expressions as they felt being accompanied by large group of other youths. Other studies also reported that variables about the learning organization could account for the significant satisfaction for both teachers and researchers of higher learning institution [38]. A concentration of university campuses may result in more socializing opportunities, leading to greater satisfaction with a campus through more socializing of young people in their generation group [2,14].

It is surprising that a number of internet accesses had a negative contribution to the happy score. This suggests more internet users in a host-city decreased the ratio of showing happy expression of youths in university campuses. Youths of intense internet users were found to have overconfidence in the web-world, but they were also reported depressive symptomatology, problem behavior, and targeting of traditional bullying in the real world [39]. Another investigation reported that adolescents as frequent internet users reported depression by isolation from their family members [40]. In addition, more internet access may result in a user’s habitual internet use and might result in a greater population of “internet addicts”. Internet addiction tended to evoke perceived stress, which our results were consistent with less happy expressions [41]. People with an internet gaming disorder manifested in above average time spent with this activity were found to have different kinds of unconscious neutral facial expressions, which depressed the expression of a smiling face [42]. Thus, these studies were all consistent with our results with the higher probability of a reduction with a happy expression in a population with higher ratio of internet users. More direct and explanatory evidence is needed to further verify these results and it is also possible the concentration on an activity itself depresses a happy face.

In our study, the sad expression score was positively correlated with the area of residential lands adjoining a university. These findings concur with those found in England [43] and South Africa [44], where residential housing-density was negative to the perceived satisfaction of neighbors. This negative relationship should be more apparent for a residence-surrounded university campus because of more open space in the campus than in resident communities. However, we detected a weak effect of the area of green space in urban parks on sad expression. Urban green space has been shown to alleviate perceived stress [16]. This may be because people in our study were mainly grouped in the university campus rather than spending time in the green space of parks at the city scale. Or if they did any residual affect was not detected in this study.

We found no environmental variables were associated with our scores of happy, sad, or the PRI. This is because our data showed an obvious ceiling that cities with heavy contamination were unlikely to be included to the database. It is not recommended to install industries with heavy contamination for a city, which has been assigned to develop mainly through intellectual promotion of local universities. A sustainable and healthy community has effective ways to dispose of garbage and sewage, which can increase human disease if not abated [45,46]. Likewise, excessive levels of air pollution and particulate matter from industrial dust can reduce human health [47]. In this study, it is likely that a person would not see these variables and as such no effect on facial images seems reasonable.

This study may have been limited by several aspects. The first may come from the number of users in the selfies. The initially collected selfies were either taken by a single person in a photo or a person cropped from selfies with a group of people. Since individuals can more accurately perceive emotions expressed than in-group members [48], our facial expressions that were analyzed from selfies should have been controlled by the individual and in-group participants. However, this was not available in this study because many factors failed to be concerned, such as the number of people that were separated from a group, genders of them (this probably matters), failure of selfies for cropping (un-intact image of face, unclear face, deflected faces), group of young and mature adults, etc. Therefore, our results may have suffered some bias from the difference between faces of individual and in-group persons. As we omitted this bias for all check-in locations, it was reasonable to assume the technical error was uniform for all campuses.

Another limit to this study comes from the negative impact from pseudoreplication on our data, which resulted in a pretended independence. According to Waller et al. [34], pseudoreplication can occur when more than one datum was observed per individual. It can also occur when data points result from the same stimulus. Both situations were also found in ecological studies [35]. Our data may have suffered a pseudoreplication impact because different types of facial expressions may have been rated from the same subject. More than one facial expression score may have been collected from the same person in a university campus. This would impact the significance of difference of facial expression scores across universities and city variables because some of the replicated observations were not independent. A gross summary suggested that the incidence of pseudoreplication ranged between 12% and 40% in studies on primate communication research [34]. The incidence in our study was far lower than this range by manual screening of selfies. Therefore, it is reasonable to assume the impact from pseudoreplication on our data points can be negligible due to the review of each person. However, we still suggest future work can employ a process of excluding more than one observation from the same individual or at the same location to eliminate any potential pseudoreplication bias.

5. Conclusions

In this study, we examined how facial expressions of young adults varied at key universities in the 211-Project across mainland China. Selfies were downloaded from SMB and photos were analyzed by the FireFACE^TM-V1.0 software to obtain scores of happy and sad expressions. Our results indicated that the geographical distribution of facial expression scores showed divisional patterns and higher positive scores were distributed in Changchun, Jinan, and Guangzhou cities. The formation of facial expression distribution among cities was related to that in universities within host-cities. Regression analysis indicated that number of regular institutions of higher learning had the positive contribution to the happy score and a number of internet accesses and area of resident lands contributed to the negative expression scores. Therefore, if the city planner aimed to promote the perceived satisfaction of young people in the campus of universities, the promotion of educational institution numbers with confining resident communities and broadband accesses would benefit the positively emotional expressions. Our study can be applied for budget and regime planners to establish sustainable development of universities with efficient evaluation using current techniques.