key: cord-0711207-cnhjp5mi authors: Park, June Young; Mistur, Evan; Kim, Donghwan; Mo, Yunjeong; Hoefer, Richard title: Toward human-centric urban infrastructure: Text mining for social media data to identify the public perception of COVID-19 policy in transportation hubs date: 2021-11-04 journal: Sustain Cities Soc DOI: 10.1016/j.scs.2021.103524 sha: 7b9bcd79906af651fcf0e9699fc052de7691bb04 doc_id: 711207 cord_uid: cnhjp5mi The COVID-19 pandemic has made transportation hubs vulnerable to public health risks. In response, policies using nonpharmaceutical interventions have been implemented, changing the way individuals interact within these facilities. However, the impact of building design and operation on policy efficacy is not fully discovered, making it critical to investigate how these policies are perceived and complied in different building spaces. Therefore, we investigate the spatial drivers of user perceptions and policy compliance in airports. Using text mining, we analyze 103,428 Google Maps reviews of 64 major hub airports in the US to identify representative topics of passenger concerns in airports (i.e., Staff, Shop, Space, and Service). Our results show that passengers express having positive experiences with Staff and Shop, but neutral or negative experiences with Service and Space, which indicates how building design has impacted policy compliance and the vulnerability of health crises. Furthermore, we discuss the actual review comments with respect to 1) spatial design and planning, 2) gate assignment and operation, 3) airport service policy, and 4) building maintenance, which will construct the foundational knowledge to improve the resilience of transportation hubs to future health crises. The study proposed the human-centric facility management to investigate and incorporate the human perceptions of COVID-19 policy for healthy and safe airports. We identified 4 representative topics (Staff, Shop, Space, and Service) from 103,428 social media reviews in 64 hub airports in the US using text mining techniques. Due to the behavioral changes required by COVID-19 policy, passengers began to identify Space and Service issues, where we can potentially improve the user experience in airports. The research framework can be used to discover the human perceptions of other urban infrastructures. While the pandemic has impacted all sectors [1, 2] , transportation hubs, 10 where people gather in large numbers when travelling between regions, re-11 main especially vulnerable to public health concerns potentially resulting in 12 the importation and exportation of the virus [3, 4] . In particular, airports 13 create international health risks as they are often overcrowded due to a large willingness to adhere to health policies [13] . Noncompliance to NPIs such 44 as mask wearing and social distancing is common in many settings, includ- Twitter users [20] . 60 The main objective of these recent studies was to identify the general 61 perception of COVID-19 policies. It is important to note that implementing 62 and ensuring compliance to such policies are critical for maintaining public The reminder of the article is organized as follows. In Section 2, we ex-151 plain the methodology of our data analysis, which includes data collection, 152 topic modeling, and data-driven discovery. Next, Section 3 summarizes the 153 results of our study. Section 4 discusses recommendations for the airport 154 management during the COVID-19 pandemic and future health crises. Fi-155 nally, we conclude the article in Section 5. 156 To identify the public perception of COVID-19 policy at airports, we em-158 ploy a data driven approach, which consists of three major steps ( Figure 1 ). As the first step, we collect text data via Google Maps reviews from the pop- In addition to the hub types, we also classified airports by their termi- archival web data detailing regional COVID-19 policies that were in use dur- In addition, we also included the COVID-19 keywords (Table 1) words. This is mainly because these words are predominantly appeared in 266 our review data due to our search method, which hinders to find the abstract In addition to our topic modeling analysis, we spatially investigate the issues from our topic modeling results. 309 We use Space Syntax, which is a set of theories to analyze spatial con- of the review data is 33.5 words. As the 75% quartile of the review data is 337 41 words, the majority of these reviews are equal to or less than a sentence 338 length. Therefore, we consider our analysis as a short text mining problem. 339 Table 2 presents the summary of our review data by Google Maps rating. In Figure 5 , we visualize and compare monthly airport passengers, the 347 number of COVID-19 policy in place, and the COVID-19 related reviews. The number of airport passengers began to fall precipitously in response 349 to the virus, developing health recommendations, and state policies as travel 350 became much rarer. However, few COVID-19 related reviews were submitted Considering that each word can occur at 5 different rates, we calculate 387 the weighted averaged of rate score of each word to understand the sentiment 388 of each word by rate. We then select the most negative and positive words, 389 whose scores were less than 2.5 and greater than 4.5, respectively (Figure 7) . Although most of the negative words contains innate negative meanings, 391 we also found that they are related to some perceptive words of the airports of Shop) contain 2,465 and 2,771 reviews, respectively, which account 70% of the total reviews. Cluster 2 (the topic of Space) and Cluster 3 (the topic 437 of Service) take the rest of the reviews. Since it is critical to understand the sentiments of passengers in the air-453 port, we will further investigate the relationship between the cluster assign- We also traced the number of reviews by cluster assignments and rate in-468 formation with the monthly interval of our data collection period in Figure 12 . Prior to May, 2020, there were almost no reviews related to the COVID-19. Shop compared to Space and Service during the COVID-19 pandemic period. As we discovered in Figure 11 , Cluster 0 and 1 have more positive reviews, CLT has more reviews in Cluster 2 in our text mining results. Our results indicate that agent movement density in LAX is evenly distributed around 531 terminal spaces. This is mainly because the terminal layout that is not 532 formed a geometric boundary but an irregular configuration. ATL and DEN 533 (Denver International Airport) have the longest passenger walking distances. The graph illustrates high-density area is only located at the center corri- 2) gate assignment and operation, 3) policy, and 4) building maintenance. 548 We detail each category with actual passengers' reviews as follows. In comments regarding gate assignment and operation, passengers com-571 plained that some gate areas were more crowded than others, which might be 572 improved by occupancy-based operation. This suggests that the airport man- Antivirus-built environment: Lessons 789 learned from covid-19 pandemic Preparing for outbreaks-implications for re-792 silient water utility operations and services, Sustainable Cities and So-793 ciety The urban political pathology of emerging infec-795 tious disease in the age of the global city Airport risk of importation and exportation 798 of the covid-19 pandemic Belief in covid-19 conspiracy 821 theories reduces social distancing over time Association 824 of covid-19 misinformation with face mask wearing and social distanc-825 ing in a nationally representative us sample Political polarization in 828 us residents covid-19 risk perceptions, policy preferences, and protective 829 behaviors Covid-19: Facemask use preva-831 lence in international airports in asia, europe and the americas Partisanship, health 834 behavior, and policy attitudes in the early stages of the covid-19 pan-835 demic Public risk perception and emotion on twitter during 837 the covid-19 pandemic Global infodemiology of covid-839 19: analysis of google web searches and instagram hashtags Leveraging twitter data to under-842 stand public sentiment for the covid-19 outbreak in singapore, Interna-843 tional Public perception of the covid-19 pandemic 845 on twitter: Sentiment analysis and topic modeling study Understanding public perception of coronavirus disease 2019 849 (covid-19) social distancing on twitter Serial interval of sars-cov-2 was shortened over time by 853 nonpharmaceutical interventions Im-856 pact of non-pharmaceutical interventions on documented cases of covid-857 The effectiveness of eight nonpharmaceutical interven-861 tions against covid-19 in 41 countries Nudge: Improving decisions about health, 863 wealth, and happiness Kjaer-865 gaard, Z. Nagy, A critical review of field implementations of occupant-866 centric building controls Introduc-869 ing iea ebc annex 79: Key challenges and opportunities in the field of 870 occupant-centric building design and operation, Building and Environ-871 ment Quantitative relationships between occupant satisfaction and satisfac-874 tion aspects of indoor environmental quality and building design Lightlearn: An adaptive 877 and occupant centered controller for lighting based on reinforcement 878 learning Develop-880 ment and implementation of a thermostat learning algorithm, Science 881 and Technology for the Human-centric infrastructure resilience: 883 Uncovering well-being risk disparity due to infrastructure disruptions in 884 disasters The role of maintenance and facility man-886 agement in logistics: a literature review Establishment of kpis for facility 888 performance measurement: review of literature Customer performance measurement in facilities 890 management: a strategic approach Linking occupant complaints to building perfor-893 mance Are users more tolerant of greenbuildings? Data mining: practical machine learning tools 897 and techniques with java implementations Application areas and 900 data requirements for bim-enabled facilities management Data mining in the construction 903 industry: Present status, opportunities, and future trends, Automation 904 in Construction Text mining for occupant perspectives on the 906 physical workplace Benchmarking operational perfor-909 mance of buildings by text mining tenant surveys, Science and Technol-910 ogy for the Automated staff 912 assignment for building maintenance using natural language processing, 913 Automation in Construction Assessment of airport service quality: A complementary 915 approach to measure perceived service quality based on google reviews Advisory circular (ac 150 / 5360-13a) Measuring citizen and government ideology in the us states: A 921 re-appraisal Correlations between covid-19 cases and 924 google trends data in the united states: A state-by-state analysis Tracking social media discourse 927 about the covid-19 pandemic: Development of a public coronavirus twit-928 ter data set Empirical study of topic modeling in twitter Nltk: the natural language toolkit COLING/ACL 2006 Interactive Presentation Sessions Natural Language Processing and Computational 935 Linguistics: A practical guide to text analysis with Python, Gensim, 936 spaCy, and Keras Foundations of statistical natural language 938 processing Latent dirichlet allocation, the 940 Short text topic modeling 942 techniques, applications, and performance: a survey, IEEE Transactions 943 on Knowledge and Data Engineering A dirichlet multinomial mixture model-based approach 945 for short text clustering The Social Logic of Space Space syntax: consolidation and transformation of an urban 951 research field Depthmap 4: A researcher's handbook Changes in airport operating procedures and implications for 957 airport strategies post-covid-19 Social distancing enhanced automated optimal design of physical spaces 962 in the wake of the covid-19 pandemic The airport 965 gate assignment problem: a survey, The scientific world journal Comprehensive analysis of the relationship between 968 thermal comfort and building control research-a data-driven literature 969 review User-led decentralized thermal comfort driven hvac opera-973 tions for improved efficiency in office buildings Hvaclearn: A reinforcement learning based 976 occupant-centric control for thermostat set-points Occupant health in buildings: Impact 981 of the covid-19 pandemic on the opinions of building professionals and 982 implications on research, Building and Environment The future of airport post covid-19 novel coronavirus (covid-19) pandemic: built 987 environment considerations to reduce transmission Four systems of policy, politics, and choice, Public admin-990 istration review A review of unsupervised statistical 992 learning and visual analytics techniques applied to performance analysis 993 of non-residential buildings Apples or or-996 anges? identification of fundamental load shape profiles for benchmark-997 ing buildings using a large and diverse dataset Agent-based analysis of urban 1000 spaces using space syntax and spatial cognition approaches: A case 1001 study in bari, italy, Sustainability 12 (2020). declare the following financial interests/personal relationships