The Harvard China Fund is continually dedicated to supporting scholars to continue their research amidst the COVID-19 pandemic. As a result, we started offering grants through an emergency funding program to help graduate students alleviate some of the costs they face when conducting their research. Today, we want to highlight one of the recipients of this program: Tianyu Su.

Tianyu Su’s research revolves around the importance of parks in the urban landscape, and the effects that parks can have on one’s physical, mental, and emotional well-being. Su analyzes behavioural data in order to give us insight into the longitudinal effects that urban parks have in Tianjin, China.

To read more about Su’s research, find the detailed report below:


As critical physical elements in urban environments, parks’ positive impacts on human physical, mental, and social well-being have been widely discussed. However, how to trace their long-term health impacts remains unclear. This research proposes an analytical framework for measuring urban parks’ longitudinal health-promoting impacts on park visitors using large-scale behavioral data. The application of the proposed methods on an empirical case study of urban parks in Tianjin reveals that online reviews can indicate parks’ positive impacts on visitors’ physical and mental well-being. However, incorporating more data sources is necessary and beneficial to determine its influence on social well-being. Meanwhile, the substantial park-level heterogeneity reveals interesting facts about both park use patterns and park characteristics, which suggests several future research directions.

Project Background

The health impacts of urban parks have been widely studied in landscape architecture, environmental studies, and public health literature. Van den Bosch and Sang (2017) provided a systematic review of reviews to investigate the concept of nature-based solutions (NBS) and its relationship with a wide range of health outcomes (e.g., all-cause mortality, birth weight, mental health, and well-being). It also discussed the pathways through which NBS intervene in health and well-being, such as stress and physical activity. In Abraham et al. (2010), over 120 studies were reviewed in the scoping study, which conceptualized that urban parks promote physical, mental, and social well-being. It also covered discussions on health-promoting pathways.

Some of the works explored what characteristics of urban parks are associated with health-promoting impacts on visitors. For example, Abraham et al. (2010) reported that the availability of public open space, vegetation richness, and pleasant perceptions could promote human health. Also, urban parks’ positive effects on child well-being are reported in some literature (Loukaitou-Sideris & Sideris, 2009; van den Bosch & Ode Sang, 2017). Based on a heuristic health-promoting framework of urban parks (Abraham et al., 2010, p. 64) and all the relevant literature reviewed above, I summarize a more comprehensive framework on the health-promoting impacts of urban parks (Figure 1), which acts as a theoretical foundation for the analytical methods proposed in this research.


Dianping Review Data

The primary data source available to this research is the user-generated online reviews of the selected urban parks from, the dominant online review platform in China. Dianping review data is a plausible data source for this research because of several reasons: First, these reviews represent a population who visited the parks and were willing to share their attitudes (perceptions, behavior, feelings, etc.). As the dominant review platform in China, Dianping has a significant influence on parks, stores, and restaurants people choose to visit.

Second, the data provides an opportunity to observe and assess the multi-year longitudinal perceived health impacts of urban parks. Moreover, the data allows us to tackle this task at one cross-sectional point.

Measuring Health Impacts via Text Mining

To understand the underlying behavioral patterns in the unstructured review texts and the implications of these patterns on perceived health impacts, I applied two text mining methods to reviews: keywords detection and topic mining. The term frequency-inverse document frequency (TF-IDF) method has been widely used in mining latent keywords of a corpus consisting of multiple text segments, such as a set of online reviews or a set of Twitter posts. In this research, the TF-IDF method is used as an exploratory approach to understanding the keywords of users’ comments.

Then, trying to expand our knowledge of the review texts from the word level to the sentence-paragraph level, I labeled the reviews in 8 dimensions (i.e., size of open space, amount of vegetation, biodiversity, pleasant perception, evoking positive emotions, increasing physical activity, promoting social integration, & promoting children physical activity) using a synonym-based topic mining model, covering both landscape characteristics and health-promoting pathways in the health-promoting framework. It results in a 8-dimension vector for each review. Each element in the vector can take the value of 1, -1, or 0, where 1 denotes a positive mentioning towards that dimension, -1 denotes a negative mentioning, and 0 denotes the dimension is not mentioned. The labeling results are then aggregated by years of each park according the following equation, where Score d,y denotes the “dimensional score” of topic d of year y , l i,d denotes the label of dimension d of the i th review, and n y denotes the number of sampled reviews of year y .


Keywords Detection Results

The application of the proposed TF-IDF keywords detection model reveals the top 20 keywords of every urban park. While the keywords vary from park to park, more than half of them are related or broadly relevant to human health. For example, “kid” and “child” are very frequently mentioned in the online reviews, indicating a certain possibility that the review texts carry information about children’s behavior in the park and the corresponding health implications. Similarly, a large proportion of reviews mentioned “play,” “exercise,” and “work out,” which inspired me to investigate how users describe their physical activities in the parks. As the

COVID-19 pandemic substantially influenced people’s life and their way of using urban open spaces, the word “pandemic” is frequently mentioned in the 2020 review data.

Topic Mining Results

Applying the proposed topic mining method, I obtain the dimensional scores of the 8 health-promoting dimensions in 2018, 2019, and 2020 of all urban parks. Figure 2 shows the results of 9 example parks.

Figure 2 reveals several interesting facts. First, these 9 parks, which differ in location, size, and design, are shown to be largely heterogeneous among the health-promoting dimensions. This fact inspires us to pay more attention to how the park characteristics and urban context affect the health-promoting mechanisms of urban parks. Second, the high percentage of reviews that mentioned (adult) physical activities and positive emotions among all parks indicate positive impacts on humans’ physical and mental well-being. However, their effects on social well-being remain unknown since we are unclear about the reasons behind the low mentioning rate from this single data source. Future investigations that incorporate more data sources or multi-site comparisons will be beneficial.

Figure 1. A framework on the health-promoting impacts of urban parks

Figure 2. Perceived health impacts of 9 example urban parks in Tianjin