An Exploration of the Association Between Residents’ Sentiments and Street Functions During Heat Waves—Taking the Five Core Urban Areas of Chengdu City as an Example

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

（1）The  introduction outline research aims but lack explicit hypotheses about how specific street functions (e.g., green spaces, industry) influence sentiments. the authors should strengthen the analytical focus.

（2）as mentioned in the paper, The study mainly relies on Sina Weibo data, which primarily reflects younger demographics, neglecting older adults and children—groups more vulnerable to heatwaves (Paragraph 1-202). so how to solve this issues.

（3）While the RoBERTa model is mentioned (Paragraphs 1-84–1-90), the preprocessing steps (e.g., stopword removal, text normalization) and training data details (e.g., JD.com corpus specifics) are insufficiently described. 

 

Author Response

Response to Reviewer 1 Comments

We would like to thank you for careful and thorough reading of this manuscript and for the thoughtful comments and constructive suggestions, which help to improve the quality of this manuscript. Here is a point-by-point response to your comments and concerns. All page and line numbers refer to the revised manuscript file.

Point 1: The introduction outline research aims but lack explicit hypotheses about how specific street functions (e.g., green spaces, industry) influence sentiments. the authors should strengthen the analytical focus.

Response 1: 

Thanks for this comment. We agree with the reviewer's suggestion, however, within the relevant field, it remains challenging to formulate a specific hypothesis based on the current body of knowledge, e.g., the presence of commercial sites positively/negatively affects people's sentiment during heat waves. Furthermore, due to variations in socio-regional-cultural contexts, it is difficult to propose a specific path hypothesis grounded in existing studies for this research. Thirdly, this is an exploratory study, and one of the primary objectives of this study is to examine the existence of relationships between variables. Therefore, while we greatly appreciate your valuable suggestion, we would like to note that it is difficult to incorporate specific hypotheses in the Introduction chapter of the current manuscript. Instead, we present four research objectives to frame the study.

Details as follows:

(Line 148-159) In sum, taking the five core urban areas of Chengdu as an example, this study focuses on the urban heatwave event in Chengdu in August 2022, aiming to achieve the following four aims:

1) conduct sentiment analysis on social media text data (Sina Weibo) during the heat wave event;

2) use spatialization methods to explore the spatial pattern of residents' sentiments during the period;

3) to classify the functional categories of streets (FCS) within the five core urban areas of Chengdu based on the Point of Interest (POI) data, and use the Spearman correlation coefficient and Sankey diagram to analyze the correlation between their distribution and the residential sentiments during heat waves (SDHW);

4) put forward relevant strategic suggestions.

 

Point 2：as mentioned in the paper, The study mainly relies on Sina Weibo data, which primarily reflects younger demographics, neglecting older adults and children—groups more vulnerable to heatwaves (Paragraph 1-202). so how to solve this issues.

Response 2:

We agree with the comment. Indeed, sentiment among populations more vulnerable to heatwaves, such as the elderly and children, are difficult to capture through Sina Weibo data, resulting in inadequate representation of these groups in our sample. This is a limitation of our study. At present, it is difficult to fill this data gap with enough additional data due to technical limitations and user selection bias. Therefore, we have explicitly acknowledged this limitation in Section 5.4 and proposed it as a direction for future work to specifically conduct more targeted investigations into how heatwaves affect the emotional well-being of different demographic groups.

Details as follows:

(Line 615-624) The method of our study still has some limitations and needs more future research. First, although sentimental expressions in social media texts can reflect group emotional tendencies, the data users were predominantly from the youth group, with insufficient sample coverage from the elderly and children’s groups. Given that the latter two groups are more emotionally vulnerable in extreme weather, the findings have limitations on the overall representativeness of urban residents' sentiments and may underestimate the negative emotional impact of heat waves on specific groups. Future studies could consider individual characteristics such as occupation, income, and age of residents to more deeply analyze how the emotional health of different populations is affected by high temperature.

 

Point 3: While the RoBERTa model is mentioned (Paragraphs 1-84–1-90), the preprocessing steps (e.g., stopword removal, text normalization) and training data details (e.g., JD.com corpus specifics) are insufficiently described.

Response 3:

Thank you for this helpful comment. In accordance with this comment and other reviewers' suggestions, we have rewritten the Method chapter. In the current version, we clearly articulate the difference between traditional NLP and the RoBERTa model, and why no preprocessing steps need in the RoBERTa. In addition, we add more training data details about JD.com corpus specifics to describe the pre-training more sufficiently.  

Details as follows:

(Line 522-534) In this study, the uer/roberta-base-finetuned-jd-binary-chinese pre-trained model was adopted to complete the sentiment analysis of Weibo. RoBERTa (Robustly Optimized BERT Approach) is a pre-trained model provided by the natural language processing (NLP) library transformers developed by Hugging Face. Based on RoBERTa, uer/roberta-base-finetuned-jd-binary-chinese is an improvement increasing training data and performance specifically for Chinese sentiment analysis tasks. It was fine-tuned during large-scale pre-training using the UER-py framework and combined with Chinese datasets from various fields. During the pre-training stage, the model used JD.com review data as training data, which included review content, reviewer nicknames or IDs, review time, review ratings (such as positive, neutral, negative reviews, etc.) and other information. By learning the emotional patterns in JD.com reviews to achieve emotional analysis and judge whether the reviews express positive or negative emotions, this model can be more adaptable to the unique characteristics of Chinese texts and better simulate the Chinese social media context.

Unlike traditional NLP that requires manual word segmentation using tools like jieba before sentiment analysis, uer/roberta-base-finetuned-jd-binary-chinese incorporates subword tokenization during the pre-training phase. When loading the model, BertTokenizer is automatically invoked to split Chinese text into subword units, where rare words are decomposed into multiple subwords while common words remain intact. Additionally, due to exposure to extensive raw text (including stopwords, numbers, special symbols and other expressions) during pre-training, the model has learned semantic patterns and associations in the presence of stopwords, such as price 100 yuan and price one hundred yuan. As a result, operations like stopword removal and text normalization are unnecessary by using the uer/roberta-base-finetuned-jd-binary-chinese. Taking above, it is considered an ideal choice for the sentiment analysis of Chinese social media in our study.

 

Additional clarifications

In addition to the above comments, all spelling and grammatical errors have been corrected. Once again, we thank you for the time you put in reviewing our manuscript and look forward to meeting your expectations.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This study used natural language processing techniques to quantify the social media text sentiment of Chengdu residents during the high temperature period; besides, spatial and correlation analyses were combined with traditional geographic information to explore the spatial heterogeneity of the sentiment during the period, and its interaction with the functional types of streets. The research topic is interesting and well-structured, with recommendations to guide practice.

However, the following issues need to be addressed before further consideration:

 

1) Some of the structures could be adapted. For example, the first subsection of (Introduction) should focus on the context of the increasing severity of heat waves, while the second subsection centers on resilient city development and trends in sentiment-related resilience research. The third subsection continues to elaborate on existing research on sentiment in the context of heat waves.

 

2) Although, the Lines 78-81 mention the change of Chengdu's high temperature in recent years, it is suggested to supplement the background data of Chengdu's transformation from a “mild climate city” to a “high-temperature sensitive city” (e.g., the increasing trend of the number of high-temperature days in the past ten years) to strengthen the necessity of the study. For example, according to the statistics of Sichuan Daily (https://3g.china.com/act/news/10000169/20250521/48364502.html), the number of days of high temperature in Chengdu was only 2 days in 2015, but reached 27 days in 2022, 11 days in 2023, and 21 days in 2024. This data visualizes that since 2022, the number of days of high temperature in Chengdu has been increasing rapidly, and hot weather is becoming more frequent.

 

3) Information about the seriousness of high temperatures in Chengdu in 2022 can be illustrated using the chart.

 

4) (Introduction, Line 125-128) "Considering the availability of detailed demographic data and alignment with existing studies, this study employs the street as the basic spatial analysis unit, since it is the smallest internal constituent unit of Chinese cities, this scale has been widely applied in relevant extreme climate research." The content for the street as the research unit should be moved to the line after 155-179. The logic should be to elaborate on the seriousness of high temperatures in Chengdu in 2022, then present the reasons for the selection of the research unit, and then present the information on streets in the five urban areas.

 

5) (Subsection 4.3, line 370-373) “In addition, the effect of traffic function on sentiments is more complicated, the best sentiment flow lines in the figure tend to flow to the higher proportion of traffic function, while the better sentiment flow to the lower proportion or even the lowest proportion of traffic function.” mentions that there are spatial differences in the effects of transportation functions on sentiment. However, subsection 5.2.2 does not analyze the reasons for this result in depth. It is recommended to analyze the correlation between noise/heat island effect due to high-density traffic and the inconvenience of traveling due to low-density traffic and sentiment respectively, taking into account the actual situation of traffic in the five urban areas of Chengdu.

 

6) (Subsection 5.2.2) The limited role of green space in extreme heat is mentioned here, and it is recommended that citations be added, e.g., Wakayama, M., Mameno, K., Owake, T., Aikoh, T., & Shoji, Y. (2025). Climate change-induced heat reduces urban green space use: Insights from mobile phone location big data. Urban Forestry & Urban Greening, 107, 128771. This study in Sapporo, Japan, found that high temperatures affect the number of people using urban green spaces; the number of people using green spaces decreased significantly above 28°C, and green spaces with water playing and indoor facilities were more highly utilized at high temperatures.

 

7) The use of the term “sentiment” should be consistent, and some usage of emotion should be changed to sentiment in the discussion section.

 

8) It should be mentioned in the limitations section: 1) The street function type scoring method based on POI data can be further validated in the future; 2) In response to the analysis results, a more specific strategy can be proposed in the future, using a certain street with sentiment-function typical characteristics during high temperatures as an example; (3) Social media use users are mainly middle-aged and young representatives of the population, and in the future, different age groups can be used to expand the scope. At the same time, more social variables such as users' education and occupation can be added to correlate with residents' emotions during the heat wave, in order to explore the social inequality of specific groups of people in specific functional types of streets during the heat wave.

 

Some writing revisions:

• (Abstract, Line 12-14) "However, the correlation mechanism between different urban functional layouts and residents' sentiments at the street scale still requires more analysis." could be optimized as “However, the associative mechanism between diverse urban functional layouts and residents’ emotions at the street scale remains underexplored.”
• (Line21-22) “the proportion of positive sentiments within the Second Ring Road is higher than that in the peripheral areas” could be optimized as “positive emotions within the Second Ring Road exhibit a higher proportion than peripheral areas" for smoother flow.”
• (line 168) Eastern New Area should be Eastern New District.

Author Response

Response to Reviewer 2 Comments

We would like to thank you for careful and thorough reading of this manuscript and for the thoughtful comments and constructive suggestions, which help to improve the quality of this manuscript. Here is a point-by-point response to your comments and concerns. All page and line numbers refer to the revised manuscript file.

OVERALL IMPRESSION

This study used natural language processing techniques to quantify the social media text sentiment of Chengdu residents during the high temperature period; besides, spatial and correlation analyses were combined with traditional geographic information to explore the spatial heterogeneity of the sentiment during the period, and its interaction with the functional types of streets. The research topic is interesting and well-structured, with recommendations to guide practice.

 

Point 1: Some of the structures could be adapted. For example, the first subsection of (Introduction) should focus on the context of the increasing severity of heat waves, while the second subsection centers on resilient city development and trends in sentiment-related resilience research. The third subsection continues to elaborate on existing research on sentiment in the context of heat waves.

Response 1:

Thank you for this helpful comment. We have restructured the Introduction to ensure a more coherent logical flow between its subsections.

Details as follows:

(Line 37-159)1. Introduction

1.1 Studies on heat waves and healthy city construction

1.2 Urban resilience and residents' sentiments

1.3 Residents' sentiments during heat waves

1.4 Urban functional layout and residents' sentiments

1.5 The validity of techniques and the purpose of study

 

Point 2: Although, the Lines 78-81 mention the change of Chengdu's high temperature in recent years, it is suggested to supplement the background data of Chengdu's transformation from a “mild climate city” to a “high-temperature sensitive city” (e.g., the increasing trend of the number of high-temperature days in the past ten years) to strengthen the necessity of the study.

Response 2:

Agree with this comment. To address this issue, we have supplemented our analysis with contextual data documenting Chengdu's transition from a “mild climate city” to a “high-temperature sensitive city”. This background information draws on statistics published by Sichuan Daily (https://www.china.com/act/news/10000169/20250521/48364502.html)."

Details as follows:

(Line 63-68) According to statistics from the Sichuan Daily (https://3g.china.com/act/news/10000169/20250521/48364502.html), there were only 2 high-temperature days (35°C and above) in 2015. However, the number of high-temperature days in Chengdu has entered double digits since 2022, with 27 days in 2022, 11 days in 2023, and 21 days in 2024. This data in Chengdu is growing rapidly, with high-temperature weather becoming increasingly frequent.

 

Point 3:

Information about the seriousness of high temperatures in Chengdu in 2022 can be illustrated using the chart.

Response 3:

Thank you for your careful examination of our manuscript. We added a comparative chart illustrating daily average temperatures in Chengdu during August 2021 versus August 2022, demonstrating the severity of extreme heat during the summer of 2022.

Details as follows:

(Line 72-75)

 please refer the Figure 1

 

Point 4:

(Introduction, Line 125-128) "Considering the availability of detailed demographic data and alignment with existing studies, this study employs the street as the basic spatial analysis unit, since it is the smallest internal constituent unit of Chinese cities, this scale has been widely applied in relevant extreme climate research." The content for the street as the research unit should be moved to the line after 155-179. The logic should be to elaborate on the seriousness of high temperatures in Chengdu in 2022, then present the reasons for the selection of the research unit, and then present the information on streets in the five urban areas.

Response 4:

Thank you for pointing this out. We have relocated the indicated text to the chapter 2 (Study Area), positioning it immediately before the statement: “This study finally determines the study area as the five core urban areas of Chengdu (Figure 2), namely Wuhou District, Qingyang District, Chenghua District, Jinjiang District, and Jinniu District.”

This reorganization establishes a logical progression within the Study Area section:

(1) First, describing the 2022 summer heatwave event in Chengdu

(2) Then, presenting the rationale for selecting street-level units as our spatial scale

(3) Finally, introducing street-level information for the five urban districts.

Details as follows:

(Line 197-206) Studies on urban heat waves generally focus on the area most significantly affected, that is, the central urban area. Considering the availability of detailed demographic data and alignment with existing studies, this study employs the street as the basic spatial analysis unit, since it is the smallest internal constituent unit of Chinese cities, this scale has been widely applied in relevant extreme climate research [19].This study finally determines the study area as the five core urban areas of Chengdu (Figure 2), namely Wuhou District, Qingyang District, Chenghua District, Jinjiang District, and Jinniu District. Taking streets as units (a total of 77), including 17 streets in Wuhou District, 14 streets in Qingyang District, 14 streets in Chenghua District, 16 streets in Jinjiang District, and 16 streets in Jinniu District.

 

Point 5：(Subsection 4.3, line 370-373) “In addition, the effect of traffic function on sentiments is more complicated, the best sentiment flow lines in the figure tend to flow to the higher proportion of traffic function, while the better sentiment flow to the lower proportion or even the lowest proportion of traffic function.” mentions that there are spatial differences in the effects of transportation functions on sentiment. However, subsection 5.2.2 does not analyze the reasons for this result in depth. It is recommended to analyze the correlation between noise/heat island effect due to high-density traffic and the inconvenience of traveling due to low-density traffic and sentiment respectively, taking into account the actual situation of traffic in the five urban areas of Chengdu.

Response 5:

Thank you for pointing this out. In subsection 5.2.2, by citing the reference "The effect of transport infrastructure, congestion and reliability on mental wellbeing: a systematic review of empirical studies", this review aims to provide an up-to-date synthesis of research evidence about the influence of transport infrastructure and operational performance (congestion, delays and reliability) on mental health/wellbeing. This further explains the reasons why a better mood flow is directed towards a lower proportion or even the lowest proportion of transportation functions, which echoes the viewpoint in subsection 4.3 and ensures the logical flow of the text.

The reference is as follow.

-Conceição, M. A., Monteiro, M. M., Kasraian, D., van den Berg, P., Haustein, S., Alves, I., … Miranda, B. (2022). The effect of transport infrastructure, congestion and reliability on mental wellbeing: a systematic review of empirical studies. Transport Reviews, 43(2), 264–302. https://doi.org/10.1080/01441647.2022.2100943

Details as follows:

(Line 513-523) In addition, according to the results, it is known that a high percentage of traffic function is more associated with positive sentiments. Previous research has found that traffic function tends to play a positive role in influencing residents' sentiments [28], and residents living in areas with good access to transportation may have a better emotional experience. But at the same time, the results also show that although the best emotional flow lines tend to flow to a higher proportion of the traffic function, the better emotions flow to a lower proportion or even the lowest proportion of the traffic function. This is mentioned in the relevant research on the impact of a high proportion of traffic on residents' emotions and mental health. Robust evidence shows traffic congestion and delays, especially high density and low speeds common in urban areas, negatively impact mental health and wellbeing assessments [34].

 

Point 6：(Subsection 5.2.2) The limited role of green space in extreme heat is mentioned here, and it is recommended that citations be added.

Response 6:

Thank you for this helpful comment. We have included the reference "Climate change-induced heat reduces urban green space use: Insights from mobile phone location big data " to support the argument that a high proportion of green spaces and parks does not lead to a very significant improvement in sentiment. This study focuses on whether the increase in extreme weather events and the rise in summer temperatures affect urban residents' entry into and use of urban green spaces.

The reference is as follow.

-Wakayama, M., Mameno, K., Owake, T., Aikoh, T., & Shoji, Y. (2025). Climate change-induced heat reduces urban green space use: Insights from mobile phone location big data. Urban Forestry & Urban Greening, 107, 128771.

Details as follows:

(Line 523-532) The low proportion of green space and park during the heat waves also led to negative sentiments, but the high proportion of green spaces and parks did not have a very significant improvement in sentiments, probably because of the extreme high temperatures during the heat waves, which made people reluctant to go outdoors even when living around green spaces, and also exacerbated the reluctance to go outside due to the electricity restrictions in public areas caused by the heat waves. A study in Sapporo, Japan, found that high temperatures affect the number of people using urban green spaces; the number of people using green spaces decreased significantly above 28°C, and green spaces with water playing and indoor facilities were more highly utilized at high temperatures [35].

 

Point 7：The use of the term “sentiment” should be consistent, and some usage of emotion should be changed to sentiment in the discussion section.

Response 7:

Thank you for your careful examination of our manuscript. We have changed the usages of "emotion" to "sentiment" in the discussion section.

 

Point 8：It should be mentioned in the limitations section: 1) The street function type scoring method based on POI data can be further validated in the future; 2) In response to the analysis results, a more specific strategy can be proposed in the future, using a certain street with sentiment-function typical characteristics during high temperatures as an example; (3) Social media use users are mainly middle-aged and young representatives of the population, and in the future, different age groups can be used to expand the scope. At the same time, more social variables such as users' education and occupation can be added to correlate with residents' emotions during the heat wave, in order to explore the social inequality of specific groups of people in specific functional types of streets during the heat wave.

Response 8:

Agree with this comment. Based on the reviewers' suggestions, we have supplemented the content related to points (1) and (2) into the "Limitations and Future Research" section. However, regarding point (3) raised in the comments, the original section already addressed substantially comparable points, and we therefore deemed no further revisions necessary.

Details as follows:

(Line 611-614) Within Chengdu, based on the representative street cases identified in this study, further exploration of the correlation between sentiments and street functions could be conducted, with corresponding strategies implemented on a small scale to monitor changes in residents' sentiments before and after renovations.

(Line 629-633) In addition, considering that the functions of different streets have different influential scopes and public awareness, the buffer zone indicators used in the street function scoring formula in this study can only provide approximate street function information. Further research should be conducted in the future on POI—street function categories calculations.

 

Additional clarifications

In addition to the above comments, all spelling and grammatical errors have been corrected. Once again, we thank you for the time you put in reviewing our manuscript and look forward to meeting your expectations.

 

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Review of Research Paper 

This study offers valuable insight into how different street functions in Chengdu shape residents’ emotional experiences during heat waves. By combining social media sentiment analysis with spatial data, the authors tackle an increasingly relevant topic in the context of urban climate resilience. The study is timely and ambitious, particularly given the challenges of extracting meaningful data from real-time digital platforms like Weibo. That said, several aspects of the methodology and analysis would benefit from greater clarity and rigor. Strengthening these areas will increase its utility for policymakers and urban planners. Below are my key recommendations aimed at refining the manuscript:

Key Recommendations for Improvement

1. Enhance Methodological Transparency

• Data Collection: Please clarify the timeframe of data collection, selection criteria for Weibo posts, and precise boundaries of the “five core urban areas.” It would also be helpful to detail how POI (point-of-interest) data were classified into street function categories, what thresholds or rules were used. 
• Sentiment Analysis: Indicate what NLP model was used ( BERT, LSTM), how it was trained, and the language resources involved (sentiment lexicons for Chinese). Provide also some validation metrics, such as confusion matrices or F1 scores. Consider also how dialectal nuances, such as Sichuanese Mandarin, might affect sentiment interpretation.

2. Strengthen Statistical Rigor

• Correlation Metrics: Include confidence intervals and p-values when reporting Spearman correlation coefficients. Sensitivity analyses also could reinforce the reliability of your findings.
• Spatial Analysis: Incorporating tests for spatial autocorrelation, such as Moran’s I or Geary’s C, and using spatial regression models like SEM, if possible, would better account for geographic clustering and reduce potential bias.
• Control for Confounders: Consider including socioeconomic variables, such as average income or age distribution, in multivariate models to better isolate the influence of street functions on sentiment.

3. Address Potential Biases

• Data Representativeness: Acknowledge the demographic limitations of Weibo users, particularly with regard to age and socioeconomic status. Cross-checking findings with surveys or heat-related complaint hotlines could help triangulate and validate observed patterns.
• Temporal Variation: if possible, explore how sentiment changes over time, for example, comparing midday versus evening tweets during heat waves. 

4. Refine Interpretation and Language

• Avoid Overstating Causality: Since this is a correlational study, I suggest avoiding causal terms such as “promotes” or “leads to.” More neutral language, such as “associated with,” would be appropriate. Also consider reverse causality possibilities, for instance, negative sentiment may reflect pre-existing deficits in green space, rather than the other way around. 
• Offer Explanatory Hypotheses: Consider briefly theorizing why certain street types, like green spaces, might correspond to positive sentiments, perceived safety, or increased social interaction, for example.

5. Improve Visual Communication

• Sankey Diagrams: Clarify node categories and flow volumes more explicitly. Color gradients could be used to reflect the intensity of sentiment polarity.
• Spatial Maps: Overlay temperature or heat index data to contextualize sentiment clusters more meaningfully. Including administrative boundary insets would also help non-local readers navigate the geography.

6. Expand on Generalizability and Policy Relevance

• Broader Context: It would be valuable to compare your findings with other cities (if possible), perhaps Chongqing with similar climatic conditions or Beijing with a different urban form.
• Urban Planning Alignment: Consider aligning your recommendations with Chengdu’s ongoing “Park City” initiative, suggesting specific, actionable design interventions such as minimum green canopy thresholds for residential areas.

 

Additional Resources for Broader Insight

To deepen the theoretical and methodological foundations of your work, I encourage integrating insights from the following two studies:

• Impacts of a New Highway on Urban Development and Land Accessibility in Developing Countries
  This study explores how transport infrastructure reshapes urban dynamics and accessibility in a rapidly growing city, Beirut, offering parallels to Chengdu's street-level interventions (Al-Shaar, W., Nehme, N., Bonin, O. et al. Impacts of a New Highway on Urban Development and Land Accessibility in Developing Countries: Case of Beirut Southern Entrance in Lebanon. Arab J Sci Eng 46, 5783–5800 (2021). https://doi.org/10.1007/s13369-020-05330-8).

 

Author Response

Response to Reviewer 3 Comments

 

We would like to thank you for careful and thorough reading of this manuscript and for the thoughtful comments and constructive suggestions, which help to improve the quality of this manuscript. Here is a point-by-point response to your comments and concerns. All page and line numbers refer to the revised manuscript file.

OVERALL IMPRESSION

This study offers valuable insight into how different street functions in Chengdu shape residents’ emotional experiences during heat waves. By combining social media sentiment analysis with spatial data, the authors tackle an increasingly relevant topic in the context of urban climate resilience. The study is timely and ambitious, particularly given the challenges of extracting meaningful data from real-time digital platforms like Weibo. That said, several aspects of the methodology and analysis would benefit from greater clarity and rigor. Strengthening these areas will increase its utility for policymakers and urban planners. Below are my key recommendations aimed at refining the manuscript:

 

Point 1: Data Collection: Please clarify the timeframe of data collection, selection criteria for Weibo posts, and precise boundaries of the “five core urban areas.” It would also be helpful to detail how POI (point-of-interest) data were classified into street function categories, what thresholds or rules were used.

Response 1:

Thank you for this comment. In accordance with this comment and other reviewers' suggestions, we have added more details and information about data collection. In addition, we added precise boundaries of the “five core urban areas.” please refer the figure 4 of the revised version.

Details as follows:

(Line 242-249) The Weibo posts obtained through web scraping need to be screened to improve the data accuracy. Firstly, ArcGIS was used to remove the posts checked in outside the five core urban areas of Chengdu. Following, for posts cleaning, the Python code was used to remove advertising posts, blank posts, and Weibo posts without text information; besides, removed the content such as "@user", "#talk hashtag#", and "http: URL link" in the Weibo texts. Finally, 19,388 valid posts were retained, as shown in Figure 5. The yellow dots in the figure represent the geographical locations of the Weibo posts with check-in information.

(Line 304-337) By weighting the POI data, the distribution of urban functional categories is determined by the percentage of the total scores of POI points representing different urban functions at street scale, so as to carry out a quantitative study of urban functions. Some POI data with insignificant functional categories, such as public restrooms and newsstands, are first excluded. Besides, considering the differences in the area of each street unit, the direct calculation of the percentage of the number of POIs in each unit cannot accurately reflect the actual functional attributes of the situation, so the frequency density index is introduced.

On the other hand, the POI data is a point without any area information of the geographic entity, and there are differences in the scope of influence of the geographic entity of different functional categories. If only the frequency density index is considered will make the calculation results biased towards the categories of POIs with a larger number in the region, such as shopping services and catering services, and it is difficult to reflect the differences in the functional categories of different regions.

Therefore, our study introduced buffer zone area indicators with different radii according to POI public awareness, and carries out further weighting and reconciliation of various categories of POIs: 200m for commercial, 250m for residential, 300m for public, 350m for industrial, 400m for transportation, and 150m for green space and park [26]. The weight of the buffer zone area of a certain category of POI within a single research unit (a street) over the area of all buffer zones within the research unit is taken as the weight of that category within the unit.

Based on the above, this study proposed a quantitative identification method for the distribution of urban functional categories, with the following calculation formula:

where k denotes the kth street unit; i denotes the ith category of POI; Sk denotes the kth street area; Si denotes the area of the buffer zone of the ith category of POI; and ni denotes the number of the ith category of POI within that street. The weight of the buffer area of a category of POI within a single street over the area of all buffers within that street is used as the weight of that category of POI within that street, and H () represents the ratio of the frequency density of the ith category of POI over the frequency density of all categories of POIs within the kth street. The proportion of POIs in the six major categories is finally calculated and the distribution of urban functional categories is mapped at the street scale. In addition, as in the case of the Weibo data mentioned above, all the functional category shares were transformed into a five-category hierarchy, categorized as lowest, lower, medium, higher, and highest.

 

Point 2: Sentiment Analysis: Indicate what NLP model was used ( BERT, LSTM), how it was trained, and the language resources involved (sentiment lexicons for Chinese). Provide also some validation metrics, such as confusion matrices or F1 scores. Consider also how dialectal nuances, such as Sichuanese Mandarin, might affect sentiment interpretation.

Response 2: 

Thank you for this comment. In the methods section, we added the specific NLP model we used and explained how to pre-train it using JD.com review data to learn content expression in a Chinese context (including dialects and other oral expressions). We also did a manual sentiment analysis check on the 2,000 data sets in the results to make sure there were no issues with the model or metrics..

Details as follows:

(Line 255-268) In this study, the uer/roberta-base-finetuned-jd-binary-chinese pre-trained model was adopted to complete the sentiment analysis of Weibo. RoBERTa (Robustly Optimized BERT Approach) is a pre-trained model provided by the natural language processing (NLP) library transformers developed by Hugging Face. Based on RoBERTa, uer/roberta-base-finetuned-jd-binary-chinese is an improvement increasing training data and performance specifically for Chinese sentiment analysis tasks. It was fine-tuned during large-scale pre-training using the UER-py framework and combined with Chinese datasets from various fields. During the pre-training stage, the model used JD.com review data as training data, which included review content, reviewer nicknames or IDs, review time, review ratings (such as positive, neutral, negative reviews, etc.) and other information. By learning the emotional patterns in JD.com reviews to achieve emotional analysis and judge whether the reviews express positive or negative emotions, this model can be more adaptable to the unique characteristics of Chinese texts and better simulate the Chinese social media context.

(Line 286-289) Finally, 17388 Weibo posts obtained sentiment scores ranging from 0.02 to 0.98 by the model and other 2,000 were selected as the validation set for manual verification, the accuracy rate reaching 87% and indicating that the results are highly reliable.

 

Point 3: Correlation Metrics: Include confidence intervals and p-values when reporting Spearman correlation coefficients. Sensitivity analyses also could reinforce the reliability of your findings.

Response 3: 

Thank you for pointing this out. In the result section, we add confidence intervals and p-values when reporting Spearman correlation coefficients. The overall research framework for this manuscript is based on the following articles and didn't include sensitivity analysis. However, sensitivity analysis is a very important technique, and we will consider it in future research.

The references are as follows.

-Dai, D., Dong, W., Wang, Y., Liu, S., & Zhang, J. Exploring the relationship between urban residents' emotional changes and built environment before and during the COVID-19 pandemic from the perspective of resilience. Cities 2023, 141, 104510.

-Wang, B., Loo, B. P., Zhen, F., & Xi, G. Urban resilience from the lens of social media data: Responses to urban flooding in Nanjing, China. Cities 2020, 106, 102884.

-Zhu, Y., Wang, J., Yuan, Y., Meng, B., Luo, M., Shi, C., & Ji, H. Spatial heterogeneities of residents' sentiments and their associations with urban functional areas during heat waves–a case study in Beijing. Computational Urban Science 2024, 4(1), 7.

Details as follows:

(Line 404-414) This section focuses on the effects of combinations of two or three urban functional types on SDHW. Table 1 lists the association rules at p-values less than 0.1 and 0.05, which show which FCS combinations have a more significant positive or negative effect on SDHW. From the table, whether or not an FCS combination is significantly associated with SDHW is mainly influenced by the share of high residential function, with green space and parks and public function also appearing more frequently in each rule.

Table 1. Correlation between SDHW and multiple urban FCS

	Variable	Correlation with SDHW
ID	Multiple urban FCS	Coefficient	p-values
1	TCS, CCS	-0.11	0.03**
2	GPCS, RCS	0.26	0.02**
3	RCS, CCS	0.19	0.09*
4	ICS, RCS	0.28	0.01**
5	GPCS, CCS, RCS	0.20	0.08*
6	GPCS, ICS, RCS	0.23	0.05*
7	GPCS, PCS, RCS	0.20	0.08*
8	CCS, ICS, RCS	0.31	0.01**
9	ICS, PCS, RCS	0.23	0.04**
10	CCS, PCS, TCS	-0.17	0.03**

1 * Significant at the 90 % level (p<0.1); ** Significant at the 95 % level (p<0.05).

2 Note: TCS, traffic category of streets; CCS, commercial category of streets; GPCS, green spaces and park category of streets; RCS, residential category of streets; ICS, industrial category of streets; PCS, public category of streets.

 

Point 4:

Spatial Analysis: Incorporating tests for spatial autocorrelation, such as Moran’s I or Geary’s C, and using spatial regression models like SEM, if possible, would better account for geographic clustering and reduce potential bias.

Response 4:

Thank you for this helpful comment. This study aims to explore the relationship between street functional categories and sentiments, so we did not use spatial auto-correlation to examine the distribution patterns of sentiments. In addition, we analyzed the spatial distribution patterns of sentiments (please refer the Figure 9 and the context), and we believe that this section is sufficient to explain the relevant content.

Details as follows:

(Line 352-366) As shown in Figure 9 (left), during the heat waves, there were obvious spatial differences in the distribution of sentiment values of the residents in Chengdu City; the proportion of the five categories of sentiment values of the streets in each district was also counted (Figure 9, right). Overall, for positive sentiment values (better and best categories), streets within the Second Ring Road are larger than streets outside the Second Ring Road, and streets in the western districts are larger than streets in the eastern districts. The main western districts are Wuhou, Qingyang and Jinniu; of them, 57% streets in Qingyang district have positive sentiment values, and 47% streets in Wuhou district have positive sentiment values. Negative sentiment values (worse and worst categories), on the other hand, are dominated by the eastern districts, for example, 71% the streets in Chenghua district have negative sentiment values, with Shahe Street, Shishishan Street, and Chenglong Road Street all having negative sentiment values.

Please refer the Figure 9. 

 

Point 5: Control for Confounders: Consider including socioeconomic variables, such as average income or age distribution, in multivariate models to better isolate the influence of street functions on sentiment.

Data Representativeness: Acknowledge the demographic limitations of Weibo users, particularly with regard to age and socioeconomic status. Cross-checking findings with surveys or heat-related complaint hotlines could help triangulate and validate observed patterns.

Response 5: 

Thank you for this comment. In accordance with this comment and other reviewers' suggestions, we have added these constructive comments to the limitations.

Details as follows:

(Line 615-624) First, although sentimental expressions in social media texts can reflect group emotional tendencies, the data users were predominantly from the youth group, with insufficient sample coverage from the elderly and children’s groups. Given that the latter two groups are more emotionally vulnerable in extreme weather, the findings have limitations on the overall representativeness of urban residents' sentiments and may underestimate the negative emotional impact of heat waves on specific groups. Future studies could consider individual characteristics such as occupation, income, and age of residents to more deeply analyze how the emotional health of different populations is affected by high temperature.

 

Point 6: Temporal Variation: if possible, explore how sentiment changes over time, for example, comparing midday versus evening tweets during heat waves.

Response 6:

Thank you for your helpful comments. Since our study topic is about the correlation between sentiments and street function, we did not continue to compare the differences in sentiments between different time periods due to space limitations.

 

Point 7: Avoid Overstating Causality: Since this is a correlational study, I suggest avoiding causal terms such as “promotes” or “leads to.” More neutral language, such as “associated with,” would be appropriate. Also consider reverse causality possibilities, for instance, negative sentiment may reflect pre-existing deficits in green space, rather than the other way around.

Response 7:

Thanks for your comments. To avoid overstate causality, we have used neutral language and checked the manuscript to ensure no writing and grammar errors.

 

Point 8: Offer Explanatory Hypotheses: Consider briefly theorizing why certain street types, like green spaces, might correspond to positive sentiments, perceived safety, or increased social interaction, for example.

Response 8:

Thank you for pointing this out. In accordance with this comment and other reviewers' suggestions, we have added relevant information in the section 5.2.2.

Details are as follows.

(Line 523-535) The low proportion of green space and park during the heat waves also led to negative sentiments, but the high proportion of green spaces and parks did not have a very significant improvement in sentiments, probably because of the extreme high temperatures during the heat waves, which made people reluctant to go outdoors even when living around green spaces, and also exacerbated the reluctance to go outside due to the electricity restrictions in public areas caused by the heat waves. A study in Japan found that high temperatures affect the number of people using urban green spaces; the number of people using green spaces decreased significantly above 28°C, and green spaces with water playing and indoor facilities were more highly utilized at high temperatures [35]. Besides, in line with the results of the previous analysis, the limited impact of green spaces on sentiment may make the provision of healthcare services and building energy efficiency more significant, discussions that still need more work to be validated.

 

Point 9: Sankey Diagrams: Clarify node categories and flow volumes more explicitly. Color gradients could be used to reflect the intensity of sentiment polarity.

Response 9: 

Thank you for pointing this out. In accordance with this comment and other reviewers' suggestions, we have added relevant information in the section 4.3.2.

Details are as follows.

(Line 415-429) Specifically, among the two combinations of FCSs, good SDHW is more likely to be found on streets with a lower proportion of commercial category (CCS) and a higher proportion of traffic category (TCS) (e.g., Figures 12-1, whose Sankey diagram shapes tend to be dominated by upward bending curves). Residential streets with a high proportion of green spaces and park (GPCS) (as in Figure 12-2, where the shape of the Sankey diagram tends to be a smooth straight line) mainly show positive SDHW in streets with a high proportion of residential category (RCS), and also tend to have positive SDHW in streets with a low proportion of commercial function or a low proportion of industrial category (ICS) (as in Figure 12-3 and 12-4, where the shape of the Sankey diagram tends to be an upward-bending curve).

Please refer the Figure 12.

 

Point 10: Spatial Maps: Overlay temperature or heat index data to contextualize sentiment clusters more meaningfully. Including administrative boundary insets would also help non-local readers navigate the geography.

Response 10: 

Thank you for your constructive suggestion. In fact, we overlaid temperature maps in our preliminary analysis, but we found that the surface temperature differences between districts during the heatwave were not significant, and all were above 35℃. Therefore, we did not add overlay maps in our preliminary analysis, but instead used temperature comparison chart (Figure 1) and sentiment distribution maps (Figure 9) to illustrate the background of the heatwave.

Details are as follows.

Please refer the Figure 1 and Figure 9.

 

Point 12: Broader Context: It would be valuable to compare your findings with other cities (if possible), perhaps Chongqing with similar climatic conditions or Beijing with a different urban form.

Response 12: 

Thank you for your helpful comment. Our current research has yielded many findings and results. Due to space limitations, it is difficult to include comparative studies with other cities in our current study. Therefore, we have added them in the limitation section.

Details are as follows.

(Line 607-614) In terms of study area, the scope could be further expanded to include different types of urban environments, including but not limited to large cities, medium-sized cities, and cities with different geographical and climatic conditions, to facilitate further comparative research and analysis. Within Chengdu, based on the representative street cases identified in this study, further exploration of the correlation between sentiments and street functions could be conducted, with corresponding strategies implemented on a small scale to monitor changes in residents' sentiments before and after renovations.

 

Point 13: Urban Planning Alignment: Consider aligning your recommendations with Chengdu’s ongoing “Park City” initiative, suggesting specific, actionable design interventions such as minimum green canopy thresholds for residential areas.

Response 13: 

Thank you for this helpful comment. First, the relationship between green space functions and sentiments is only one part of this study, and the results show that the relationship between green space and sentiments during heat waves is unclear. Therefore, our analysis does not emphasize the effect of park city construction on residents’ sentiments. Second, our current data cannot propose minimum green canopy thresholds, which require empirical research results in the fields of thermal environment or thermal comfort. Our current study can only test whether there is a relationship between green space and sentiments and the degree of that relationship.

 

Point 14: Additional Resources for Broader Insight

To deepen the theoretical and methodological foundations of your work, I encourage integrating insights from the following two studies:

Impacts of a New Highway on Urban Development and Land Accessibility in Developing Countries

This study explores how transport infrastructure reshapes urban dynamics and accessibility in a rapidly growing city, Beirut, offering parallels to Chengdu's street-level interventions (Al-Shaar, W., Nehme, N., Bonin, O. et al. Impacts of a New Highway on Urban Development and Land Accessibility in Developing Countries: Case of Beirut Southern Entrance in Lebanon. Arab J Sci Eng 46, 5783–5800 (2021). https://doi.org/10.1007/s13369-020-05330-8).

Response 14: 

Thank you for providing us with valuable literature, but we apologize that after carefully reading the paper, we found that it had little relevance to our research, so we did not add this reference in the revised version.

 

Additional clarifications

In addition to the above comments, all spelling and grammatical errors have been corrected. Once again, we thank you for the time you put in reviewing our manuscript and look forward to meeting your expectations.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

The originality of the article is reliant on the abstract stating that the association between sentiment, heat waves and street function requires more analysis. It is from this article assumed as an original article with signficant findings. The findings are interesting. The methods require further explanation and discussion could be more advanced by implication of different function, negative or positive sentiments, to lead to balanced and well reasoned recommendations. Examples are in green spaces, building efficiency and medical services, particularly for recommendations for streets of limited function. The conclusion could include a summary of most significant findings.

3.2 Method

What text information on weibo is posted, or how are posts understood as associated with sentiment being measured. How are these measures specifically similar to measures of the other softwares used.

Is the measure of sentiment in heatwaves by different locations influenced by green spaces different for normal temperatures. Is it measured for this study. If it isn’t could it be a limitation. Can the authors provide more information about the natural language processing.

Sentiment values require more explanation, including how roberta-base-finetuned-jd-binary-chinese corpus, and BERT measure, by indicators of sentiment. Or examples of how specific texts are understood by sentiment measures, past the worst to best measure.

Line 477-495: This limited influence on emotion of green spaces could be further recognised as measured by the methods used. Location of weibo posts and natural language analysis as indoors or outdoors is an example. Is it a limited influence because of awareness, and does it take the general influence of green spaces on temperature in terms of cooling into account. Or does this point make the provision of medical services, building efficiencies more important. Could these questions be addressed in a general discussion, past this section.

This study associates public function with sentiment during a heatwave. The recommendations for improvement are only specific to public function and well known improvements as mitigative for heatwaves. These recommendations are good for planning but there could be more about the function and sentiment measures, and then about whether optimising function for all streets is a proven strategy. While recommendations do seem realistic and rational for improvement, it could be better to provide example of how often hospital admissions, and medical service is required. It could also be more advanced by recommendation to bring these recommendations together and even consider preventative measures like green spaces, as compared to medical services. Whether green spaces for cooling, elevators etc.. can significantly reduce need for medical services, then a measure of function. This measure of function could eventually be measured using sentiment.

Line 502: Include the functional categories of each of the three streets.

Line 513-514: this is a different point to green spaces. Could building efficiency be a subheading?

Some recommendations including medical services for limited public use streets, diverse UGS by plant selections, and elevators need literature support. They are all well supported and studied, and are good recommendations.

5.4

The limitations are very good, and begin to and would improve how the method of using weibo to measure emotional state is or isn’t limited.

Conclusion

Line 578: The study could more clearly explain the functional category associated with sentiments in different streets. More specifically, which sentiments, if there can be more than a worst to best measure. This point supports the comment for methods, and indicators of sentiment.

Briefly explain which function types were of most negative, worst sentiments, commercial and no function. This point could be summarised with recommendations as a sentence or two. The variable influence of green spaces while being recommended might need some addressing in limitations of discussion. A subheading for energy efficient buildings, and further information about this function particularly for balanced recommendation.

The conclusion could explain and summarise the study, including heat wave patterns for the study year. A reader can then get a comprehensive understanding of the meaning and significance of the study.

Comments on the Quality of English Language

The article needs a proof read for grammatical corrections.

examples include:

Abstract

Global warming is a relevant point but the sentence could start with

‘The impact of heat waves….’

 Line 23: show instead of shown..

Line 102: and instead of is.

Line 109: take instead of taken.

Line 355: delete a ,

line 518: waves. It analyses 

Some of the figures need spelling corrections.m 

 

 

Author Response

Response to Reviewer 4 Comments

We would like to thank you for careful and thorough reading of this manuscript and for the thoughtful comments and constructive suggestions, which help to improve the quality of this manuscript. Here is a point-by-point response to your comments and concerns. All page and line numbers refer to the revised manuscript file.

 

OVERALL IMPRESSION

The originality of the article is reliant on the abstract stating that the association between sentiment, heat waves and street function requires more analysis. It is from this article assumed as an original article with significant findings. The findings are interesting. The methods require further explanation and discussion could be more advanced by implication of different function, negative or positive sentiments, to lead to balanced and well reasoned recommendations. Examples are in green spaces, building efficiency and medical services, particularly for recommendations for streets of limited function. The conclusion could include a summary of most significant findings.

 

Point 1: What text information on weibo is posted, or how are posts understood as associated with sentiment being measured. How are these measures specifically similar to measures of the other softwares used.

Response 1:

Thank you for this thoughtful insight. In accordance with this comment and other reviewers' suggestions, we have rewritten the method section. In the current version of the method, we clearly articulate the difference between traditional NLP and the RoBERTa model. In addition, the text information on Weibo is mainly about users’ experience and thoughts during the heat wave in August,2022, however we can’t present them completely considering the large data volume. Therefore, we add an example of the post text and its final sentiment score to make the method more understandable.

Details as follows:

(Line 269-280) Unlike traditional NLP that requires manual word segmentation using tools like jieba before sentiment analysis, uer/roberta-base-finetuned-jd-binary-chinese incorporates subword tokenization during the pre-training phase. When loading the model, BertTokenizer is automatically invoked to split Chinese text into subword units, where rare words are decomposed into multiple subwords while common words remain intact. Additionally, due to exposure to extensive raw text (including stopwords, numbers, special symbols and other expressions) during pre-training, the model has learned semantic patterns and associations in the presence of stopwords, such as price 100 yuan and price one hundred yuan. As a result, operations like stopword removal and text normalization are unnecessary by using the uer/roberta-base-finetuned-jd-binary-chinese. Taking above, it is considered an ideal choice for the sentiment analysis of Chinese social media in our study.

(Line 284-289) For example, the score was 0.02 with the comment “大热天在外面找了半个小时的咖啡店，热出一身汗” (I spent half an hour looking for a coffee shop outside in the heat, and worked up a sweat). Finally, 17388 Weibo posts obtained sentiment scores ranging from 0.02 to 0.98 by the model and other 2,000 were selected as the validation set for manual verification, the accuracy rate reaching 87% and indicating that the results are highly reliable.

 

Point 2: Is the measure of sentiment in heatwaves by different locations influenced by green spaces different for normal temperatures. Is it measured for this study. If it isn’t could it be a limitation. Can the authors provide more information about the natural language processing.

Response 2: 

The aim of this study was to explore the relationship between the different functions of urban streets and the sentiment of residents during a heatwave. We strongly agree that comparisons between heatwave sentiment influenced by green spaces and sentiment during normal period should be measured. However, this study has already captured 19,388 posts during the August 2022 heatwave alone, which is an extremely large work. Therefore, if a comparison between the data during normal period is needed, additional work would be consumed to collect, clean and analyze the data. This would be a huge challenge for us and would distract from the focus of the study since the green fields are only one part of the analysis.

We respect this suggestion from the reviewers and have therefore added this information to the limitations of the manuscript to inspire more future research.

Details as follows:

(Line 638-641) At the same time, since the sentiment-enhancing effect of green space functions is inconsistent with some previous studies, we will analyze the effect of green space on residents’ sentiments under normal temperature conditions to compare changes during different periods.

 

Point 3: Sentiment values require more explanation, including how roberta-base-finetuned-jd-binary-chinese corpus, and BERT measure, by indicators of sentiment. Or examples of how specific texts are understood by sentiment measures, past the worst to best measure.

Response 3: 

Thank you for pointing this out. In the method section, we add more information about roberta-base-finetuned-jd-binary-chinese and training data details about JD.com corpus specifics to describe the pre-training more sufficiently. As the comment, we need to present examples of how specific texts are understood by sentiment measures. Therefore, we add an example of the post text and its final sentiment score, also provide the final score range of sentiment based on all Weibo posts.

Details as follows:

(Line 255-268) In this study, the uer/roberta-base-finetuned-jd-binary-chinese pre-trained model was adopted to complete the sentiment analysis of Weibo. RoBERTa (Robustly Optimized BERT Approach) is a pre-trained model provided by the natural language processing (NLP) library transformers developed by Hugging Face. Based on RoBERTa, uer/roberta-base-finetuned-jd-binary-chinese is an improvement increasing training data and performance specifically for Chinese sentiment analysis tasks. It was fine-tuned during large-scale pre-training using the UER-py framework and combined with Chinese datasets from various fields. During the pre-training stage, the model used JD.com review data as training data, which included review content, reviewer nicknames or IDs, review time, review ratings (such as positive, neutral, negative reviews, etc.) and other information. By learning the emotional patterns in JD.com reviews to achieve emotional analysis and judge whether the reviews express positive or negative emotions, this model can be more adaptable to the unique characteristics of Chinese texts and better simulate the Chinese social media context.

(Line 281-289) During the sentiment analysis process, the roberta-base-finetuned-jd-binary-chinese corpus was called to conduct sentiment analysis on the cleaned Weibo texts and obtain the final sentiment score for each Weibo post. The score ranges from 0 to 1, and the closer to 1 the more positive; or, the more negative it is. For example, the score was 0.02 with the comment “大热天在外面找了半个小时的咖啡店，热出一身汗” (I spent half an hour looking for a coffee shop outside in the heat, and worked up a sweat). Finally, 17388 Weibo posts obtained sentiment scores ranging from 0.02 to 0.98 by the model and other 2,000 were selected as the validation set for manual verification, the accuracy rate reaching 87% and indicating that the results are highly reliable.

 

Point 4:

Line 477-495: This limited influence on emotion of green spaces could be further recognised as measured by the methods used. Location of weibo posts and natural language analysis as indoors or outdoors is an example. Is it a limited influence because of awareness, and does it take the general influence of green spaces on temperature in terms of cooling into account. Or does this point make the provision of medical services, building efficiencies more important. Could these questions be addressed in a general discussion, past this section.

Response 4:

Thank you for this thoughtful insight. In the discussion section about green space, we note that supplementing analysis with geotagged post locations could potentially serve as a validation method for distinguishing indoor or outdoor environments. However, this approach suffers from two critical limitations: 1)the currently available geolocation metadata lacks sufficient precision to reliably determine whether posts originate from indoor or outdoor settings; 2) the recorded posting location may not correspond to the actual space of environmental experience due to potential discrepancies between posting behavior and physical presence—a confounding variable in spatial analysis.

Concerning "Does it take the general influence of green spaces on temperature in terms of cooling into account?" It is true that green spaces have an overall effect on urban temperatures in terms of cooling. However, this study focuses on how urban functions affect the sentiment of its residents, and the calculation of the cooling effect of green spaces belongs to research work in the field of urban thermal environments, which is not in line with our topic, and therefore we will conduct another round of analysis on this in future work.

At last, as to whether this point make the provision of medical services, building efficiencies more important. We recognize that this is an important piece of information, so we are adding it to the general discussion.

Details as follows:

(Line 523-535) The low proportion of green space and park during the heat waves also led to negative emotions, but the high proportion of green spaces and parks did not have a very significant improvement in emotions, probably because of the extreme high temperatures during the heat waves, which made people reluctant to go outdoors even when living around green spaces, and also exacerbated the reluctance to go outside due to the electricity restrictions in public areas caused by the heat waves. Besides, in line with the results of the previous analysis, the limited impact of green spaces on sentiment may make the provision of healthcare services and building energy efficiency more significant, discussions that still need more work to be validated.

 

Point 5: This study associates public function with sentiment during a heatwave. The recommendations for improvement are only specific to public function and well known improvements as mitigative for heatwaves. These recommendations are good for planning but there could be more about the function and sentiment measures, and then about whether optimising function for all streets is a proven strategy.

Response 5: 

Thanks for this thoughtful suggestion. The purpose of this study was to explore the relationship between different functions of the urban street and residents' sentiment during a heat wave, and to provide recommendations for improvements in street function based on the results. We agree that more attention needs to be paid to the measurement of functional and mood indicators, as well as to the issue of optimizing all street functions. However, we have already analyzed a large number of links between street functions and sentiments during heat waves, and it is difficult for space reasons to expand on this discussion. Therefore, we have added these constructive comments to the limitations.

Details as follows:

(Line 611-614) Within Chengdu, based on the representative street cases identified in this study, further exploration of the correlation between sentiments and street functions could be conducted, with corresponding strategies implemented on a small scale to monitor changes in residents' sentiments before and after renovations.

 

Point 6: While recommendations do seem realistic and rational for improvement, it could be better to provide example of how often hospital admissions, and medical service is required. It could also be more advanced by recommendation to bring these recommendations together and even consider preventative measures like green spaces, as compared to medical services. Whether green spaces for cooling, elevators etc. can significantly reduce need for medical services, then a measure of function. This measure of function could eventually be measured using sentiment.

Response 6:

Thank you for this helpful comment. However, due to the sensitive nature of medical data and our lack of formal partnerships with healthcare institutions, we were unable to access information such as hospital admission frequencies or medical service demand. We recognize this as a significant potential information and have noted your valuable suggestion in the Limitations section of our manuscript.  

Details as follows:

(Line 641-644) In future research, we will collaborate with community hospitals to obtain medical information to corroborate the findings of this study that the medical function-oriented public category of streets helps promote positive emotions among residents during heat waves.

 

Point 7: Line 502: Include the functional categories of each of the three streets.

Response 7:

Thank you for your careful examination of our manuscript. We sincerely apologize, but despite careful examination of line 502 and its surrounding context, we have been unable to locate the information regarding the three streets mentioned. Maybe it is a wrong line number? Could we humbly request that you provide more specific details or contextual references in your next review round?

 

Point 8: Line 513-514: this is a different point to green spaces. Could building efficiency be a subheading?

Response 8:

Agree with this comment. We have revised the title from "5.3.1 Improvement of green coverage" to "5.3.1 Improvement of green coverage and building efficiency". Besides, the figure is also be adjusted.

Details as follows:

5.3.1. Improvement of green coverage and building efficiency

 

Point 9: Some recommendations including medical services for limited public use streets, diverse UGS by plant selections, and elevators need literature support. They are all well supported and studied, and are good recommendations.

Response 9:

Thank you for pointing this out. We have added some reference and government news to support our recommendations in the Implications section to prove them effective.

The references are as follows.

-Salih, K., & Báthoryné Nagy, I. R. Review of the Role of Urban Green Infrastructure on Climate Resiliency: A Focus on Heat Mitigation Modelling Scenario on the Microclimate and Building Scale. Urban Science 2024, 8(4), 220.

-Guo, F., DONG, J., & GUO, R. A New Evaluation Method for Heat Exposure Risk Integrated With Outdoor Spatial Behaviour. World Architecture 2022, (09), 92-96.

-Government news retrieved from http://sh.people.com.cn/n2/2024/0816/c138654-40946222.html

 

Point 10: Line 578: The study could more clearly explain the functional category associated with sentiments in different streets. More specifically, which sentiments, if there can be more than a worst to best measure. This point supports the comment for methods, and indicators of sentiment.

Response 10: 

Thank you for this thoughtful insight. In accordance with this comment and previous study (Zhu et al., 2024) suggestions, the current method (uer/roberta-base-finetuned-jd-binary-chinese pre-trained model) can only identify whether a sentiment is positive or negative. In the future, we will train the model through supervised learning so that it can assign corresponding scores to specific sentiments. We have added this content to the Limitations and future research section.

The references are as follows.

-Zhu, Y., Wang, J., Yuan, Y., Meng, B., Luo, M., Shi, C., & Ji, H. Spatial heterogeneities of residents' sentiments and their associations with urban functional areas during heat waves–a case study in Beijing. Computational Urban Science 2024, 4(1), 7.

Details are as follows.

(Line 624-627) Second, the deep learning-based sentiment classification model only achieves a binary division of positive and negative sentiments, failing to refine specific sentiment dimensions such as anger and sadness, and the accuracy of sentiment measurement needs to be improved. Future studies can introduce a more refined sentiment classification system to deepen the research on sentiment cognition under high temperature and heat waves [29].

 

Point 11: Briefly explain which function types were of most negative, worst sentiments, commercial and no function. This point could be summarised with recommendations as a sentence or two. The variable influence of green spaces while being recommended might need some addressing in limitations of discussion. A subheading for energy efficient buildings, and further information about this function particularly for balanced recommendation.

Response 11: 

Thank you for your constructive suggestion. Following this suggestion, we have briefly described the correlation between business functioning and sentiment in the conclusions chapter. However, due to the composite nature of street functions, it is difficult to state directly in which function type emotions are worst.

Besides, about the variable influence of green spaces and subheading have been clarified above, please refer the point 4 and 8.

Details are as follows.

(Line 655-658) We found that the street categories of green space and public show a significant promoting role on residents' positive sentiments but industrial and commercial category often correlate with negative sentiments.

 

Point 12: The conclusion could explain and summarise the study, including heat wave patterns for the study year. A reader can then get a comprehensive understanding of the meaning and significance of the study.

Response 12: 

Thank you for this thoughtful insight. Indeed, we should clarify the study year and results like sentiment patterns. Therefore, we have rewritten the conclusion section.

Details are as follows.

(Line 649-660) Based on Sina Weibo social media data, this study explores the spatial distribution of residents' sentiments in five core urban areas of Chengdu by NLP under high-temperature and heat waves in August, 2022 that positive emotions within the Second Ring Road exhibit a higher proportion than peripheral areas, negative sentiments are more gathered in the eastern area. In addition, we analyze its association with different functional categories at the street scale, and reveals which functional category is associated with sentiments in different streets. We found that the street categories of green space and public show a significant promoting role on residents' positive sentiments but industrial and commercial category often correlate with negative sentiments. The combination of residential category and other functional categories has a strong correlation with sentiments, indicating that a reasonable functional combination within residential areas plays a crucial role in promoting residents' positive sentiments.

 

Point 13: The article needs a proof read for grammatical corrections.

examples include:

Global warming is a relevant point but the sentence could start with

‘The impact of heat waves….’

 Line 23: show instead of shown..

Line 102: and instead of is.

Line 109: take instead of taken.

Line 355: delete a ,

line 518: waves. It analyses

Some of the figures need spelling corrections.m

Response 13: 

Thanks for this comment. We have revised and checked the manuscript and ensure no visual, writing and grammar errors. Please refer the revised version for more details.

 

Additional clarifications

In addition to the above comments, all spelling and grammatical errors have been corrected. Once again, we thank you for the time you put in reviewing our manuscript and look forward to meeting your expectations.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

I have no further comments.