A Spatial Information Extraction Method Based on Multi-Modal Social Media Data: A Case Study on Urban Inundation

Round 1

Reviewer 1 Report

The paper proposes a method for extracting spatiotemporal information of urban events from multimodal social media data, such as text, images, and videos. The method consists of three steps: data crawling and preprocessing, coarse-grained spatiotemporal information extraction, and fine-grained spatial information extraction. The paper uses the “July 20th Zhengzhou Heavy Rainfall” incident as a case study to evaluate and analyze the effectiveness of the method. The paper claims that the method can achieve high spatial precision and accuracy at both coarse and fine levels and demonstrates the advantages of multimodal data in enhancing the spatiotemporal information extraction. Here are some suggestions for improvement:

 

 

1.     The paper lacks a clear research question or hypothesis that guides the study. It is not clear what the specific objectives or contributions of the paper are, or how they address a gap in the existing literature.

 

2.     A sufficient literature review or background information on the related work and concepts should be provided. The paper only cites a few studies and does not discuss their strengths and weaknesses or compare them with the proposed method.

 

3.     Lines 37-38 “The extensive nature of information dissemination on social media renders it a rich source of spatiotemporal data.”: References should be citied to support the statements here, for example, a paper titled “Dynamic assessment of PM2. 5 exposure and health risk using remote sensing and geo-spatial big data”.

 

4.     A clear rationale or procedure for selecting the 23 pairs of high-quality image-text data for the fine-grained spatial information extraction step should be provided. The paper does not explain how the quality or relevance of the images is determined or measured. The paper also does not discuss how the sample size or selection may affect the generalizability or validity of the results.

 

 

5.     The paper does not discuss the limitations, challenges, or ethical implications of the proposed method. The paper does not acknowledge any potential sources of error, bias, uncertainty, or noise in the data or the method. The paper also does not consider any ethical issues related to using social media data for spatiotemporal information extraction, such as privacy, consent, ownership, etc.

Minor editing of English language required

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

This paper presents a technological approach for extracting the event‘s spatial and temporal information from social media data. One interesting point is that the authors employ NLP technology to obtain fine-grained information. Overall, the method is convincible according to the results. One major issue is that the introduction section needs more literature review and one-Chinese publication. The length of the paper is too long to read. Major academic contribution and novelty is unclear which should be addressed in the front part.

None

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Social media multimodal data evaluation is important for research, especially for disaster incidences.  I found this research topic important yet, several points required to be clarified.

Abstract.
The "MIST-SMMD" should be written in the long form and you can put an abbreviation near to it.

Introduction.
We know this is not the first study for SM multimodal data evaluation for disaster incidence mapping. Please enhance your introduction with the related studies. Which one is the long form of the model line 109 or line 138?line 122-123 propose a set of strict standardization rules for .. which rules are they? connect the methodology part to define these. 
line 124 idea blurs there please clearly write what are standardization rules for?
line 125 LSGL should be written in long form.

Mehodology.

In general, the part requires a bit of structuring since it is not easy to follow. Figure 1 looks good but it is not well connected with the paragraphs. You might connect each sub-part in steps with the paragraph by giving a number. line 162-163 text classification model can be used or used in this study? Have you considered the fine-grained filtering techniques or not? If not, why?

Table 1 has a blog post with text. The text should be translated into English otherwise most of the people reading this can not understand the concept.

The abbreviations GPE, LOC, and FAC should be given in the long form at first then abbreviations can be used in the text. 

Line 206 why JioNLP is used for parsing, why this is used? Other parsing algorithm examples? While reading having pseudo code or algorithm can be helpful to follow. The methodology starts to blur line 219. Figure 2 should be well connected with the text. Figure 3 has text in Chinese that should be translated into English.

 

3. Experimental Setup

Figure 6 is explaining images well especially the cat photo in the negatives is awesome =)

Figure 7 should be constructed again. Dot maps (a, b) are not readable. The sizes do not readable and seem all the same in (a). Changing the color other than blue for the dots could be good as well (a, b). 

Spacial Precision is it really "spacial"?  or mean spatial? The "spacial" word repeats several times in the text. Please fix it. The size of the image-text data (23 paris) that is used in the study seems questioning. But still, the results are there. 

* There should be additional comment lines in English in the github code if you want to reach out the worldwide researchers.

 

 

 

In some points, English or the way you write seems to blur the idea. Especially in the methodology and experiment part. Some of the parts are not easy to understand, the reader should use imagination to complete the idea. Please make your research read anyone else from other fields or out of the research team and take advice.

Author Response

Please see the attachment.

Author Response File: Author Response.docx