Classification of Seismaesthesia Information and Seismic Intensity Assessment by Multi-Model Coupling

Round 1

Reviewer 1 Report

The manuscript presents a novel multi-model coupled to assess seismic intensity based on BERT-CNN model to classify the seismic level of microblog disaster data, the seismaesthesia intensity attenuation model to solve the problem of subjectivity of microblog data and the fitting elliptic interpolation method to solve the problem of insufficient data in some disaster areas in order to improve the accuracy and universality of the seismic intensity assessment model.

The manuscript is well written and organized, but the following aspects should be corrected or introduced and clarified:

Point 1: line 16, the double hyphens should be replaced with a single hyphen.

Point 2: line 39, please delete one of the two commas.

Point 3: line 47, please use a correct format for citation in the test.

Point 4: section 2.1, the authors should better explain in which way it is possible to extract and to systematize the different type of data, this phase of procedure is presented with very poor level of detal

Author Response

Dear reviewer:
Firstly, thank you very much for your comments which will make positive contributions to the submitted paper and improve it. We have carefully revised the manuscript in the light of your comments and the detailed corrections and responses are listed below point by point. All the changes highlight by red color in our revised manuscript.Please see the attachment for details，thank.

Author Response File: Author Response.doc

Reviewer 2 Report

Kindly revise the abstract (minor) and check the references after the final assessment.

Author Response

Dear reviewer
Firstly, thank you very much for your comments which will make positive contributions to the submitted paper and improve it. We have carefully revised the manuscript in the light of your comments and the detailed corrections and responses are listed below point by point. All the changes highlight by red color in our revised manuscript.Please see the attachment for details，thank.

Author Response File: Author Response.doc

Reviewer 3 Report

This study attempts to establish an evaluation method for earthquake influence field assessment by using subjective microblog disaster data. In general, the method is suitable and clear, and the results have certain reference significance for the determination of earthquake affected areas. However, there are some problems in this study which need be further revised and improved.

 1. Influence of uneven distribution of subjective microblog data on results. Although the author emphasized the multi-model approach, the basic microblog data inevitably has the problem of uneven distribution. Is it appropriate to use inverse distance interpolation to process subjective perception data?

 2.  The subjective feelings of different people are not the same. People's feelings of the same objective influence of an event vary. How to balance this effect?

 3. Figure 4a is an obvious problem. The author also pointed out these unreasonable places in the paper. For example, the feedback of the masses is very intensive, and there are all kinds of intensity of feelings in the areas far from the epicenter. Then, just because of the distance from the epicenter, the data should not be considered? In contrast, near the epicenter, there are very few data points, but based on such a small number of points, the seismic impact field is assessed?

How should this problem be considered in different earthquake cases? In fact, the more earthquake-stricken areas are, the less such perception data can be obtained. On the contrary, more such data can be obtained in the less affected areas, because the local people have the time and conditions to release the perception information.

 4. Therefore, in view of the characteristics of microblog data, I personally believe that it may not be mature to analyze the earthquake impact field by using such disaster feeling information as microblog. It will not be solved by only adding some methods such as machine learning and so on. The main limitation is data defects.

 5. In addition, abbreviations that appear for the first time need to be marked with the full text. At present, some scholars have tried to use mobile phone signaling data in the study of disaster affect. It may be inspiration for the authors to understand these problems.

Author Response

Dear reviewer:
  Firstly, thank you very much for your comments which will make positive contributions to the submitted paper and improve it. We have carefully revised the manuscript in the light of your comments and the detailed corrections and responses are listed below point by point. All the changes highlight by red color in our revised manuscript. Please see the attachment for details, thank.

Author Response File: Author Response.pdf