Key Influencing Factors Identification in Complex Systems Based on Heuristic Causal Inference
Round 1
Reviewer 1 Report
The paper highlights the significance of identifying key influencing factors in complex systems for better comprehension of their evolution and for informed decision-making. The paper could be further improved if the following aspects could be considered in a revised version.
. In the future work, authors should add details about how will the deep integration of theories and methods related to complex networks and causal inference would truly result in a deeper understanding of other complex systems?
2. In the figures, meaningful axis titles shall be added which are missing through the paper.
3. It would be better, if section 3.2.1. Global Causal Network Learning could be presented in the form of an algorithm.
4. The section 3.1. Technical Framework needs further details to explain the framework for better understanding of the readers.
5. The abstract highlights heuristic causal inference model which needs further description of how the proposed heuristic causal inference model works, which steps are involved, and its novelty in comparison to existing methods. This would enhance the clarity of the paper.
6. The abstract and paper mentions the use of the FCI algorithm and heuristic causal inference method, it doesn't elaborate on why these specific approaches were chosen or how they contribute to addressing the identified limitations of other methods.
7. Literature shall be updated with latest and relevant literature.
8. The grammar shall be improved throughout the paper.
English shall be improved.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Dear Authors
I congratulate you on the research reported in the manuscript applsci-2587692. In this research, the authors state that in complex systems conditioned by multiple factors, it is of great importance to accurately identify the main influencing factors in order to master the law of evolution and development of the system and obtain suggestions or scientific decision-making schemes. The authors pointed out that at present, the method based on experimental simulation is limited by the difficulty of constructing the system model. However, the method based on decision testing and evaluation laboratory (DEMATEL) involves a wide range of subjects and is greatly influenced by subjective factors. They therefore proposed a new model based on heuristic causal inference. The model uses the FCI algorithm with prior knowledge to learn the global causal network between multiple factors in the complex system. The proposal is robust and brings a new approach to dealing with the issue in question. I believe that as an application, it has value in resolving this issue. Below I will make some considerations and suggestions for improvement:
1) The abstract is the gateway to knowledge of your research. In this sense, it presents a clear objective, the development of the methodology and the authors emphasize the results. I thought it was very good. However, I would suggest informing the reader of the practical and social implications, if any, as well as indicating future research, in order to fill in the gaps that the authors did not address;
2) The work is very well structured. In section 2.2, the authors briefly touched on the uses of DEMATEL. Dematel is part of a set of multi-criteria methods. I suggest that the authors include a paragraph or two in the introduction giving the reader a brief summary of the existing methods and the reasons for choosing DEMATEL. For this task, I suggest they consider the following review of multicriteria methods:https://doi.org/10.3390/electronics11111720 ;
3) Figures 1 and 2 clearly translate the methodology for the reader. Very good. Regarding Figure 4, the quality should be improved, the numbers are too small, I suggest including the figure on a page in landscape format;
4) In section 2, the authors provide a literature review. Very good. However, there is a lack of information on the keywords used to select the articles; the databases in which they were searched, as well as the criteria for inserting and excluding articles in the database for reading. I suggest you include this information.
5) The authors created an algorithm to run the model or used specific software. Please state this in the text. If they have created an algorithm, provide the reader with the link to obtain the scripts;
Happy proofreading!
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
The manuscript authors suggest a model to identify the significant factors in complex systems. The model employs heuristic causal inference and comprises three primary modules: causal network learning, heuristic causal effect calculation, and identification of influential factors. The analysis primarily relies on the number of causal pathways, average length of causal pathways, and degree of contribution to causal pathways. To validate the method, the study utilizes a single experimental dataset called SECOM.
1)The manuscript is based on the DEMATEL method and additional cited methods (line 76). The authors should provide a clearer explanation of the novelty of their work, as the novel contributions mentioned in line 54 are too general. It is recommended that they clarify this point (in Sections 1 Introduction and 6 Conlusions).
2) Line 102. You might reference Fontela's and Gabus's original article or mention that [19-21] are review articles.
3) Lines 157-159. What is the definition of influence degree? (or Lines 250-251: "causal path contribution degree". This is defined on line 304, but the concept has been used several times before that. Please either provide an earlier definition or revise the text accordingly.) Please ensure that all variables and concepts are defined when they are first mentioned in the text. Restructuring of the text may be needed.
4) Other concepts, such as "adjacent local causal network", "Markov blanket", "Markov boundary", "causal skeleton", "orientation of V-structure" and "order ?" may need more detailed explanations.
5) Does the quantity influence degree consider the distance (decay/attenuation of influence) from the target node? There are other methods that take this effect into account. This can be a major deficiency/limitation of the DEMATEL method. Please discuss and compare the method with other methods in the literature.
6) Please extend the Introduction/Discussion sections. Instructions for authors (https://www.mdpi.com/journal/applsci/instructions): "The findings and their implications should be discussed in the broadest context possible and limitations of the work highlighted."
There are many methods, including principal component analysis, factor analysis, and regression analysis. How do these relate to the current work? Are they alternative methods to the proposed method of your work? I recommend revising the Introduction and Discussion sections.
7) Lines 154, 194, 222, 230, 241, 284, 313, 315, 319, 325. Misplacement of ".".
8) Lines 234, 235, 237, 243, 268, 283, 287, 293, 294, 303, 307, 308, 318, 327, 333, 334. Misplacement of ",".
9) All appreciations are not explained, for example, FCI in line 15 (Fast Casual Inference?).
10) Line 214. "...included in itself"?
11) "." in the formulas missing. Also lines 329 and 333 and so on.
12) Line 326. "in's?k-order". Please check and correct.
13) Please check and correct the format of figure captions (the use of "." and capital letters).
14) In Figure 5, there seems to be an excessive number of decimals. Additionally, could you kindly verify the bolding in the same figure?
15) Has the SECOM dataset been analysed with other methods in the literature? If analyses have been performed, please compare the results with your study. In addition, it is recommended that you include a brief summary of the results obtained from other relevant methods (Section 2 Related Work) with other datasets.
16) Line 343. Please check "funded". Do you mean "found"?
17) Please provide a better error analysis of the results in Section 4 (50% of the columns are deleted from the data, lines 355 and 356). In my opinion, Sections 4.6 and 5 only partly answer the question. It would be beneficial to extend the research by using additional datasets or conducting a deeper analysis of the SECOM dataset. For instance, analyzing the SECOM dataset with different parameter values of alpha and beta, perhaps less than 1 (0.5?), would be useful. If the results are similar to those obtained with alpha=1, and/or beta=1, a shorter description in the text would suffice, or they could be presented in an appendix.
18) Table 1. "Average neighborhood nodes" or "Average number of neighborhood nodes"? "Characteristic path length" is this the same as the pathway length in the main text. Please check the terms and clarify if needed. What is "Network density"?
19) How are the results normalised (lines 357, 393)?
20) Could you please provide more information about what is meant by "through a certain cause" in line 403?
21) The term "correctness" in line 414 may be too strong. Perhaps it could be rephrased.
Proofreading and minor corrections are needed.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
I have no more comments.
Reviewer 3 Report
The authors have responded to my questions. Thank you very much!
-
