Identifying Critical Success Factors of an Emergency Information Response System Based on the Similar-DEMATEL Method

Abstract

An emergency information response system (EIRS) is a system that utilizes various intelligence technologies to effectively handle various emergencies and provide decision support for decision-makers. As critical success factors (CSFs) in EIRS play a vital role in emergency management, it is necessary to study the CSFs of EIRS. Most previous studies applied the Decision Experiment and Decision-Making Trial and Evaluation Laboratory (DEMATEL) method with complete evaluation information to identify CSFs. Due to the complexity of the decision-making environment when identifying CSFs of EIRS, decision-makers sometimes cannot provide complete evaluation information during the decision-making process. To fill this gap, this paper provided a Similar-DEMATEL method to impute the missing values and identify CSFs of EIRS, which may avoid the dilemma of decision distortion and make decision-making results more accurate. It is found that the factors of Information mining capability, Equipment support capability, Monitoring and early warning capability, and Organization participation capability are the CSFs in EIRS. Our proposed method differs from previous research, such as the mean imputation method, to impute the missing values. We compared the differences between the proposed method and the mean imputation method and gave the advantages of the proposed method. Our method focuses more on uncertain decision-making environments, which is conducive to improving the efficiency of EIRS in emergency management, and therefore it is more widely adopted.

1. Introduction

The COVID-19 epidemic is a major international public emergency. Although the epidemic has passed away in China, its harm to the economy and society far exceeded people’s expectations. The emergency information response system (EIRS) is a system that can quickly and efficiently collect and analyze information by using various intelligence technologies, helping organizations make quick and effective responses to emergencies, which played a vital role in the control of emergencies in China. For example, it can accurately monitor the track of affected people in China through big data technology. From the content contained in EIRS, it can be seen that in urban construction and sustainable development, EIRS also played an important role in quickly restoring the original functions of the city when the city faced various disasters.

Many scholars have conducted research on EIRS, and their views are as follows: Excellent EIRS can achieve a series of goals in emergency management by exploring the essence of emergency management [1]. Therefore, when facing emergency problems, the government needs to have a highly unified information system and a high degree of organizational flexibility. Therefore, in order to better promote the development of EIRS, many countries have conducted in-depth research on key technologies such as information integration mechanisms and information security of emergency platforms [2,3,4,5], improving their financial support capabilities, including personnel costs, equipment costs, management costs, equipment consumption, and maintenance costs, thereby greatly improving their security capabilities when facing emergency problems [6]. The widespread use of devices such as mobile phones and robots has made it easier to collect more and more accurate emergency information [7,8]. The full application of IT applications has also improved the ability to mine information [9], further unleashed organizational flexibility [10,11], and optimized the government’s emergency management decision-making ability.

Prior research laid the foundation for improving the capability of EIRS. However, EIRS sometimes cannot produce a desirable efficiency, as decision-makers may not be able to effectively utilize EIRS in certain situations, such as excessive information mining or misuse of information security [4,12,13]. Thus, a scientific way is needed to identify critical success factors (CSFs) and to clarify the relationships among factors. Decision-makers should pay more attention to the CSFs. Less attention should be paid to those non-CSFs to avoid negative effects on emergency management.

Previous studies used methods such as unstructured interviews [7] and questionnaires [14] to explore the potential and important factors, which did not adequately identify the CSFs of EIRS because of decision-makers’ (DMs) incomplete knowledge structure and different educational backgrounds. Therefore, it is critical to apply an appropriate decision-making method to identify the CSFs of EIRS. The Decision Experiment and Decision-Making Trial and Evaluation Laboratory (DEMATEL) method has been used to identify CSFs in many studies [15]. Like several methods to enhance the efficiency of emergency management [16,17,18,19,20], the DEMATEL method needs experts to evaluate the interrelation among factors. However, DMs may provide hybrid evaluation values, and the evaluation values may contain some missing values. Unfortunately, scholars have not conducted in-depth research on this kind of DEMATEL method in these situations.

To address these limitations, a novel Similar-DEMATEL method with hybrid assessment scales of crisp value and interval value was proposed to identify CSFs of EIRS in the decision environments containing missing values.

The main contributions of this study are as follows:

(1)

A total of 10 factors of EIRS, which are utilized to explore the causal relationships between factors, are summarized from existing literature.

(2)

A matrix with missing values for DEMATEL was constructed and compensates for missing values through similarity degree.

(3)

CSFs of EIRS were identified by the novel method. Four factors were identified as CSFs, and six factors were identified as non-CSFs. Based on the results, suggestions were proposed in the discussion that would enhance the efficiency of governmental emergency management.

The remainder of this paper is organized as follows: Section 2 reviews the previous related work. Section 3 introduces the preliminaries, including factors of EIRS and symbol descriptions. Section 4 presents the proposed Similar-DEMATEL method. Section 5 illustrates the result of decision-making. Section 6 provides decision-making suggestions. The conclusions are provided in the concluding section (Section 7).

2. Literature Review

There are two main types of literature related to this article: one is fuzzy DEMATEL, and the other is missing values.

2.1. Hybrid DEMATEL

In the typical multi-criteria decision-making (MCDM) method, the DMs preferences were measured using single assessments such as crisp values or fuzzy membership degrees. However, single assessments such as crisp values or interval values sometimes cannot express human thinking [21,22]. The DMs preferences sometimes need to be measured using two or more kinds of assessment scales owing to complex human judgment under uncertain circumstances [23]. For instance, Guo et al. (2012) developed an attitudinal-based method in which the attribute expressed in different assessments in hybrid MCDM [24]. Büyüközkan and Tüfekçi (2021) proposed a new hybrid evaluation model for the blockchain based on multiple assessments in a fuzzy environment [25]. Liu et al. (2021) provided a new group of decision-making methods incorporating regret theory with the assessment of real numbers, interval numbers, and linguistic Z-numbers, which are used to evaluate cloud service [26].

In the 1970s, Fontela and Gabus (1974) proposed the DEMATEL method, which could uncover CSFs by exploring the relationship among factors [15]. The DEMATEL method has been widely used in knowledge management, emergency decision-making, and other fields [27,28,29]. In recent years, the extended application of DEMATEL has become increasingly widespread. One extension of DEMATEL is to apply DEMATEL to other methods to propose a hybrid MADM framework. Tadić (2014) integrated DEMATEL, ANP, and VIKOR for selecting the CL concept, which is successfully performed in this paper for the City of Belgrade [30]. Du (2013) developed the Kano-DEMATEL-TOPSIS method and applied it to logistics service supply chain (LSSC) management. The hybrid method can better solve the problem of benefit distribution, which comprises segmented task evaluation and individual risk compensation in LSSC [31]. Like many other typical MCDM methods, the typical DEMATEL needs DMs to provide assessments against criteria using a single assessment scale. Thus, the other extension of DEMATEL is to modify the assessment scale. One way to modify the assessment scale is to expand it into a diverse set of language terms such as Pythagorean, triangular fuzzy numbers, intuitionistic fuzzy numbers, etc. [32,33,34,35]. The other way to modify the assessment scale is to expand the scope of assessment scales. For example, in the literature [36], the scope of assessment scales can be any natural number.

Although DEMATEL is widely used, the application of this effective structural modeling tool to evaluate EIRS needs to be explored [37]. Considering a single assessment scale cannot express human thinking, it needs two or more kinds of assessment scales for the DEMATEL method to identify the CSFs of EIRS. However, the DEMATEL method using hybrid evaluation assessments is still in its early stages. Jin et al. (2020) put forward a hybrid DEMATEL method with the assessment of a real number, an interval number, and a triangular fuzzy number to identify the CSFs of emergency supply support capacity [38]. Jin and Zhang (2023) provided an IFS-IVIFS-DEMATEL method with the assessment of IFS and IVIFS to identify the CSFs [39]. These studies using hybrid evaluation assessments have played a good role in identifying CSFs. Therefore, a Hybrid DEMATEL with two kinds of assessment scales will be applied to identify the CSFs of EIRS.

2.2. Missing Values Imputation

The evaluation information may contain some missing values due to DMs incomplete knowledge structure and different educational backgrounds. In general, the simplest way to solve this problem is to delete the evaluation information and datasets with missing values. However, if the missing values are very important for DMs, deleting the evaluation information and dataset may result in an incomplete reflection of the real problem, which could distort the DMs thoughts [40]. The imputation method is then utilized to obtain the missing values [41]. The imputation method generally contains two groups: the statistical imputation method and the machine learning-based imputation method [42].

The simplest statistical imputation method is mean and mode imputation [43,44]. For mean imputation, the missing value is filled in by the average value of all the observed data [45,46]. For the mode method, the missing value is filled in by the frequent value of the attribute [47]. The machine learning method is another typical imputation method, which includes K nearest neighbor imputation (KNN), class center-based missing values imputation (CCMVI), fuzzy c-mean clustering method, et al. The KNN method is a typical machine learning method in which the missing values are imputed by the values calculated from the nearest observed data of k [48]. The CCMVI method means that the missing values are imputed by the values of distance based on measuring the class center of each class [43]. The fuzzy c-mean clustering method means that the missing values are imputed by the values using the selected shorter interval based on the cluster membership value [44]. In recent years, new machine learning methods have been proposed by a growing number of scholars and have been applied in increasing fields [41]. For example, Ramezani et al. (2018) proposed a hybrid classifier method named logistic adaptive network-based fuzzy inference system (LANFIS). It is a combination of logistic regression and an adaptive network-based fuzzy inference system, which is also applied in diagnosing diabetic diseases [41]. Wang et al. (2019) proposed a method of processing missing values based on the combination of denoising autoencoders (DAE) and generative adversarial networks (GAN), aiming at noise interference in industrial scenes [49]. Ali et al. (2022) proposed a method of fuzzy k-top matching value (FKTM) for missing value imputation that was compared with multivariate imputation by chained equations (MICE) utilizing a support vector machine [50]. Nevertheless, there are few studies about the missing values of assessment for the DEMATEL method.

3. Preliminaries

To better lay the foundation for identifying the CSFs of EIRS, we must obtain the influencing factors of EIRS and further provide some symbols.

3.1. Factors of EIRS

In order to identify the CSFs of EIRS, it is reasonable to extract some factors from prior literature. According to the situation of EIRS in China, some factors, such as safe management [6] and forecasting techniques [51], are integrated. The other factors are selected, such as the information fund support capability [6] and the information coordination capability [11]. After this, the opinions of specialists are sought using a questionnaire survey. Then, factors of EIRS are obtained.

Table 1. Factors of EIRS.

Factors	Description
F1, Information Collection Capability	It uses a relevant approach to collect massive amounts of information scattered to ensure the comprehensiveness, accuracy, and timeliness of the information source [8].
F2, Information Integrating Capability	It integrates the collected information into the EIRS, which can subsequently provide services for information mining and decision-making coordination [52,53].
F3, Information Plan Capability	The scenarios, types, and levels of various emergencies are different. An effective Information Plan Capability can have basic support capabilities for different business plans [51,54,55].
F4, Information Coordination Capability	It establishes the decision-making and coordination mechanism of EIRS, which can effectively respond to emergencies and reduce the harm caused by emergencies [11].
F5, Fund Support Capability	Talents and technologies in emergencies need sufficient financial support to maximize their effectiveness. Investing funds in EIRS can effectively reduce the harm and loss caused by emergencies [6].
F6, Information Mining Capability	EIRS needs to comprehensively use a series of information technologies to mine and utilize various information resources to promote the improvement of emergency information response capability [23,56].
F7, Equipment Support Capability	It refers to the capability to develop, share, integrate, and mine heterogeneous data using appropriate equipment [2,57].
F8, Monitoring and Early Warning Capability	EIRS needs to monitor trend data of epidemic situation and development in real time and predict the change trend serving for event decision-making [2,51].
F9, Information Security Capability	It refers to the generation, transmission, and storage of EIRS information to ensure the reliability and security of data [4].
F10, Organizational Participation Capability	It refers to the ability of relevant departments to coordinate their participation in EIRS [58,59].

lists the factors in EIRS.

3.2. Symbols Description

Human assessments for interrelationship comparisons among factors are generally given by crisp values. However, assessments with preferences are often vague and difficult to estimate with crisp values owing to the complex surroundings. In this study, the interrelationship comparisons among factors are given in crisp values or interval values.

Let $H=(H_1,H_2,\dots ,H_n)$ be a set of factors of EIRS, suppose that $E=(E_1,E_2,\dots ,E_K)$ is a set of experts. $c_{ij}^k$ indicates that it is the assessment of $H_i$ against $H_j$ by the $k\mathrm{th}$ expert. $s_{ij}^k$ indicates that the $k\mathrm{th}$ expert expresses his/her preferences for interrelationship comparison by a crisp number, $1\le k\le K$. $[x_{ij}^{kl},x_{ij}^{ku}]$ indicates the $k\mathrm{th}$ expert expresses his/her preferences for interrelationship comparison by interval numbers. $q_{ij}^k$ indicates that the $k\mathrm{th}$ expert has not expressed his/her preferences for interrelationship comparisons, and it is a missing value. Therefore, the $k\mathrm{th}$ expert constructs a hybrid initial direct correlation matrix (HIDCM). that is, ${\mathrm{C}}^k={[c_{ij}^k]}_{n\times n}=[x_{ij}^{kl},x_{ij}^{ku}]\cup [s_{ij}^k]\cup [q_{ij}^k]$.

4. The Proposed Method

Abdullah (2022) argued that the accuracy and stability of data predictions can be effectively improved by using historical scenarios in similar situations [60]. In this paper, we suppose that under similar situations, the experts who hold the same idea have an identical pairwise comparison of two factors. Based on this point, the following decision-making process is presented:

Step 1: $K$ experts from different fields were invited to evaluate interrelationship comparisons among factors. Each expert ($E^k,k=1,2,\dots ,K$) constructs a HIDCM.

Let $S{D}^{ki}$ be a vector of assessments of comparison for $H_i$ against $H_j$ for expert $E^k$ in HIDCM. That is to say, $S{D}^{ki}=[c_{i1}^k,c_{i2}^k,\dots ,c_{in}^k],k=1,2,\dots ,K,i=1,2,\dots ,n$, $c_{ij}^k$ is an evaluation value of HIDCM by $E^k$. We called $S{D}^{ki}$ a KHCV (correlation vector of expert $E^k$ for factor $H_i$). For instance, the correlation vector of $E_1$ for $H_1$ is $S{D}^{11}=[c_{11}^1,c_{12}^1,\dots ,c_{1n}^1]$. Thus, $K$ vectors about KHCV were formed for $H_i$ by all experts. Let $S{D}^{pm}$ be KHCV with missing values for $H_m$ by $E^P$. We called $S{D}^{pm}$ a MVV. Assume that a HIDCM for $E^P$ has $t{c}^p$ MVV.

For $H_i$, we constructed a matrix $C{C}^i={[S{D}^{1i},S{D}^{2i},\dots ,S{D}^{Ki}]}^{\mathrm{T}}$, where $T$ is transposition of the matrix. We called $C{C}^i$ a HVM (virtual matrix of $H_i$). For example, the HVM of a factor $H_1$ is $C{C}^1={[S{D}^{11},S{D}^{21},\dots ,S{D}^{K1}]}^{\mathrm{T}}={[[c_{11}^1,c_{12}^1,\dots ,c_{1n}^1],[c_{11}^2,c_{12}^2,\dots ,c_{1n}^2],\dots ,[c_{11}^K,c_{12}^K,\dots ,c_{1n}^K]]}^{\mathrm{T}}$. Therefore, there are n HVMs in our decision-making situation. Assume that HVM of $H_i$ contains $S^i$ MVV. Obviously, the sum of the MVV of HIDCM by all experts equals the sum of the MVV of all HVMs, that is to say, ${\displaystyle \sum _{i=1}^n{S}^i}={\displaystyle \sum _{k=1}^{K}t{c}^k}$.

Step 2: Convert the crisp values to interval values and remove the MVV. The crisp values in HIDCM ($C^k,k=1,2,\dots ,K$) are converted into interval values according to the literature [61].

The formula is,

$$y_{ij}^{kl}=s_{ij}^k, y_{ij}^{ku}=s_{ij}^k, i=1\dots n, j=1\dots n$$

(1)

Thus, the hybrid initial direct correlation matrix (HIDCM) becomes a new matrix ${\mathrm{C}}^k={[c_{ij}^k]}_{n\times n}=[x_{ij}^{kl},x_{ij}^{ku}]\cup [y_{ij}^{kl},y_{ij}^{ku}]\cup [q_{ij}^k]$. Therefore, $C{C}^i$ of $H_i$ is composed of interval numbers in addition to MVV.

Then, we delete the $S{D}^{pm}$ from $C{C}^i$ of $H_i$. The remaining HVM of $H_i$ is $CC{S}^i={[cc{s}_{kj}^i]}_{(K-S^i)\times n}={[cc{s}_{kj}^{il},cc{s}_{kj}^{iu}]}_{(K-S^i)\times n}, i=1,2,\dots ,n$.

Step 3: Introduce the similarity degree of literature [60,62], the similarity degree of $H_i$ for expert $E^k$ against expert $E^r$ is,

$$si{m}_i^{kr}=\frac{1}{n}{\displaystyle \sum _{j=1}^n\frac{cc{s}_{kj}^{il}cc{s}_{rj}^{il}+cc{s}_{kj}^{iu}cc{s}_{rj}^{iu}}{\sqrt{{(cc{s}_{kj}^{il})}^2+{(cc{s}_{kj}^{iu})}^2}\sqrt{{(cc{s}_{rj}^{il})}^2+{(cc{s}_{rj}^{iu})}^2}}}, k=1,2,\dots ,K-S^i, r=1,2,\dots ,K-S^i$$

(2)

Since the interrelation comparison of chosen factors on themselves is 0, namely, $cc{s}_{kj}^{il}=c{s}_{kj}^{iu}=cc{s}_{rj}^{il}=cc{s}_{rj}^{iu}=0$.

Step 4: The comprehensive similarity degree of $H_i$ for expert $E^k$ is,

$$si{m}_i^k=({\displaystyle \sum _{r=1}^{K-S^i}si{m}_i^{kr})/(K-S^i)}$$

(3)

Similarly, the comprehensive similarity degree of $H_i$ for expert $E^q$ is,

$$si{m}_i^q=({\displaystyle \sum _{q=1}^{K-S^i}si{m}_i^{\hspace{0.33em}qr}})/(K-S^i)$$

(4)

Therefore, comprehensive similarity sequences $si{m}_i^1,si{m}_i^2,\dots ,si{m}_i^k,\dots ,si{m}_i^q\dots ,si{m}_i^{K-S^i}$ of factors $H_i$ are constructed.

Step 5: Similarly, comprehensive similarity sequences of other factors are constructed. Because the remaining HVM of $H_i$ is constructed by the KHCV of the remaining experts, of $H_i$. Thus, the corresponding position of the comprehensive similarity sequence of $H_i$ is supplemented with 0. In this way, the synthesis matrix (${\mathrm{CZ}=[\mathrm{cz}}_{ik}{]}_{n\times K}$) is constructed, where, $i$ indicates the factors, $i=1,2,\dots ,\mathrm{n}$, $k$ indicate the experts, $k=1,2,\dots ,\mathrm{K}$.

Step 6: Calculate the average preference value of each expert.

$$y{s}^k={\displaystyle \sum _{i=1}^{n}si{m}_i^k}/(K-t{c}^k)$$

(5)

Step 7: To calculate the expert proximity value, the formula is $\left|y{s}^k-y{s}^p\right|$. If the expert proximity value between $E^k$ and $E^p$ has the smallest deviation in all values, we thought the two experts had the same idea in this situation. Then, the missing values of $E^p$ are imputed by an assessment of $E^k$ in the same position. The formula is $[q_{ij}^P]={[y{s}_{ij}^{kl},y{s}_{ij}^{ku}]}_{n\times n}$.

Step 8: Therefore, a new direct incidence matrix $C^k={[c_{ij}^k]}_{n\times n}$ is constructed.

$$C^k={[c_{ij}^k]}_{n\times n}={[x_{ij}^{kl},x_{ij}^{ku}]}_{n\times n}\cup {[y_{ij}^{kl},y_{ij}^{ku}]}_{n\times n}\cup {[y{s}_{ij}^{kl},y{s}_{ij}^{ku}]}_{n\times n}={[c_{ij}^{kl},c_{ij}^{ku}]}_{n\times n}$$

(6)

Step 9: Weight the new direct incidence matrix according to the number of experts.

$$NQ={[n{q}_{ij}]}_{n\times n}={[n{q}_{ij}^l,n{q}_{ij}^u]}_{n\times n}=[({\displaystyle \sum _{k=1}^K{c}_{ij}^{kl}})/K,{({\displaystyle \sum _{k=1}^K{c}_{ij}^{ku})}/K]}_{n\times n}$$

(7)

Normalize the weighted direct incidence matrix, that is,

$$\tilde{G}={[g_{ij}^l,g_{ij}^u]}_{n\times n}=[n{q}_{ij}^l,n{q}_{ij}^u]/\underset{1\le j\le n}{\max }{\displaystyle \sum _{i=1}^n\sqrt{[{(n{q}_{ij}^l)}^2+{(n{q}_{ij}^u)}^2]}}$$

(8)

Step 10: Calculate the comprehensive impact incidence matrix $\tilde{T}$, where “$I$“ is the unit matrix.

$$\tilde{T}=\underset{l\to \infty }{\lim }(\tilde{G}+{\tilde{G}}^2+\cdots +{\tilde{G}}^l)=\tilde{G}{(I-\tilde{G})}^{-1}$$

(9)

Obtain a degree of influential impact $\tilde{R}$ and a degree of influenced impact $\tilde{D}$.

$$\tilde{R}={\displaystyle \sum _{i=1}^n\tilde{T}}={\displaystyle \sum _{i=1}^n[t_{ij}^L,t_{ij}^U]}=[{\displaystyle \sum _{i=1}^n{t}_{ij}^L,{\displaystyle \sum _{i=1}^n{t}_{ij}^U]}}=[R^L,R^U]$$

(10)

$$\tilde{D}={\displaystyle \sum _{j=1}^n\tilde{T}}={\displaystyle \sum _{j=1}^n[t_{ij}^L,t_{ij}^U]}=[{\displaystyle \sum _{j=1}^n{t}_{ij}^L,{\displaystyle \sum _{j=1}^n{t}_{ij}^U]}}=[D^L,D^U]$$

(11)

Step 11: Obtain a probability matrix of $\tilde{R}$. we use the ranking method of interval numbers proposed by [63]. The formula is,

$$R_{ij}=\frac{\max \left[0,l(Q_i)+l(Q_j)-\max (0,R_j^U-R_i^L)\right]}{l(Q_i)+l(Q_j)}$$

(12)

where

$$l(Q_i)=R_i^{+}-R_i^{-}, l(Q_j)=R_j^{+}-R_j^{-},$$

(13)

The formula for the probability matrix of $\tilde{D}$ is,

$$D_{ij}=\frac{\max \left[0,l(Q_i)+l(Q_j)-\max (0,D_j^U-D_i^L)\right]}{l(Q_i)+l(Q_j)}$$

(14)

where

$$l(Q_i)=D_i^{+}-D_i^{-}, l(Q_j)=D_j^{+}-D_j^{-},$$

(15)

Calculate the crisp values of the degree of influence. The formula is,

$$R_i={\displaystyle \sum _{k=1}^n{R}_{ik}},i=1,2,\cdots n$$

(16)

Calculate the crisp values of the degree of influence. The formula is,

$$D_i={\displaystyle \sum _{k=1}^n{D}_{ik}},i=1,2,\cdots n$$

(17)

Step 12: Calculate the central degree. The formula is,

$$P_i=R_i+D_i$$

(18)

Calculate the causal degree. The formula is,

$$H_i=R_i-D_i$$

(19)

Factors can be divided into two categories by causal degree. A factor greater than or equal to zero is considered the cause factor. A factor less than zero is regarded as the effect factor. The cause factor can influence other factors and is usually considered a CSF when its central degree does not rank too low, while the effect factor is passive; therefore, it is not easily considered a CSF.

The decision flowchart is as follows (Figure 1).

5. The Result of Decision-Making

Firstly, 11 experts in related fields (i.e., information, emergency management) were invited to conduct a questionnaire survey on the influential factors of EIRS. Secondly, in order to meet the expectations of experts to a greater extent, experts were allowed to give crisp values or interval values to evaluate the interrelationship among factors. Dytczak&Ginda (2013) considered that the assessment scale can be N scales (N is a natural number) [36]. The assessment from 0 to 9 (“None”, “Very weak”, “Relatively weak”, “Somewhat weak”, “Slightly weak”, “Moderate”, “Slightly strong”, “Somewhat strong”, “Relatively strong” and “Very strong”) in this evaluation is applied to describe the interrelationship. In addition, the interrelationship comparisons among factors by interval values are in [0, 9]. Thirdly, if an expert cannot give an assessment of the interrelationship comparisons among certain factors, he/she may not give the evaluation value. We asked each expert not to exceed two missing values during the evaluation process.

5.1. Step of Decision-Making

Step 1: Consider the limited space; the hybrid initial direct correlation matrix of Expert 1 is shown in

Table 2. Hybrid initial direct correlation matrix for Expert 1.

	F1	F2	F3	F4	F5	F6	F7	F8	F9	F10
F1	0	5	3	2	6	3	3	3	3	[3, 5]
F2	[2, 5]	0	[1, 3]	3	■	1	6	[2, 4]	3	7
F3	3	4	0	6	5	7	[2, 5]	9	[2, 5]	4
F4	2	[2, 4]	6	0	3	3	[6, 9]	4	2	[2, 3]
F5	3	[3, 6]	3	[1, 3]	0	6	4	5	4	5
F6	2	■	5	3	[3, 6]	0	4	6	6	3
F7	[3, 7]	2	[6, 9]	6	5	[8, 9]	0	2	2	[3, 6]
F8	6	3	4	6	7	3	[2, 3]	0	5	8
F9	5	6	[1, 4]	[3, 5]	3	2	4	[6, 9]	0	5
F10	6	[3, 4]	3	5	4	9	5	6	[3, 6]	0

Note: ■ indicates the missing values.

.

For convenience, we only listed the virtual matrix of factor $F_1$, as shown in

Table 3. Virtual matrix of factor F1.

	F1	F2	F3	F4	F5	F6	F7	F8	F9	F10
F1 of Expert 1	0	5	3	2	6	3	3	3	3	[3, 5]
F1 of Expert 2	0	5	4	2	6	3	[3, 5]	■	2	2
F1 of Expert 3	0	5	[2, 3]	2	6	4	4	3	[3, 5]	3
F1 of Expert 4	0	[6, 8]	2	[1, 3]	[6, 9]	3	[1, 4]	4	2	[2, 4]
F1 of Expert 5	0	5	3	2	4	■	5	[3, 5]	2	[3, 5]
F1 of Expert 6	0	5	2	1	5	3	[1, 4]	1	4	2
F1 of Expert 7	0	5	4	1	5	3	2	[3, 6]	4	2
F1 of Expert 8	0	5	3	■	7	4	5	2	3	3
F1 of Expert 9	0	4	[2, 4]	3	6	3	3	3	2	2
F1 of Expert 10	0	[5, 6]	[3, 4]	3	6	[2, 4]	4	3	[2, 3]	[1, 5]
F1 of Expert 11	0	5	5	■	6	3	4	[3, 5]	3	3

Note: ■ indicates the missing values.

.

Step 2: It can be seen that the HVM of factor F1 by experts 2, 5, 8, and 11 has missing values from Table 3. We deleted these MVVs. Then, we converted the crisp values into interval values and obtained the interval virtual matrix of factor F1. They are shown in

Table 4. Remaining HVM of factor F1.

	F1	F2	F3	F4	F5	F6	F7	F8	F9	F10
F1 of Expert 1	[0, 0]	[5, 5]	[3, 3]	[2, 2]	[6, 6]	[3, 3]	[3, 3]	[3, 3]	[3, 3]	[3, 5]
F1 of Expert 3	[0, 0]	[5, 5]	[2, 3]	[2, 2]	[6, 6]	[4, 4]	[4, 4]	[3, 3]	[3, 5]	[3, 3]
F1 of Expert 4	[0, 0]	[6, 8]	[2, 2]	[1, 3]	[6, 9]	[3, 3]	[1, 4]	[4, 4]	[2, 2]	[2, 4]
F1 of Expert 6	[0, 0]	[5, 5]	[2, 2]	[1, 1]	[5, 5]	[3, 3]	[1, 4]	[1, 1]	[4, 4]	[2, 2]
F1 of Expert 7	[0, 0]	[5, 5]	[4, 4]	[1, 1]	[5, 5]	[3, 3]	[2, 2]	[3, 6]	[4, 4]	[2, 2]
F1 of Expert 9	[0, 0]	[4, 4]	[2, 4]	[3, 3]	[6, 6]	[3, 3]	[3, 3]	[3, 3]	[2, 2]	[2, 2]
F1 of Expert 10	[0, 0]	[5, 6]	[3, 4]	[3, 3]	[6, 6]	[2, 4]	[4, 4]	[3, 3]	[2, 3]	[1, 5]

.

Step 3: Calculate the comprehensive similarity degree of F1, as shown in

Table 5. Comprehensive similarity degree of F1.

Factor F1	Expert 1	Expert 3	Expert 4	Expert 6	Expert 7	Expert 9	Expert 10
Expert 1	0	8.8409	8.8039	8.0394	8.3994	8.6443	9.1002
Expert 3	8.8409	0	8.5035	8.1043	8.3093	8.6972	8.8412
Expert 4	8.8039	8.5035	0	9.1214	8.3243	8.5493	9.075
Expert 6	8.0394	8.1043	9.1214	0	8.2679	7.8566	7.7947
Expert 7	8.3994	8.3093	8.3243	8.2679	0	8.1797	8.2076
Expert 9	8.6443	8.6972	8.5493	7.8566	8.1797	0	8.9672
Expert 10	9.1002	8.8412	9.075	7.7947	8.2076	8.9672	0

.

Step 4: Obtain the comprehensive similarity sequences of each expert for F1, as shown in

Table 6. Comprehensive similarity sequences of each expert for F1.

	Expert 1	Expert 3	Expert 4	Expert 6	Expert 7	Expert 9	Expert 10
F1	7.404	7.328	7.4825	7.0263	7.0983	7.2706	7.4265

.

Step 5: Similarity: we obtain the comprehensive similarity sequences of every expert for F2, F3, F4, F5, F6, F7, F8, F9, and F10. We supplemented with 0 in the corresponding vacant position. The synthesis matrix is shown in

Table 7. Synthesis matrix.

	Expert 1	Expert 2	Expert 3	Expert 4	Expert 5	Expert 6	Expert 7	Expert 8	Expert 9	Expert 10	Expert 11
F1	7.404	0	7.328	7.4825	0	7.0263	7.0983	0	7.2706	7.4265	0
F2	0	8.1455	8.1445	8.1445	8.1325	8.2136	8.141	8.1727	7.9824	8.1402	8.1945
F3	8.0251	8.0845	8.0274	7.979	8.0514	8.0062	8.0514	8.0168	8.0308	0	7.961
F4	7.8957	8.0399	7.9784	7.8261	7.8452	8.1011	8.1877	8.1011	7.9236	8.0975	8.0362
F5	7.9619	7.8852	7.9105	7.9803	7.7847	0	8.0826	7.9792	7.940	7.8969	7.897
F6	0	7.9128	7.9619	7.9089	8.0005	7.9286	7.9572	7.9809	8.0322	8.049	7.9465
F7	8.0612	8.0692	7.7845	7.9951	8.0531	8.0494	8.0354	8.0232	8.0692	7.7902	7.9955
F8	7.8265	7.6719	0	7.7605	7.6222	7.8105	7.6845	0	7.8409	7.5748	7.7213
F9	8.1439	7.9892	7.9242	7.7648	7.9466	8.0278	8.0024	7.8047	8.0278	8.0318	8.111
F10	8.0683	8.1098	8.0295	8.0915	8.1185	8.1221	8.0988	8.1084	7.9235	8.0757	8.0911

.

Step 6: Calculate the mean preference value of each expert, as shown in

Table 8. The average preference value of each expert.

	Expert 1	Expert 2	Expert 3	Expert 4	Expert 5	Expert 6	Expert 7	Expert 8	Expert 9	Expert 10	Expert 11
preference mean value	8.9233	8.8812	8.8079	8.8247	8.8645	8.8676	8.8636	8.8669	8.8224	8.85	8.9124

.

Step 7: Obtain the expert proximity value, as shown in

Table 9. Expert proximity value.

	Expert 1	Expert 2	Expert 3	Expert 4	Expert 5	Expert 6	Expert 7	Expert 8	Expert 9	Expert 10	Expert 11
Expert 1		0.0665	0.0246	0.03	0.0272	0.0027	0.0106	0.1	0.0192	0.0253	0.0716
Expert 2			0.091	0.0965	0.0393	0.0692	0.0558	0.0336	0.0857	0.0917	0.0051
Expert 3				0.0054	0.0518	0.0219	0.0352	0.1246	0.0053	0.0007	0.0961
Expert 4					0.0572	0.0273	0.0406	0.1301	0.0108	0.0047	0.1016
Expert 5						0.0299	0.0166	0.0729	0.0464	0.0525	0.0444
Expert 6							0.0133	0.1028	0.0165	0.0226	0.0743
Expert7								0.0894	0.0298	0.0359	0.061
Expert 8									0.1193	0.1253	0.0285
Expert 9										0.006	0.0908
Expert 10											0.0968
Expert 11

.

The difference between the proximity values of Expert 1 and Expert 6 is 0.0027. It shows that Expert 1 and Expert 6 have the most consistent ideas among all experts. Therefore, we apply the assessment of Expert 6 to fill in the missing values of Expert 1 in the corresponding position of HIDCM. That means that Expert 1 obtained the assessment of the interrelation of F2 against F5 and F6 of 2 and 5, respectively.

The hybrid initial direct correlation matrix built by Expert 4, Expert 7, and Expert 9 has no missing values. The missing values of HIDCM by other experts are imputed in a similar way, as shown in

Table 10. Imputed values of experts.

Expert	Missing Value Position	Expert with Same Idea	Imputed Values
Expert 1	F2/F5 and F6/F2	Expert 6	2 and 5
Expert 2	F1/F8	Expert 11	4
Expert 3	F8/F3	Expert 10	[4, 5]
Expert 5	F1/F6	Expert 7	3
Expert 6	F5/F7	Expert 1	3
Expert 8	F1/F4 and F8/F3	Expert 11	1 and 4
Expert 10	F3/F9	Expert 3	3
Expert 11	F1/F4	Expert 2	[2, 3]

.

Step 8: Convert the assessments into interval numbers. Therefore, the new direct incidence matrix for Expert 1 is shown in

Table 11. New direct incidence matrix of Expert 1.

	F1	F2	F3	F4	F5	F6	F7	F8	F9	F10
F1	[0, 0]	[5, 5]	[3, 3]	[2, 2]	[6, 6]	[3, 3]	[3, 3]	[3, 3]	[3, 3]	[3, 5]
F2	[2, 5]	[0, 0]	[1, 3]	[3, 3]	[2, 2]	[1, 1]	[6, 6]	[2, 4]	[3, 3]	[7, 7]
F3	[3, 3]	[4, 4]	[0, 0]	[6, 6]	[5, 5]	[7, 7]	[2, 5]	[9, 9]	[2, 5]	[4, 4]
F4	[2, 2]	[2, 4]	[6, 6]	[0, 0]	[3, 3]	[3, 3]	[6, 9]	[4, 4]	[2, 2]	[2, 3]
F5	[3, 3]	[3, 6]	[3, 3]	[1, 3]	[0, 0]	[6, 6]	[4, 4]	[5, 5]	[4, 4]	[5, 5]
F6	[2, 2]	[4, 4]	[5, 5]	[3, 3]	[3, 6]	[0, 0]	[4, 4]	[6, 6]	[6, 6]	[3, 3]
F7	[3, 7]	[2, 2]	[6, 9]	[6, 6]	[5, 5]	[8, 9]	[0, 0]	[2, 2]	[2, 2]	[3, 6]
F8	[6, 6]	[3, 3]	[4, 4]	[6, 6]	[7, 7]	[3, 3]	[2, 3]	[0, 0]	[5, 5]	[8, 8]
F9	[5, 5]	[6, 6]	[1, 4]	[3, 5]	[3, 3]	[2, 2]	[4, 4]	[6, 9]	[0, 0]	[5, 5]
F10	[6, 6]	[3, 4]	[3, 3]	[5, 5]	[4, 4]	[9, 9]	[5, 5]	[6, 6]	[3, 6]	[0, 0]

as an example.

Step 9: The weighted new direct incidence matrix is obtained, as shown in

Table 12. Weighted new direct incidence matrix.

	F1	F2	F3	F4	F5	F6	F7	F8	F9	F10
F1	[0, 0]	[5, 5.27]	[3, 3.36]	[1.73, 2]	[5.73, 6]	[3.09, 3.27]	[3.27, 3.82]	[2.82, 3.45]	[2.73, 3]	[2.36, 3.27]
F2	[3.73, 4.09]	[0, 0]	[2.64, 3.27]	[2.91, 3.09]	[2.18, 2.27]	[1, 1.18]	[5.73, 5.91]	[1.91, 3.27]	[3, 3.18]	[6.82, 7]
F3	[2.91, 3.18]	[4.09, 4.09]	[0, 0]	[5.82, 6]	[5, 5]	[3.27, 3.36]	[2.82, 3.18]	[3, 3]	[1.91, 2.36]	[3.82, 3.82]
F4	[2, 2]	[1.91, 2.18]	[5.82, 5.82]	[0, 0]	[5.18, 5.27]	[1.18, 1.36]	[7.55, 7.91]	[2, 2]	[2, 2]	[1.73, 2.73]
F5	[2.73, 3.18]	[6.55, 6.91]	[3, 3]	[1.09, 1.55]	[0, 0]	[3.27, 3.27]	[3.91, 3.91]	[4, 4.18]	[4.09, 4.09]	[4.73, 4.82]
F6	[2.09, 2.09]	[4.09, 4.09]	[4.82, 4.82]	[3.09, 3.09]	[2.73, 3.18]	[0, 0]	[4, 4]	[5.64, 5.91]	[5.73, 6.18]	[2.64, 3.18]
F7	[4.82, 5.27]	[2.09, 2.09]	[7.45, 8]	[5.91, 5.91]	[4.45, 5.09]	[7.73, 7.82]	[0, 0]	[2, 2]	[2, 2]	[3.27, 3.55]
F8	[5.45, 6.27]	[3.18, 3.82]	[3.73, 4.36]	[5.09, 5.36]	[7, 7.18]	[3.09, 3.27]	[1.91, 2.09]	[0, 0]	[5.64, 5.91]	[8.09, 8]
F9	[4.73, 5.09]	[4.91, 5.09]	[1.91, 2.45]	[3.27, 3.45]	[2.27, 2.27]	[2.82, 2.82]	[2.45, 3]	[7.91, 8.18]	[0, 0]	[5.09, 5.09]
F10	[5.91, 6.09]	[4.18, 4.27]	[3, 3.09]	[4.27, 4.27]	[4, 4.18]	[8.64, 8.91]	[5, 5]	[6.45, 6.45]	[5.73, 6]	[0, 0]

.

Step 10: Calculate the interval values of the degree of influential impact and the degree of influenced impact. The interval values of degree of influential impact are: F1 = [2.58, 3.23], F2 = [2.69, 3.3], F3 = [2.81, 3.24], F4 = [2.56, 3.02], F5 = [2.95, 3.41], F6 = [3.04, 3.55], F7 = [3.34, 3.87], F8 = [3.73, 4.39], F9 = [3.14, 3.68], and F10 = [4.02, 4.57]. The interval values of degree of influenced impact are: F1 = [3, 3.6], F2 = [3.13, 3.64], F3 = [3.06, 3.65], F4 = [2.89, 3.35], F5 = [3.32, 3.87], F6 = [3, 3.45], F7 = [3.17, 3.7], F8 = [3.09, 3.68], F9 = [2.87, 3.37], and F10 = [3.33, 3.96].

Step 11: Calculate the crisp values of the degree of influential impact and the degree of influenced impact. The crisp values of degree of influential impact are: F1 = 2.408; F2 = 2.9345; F3 = 3.0622; F4 = 1.5197; F5 = 4.2246; F6 = 5.1215; F7 = 7.0233; F8 = 8.6855; F9 = 5.8247; and F10 = 9.196. The crisp values of degree of influenced impact are: F1 = 4.5506, F2 = 5.3188, F3 = 5.0379, F4 = 2.7107, F5 = 7.2401, F6 = 3.7541, F7 = 5.7815, F8 = 5.3045, F9 = 2.7652, and F10 = 7.5365.

Step 12: Obtain the central degree and the causal degree. The central degrees are: F1 = 6.9586, F2 = 8.2533, F3 = 8.1001, F4 = 4.2304, F5 = 11.4647, F6 = 8.8756, F7 = 12.8048, F8 = 13.99, F9 = 58.5899, and F10 = 16.7325. The causal degrees are: F1 = −2.1426, F2 = −2.3844, F3 = −1.9757, F4 = −1.191, F5 = −3.0155, F6 = 1.3674, F7 = 1.2417, F8 = 3.3811, F9 = 3.0595, and F10 = 1.6594.

The type of central degree is $F10>F8>F7>F5>F6>F9>F2>F3>F1>F4$. The sort of causal degree is $F8>F9>F10>F6>F7>F4>F3>F1>F2>F5$. Cause factors are $F6,F7,F8,F9,F10$. Effect factors are $F1,F2,F3,F4,F5$. They are shown in Figure 2.

5.2. Comparison with the Results of the Mean Input Method

In this paper, the mean imputation method is used to make a comparison with the proposed method. The missing values of one expert are filled by the weighted average value of the initial direct correlation matrix given by other experts, as shown in

Table 13. Imputed values of other experts.

Expert	Missing Value Position	Expert with Same Idea	Imputed Values
Expert 1	F2/F5 and F6/F2	Expert 6	[2.2, 2.3] and 4
Expert 2	F1/F8	Expert 11	[2.8, 3.5]
Expert 3	F8/F3	Expert 10	[3.67, 4.33]
Expert 5	F1/F6	Expert 7	[3.1, 3.3]
Expert 6	F5/F7	Expert 1	4
Expert 8	F1/F4 and F8/F3	Expert 11	[1.89, 2.11] and [3.67, 4.33]
Expert 10	F3/F9	Expert 3	[1.9, 2.4]
Expert 11	F1/F4	Expert 2	[1.89, 2.11]

.

Then, we calculated the degree of influential impact, the degree of influenced impact, the central degree, and the causal degree. The degrees of influential impact are: F1 = 2.4931, F2 = 2.9429, F3 = 3.0609, F4 = 1.5132, F5 = 4.2934, F6 = 5.0366, F7 = 7.0086, F8 = 8.679, F9 = 5.7785, and F10 = 9.1937. The degrees of influenced impact are: F1 = 4.5392, F2 = 5.2179, F3 = 4.9998, F4 = 2.8411, F5 = 7.2063, F6 = 3.7821, F7 = 5.8592, F8 = 5.2864, F9 = 2.8004, and F10 = 7.4675.

The central degrees are: F1 = 7.0323, F2 = 8.1609, F3 = 8.0607, F4 = 4.3544, F5 = 11.4997, F6 = 8.8187, F7 = 12.8679, F8 = 13.9655, F9 = 58.5789, and F10 = 16.6612. The causal degrees are: F1 = −2.0462, F2 = −2.275, F3 = −1.9389, F4 = −1.3279, F5 = −2.9129, F6 = 1.2545, F7 = 1.1494, F8 = 3.3926, F9 = 2.9781, and F10 = 1.7262.

It can be seen that the type of central degree and causal degree have not changed as per our proposed method. The sort of central degree is $F10>F8>F7>F5>F6>F9>F2>F3>F1>F4$, and the sort of causal degree is $F8>F9>F10>F6>F7>F4>F3>F1>F2>F5$. Cause factors are $F6,F7,F8,F9,F10$ and effect factors are $F1,F2,F3,F4,F5$.

We carefully observe the difference between our proposed method and the mean imputation method. The maximum value of the central degree using the proposed method is 16.7325, the minimum value is 4.2304, and the difference is 12.5021. The maximum value of the central degree using the mean imputation method is 16.6612, the minimum value is 4.3544, and the difference is 12.3068. The maximum value of causal degree using the proposed method is 3.3811, the minimum value is −3.0155, and the difference is 6.3966. The maximum value of causal degree using the mean imputation method is 3.3926, the minimum value is −2.9129, and the difference is 6.3055. It can be easily observed that the difference between the central degree and the causal degree in our proposed method is larger than the mean imputation method.

In addition, we take the HIDCM of Expert 1 as an example to observe the characteristics of the imputation, as shown in

Table 14. The new hybrid initial direct incidence matrix by the proposed method.

	F1	F2	F3	F4	F5	F6	F7	F8	F9	F10
F1	0	5	3	2	6	3	3	3	3	[3, 5]
F2	[2, 5]	0	[1, 3]	3	2	1	6	[2, 4]	3	7
F3	3	4	0	6	5	7	[2, 5]	9	[2, 5]	4
F4	2	[2, 4]	6	0	3	3	[6, 9]	4	2	[2, 3]
F5	3	[3, 6]	3	[1, 3]	0	6	4	5	4	5
F6	2	5	5	3	[3, 6]	0	4	6	6	3
F7	[3, 7]	2	[6, 9]	6	5	[8, 9]	0	2	2	[3, 6]
F8	6	3	4	6	7	3	[2, 3]	0	5	8
F9	5	6	[1, 4]	[3, 5]	3	2	4	[6, 9]	0	5
F10	6	[3, 4]	3	5	4	9	5	6	[3, 6]	0

Note: Italic bold means missing values supplement.

and

Table 15. The new hybrid initial direct incidence matrix by mean imputation.

	F1	F2	F3	F4	F5	F6	F7	F8	F9	F10
F1	0	5	3	2	6	3	3	3	3	[3, 5]
F2	[2, 5]	0	[1, 3]	3	[2.2, 2.3]	1	6	[2, 4]	3	7
F3	3	4	0	6	5	7	[2, 5]	9	[2, 5]	4
F4	2	[2, 4]	6	0	3	3	[6, 9]	4	2	[2, 3]
F5	3	[3, 6]	3	[1, 3]	0	6	4	5	4	5
F6	2	4	5	3	[3, 6]	0	4	6	6	3
F7	[3, 7]	2	[6, 9]	6	5	[8, 9]	0	2	2	[3, 6]
F8	6	3	4	6	7	3	[2, 3]	0	5	8
F9	5	6	[1, 4]	[3, 5]	3	2	4	[6, 9]	0	5
F10	6	[3, 4]	3	5	4	9	5	6	[3, 6]	0

Note: Italic bold means missing values supplement.

. Table 14 lists the new hybrid initial direct incidence matrix by the proposed method, and Table 15 lists the new hybrid initial direct incidence matrix by the mean imputation method.

It can be seen that the assessment of F2 against F5 is [2.2, 2.3]. The imputation values of this method are integers or interval integers, while the values of the mean imputation method are decimals. It is different from the traditional DEMATEL assessment scale, in which the boundaries of crisp values or interval values are all integers.

6. Discussion

First of all, the causal degree of Information mining capability (F6), Equipment support capability (F7), Monitoring and early warning capability (F8), Information security capability (F9), and Organization participation capability (F10) are positive, which indicates that these factors will affect other factors. Secondly, the ranking of the central degree of Information security capability (F9) is low in all factors, which indicates that this factor is of low importance. Therefore, the CSFs of EIRS are F6, F7, F8, and F10. Therefore we propose the following suggestions:

6.1. Monitoring and Early Warning Capability

DMs should focus on information monitoring and early warning. ① Establish an early warning mechanism. DMs should establish the early-warning situation of where and what kind of emergency and report to who and which organizations. In this way, the high-level organization reasonably establishes the early-warning trigger threshold, which is, what kind of early-warning trigger threshold can be issued by which organizations. ② Improving the assessment, reward, and punishment systems It stimulates the enthusiasm and initiative of personnel in early warning work, effectively improves the response to all kinds of emergencies, and provides more accurate and intuitive information for scientific decision-making and scheduling.

6.2. Organization Participation Capability

The timely response of EIRS needs the cooperation of relevant organizations in different fields. Therefore, it is necessary to build a strong organizational structure to gather various organizations. In the process of strengthening organizational participation, DMs should pay attention to the following three points. ① Promote the profession of the organization. By fully integrating the knowledge of various fields into the EIRS, we can better respond to emergencies. Therefore, the organization should be jointly constructed by health, meteorological, public security, and other organizations, etc. ② Enhance the cooperation of the organizations. Different emergency scenarios, such as snow disasters, rainstorms, and epidemic situations, require the active participation of different organizations. It would improve the cooperation abilities of various organizations by collecting and actively combining information. ③ Foster the relevance of the daily work of various organizations. In their daily work, the high-level organization should pay attention to the correlation of organizations in various organizations, such as establishing relatively safe online communication channels, including QQ groups. This is helpful to maintain the relevance of cooperative organizations and to ensure more cooperation when emergencies arise.

6.3. Information Mining Capability

The accuracy of information mining affects the timely response to information. Therefore, we should conduct in-depth mining of information capability, which can effectively prevent risks and ensure the timeliness and accuracy of information responses. Therefore, a new generation of information technology analysis approaches (i.e., big data, artificial intelligence, and blockchain) are used to conduct in-depth mining of intelligence information and predict the changing trend of emergencies. In this way, DMs can better improve the effectiveness of EIRS and control the whole process of emergencies.

6.4. Equipment Support Capability

DMs should attach importance to the support capability of information equipment. ① Increase hardware investment. The storage and sharing of intelligence and the integration and mining of heterogeneous data depend on the input of hardware. If the input of hardware equipment is reasonable, DMs could better monitor and early-warn emergencies. ② Promote the fine management of equipment. DMs should standardize and improve the operation, maintenance, repair, and maintenance workflow of information equipment and establish a reasonable account system. Only in this way can the management and operation of information equipment be effectively improved. ③ Strengthen this study and assessment of information equipment users. DMs should fully improve the learning efficiency of equipment users, enable them to flexibly master new products and technologies of relevant information equipment, and improve the management and assessment mechanisms of equipment users.

7. Conclusions

EIRS plays a vital role in emergency management, especially its CSFs. Scholars identify the CSFs of EIRS through various methods. Due to the complex decision environment and the incomplete knowledge structure of experts, there are some missing values in the interrelationship comparisons among factors. In addition, we allow the expert to give two assessments: crisp value and interval value. Therefore, a Similar-DEMATEL method based on the same idea of an expert is proposed to fill in the missing evaluation value and identify the CSFs of EIRS.

In this paper, we found that the CSFs of EIRS are Organization participation, Monitoring and early warning ability, Information mining ability, and Equipment support ability. We would attach importance to these CSFs to improve the efficiency of EIRS. It provides a foundation for risk assessment and scientific decision-making and brings possibilities for building sustainable cities.

The Similar-DEMATEL method is different from the mean imputation method for imputing the missing values. It has a better chance of focusing more on the uncertain decision-making circumstances. It offers a scientific and effective decision-making method for identifying the CSFs of EIRS to deal with emergencies. This method can also be used to identify CSFs in more complex and uncertain environments, such as knowledge management, fault identification, and other fields.

However, the limitation of this study is to identify CSFS through expert preferences. Future research can also specify the relationship between CSFs through methods such as structural equation models or interpretive structural models.