Large group two-stage risk emergency decision-making method based on big data analysis of social media

Abstract

In decision-making for major emergencies, the risk of a decision is caused by both the uncertainty of the decision maker and his or her deviation from group consistency. In this paper, a two-stage risk emergency decision-making method for large groups based on social media big data is hence proposed. In this method, the user-generated content related to major emergencies is first collected from social media. Then, an emergency decision attribute system is constructed based on the public opinion information about events, which is obtained by natural language processing. Term frequency-inverse document frequency and expert evaluation are used to determine the weights of the attributes. Second, an open two-stage decision-making process is designed to quantify decision risk according to the reliability and accuracy of decision makers’ opinions. Here, clustering is used to calculate the value of group members. Then, the technique for order preference by similarity to an ideal solution (TOPSIS) method is employed to rank decision alternatives. Finally, a case analysis and comparison of the major explosions in Tianjin Port on August 12, 2015, demonstrates the feasibility and effectiveness of the proposedmethod.

Keywords

User-generated content (UGC) big data large group emergency decision-making decision risk

1. Introduction

In recent years, the occurrence of major emergencies has become more frequent, resulting in an increase in the loss of life and property. For instance, in the Tianjin Port Explosions, a major explosion and fire accident occurred in the warehouses of Tianjin Port Ruihai Company on August 12, 2015. A total of 165 victims were killed by the explosions, 798 people were injured, and the direct economic losses reached 6.866 billion RMB. The explosion had a strong influence on society and attracted a large amount of attention from the Chinese government. After the accident, experts from different fields rushed to Tianjin to participate in the urgent disposal work. However, owing to the delay and incompleteness of rescue site information, disaster information could not be quickly collected and comprehensively summarized for the decision makers (DMs), leading to a series of problems such as the lack of awareness of risk and the difficulty of judging group behavior in decision-making analysis. Therefore, how to use big data information generated at the initial stage of an emergency so that safer decisions can be quickly is an urgent problem to be solved.

In the past, multi-attribute emergency decision-making processes mainly relied on expert knowledge and experience [1, 2]. However, for some emergency decision-making in certain scenarios, such as when the public is strongly concerned after an event occurs or the information available to experts is not sufficient, the information owned by the public needs to be introduced into the decision-making process to reduce the risk caused by information asymmetry between experts and the public. The rapid development of social media has provided an important platform for the expression of public wisdom as well as for data mining and information extraction in recent years [3]. After a major emergency, hundreds of millions of “netizens” can quickly gather and participate on the Internet, and their ideas and knowledge are posted on the network in the form of words, pictures, audio, video, and other media, forming a huge volume of user-generated content (UGC). Although these sources of non-structural data are scattered, they contain a tremendous amount of valuable information that can support emergency decision-making. Crowdsourcing is an emerging model that has been widely applied in emergency management [4 –6]. This model obtains an integrated result and exploits the “wisdom-of-crowds” effect by assigning tasks to relevant groups of people such as Twitter or Weibo users. Public opinions regarding emergencies can be collected and analyzed using online social media as a crowdsourcing platform. Current studies on UGC for emergency management mainly focus on emergency detection and early warning [7, 8], monitoring and description of events [9], assessment of the damage caused by accidents [10, 11], modes of information dissemination and distribution [12, 13], and individual or group behaviors in the context of emergencies [14, 15]. However, there is a lack of literature on employing UGC in the multi-attribute emergency decision-making process. This paper argues that UGC contains the public perception of the damage caused by the event and the trends of its development. The DMs need to pay attention to the public’s concerns about an event and try to reduce the impacts of the event. Using data mining technology to process UGC to obtain timely and useful information helps DMs to make decisions from a wider perspective.

In contrast, due to the complexity of major emergencies, emergency decision-making often requires the joint involvement of experts in various fields to ensure the rationality of the results. Hence, large group decision-making problems occur in emergency management [16 –19]. When the number of experts involved is greater than 11, they are defined as a large decision-making group [20]. However, the unpredictable development of situations, the incompleteness of decision-making information, and the subjective influence of DMs as “bounded rational” persons [21] make it hard for DMs to give a comprehensive and reasonable judgment in emergencies. At present, researchers in China and around the world have carried out a great deal of relevant research on this issue. Considering the subjective attitudes of decision-makers, Zhu [22] proposed a multi-attribute risk group decision-making method based on prospect theory. This method sets multiple reference points for both static and dynamic situations. Based on the emergency evacuation of villages downstream from a reservoir dam, Liu et al. [23] proposed an analytical method based on cumulative prospect theory to determine risk in decision-making problems in an emergency response. Based on the idea of regret theory, Yuan et al. [24] proposed corresponding risk decision-making methods in view of the problems that could arise in different scenarios and losses after emergency plans have been implemented. Xuet al. [25] proposed a method of dynamic-risk emergency decision-making for large groups based on a Markov chain considering the preference-transfer problem of DMs in the decision-making process under complex environments. Liu et al. [26] proposed a risk decision-making method based on fault tree analysis in view of such characteristics as the dynamic evolution and multi-scenario of emergency decision-making. Tang [27] built a Bayesian network model based on survey data, statistical analysis and domain expert knowledge to analyze the risks of sudden water pollution accidents in the South-to-North Water Diversion Project. Although the above research provides risk emergency decision-making methods based on the perspective of psychological behavior or mature mathematical methods, these methods are seldom applied to large group decision-making, and the preference information is given independently by DMs in the decision-making process, with the results unavailable and invisible to other DMs. However, in many real scenarios (such as the expert meeting method), the opinions of DMs are openly displayed, that is, DMs can observe other members’ preferences and then express their own opinions [28, 29]. Hence, DMs are easily influenced by the preferences of other DMs in this situation. At the same time, the emergency decision-making process needs to produce a result quickly. The discrepancy between a DM’s own cognition level and experience may cause their preference to differ from those of the others. Moreover, a DM may do his least to finish the decision task, that is, he may works perfunctorily, not responsibly during the actual process, because there are more members in large group decision-making. The uncertainty and deviation from group preference consistency that these less-specialized people may show could increase the risk of a decision and have a greater impact on the final result. Therefore, it is necessary to effectively identify and reasonably deal with the individual subjective decision risks of DMs.

To address the above shortcomings, the proposed method combines the UGC data of major emergencies with large group emergency decision-making, and proposes a two-stage risk emergency decision-making method for large groups based on the analysis of social media big data. First, natural language processing is used to obtain public opinion information from the UGC and the attribute system is constructed using on expert knowledge. A comprehensive fuzzy evaluation is applied to determine the level of damage of the event, and the weights of decision attributes are determined. Then, two decision-making stages are used to collect the DM’s preference information. A DM’s decision risk coefficient is calculated according to the difference between the preference information of the same DM collected at two different times, and the DM weights are determined by clustering. Based on these results, a comprehensive ranking of the alternatives is realized.

The remainder of this paper is organized as follows. Section 2 introduces the design of the method, conversion of a language preference matrix, distribution of word weights, and the weights of the attributes and DMs in large group decision-making are determined. In Section 3, an example is used to illustrate the utility and applicability of the proposed model. Finally, the main conclusions are presented in Section 4.

2. Preliminaries

This section describes the conversion of a language preference matrix and the distribution of word weights. These are the basic steps of the decision method proposed in this paper.

2.1. Conversion of language preference matrix

Because of the complexity and fuzziness of emergency decision-making problems, it is difficult for DMs to use quantitative values to describe all the factors affecting a decision. Moreover, the differences among attributes mean that it is difficult to apply the same set of linguistic terms to all of them. Therefore, the proposed method adopts the multi-granularity generalized 2-tuple semantic model [30], which can effectively avoid the loss and distortion of language evaluation information and calculates language information accurately and practically.

Definition 1. [30]: Let S ={ s₀, s₁, ⋯ , s_g } be a set of linguistic terms and β ∈ [0, 1] be the integrated operational results. The function Δ that transforms a 2-tuple semantic variable (s_r, α) into the real value β is expressed as follows: $Δ : [0, 1] \to S \times [- \frac{1}{2 g}, \frac{1}{2 g})$

$\begin{matrix} Δ (β) = (s_{r}, α), with \\ {\begin{matrix} s_{r}, r = round (β \cdot g) \\ α = β - \frac{r}{g}, α \in [- \frac{1}{2 g}, \frac{1}{2 g}) \end{matrix} \end{matrix}$ (1) where round() operation yields the integer r that is closest to β.

The 2-tuple semantic variable can be converted into the corresponding real value β ∈ [0, 1] as follows: $Δ^{- 1} : S \times [- \frac{1}{2 g}, \frac{1}{2 g}) \to [0, 1]$

$Δ^{- 1} (s_{r}, α) = \frac{r}{g} + α = β$ (2)

Definition 2. [30]: Let S ={ s₀, s₁, ⋯ , s_g } be a set of linguistic terms and [(s_a, α_a) , (s_z, α_z)] represent an interval of 2-tuple semantic variables, where (s_a, α_a) ≤ (s_z, α_z). The function Δ that transforms the interval 2-tuple semantic variable (s_r, α) into interval value [β₁, β₂] (β₁, β₂ ∈ [0, 1] , β₁ ≤ β₂) is expressed as follows: $Δ [β_{1}, β_{2}] = [(s_{a}, α_{a}), (s_{z}, α_{z})]$

$with {\begin{matrix} s_{a}, a = round (β_{1} \cdot g) \\ s_{z}, z = round (β_{2} \cdot g) \\ α_{a} = β_{1} - \frac{a}{g}, α_{a} \in [- \frac{1}{2 g}, \frac{1}{2 g}) \\ α_{z} = β_{2} - \frac{z}{g}, α_{z} \in [- \frac{1}{2 g}, \frac{1}{2 g}) \end{matrix}$ (3)

The inverse function Δ^-1 is defined as follows:

$- Δ^{- 1} [(s_{a}, α_{a}), (s_{z}, α_{z})] = [\frac{a}{g} + α_{a}, \frac{z}{g} + α_{z}]$ (4)

To make it easier for the DMs to express preference information, they can adopt either language values or interval language values (for example, s₁ or [s₁, s₃]). The methods to convert them into interval 2-tuple semantic preferences are respectively as follows: $s_{1} \to [(s_{1}, 0), (s_{1}, 0)], (s_{1}, s_{3}) \to [(s_{1}, 0), (s_{3}, 0)]$

2.2. Word weight distribution

Term frequency-inverse document frequency (TF-IDF) is a classic method in the vector space model that is used to calculate the feature weight of documents. It was proposed by Salton in 1988 [31, 32]. The TF-IDF of a term is the dot product of the term frequency and the inverse document frequency. The term frequency refers to the frequency at which a term (word or phrase) appears in a particular document, and the inverse document frequency refers to how frequent a term is in all the examined documents. The formula of TF-IDF is as follows: $w_{td} = {tf}_{td} \times log (\frac{N}{{df}_{t}})$ (5) where w_td is the weight for term t in document d, tf_td is the frequency of term t in document d, N is the total number of documents in the document corpus, and df_t is the number of documents containing term t in the document corpus.

The advantage of the TF-IDF approach lies in its ability to statistically analyze and deal with big unstructured text data. High weights can be generated by a high frequency word in a particular document as well as that word’s low frequency in other documents. Therefore, TF-IDF tends to filter out common words and keep some less frequent but important words. For example, in general documents, the word “explosion,” which is common in many texts that discuss explosions, will have low inverse document frequency and a correspondingly low weight.

3. Proposed method

In this section, the large group emergency decision-making problem is described and the principles and procedures of the proposed decision-making method are given.

3.1. Description of the problem

In the process of emergency decision-making, let X = {x₁, x₂, ⋯ , x_P} be a set of alternatives, E = {e₁, e₁, . . . , e_M} (M ≥ 11) be a large group of DMs and U = {u₁, u₂, . . . , u_N} be a set of attributes. The weight vector of the DMs is ω = {ω₁, ω₂, ⋯ ω_M}, where ω_i ≥ 0 and $\sum_{i = 1}^{M} ω_{i} = 1$ . The weight vector of the attributes is W = {w₁, w₂, . . . , w_N}, where w_j ≥ 0 and . A set of linguistic terms S ={ s₀, s₁, ⋯ , s_g } is used by the DMs to give the initial preference matrix $V_{i}^{(1)} = (B_{lj}^{i})_{P \times N}$ , where $B_{lj}^{i}$ is the language preference value given by the ith DM for the attribute u_j ∈ U of alternative x_l ∈ X.

In an environment in which big data is available, when major emergencies occur, the public’s concerns and opinion information about the event should be fully tapped. Moreover, the risk caused by the uncertainty and deviation from group consistency of DMs’ preferences can be measured using to the reliability and accuracy of DMs’ opinion. On this basis, the scientific ranking of decision alternatives can be realized.

3.2. Process of the method

To take public information into consideration in the process of decision-making, the method proposed in this study starts from the emergency UGC big data obtained by data mining. First, decision-making experts construct an attribute system with the extracted event keywords and then evaluate the prepared alternatives. Because the members of decision-making groups usually have strong dependence behaviors in reality, “herd behavior” is most likely to occur when other members’ decisions are observable and the decision order is determined [33]. In this case, DMs are apt to follow others’ judgment, not expressing their true preferences, and the decision information clearly becomes less reliable. By using this behavioral feature we can let DMs to make decisions in open-ended situations and identify those group members who are more uncertain about their preferences by documenting any changes in preference information. Therefore, the proposed method collects DMs’ preference information in two stages, that is, each DM first gives his or her individual preference information independently, and then re-gives this information with respect to the opinions of the other DMs. Finally, the DMs’ evaluation information is used to rank the alternatives according to the attribute weights and membership weights.

3.3. Construction of the decision attribute system and calculate of the attribute weights

When the data of a major emergency scene is reported to higher levels of management layer by layer, the emergency information becomes delayed and incomplete. It hence becomes more difficult to carry out dynamic real-time monitoring of events, which greatly increases the risk of emergency decision-making analysis. With the development of information technology and the increasing public attention to emergencies, network information resources have become an important part of emergency management in recent years. The occurrence of such events generates a large amount of public information about these events on social networking platforms. This UGC not only reflects the current concerns of the public, but also contains the public’s description, mentality, and perception of the damage caused by the event. Therefore, it is necessary to consider the opinion information of the public in social media to determine the attribute weights for emergency decision alternatives.

(1) Construction of the decision attribute system

The advantage of UGC lies in its immediate response to emergencies, but this information is also very scattered. Using a combination of expert advice and guidance, this problem can be effectively addressed. In the proposed method, the evaluation attribute system of an emergency decision scheme based on UGC consists of two parts: i) the traditional evaluation attributes of emergency decision alternatives and ii) keywords in the real time information of social media users. Specifically, the former must be determined by experts through group discussion and as a framework of the attribute system; The latter are found by first obtaining UGC through data mining, and then extracting several keywords that represent the text features and can be used as a generalization of public opinion information. Finally, the attribute system is constructed using the semantic relationships between keywords and attributes.

The TF-IDF algorithm is adopted to sort the terms after text segmentation for keyword extraction. Then, the terms with higher weights are selected as keywords. The text data extracted by crawling social media sites functions as the document corpus, with each piece of textual information treated as a document. The weight w_h of keyword h is obtained by arithmetically averaging the numbers of all the documents containing the word, as follows: $w_{h} = \frac{1}{{df}_{h}} \times (\sum_{d \in D_{h}} {tf}_{hd}) \times log (\frac{N}{{df}_{h}})$ (6)

(2) Damage assessment and attribute weight calculation

Using the established attribute system, a fuzzy comprehensive evaluation can assess the damage caused by events in real time, and provide reference information for DMs. The level of damage is used to describe the impact of emergencies on the local community, people’s property, and the public mindset. Because the keyword acquisition method focuses on statistical analysis rather than content analysis, its weight only represents the importance of its frequency in the documents, but the semantics of different words also reflect different degrees of damage. Hence, the damage assessment of an event requires experts to give each keyword a semantic score S_h to characterize the severity of damage. For example, the word “death” is more serious than “wounded,” so the score assigned to “death” should be greater than that of “wounded.” The score ranges from 0 to 5.

The level of damage reflected in different attributes can be obtained by a weighted sum of each corresponding keyword’s semantic score. The level of damage of attribute j is calculated as follows: ${Da}_{j} = \sum_{h = 1}^{y} w_{h} S_{h} (h = 1, 2, \dots, y)$ (7) where y represents the number of keywords corresponding to attribute j.

Because the calculation of the level of damage comes from the subjective perception and the preference information of the public, which affects the event, when the level of damage of a certain attribute is greater, the public has been greatly affected by or given more attention to this factor. To avoid additional serious consequences, DMs must attach more importance to this factor. Therefore, the weight of an attribute is determined by the level of damage of each attribute, as follows: $w_{j} = \frac{{Da}_{j}}{\sum_{j = 1}^{N} {Da}_{j}} (j = 1, 2, \dots, N)$ (8)

3.4. Weights of large-group DMs

With the development of computer network and information technology, many specialized decision support systems have been developed, which make it more convenient for a DM to collect, process, exchange, and share information. Because of the large number of DMs in major emergency decision-making, the subjective decision-making ability of different members is an important factor for determining the risk of a decision, and this directly affects the decision-making results. Therefore, the subjective decision risk factors of DMs need to be included in the membership weight determination. The proposed method measures the risk of a DM’s decision based on the two-stage decision-making process and quantifies this risk by defining the reliability and accuracy of his or her opinions. The weight of each member is revised according to the value of his or her decision risk.

(1) Collection of decision preference information

When a major emergency event occurs, DMs make use of the attribute system established using the UGC and the damage assessment information of the event. In the first stage, to formulate multiple sets of alternatives, DM e_i gives his or her language preference information $B_{lj}^{i}$ independently for the jth attribute of the l th alternative. These data are used to form independent language decision preference matrix $V_{i}^{(1)} = (B_{lj}^{i})_{P \times N}$ , where $B_{lj}^{i}$ is defined as the Da_j independent opinion of DM e_i. In the second stage, the independent opinions of the DMs are made public. DM e_i can obtain and observe the preferences of other DMs, and choose to modify or maintain his or her preference information given in the first stage. Then, language preference information $Q_{lj}^{i}$ for each alternative is resubmitted and used to form reference language preference matrix $V_{i}^{(2)} = (Q_{lj}^{i})_{P \times N}$ , where $Q_{lj}^{i}$ is defined as the reference opinion of DM e_i.

During this process, the opinions of other members in the group are considered by each DM as a feedback system, so the opinions in the second phase can be considered as those with respect to other DMs’ opinions. At the same time, the opinions of some members may be affected by those of the others in the second stage. For example, those less-professional DMs who are not sure about the preference information they have given in the first stage will modify their initial preferences after observing the preference information of others. Alternatively, because of crowd psychology, some DMs tend to have opinions that are consistent with most opinions, while some may choose to remain unchanged from their initial preference information. Generally speaking, such members who insist on their first opinions have a higher degree of confidence in making judgments, and the more certain their decision-making preferences are. On the contrary, those who have deviated from their initial opinions are less confident, and the more uncertain their decision-making preferences are.

(2) Measurement of the decision risk coefficient considering the reliability and accuracy of DMs.

To measure the discrepancy between DMs’ independent opinions and reference opinions, the concept of a confidence gap is defined. First, to facilitate processing, after the initial language preference information given by DMs is collected, it is transformed into the form of interval 2-tuple semantics, and then converted into the corresponding interval preference value according to Definition 2. Thus, interval preference matrix ${\tilde{V}}_{i}^{(1)} = (b_{lj}^{i})_{P \times N}$ of the independent opinions is obtained, where $b_{lj}^{i} = [b_{lj}^{iL}, b_{lj}^{iU}]$ indicates the independent opinion interval preference value and $b_{lj}^{iU} \geq b_{lj}^{iL} \geq 0$ ; reference opinion interval preference matrix ${\tilde{V}}_{i}^{(2)} = (q_{lj}^{i})_{P \times N}$ is also obtained, where $q_{lj}^{i} = [q_{lj}^{iL}, q_{lj}^{iU}]$ indicates the reference opinion interval preference value and $q_{lj}^{iU} \geq q_{lj}^{iL} \geq 0$ .

Let $D (b_{lj}^{i}, q_{lj}^{i})$ , which is the distance of the interval preference values of the independent and reference opinions, be the confidence gap of DM e_i. When this value is smaller, it indicates that the independent opinion of DM e_i is closer to the reference opinions, which suggests that the DM is more confident about his or her opinion; on the contrary, a more distant gap between the DM’s independent opinion and reference opinion indicates that he or she is less confident. The calculation of this confidence gap is as follows: $D (b_{lj}^{i}, q_{lj}^{i}) = \frac{\sqrt{2}}{2} \sqrt{{(b_{lj}^{iL} - q_{lj}^{iL})}^{2} + {(b_{lj}^{iU} - q_{lj}^{iU})}^{2}}$ (9) where $D (b_{lj}^{i}, q_{lj}^{i})$ satisfies the following properties: 1) boundedness: $0 \leq D (b_{lj}^{i}, q_{lj}^{i}) \leq 1$ , 2) symmetry: $D (b_{lj}^{i}, q_{lj}^{i}) = D (q_{lj}^{i}, b_{lj}^{i})$ , and 3) reflexivity: $D (b_{lj}^{i}, q_{lj}^{i}) = 0$ , if $b_{lj}^{i} = q_{lj}^{i}$ . Properties 1–3 can be easily proved.

Then, the mean value of the attributes of the alternatives is calculated to reflect the group opinion. For example, the averages of the jth attribute of the lth alternative in the independent and reference decision-making processes are respectively as follows: ${\bar{b}}_{lj} = [{\bar{b}}_{lj}^{L}, {\bar{b}}_{lj}^{U}] = [\frac{1}{M} \sum_{i = 1}^{M} b_{lj}^{iL}, \frac{1}{M} \sum_{i = 1}^{M} b_{lj}^{iU}]$

${\bar{q}}_{lj} = [{\bar{q}}_{lj}^{L}, {\bar{q}}_{lj}^{U}] = [\frac{1}{M} \sum_{i = 1}^{M} q_{lj}^{iL}, \frac{1}{M} \sum_{i = 1}^{M} q_{lj}^{iU}]$ (10)

For large group decision members, in past studies on multi-attribute group decision-making, the reliability of a DM’s preference information was rarely considered, or it was simply assumed that all preference information is completely reliable. In fact, no DM is entirely rational, and the different degrees of reliability of the judgements will have an impact on the decision result.

Let ${Ra}_{lj}^{i}$ be the reliability of a DM’s opinions. It indicates the level of certainty of the preference information for the jth attribute in the l th alternative given by DM e_i, and is calculated as follows: ${Ra}_{lj}^{i} = \frac{1 + D ({\bar{b}}_{lj}, {\bar{q}}_{lj})}{1 + D (b_{lj}^{i}, q_{lj}^{i})}$ (11) where the meanings of $D ({\bar{b}}_{lj}, {\bar{q}}_{lj})$ are given in Equations (9)–(10), which indicate the differences of the two group opinions in independent and reference situations. If the value of ${Ra}_{lj}^{i}$ is zero, this indicates that most members of the group adhere to the opinions of the first phase. Because $D ({\bar{b}}_{lj}, {\bar{q}}_{lj})$ is a constant value for all DMs, the ratio of $D ({\bar{b}}_{lj}, {\bar{q}}_{lj})$ to the confidence gap of each DM can reflect the degree of certainty of that DM’s opinion. If a DM’s confidence gap $D (b_{lj}^{i}, q_{lj}^{i})$ is bigger, the ratio will be smaller, indicating that the opinion given by that DM has a lower degree of reliability, that is, his or her preference information is of higher uncertainty. In contrast, when a DM’s confidence gap is smaller, the ratio will be larger, indicating that his or her opinion is of higher certainty. At the same time, to avoid allowing the denominator to be zero, one is added to both the numerator and denominator.

In addition, the consensus reaching of group opinions in group decision-making is also a matter of concern [34 –37]. It is often more practical to determine whether the outcome of a decision can be accepted or to what extent it can be accepted by a group because it relates to whether the alternative can be implemented quickly. When individual opinions deviate from group opinions to a greater extent, this shows that their resistance to group consistency has increased, and the decision risk brought by member preference conflicts will be triggered. To enable the decision-making result to reach a higher level of consensus, this paper argues that the majority opinion is more accurate than the minority opinion, that is, when the preference information given by a DM is closer to the opinion of the group, it indicates that the DM is more professional.

Let ${Ac}_{lj}^{i}$ be the accuracy of a DM’s opinions. It indicates the degree of consistency of preference given by DMe_iwith respect to those of other members for the jth attribute in the l th alternative and is calculated as follows: ${Ac}_{lj}^{i} = \frac{1}{1 + D (q_{lj}^{i}, {\bar{q}}_{lj})}$ (12) where $D (q_{lj}^{i}, {\bar{q}}_{lj})$ represents the distance between the individual opinion of a DM and the group opinion in the second stage. If $D (q_{lj}^{i}, {\bar{q}}_{lj})$ is smaller, then a larger ${Ac}_{lj}^{i}$ is obtained, which means that the closer the DM’s opinion is to most people, the more contribution they make to the group consistency. Conversely, if $D (q_{lj}^{i}, {\bar{q}}_{lj})$ is larger and ${Ac}_{lj}^{i}$ is smaller, this indicates that the opinions he or she gives are deviate more from the group consistency.

For DMs, the risk caused by their own decision-making ability is unavoidable. A high degree of reliability alone does not guarantee a lower-risk decision will be made. When lacking knowledge or experience about an event, a DM may insist on sticking with his or her initial observations to maintain the illusion of his or her professionalism. Therefore, the decision risk of DMs is measured by both the reliability and accuracy of their opinions.

First, the reliability of Definition 4 and the accuracy of Definition 5 are both benefit type indicators but they have different ranges. To eliminate the influence of different dimension scales on the decision-making results and to facilitate judgment, a non-linear transformation of the reliability and accuracy is conducted to obtain standardized values, calculated as follows:

$- {\begin{matrix} {Ra}_{lj}^{i}' = \frac{{Ra}_{lj}^{i} - min_{i} {Ra}_{lj}^{i}}{max_{i} {Ra}_{lj}^{i} - min_{i} {Ra}_{lj}^{i}} \\ {Ac}_{lj}^{i}' = \frac{{Ac}_{lj}^{i} - min_{i} {Ac}_{lj}^{i}}{max_{i} {Ac}_{lj}^{i} - min_{i} {Ac}_{lj}^{i}} \end{matrix} i \in 1, 2, \dots, N$ (13)

Through the above explanation, the decision risk coefficient of a DM can be defined as λⁱ, indicating the risk degree of both the uncertainty and deviation from group consistency of a DM’s preference, as follows:

$\begin{matrix} λ^{i} & = \frac{1}{P \cdot N} \sum_{l = 1}^{P} \sum_{j = 1}^{N} [θ (1 - {Ra}_{lj}^{i}') \\ + (1 - θ) (1 - {Ac}_{lj}^{i}')] \end{matrix}$ (14)

where θ and (1 - θ) represent the weights corresponding to reliability and accuracy, respectively. These weights are determined by the DMs according to the decision-making problem and the actual situation. If there is no special preference, the value is set as θ = 1/2, indicating that reliability and accuracy are given the same weight.

For the acquisition process of λⁱ, first, the decision risk coefficient of a DM’s preference given for the attributes of different alternatives is calculated, and then the decision risk coefficients of all members are obtained by simple weighted arithmetic averaging. Obviously, when a DM’s opinion is more reliable or accurate, he or she is given a smaller the decision risk coefficient, and his or her decision risk is smaller. On the contrary, a greater the value of λⁱ indicates a greater the decision risk.

(3) Large group preference clustering and the weight of members.

After the decision risk has been quantified for large-group members, the weight of the members needs to be determined. Using opinion clustering, DMs’ opinions are analyzed and managed, with the large number of members involved in decision-making considered. In the proposed method, the clustering method in [38] is used to cluster adjusted reference opinion interval preference matrix ${\tilde{V}}_{i}^{(2)} = (q_{lj}^{i})_{P \times N}$ , and the decision groups can be clustered into K clusters, where n_k is the number of members in cluster C^k. Using this clustering method, DMs in the same cluster can be assigned the same weight because they have similar preference information. At the same time, DMs in larger-scale clusters should be assigned higher weights based on the majority principle. Hence, initial weights of members obtained from the clustering result are as follows:

$ω_{i}^{'} = ω_{n k} = n k / \sum_{k = 1}^{K} n_{k}^{2}, i = 1, 2…, M$ (15)

Next, the initial weights of the members are adjusted by the member’s decision risk coefficient λⁱ. A greater value of λⁱ indicates more uncertainty or deviation from the consistency goal for the information provided by the DM. Hence, a lower weight should be given to the DM. Finally, the weight of decision risk of DM e_i is calculated as follows:

$ω_{i}^{'} = ({ω^{'}}_{i} \cdot \frac{1}{λ^{i}}) / \sum_{i = 1}^{M} ({ω^{'}}_{i} \cdot \frac{1}{λ^{i}})$ (16)

3.5. Proposed method steps

Briefly, the steps of the proposed method is as follows:

Step 1: UGC related to emergencies is extracted by crawling social media sites, and the attribute system is constructed using the attribute framework developed by experts and the keywords extracted from the UGC.

Step 2: Using the TF-IDF weight of keyword w_h obtained from Equation (6) and semantic evaluation value S_h determined by the expert scoring method, the level of damage of each attribute Da_j is calculated according to Equation (7) and the weight of attribute w_j is obtained after normalization.

Step 3: Using the two-stage decision-making process, independent opinion language preference matrix $V_{i}^{(1)} = (B_{lj}^{i})_{P \times N}$ , and reference opinion language preference matrix $V_{i}^{(2)} = (Q_{lj}^{i})_{P \times N}$ given by the DMs are collected. They are respectively transformed into independent opinion interval preference matrix ${\tilde{V}}_{i}^{(1)} = (b_{lj}^{i})_{P \times N}$ and reference opinion interval preference matrix ${\tilde{V}}_{i}^{(2)} = (q_{lj}^{i})_{P \times N}$ using Equations (3) and (4) as well as Definition 2.

Step 4: Reliability ${Ra}_{lj}^{i}$ and accuracy ${Ac}_{lj}^{i}$ of the preference information given by each DM are respectively calculated by Equations (11) and (12) and normalized by Equation (13). Then, the decision risk coefficient λⁱ of each member in the decision-making group is obtained by Equation (14).

Step 5: The reference opinion preference matrix for all DMs is clustered. Then, the initial weight of each member in cluster C^k is obtained by Equation (15). The decision risk weights of all group members are determined and combined the decision risk coefficient.

Step 6: Alternatives are ranked using the TOPSIS method.

(1) The DMs’ reference opinion interval preference matrix is clustered using attribute weights to obtain the weighted decision matrix

${\tilde{V}}_{i}^{(2)} = [\begin{matrix} v_{11}^{i} v_{12}^{i} \dots v_{1 N}^{i} \\ v_{21}^{i} v_{22}^{i} \dots v_{2 N}^{i} \\ \dots \dots ⋱ \dots \\ v_{P 1}^{i} v_{P 2}^{i} \dots v_{PN}^{i} \end{matrix}]$ where $v_{lj}^{i} = q_{lj}^{i} \times w_{j}$ , and 1 ≤ i ≤ M, 1 ≤ j ≤ N.

(2) The positive and negative ideal solutions are obtained for each alternative. $\begin{matrix} V^{+} = {v_{1}^{+}, v_{2}^{+}, \dots, v_{N}^{+}}, V^{-} = {v_{1}^{-}, v_{2}^{-}, \dots, v_{N}^{-}} \end{matrix}$ $\begin{matrix} V_{j}^{+} = [v_{j}^{+ L}, v_{j}^{+ U}] = [min (v_{lj}^{iL}), min (v_{lj}^{iU})], j \in N \end{matrix}$ $\begin{matrix} V_{j}^{-} = [v_{j}^{- L}, v_{j}^{- U}] = [max (v_{lj}^{iL}), max (v_{lj}^{iU})], j \in N \end{matrix}$

(3) The distance between the positive and negative ideal solutions is calculated.

$\begin{matrix} D_{l}^{i +} & = \sqrt{\sum_{j = 1}^{M} {(V_{lj}^{i} - V_{j}^{+})}^{2}} \\ = \sqrt{\sum_{j = 1}^{M} {(V_{lj}^{iL} - V_{j}^{+ L})}^{2} + {(V_{lj}^{iU} - V_{j}^{+ U})}^{2}} \end{matrix}$ $\begin{matrix} D_{l}^{i -} & = \sqrt{\sum_{j = 1}^{M} {(V_{lj}^{i} - V_{j}^{-})}^{2}} \\ = \sqrt{\sum_{j = 1}^{M} {(V_{lj}^{iL} - V_{j}^{- L})}^{2} + {(V_{lj}^{iU} - V_{j}^{- U})}^{2}} \end{matrix}$

(4) The closeness degree of each alternative is calculated.

$C_{l} = \frac{D_{l}^{-}}{D_{l}^{+} + D_{l}^{-}}$ (17)

where $D_{l}^{+} = \sum_{j = 1}^{M} ω_{i} D_{l}^{i +}$ and $D_{l}^{-} = \sum_{j = 1}^{M} ω_{i} D_{l}^{i -}$ .

Finally, the alternatives are ranked based on the value of C_l and the optimal alternative can be selected.

4. Case study

4.1. Case background

At 23:30 h on August 12, 2015, an major fire and explosion accident occurred in the warehouse of Ruihai Company in Tianjin Port. According to seismometer data, the intensities of the two explosions were equivalent to 3 tons and 21 tons of TNT, respectively. The explosion caused heavy casualties and triggered a large number of container fires. The local fire command center responded quickly after receiving the emergency call. By 11:00 h on August 13, Tianjin Fire Corps had deployed 143 fire trucks and more than 1,000 firefighters for search and rescue at the scene. At the same time, the accident sparked strong concern from the public, who expressed it through social media. A large number of Internet users expressed their opinions and thoughts on social networks. As a large number of dangerous goods were stored on the scene and the specific explosives could not be determined, to avoid further aggravation of the situation, Tianjin Emergency Decision-making and Command Center convened 16 emergency management experts on the network and formulated the following three emergency alternatives according to the current accident status:

x₁: Continue to carry out environmental emergency monitoring and increase the number of firefighters and fire trucks at the support site to enhance fire control.

x₂: Continue to carry out environmental emergency monitoring, inform the fire-fighting authorities to evacuate the scene, and dispatch an anti-chemical team with professional equipment for search and rescue at the scene.

x₃: Continue to carry out environmental emergency monitoring, suspend the fire-fighting and search and rescue operations, and urgently identify the content, quantity, and storage of dangerous chemicals before making further decisions. The expert group identified six factors as the evaluation attributes of the alternatives based on past experience of response to such emergencies as follows: the cost of alternatives (u₁), casualties (u₂), property loss (u₃), degree of public panic (u₄), impact on the environment (u₅), and impact on life (u₆). Among them, for attributes u₁ –u₄, language term set S¹ with a granularity of 9 was adopted, and for attributes u₅ –u₆ language term set S² with a granularity of 7 was adopted. They are expressed as follows: $\begin{matrix} S^{1} = {\begin{matrix} s_{0} = extremelylow, s_{1} = verylow, s_{2} = low, \\ s_{3} = slightlylow, s_{4} = medium, s_{5} = slightly \\ high, s_{6} = high, s_{7} = veryhigh, s_{8} = extremelyhigh \end{matrix}} \end{matrix}$

$\begin{matrix} S^{2} = {\begin{matrix} s_{0} = verysmall, s_{1} = small, s_{2} = slightlysmall, s_{3} = \\ medium, s_{4} = slightlybig, s_{5} = big, s_{6} = verybig \end{matrix}} \end{matrix}$

4.2. Decision-making steps

Step 1. Large quantities of text messages on social networks were collected to support emergency decision-making and Sina Weibo was selected as the source of event-related UGC data. “Tianjin Explosion” and “Tanggu Explosion” were used as the search keywords. With the API interface provided by Sina Weibo, microblogs within the time range of 23:30 h on August 12 to 15:00 h on August 13 were captured. The recorded fields include the author’s name, microblogging content, post time, and comment content. Because of the inclusion of non-public-generated content or noisy data, a total of 79,211 microblog posts were retained after manual screening to remove news releases from the government or mainstream media accounts and messages with fewer than two words.

After the collection of the UGC text data from Sina Weibo, the data were preprocessed to remove hashtags, @ content, URLs, and other unwanted content. Then, then the text was segmented into words, stop words were removed, and part of speech tagging was performed. Finally, the frequency of words in the data was analyzed. In the implementation of the proposed method, keyword extraction is implemented in the Python, and the Python-based jieba word segmentation software package was used to divide the text into terms. The 500 terms with the highest TF-IDF values were selected as candidate keywords. Because there are several words in the Chinese language that express the same meaning, keywords with the same semantic meaning, such as “death,” “killed,” “remains,” and “pass away” were merged into the keyword “death.” Similarly, keywords “horrible,” “terrible,” and “fear” were merged into “terrible.” The TF-IDF value of the merged words was accumulated. Using the attribute framework determined by the experts, the keywords corresponding to each attribute were listed based on their semantic relationships, thus forming the evaluation system of the emergency decision-making scheme of the 8/12 Tianjin Port major explosion, as shown in Table 1.

Table 1
Emergency decision-making evaluation attribute system of the 8/12 Tianjin Port major explosion

Attribute (u) Keywords

Cost of alternatives (u₁) At all costs, lose no time, cloud seeding, unmanned aerial vehicle, fire engine

Casualty (u₂) Death, sacrifice, missing, rescue, save, hurt, safety

Property loss (u₃) Burn, ruin, collapse, broken, car, melt, on fire, mushroom cloud

Degree of public panic (u₄) Tears, heartrending, terrible, sad, shocked, worried, bless, salute

Impact on the environment (u₅) Chemical, poisonous, pungent, harmful gas, pollution, cyanide, leakage

Impact on the life (u₆) Evacuation, evacuate, no water, power outage, traffic jam, out of service, give way

Attribute (u)	Keywords
Cost of alternatives (u₁)	At all costs, lose no time, cloud seeding, unmanned aerial vehicle, fire engine
Casualty (u₂)	Death, sacrifice, missing, rescue, save, hurt, safety
Property loss (u₃)	Burn, ruin, collapse, broken, car, melt, on fire, mushroom cloud
Degree of public panic (u₄)	Tears, heartrending, terrible, sad, shocked, worried, bless, salute
Impact on the environment (u₅)	Chemical, poisonous, pungent, harmful gas, pollution, cyanide, leakage
Impact on the life (u₆)	Evacuation, evacuate, no water, power outage, traffic jam, out of service, give way

Step 2. The TF-IDF weights of the keywords were calculated by Equation (6) using the Python program. Then, using the expert scoring method, 10 experts were asked to give scores for the keywords in the attribute system according to their severity, and the semantic score of each keyword was obtained after statistical averaging. The level of damage of each attribute was calculated by Equation (7). An example of the calculation of the level of damage of attribute u₂ is shown in Table 2, with the result of 21.206.

Table 2

Level of damage of attribute u₂

Attribute	Keywords	Semantic score	TF-IDF value	Damage degree
Casualty (u₂)	Death	5.0	1.838	9.190
	Sacrifice	5.0	0.499	2.495
	Missing	4.0	0.506	2.024
	Rescue	3.5	1.094	3.829
	Save	3.0	0.783	2.349
	Hurt	2.5	0.233	0.583
	Safety	1.0	0.736	0.736
Summation				21.206

After the calculation of the level of damage of each attribute, the weight of the attributes was obtained by Equation (8), resulting in W = {0.07, 0.25, 0.16, 0.23, 0.20, 0.08}.

Step 3. In accordance with the two-stage decision-making process, DMs gave their preference information of independent opinion and reference opinion according to the linguistic sets of the six attributes. In this process, part of the DM’s preference information was modified in the second phase, except for that of DMs e₃, e₁₃, and e₁₆. Definition 2 was used convert the language preference matrices into interval preference matrices, as shown in Tables 3–6.

Table 3

DMs’ independent opinion language preference matrix

		u ₁	u ₂	u ₃	u ₄	u ₅	u ₆		u ₁	u ₂	u ₃	u ₄	u ₅	u ₆
e ₁	x ₁	[s₅,s₈]	[s₇,s₈]	[s₂,s₄]	[s₃]	[s₂,s₄]	[s₂,s₅]	e ₂	x ₁	[s₁,s₃]	[s₅,s₇]	[s₃]	[s₅,s₇]	[s₃,s₅]	[s₅,s₆]
	x ₂	[s₃,s₆]	[s₁,s₃]	[s₃,s₈]	[s₅,s₇]	[s₃,s₅]	[s₃,s₅]		x ₂	[s₂,s₅]	[s₄,s₆]	[s₆,s₇]	[s₆,s₈]	[s₃,s₆]	[s₁,s₂]
	x ₃	[s₁,s₃]	[s₄]	[s₄,s₆]	[s₇,s₈]	[s₁,s₅]	[s₂,s₄]		x ₃	[s₆,s₇]	[s₀,s₆]	[s₃,s₄]	[s₂,s₄]	[s₁]	[s₀,s₂]
⋯
e ₁₅	x ₁	[s₃,s₅]	[s₆,s₇]	[s₁,s₄]	[s₄,s₆]	[s₄,s₆]	[s₅,s₆]	e ₁₆	x ₁	[s₄,s₆]	[s₄,s₆]	[s₃,s₅]	[s₁]	[s₂,s₄]	[s₃,s₅]
	x ₂	[s₂,s₄]	[s₄,s₆]	[s₂,s₅]	[s₃,s₅]	[s₃]	[s₂,s₃]		x ₂	[s₅]	[s₁,s₄]	[s₄,s₅]	[s₅,s₇]	[s₃,s₆]	[s₃,s₅]
	x ₃	[s₁]	[s₃,s₄]	[s₄,s₇]	[s₄,s₇]	[s₁,s₂]	[s₀,s₂]		x ₃	[s₃,s₄]	[s₂,s₅]	[s₅,s₈]	[s₂,s₅]	[s₂,s₅]	[s₁,s₄]

Table 4

DMs’ reference opinion language preference matrix

		u ₁	u ₂	u ₃	u ₄	u ₅	u ₆		u ₁	u ₂	u ₃	u ₄	u ₅	u ₆
e ₁	x ₁	[s₅,s₇]	[s₇,s₈]	[s₂,s₄]	[s₂,s₄]	[s₃,s₄]	[s₂,s₅]	e ₂	x ₁	[s₅,s₆]	[s₅,s₆]	[s₃]	[s₂,s₄]	[s₃,s₄]	[s₅,s₆]
	x ₂	[s₁,s₃]	[s₁,s₃]	[s₄,s₆]	[s₆,s₇]	[s₁,s₅]	[s₂,s₄]		x ₂	[s₂,s₅]	[s₂,s₆]	[s₆,s₇]	[s₃,s₆]	[s₃,s₄]	[s₁,s₂]
	x ₃	[s₁,s₃]	[s₄]	[s₄,s₆]	[s₇,s₈]	[s₁,s₅]	[s₂,s₄]		x ₃	[s₁,s₃]	[s₀,s₆]	[s₅,s₇]	[s₆,s₇]	[s₁]	[s₁,s₂]
⋯
e ₁₅	x ₁	[s₃,s₅]	[s₅,s₆]	[s₁,s₄]	[s₄,s₆]	[s₄,s₆]	[s₅,s₆]	e ₁₆	x ₁	[s₄,s₆]	[s₄,s₆]	[s₃,s₅]	[s₁]	[s₂,s₄]	[s₃,s₅]
	x ₂	[s₂,s₄]	[s₄,s₆]	[s₄,s₅]	[s₃,s₅]	[s₃]	[s₂,s₃]		x ₂	[s₅]	[s₁,s₄]	[s₄,s₅]	[s₅,s₇]	[s₃,s₆]	[s₃,s₅]
	x ₃	[s₁]	[s₃,s₄]	[s₄,s₇]	[s₄,s₇]	[s₁,s₂]	[s₀,s₂]		x ₃	[s₃,s₄]	[s₂,s₅]	[s₅,s₈]	[s₂,s₅]	[s₂,s₅]	[s₁,s₄]

Table 5

DMs’ independent opinion interval preference matrix

		u ₁	u ₂	u ₃	u ₄	u ₅	u ₆		u ₁	u ₂	u ₃	u ₄	u ₅	u ₆
e ₁	x ₁	[0.625,	0.875,	0.250,	0.375,	0.333,	0.333,	e ₂	x ₁	0.125,	0.625,	0.375,	0.625,	0.500,	0.833,
		1.000]	1.000]	0.500]	0.375]	0.667]	0.833]			0.375]	0.875]	0.375]	0.875]	0.833]	1.000]
	x ₂	[0.375,	0.125,	0.375,	0.625,	0.500,	0.500,		x ₂	0.250,	0.500,	0.750,	0.750,	0.500,	0.167,
		0.750]	0.375]	1.000]	0.875]	0.833]	0.833]			0.625]	0.750]	0.875]	1.000]	1.000]	0.333]
	x ₃	[0.125,	0.500,	0.500,	0.875,	0.167,	0.333,		x ₃	0.750,	0.000,	0.375,	0.250,	0.167,	0.000,
		0.375]	0.500]	0.750]	1.000]	0.833]	0.667]			0.875]	0.750]	0.500]	0.500]	0.167]	0.333]
⋯
e ₁₅	x ₁	[0.375,	0.750,	0.125,	0.500,	0.667,	0.833,	e ₁₆	x ₁	0.500,	0.500,	0.375,	0.125,	0.333,	0.500,
		0.625]	0.875]	0.500]	0.750]	1.000]	1.000]			0.750]	0.750]	0.625]	0.125]	0.667]	0.833]
	x ₂	[0.250,	0.500,	0.250,	0.375,	0.500,	0.333,		x ₂	0.625,	0.125,	0.500,	0.625,	0.500,	0.500,
		0.500]	0.750]	0.625]	0.625]	0.500]	0.500]			0.625]	0.500]	0.625]	0.875]	1.000]	0.833]
	x ₃	[0.125,	0.375,	0.500,	0.500,	0.167,	0.000,		x ₃	0.375,	0.250,	0.625,	0.250,	0.333,	0.167,
		0.125]	0.500]	0.875]	0.875]	0.333]	0.333]			0.500]	0.625]	1.000]	0.625]	0.833]	0.667]

Table 6

DMs’ reference opinion interval preference matrix

u ₁	u ₂	u ₃	u ₄	u ₅	u ₆	u ₁	u ₂	u ₃	u ₄	u ₅	u ₆
e ₁	x ₁	[0.625,	0.875,	0.250,	0.250,	0.500,	0.333,	e ₂	x ₁	0.625,	0.625,	0.375,	0.250,	0.500,	0.833,
		0.875]	1.000]	0.500]	0.500]	0.667]	0.833]			0.750]	0.750]	0.375]	0.500]	0.667]	1.000]
	x ₂	[0.375,	0.125,	0.375,	0.625,	0.333,	0.333,		x ₂	0.250,	0.250,	0.750,	0.375,	0.500,	0.167,
		0.750]	0.375]	0.625]	0.875]	0.833]	0.833]			0.625]	0.750]	0.875]	0.750]	0.667]	0.333]
	x ₃	[0.125,	0.500,	0.500,	0.750,	0.167,	0.333,		x ₃	0.125,	0.000,	0.625,	0.750,	0.167,	0.167,
		0.375]	0.500]	0.750]	0.875]	0.833]	0.667]			0.375]	0.750]	0.875]	0.875]	0.167]	0.333]
⋯
e ₁₅	x ₁	[0.375,	0.625,	0.125,	0.500,	0.667,	0.833,	e ₁₆	x ₁	0.500,	0.500,	0.375,	0.125,	0.333,	0.500,
		0.625]	0.750]	0.500]	0.750]	1.000]	1.000]			0.750]	0.750]	0.625]	0.125]	0.667]	0.833]
	x ₂	[0.250,	0.500,	0.250,	0.375,	0.500,	0.333,		x ₂	0.625,	0.125,	0.500,	0.625,	0.500,	0.500,
		0.500]	0.750]	0.625]	0.625]	0.500]	0.500]			0.625]	0.500]	0.625]	0.875]	1.000]	0.833]
	x ₃	[0.125,	0.375,	0.500,	0.500,	0.167,	0.000,		x ₃	0.375,	0.250,	0.625,	0.250,	0.333,	0.167,
		0.125]	0.500]	0.875]	0.875]	0.333]	0.333]			0.500]	0.625]	1.000]	0.625]	0.833]	0.667]

Step 4. According to the differences between each DM’s opinions in the independent and reference situations as well as the differences between his or her individual opinions and the group opinions, the DM’s reliability and accuracy were calculated respectively by Equations (11) and (12), and then normalized by Equation (13). The results for DM e₁ are shown in Tables 7 and 8. Then, the DM’s decision risk coefficient in the group was calculated by Equation (14), as shown in Table 9.

Table 7

Reliability of DM e₁’s opinion

	u ₁	u ₂	u ₃	u ₄	u ₅	u ₆
x ₁	0.735	1	1	1	0	1
x ₂	1	1	0.232	1	1	1
x ₃	1	1	1	0.637	1	1

Table 8

Accuracy of DM e₁’s opinion

	u ₁	u ₂	u ₃	u ₄	u ₅	u ₆
x ₁	0.623	0.118	1	1	0.656	0.690
x ₂	0.746	0.646	0.834	0.479	0.496	0.882
x ₃	1	0.541	0.499	0.512	0.548	0.998

Table 9

Decision risk coefficients of group members

	λ ⁱ		λ ⁱ		λ ⁱ		λ ⁱ
e ₁	0.226	e ₅	0.255	e ₉	0.298	e ₁₃	0.219
e ₂	0.455	e ₆	0.221	e ₁₀	0.403	e ₁₄	0.261
e ₃	0.264	e ₇	0.218	e ₁₁	0.333	e ₁₅	0.242
e ₄	0.414	e ₈	0.242	e ₁₂	0.270	e ₁₆	0.188

Step 5. The method in [38] was used to cluster the reference opinion preference matrix of the 16 DMs. Threshold γ was set to 0.85, and four clusters and their weights were obtained. The clustering results are shown in Table 10.

Table 10

Clustering results

Clusters	Number of members	Member	Cluster weight
C ¹	7	e₁, e₅, e₈, e₁₁, e₁₃, e₁₄, e₁₆	0.630
C ²	4	e₂, e₆, e₉, e₁₅	0.204
C ³	2	e₃, e₁₀	0.052
C ⁴	3	e₄, e₇, e₁₂	0.114

The initial weights of DMs were calculated by Equation (15). Combined with the decision risk coefficient, the decision risk weight of each DM was calculated by Equation (16), as shown in Table 11.

Table 11

Initial weights and decision risk weights of DMs

Member	Initial	Decision	Member	Initial	Decision
	weight	risk weight		weight	risk weight
e ₁	0.090	0.102	e ₉	0.051	0.044
e ₂	0.051	0.029	e ₁₀	0.026	0.016
e ₃	0.026	0.025	e ₁₁	0.090	0.069
e ₄	0.038	0.023	e ₁₂	0.038	0.036
e ₅	0.090	0.090	e ₁₃	0.090	0.105
e ₆	0.051	0.059	e ₁₄	0.090	0.088
e ₇	0.038	0.044	e ₁₅	0.051	0.054
e ₈	0.090	0.095	e ₁₆	0.090	0.122

Step 6. The closeness degree of each alternative was calculated by Equation (17) and the ranked results are shown in Table 12.

Table 12

Closeness degree of alternatives and the ranked results

Alternatives	$D_{l}^{+}$	$D_{l}^{-}$	D _l
x ₁	0.2695	0.1820	0.4031
x ₂	0.1970	0.1971	0.5002
x ₃	0.2191	0.1963	0.4726
Ranking	x₂ > x₃ > x₁

Table 13

Results obtained by [39]

Alternatives	$D_{l}^{+}$	$D_{l}^{-}$	D _l
x ₁	0.2096	0.1713	0.4497
x ₂	0.2155	0.2183	0.5032
x ₃	0.2474	0.2116	0.4610
Ranking	x₂ > x₃ > x₁

It can be concluded from the result that alternative x₂ should be selected to handle the accident, i.e., to continuously carry out environmental emergency monitoring, inform the fire-fighting force to evacuate the scene, and dispatch an anti-chemical team with professional equipment for search and rescue at the scene.

The results of the case analysis reflect the fact that people pay more attention to the casualties and impact on the environment. Moreover, the explosion accident had a certain impact on the public psychology, which is also an important reason for choosing the scheme with the minimum risk of casualties. Therefore, the government should also collect public opinion after the accident using reasonable means, understand the trends in public opinion to assist scientific decision-making, release the latest progress of the incident, and prevent secondary public sentiment from causing greater impact.

4.3. Comparison and discussion

To validate the effectiveness and illustrate the advantages of the proposed method, the results are compared with the expert weight approach proposed in [39]. Only the weight of internal DMs changes, as shown in Table 13; the two methods have the same ranking result.

Ref. [39] conducted a clustering analysis for large group decision-making and proposed a two-layer weight model, which holds that the weight of clustered internal members is related to a DM’s contribution to cluster consistency. In the proposed method, the adjustment of a DM’s weight is based on the group consistency. At the same time, the risk caused by the uncertainty of their preference is also considered. By measuring the reliability and accuracy of the preference information of DMs, the less-professional DMs can be effectively identified, and the determination of each member’s weight will be more reasonable. This article does not compare the methods of determining the weights of attributes. Because the determination of attribute weights in the current literature mostly uses an expert preference matrix or is subjectively given by experts, the methods are not comparable. However, in terms of ideas, the proposed method has a strong advantage in that obtaining the attributes from social media data is an objective approach.

5. Conclusion

To make use of the UGC big data generated on social media after a major emergency as well as to consider the subjective decision risk caused by both the uncertainty and the deviation of DMs’ preferences from group consistency, a method consisting of two-stage risk emergency decision-making for large groups based on the analysis of social media big data was presented in this paper. On the one hand, this method introduces the UGC data of social media into the selection of emergency decision-making schemes and uses the TF-IDF method to extract keywords to obtain public opinion information. It then evaluates the level of damage of the event to determine the attribute weights. On the other hand, considering the adjusted preferences of DMs’ behavior in an open decision-making scenario may cause decision-making risk, so a two-stage decision-making process was designed to collect the preference information of DMs, and the weights of the DMs are determined by clustering method.

Compared with the traditional method, the weighting of attributes takes into account the trends of public interest in the information, which can make up for the lack of subjectivity of experts when determining weights, increase the public’s satisfaction with the decision, and reduce the social damage that the event may risk. The extraction of keywords does not involve complicated syntax information and semantic analysis and can be completed quickly and efficiency, which can satisfy the real-time needs of emergency decision-making. At the same time, the decision-making risk is quantified through the reliability and accuracy of DM opinions, and the members who are more uncertain about their preferences or deviate from the group can be identified, which makes the decision-making result more reasonable.

Of course, the method proposed in this paper also has certain limitations. For example, the determination of the attribute weight is related to the quality of the UGC. When there is less UGC related to the event at the beginning or more noisy data in the text, this will affect the decision result. In future studies, additional risk factors that influence decision-making should be taken into account, and we will attempt to automatically extract the attribute system from UGC and combine it with a more comprehensive spatio-temporal pattern analysis to achieve a dynamic response for emergency decision-making.

Footnotes

Acknowledgments

This research is supported by grants from the National Natural Science Foundation of China (No. 71671189), the Major Project for National Natural Science Foundation of China (71790615) and the Key Project of Natural Science Foundation of China (No. 71431006), the Key program for Financial Research Institute Foundation of Wenzhou University.

References

Yu and

K.K.

Lai , A distance-based group decisionmaking methodology for multi-person multi-criteria emergency decision support, Decision Support Systems 51(2) (2011), 307–315.

J.P.

Liu ,

H.Y.

Chen ,

Xu ,

L.G

Zhou and

Z.F.

Tao , Generalized ordered modular averaging operator and its application to group decision making, Fuzzy Sets & Systems 299(C) (2016), 1–25.

Deng ,

Liu ,

Zhang ,

Deng and

Ma , A new crowdsourcing model to assess disaster using microblog data in typhoon Haiyan, Natural Hazards Journal of the International Society for the Prevention & Mitigation of Natural Hazards 84(2) (2016), 1–16.

X.T.

Guo ,

H.Z.

Wang ,

Y.Q.

Song and

Gao , Review: Brief survey of crowdsourcing for data mining, Expert Systems with Applications 41(17) (2014), 7987–7994.

Ludwig ,

Kotthaus ,

Reuter ,

van Dongen and

Pipek , Situated crowdsourcing during disasters: Managing the tasks of spontaneous volunteers through public displays, International Journal of Human-Computer Studies 102 (2017), 103–121.

K.M.

Carley ,

Malik ,

P.M.

Landwehr ,

Pfeffer and

Kowalchuck , Crowd sourcing disaster management: The complex nature of Twitter usage in Padang Indonesia, Safety Science 90 (2016), 48–61.

Pohl ,

Bouchachia and

Hellwagner , Online indexing and clustering of social media data for emergency management, Neurocomputing 172 (2016), 168–168.

Xu ,

L.Y.

Liu and

M.M.

Wang , Emergency event detection and analysis for emergence management based on cross-media analytics, Systems Engineering-Theory & Practice 35(10) (2015), 2550–2556.

Xu ,

Liu ,

Zhang ,

Luo ,

Mei and

Hu , Building the multi-modal storytelling of urban emergency events based on crowdsensing of social media analytics, Mobile Networks and Applications 22(2) (2017), 218–227.

10.

Kryvasheyeu ,

Chen ,

Obradovich ,

Moro ,

Van Hentenryck ,

Fowler and

Cebrian , Rapid assessment of disaster damage using social media activity, Science Advances 2(3) (2016), 1–11.

11.

S.Q.

Shan ,

Lin and

Zhang , Dynamic City Disaster Loss Assessment Model Based on User Generated Content, Journal of WUT (Information & Management engineering) 38(1) (2016), 505–511.

12.

Wang ,

Ye ,

Zhu and

Lee , Using social media for emergency response and urban sustainability: A case study of the 2012 Beijing rainstorm, Sustainability 8(1) (2015), 25.

13.

Y.D.

Wang ,

Li ,

Wang and

J.Q.

Zhu , The mining and analysis of emergency information in sudden events based on social media, Journal of WUT (Information & Management Engineering) 41(03) (2016), 290–297.

14.

Lin ,

K.A.

Lachlan and

P.R.

Spence , Exploring extreme events on social media: A comparison of user repost-ing/retweeting behaviors on Twitter and Weibo, Computers in Human Behavior 65 (2016), 576–581.

15.

Bai and

Yu , A Weibo-based approach to disaster informatics: incidents monitor in post-disaster situation via Weibo text negative sentiment analysis, Natural Hazards 83(2) (2016), 1177–1196.

16.

X.H.

Xu ,

Z.J.

Du and

X.H.

Chen , Consensus model for multi-criteria large-group emergency decision making considering non-cooperative behaviors and minority opinions, Decision Support Systems 79 (2015), 150–160.

17.

X.H.

Xu ,

X.Y.

Zhong ,

X.H.

Chen and

Y.J.

Zhou , A dynamical consensus method based on exit-delegation mechanism for large group emergency decision making, Knowledge-Based Systems 86 (2015), 237–249.

18.

H.J.

Zhang ,

Y.C

Dong , Herrera-Viedma

Consensus Building for the Heterogeneous Large-Scale GDM With the Individual Concerns and Satisfactions, IEEE Transactions on Fuzzy Systems 26(2) (2018), 884–898.

19.

Y.C.

Dong ,

S.H.

Zhao ,

H.J.

Zhang ,

Chiclana and

Herrera-Viedma , A self-management mechanism for non-cooperative behaviors in large-scale group consensus reaching processes, IEEE Transactions on Fuzzy Systems, in press, DOI: 10.1109/TFUZZ.2018. 2818078.

20.

G.X.

Song and

Yang , The decision-making behavior of group decision analysis, Analysis of academic exploration 13(3) (2000), 48–49.

21.

Kahneman and

Tversky , Prospect theory: An analysis of decision under risk, Econometrica 47(2) (1979), 263–291.

22.

Zhu ,

Ma ,

Wang and

Chen , Risk decision-making method using interval numbers and its application based on the prospect value with multiple reference points, Information Sciences 385 (2017), 415–437.

23.

Liu ,

Z.P.

Fan and

Zhang , Risk decision analysis in emergency response: A method based on cumulative prospect theory, Computers & Operations Research 42 (2014), 75–82.

24.

Yuan ,

Liu and

Z.P.

Fan , Risk decision making method for emergency response considering regret aversion, Systems Engineering-Theory & Practice 35(10) (2015), 2630–2636.

25.

X.H.

Xu and

Y.S.

Yang , Method of dynamic emergency decision for risk type of large group based on cumulative prospect theory, Control & Decision 32(11) (2017), 1957–1965.

26.

Liu ,

Z.P.

Fan ,

Yuan and

Li , A FTA-based method for risk decision-making in emergency response, Computers & Operations Research 42 (2014), 49–57.

27.

Tang ,

Yi ,

Yang and

Sun , Risk analysis of emergent water pollution accidents based on a Bayesian Network, Journal of Environmental Management 165 (2016), 199–205.

28.

S.L.

Yang ,

K.Y.

Zhu ,

Fu and

G.Y.

Lu , Simulation of the group decision conformity based on cellular automata model, Systems Engineering-Theory & Practice 29(9) (2009), 115–124.

29.

Chatterjee ,

Mukhopadhyay and

Bhattacharyya , Dependent judgment analysis: A Markov chain based approach for aggregating crowdsourced opinions, Information Sciences 396 (2017), 83–96.

30.

Zhang , The multiattribute group decision making method based on aggregation operators with interval-valued 2-tuple linguistic information, Mathematical and Computer Modelling 56(1) (2012), 27–35.

31.

Salton and

Buckley , Term-weighting approaches in automatic text retrieval, Information Processing & Management 24(5) (1988), 513–523.

32.

W.D.

Abilhoa and

L.N.D.

Castro , A keyword extraction method from twitter messages represented as graphs, Applied Mathematics & Computation 240(4) (2014), 308–325.

33.

A.V.

Banerjee , A simple model of herd behavior, The Quarterly Journal of Economics 107(3) (1992), 797–817.

34.

H.J.

Zhang ,

Y.C

Dong and

Chen , The 2-rank consensus reaching model in the multigranular linguistic multiple-attribute group decision-making, IEEE Transactions on Systems Man & Cybernetics Systems PP(99) (2017), 1–15.

35.

H.J.

Zhang ,

Palomares ,

Y.C.

Dong and

W.W.

Wang , Managing non-cooperative behaviors in consensus-based multiple attribute group decision making: An approach based on social network analysis, Knowledge-Based Systems (2018), in press, doi: 10.1016/j.knosys.2018.06.008.

36.

J.P.

Liu ,

Xu ,

H.Y.

Chen ,

L.G

Zhou ,

J.M

Zhu and

Z.F.

Tao , Group decision making with interval fuzzy preference relations based on DEA and stochastic simulation, Neural Computing & Applications 6 (2017), 1–12.

37.

J.F.

Pang ,

J.Y.

Liang and

Song , An adaptive consensus method for multi-attribute group decision making under uncertain linguistic environment, Applied Soft Computing 58 (2017), 339–353.

38.

X.H.

Xu and

X.H.

Chen , Research of a kind of method of multi-attributes and multi-schemes large group decision making, Journal of Systems Engineering 23(2) (2008), 137–141.

39.

X.H.

Xu ,

Sun ,

Pan and

B.S.

Liu , Two-layer weight large group decision-making method based on multi-granularity attributes, Journal of Intelligent & Fuzzy Systems 33(3) (2017), 1797–1807.