Human-Centric Cyber Social Computing Model for Hot-Event Detection and Propagation

Microblogging networks have gained popularity in recent years as a platform enabling expressions of human emotions, through which users can conveniently produce contents on public events, breaking news, and/or products. Subsequently, microblogging networks generate massive amounts of data that carry opinions and mass sentiment on various topics. Herein, microblogging is regarded as a useful platform for detecting and propagating new hot events. It is also a useful channel for identifying high-quality posts, popular topics, key interests, and high-influence users. The existence of noisy data in the traditional social media data streams enforces to focus on human-centric computing. This paper proposes a human-centric social computing (HCSC) model for hot-event detection and propagation in microblogging networks. In the proposed HCSC model, all posts and users are preprocessed through hypertext induced topic search (HITS) for determining high-quality subsets of the users, topics, and posts. Then, a latent Dirichlet allocation (LDA)-based multiprototype user topic detection method is used for identifying users with high influence in the network. Furthermore, an influence maximization is used for final determination of influential users based on the user subsets. Finally, the users mined by influence maximization process are generated as the influential user sets for specific topics. Experimental results prove the superiority of our HCSC model against similar models of hot-event detection and information propagation.


Human-Centric Cyber Social Computing Model for Hot-Event Detection and Propagation
Lei-Lei Shi, Lu Liu , Member, IEEE, Yan Wu, Liang Jiang , Muhammad Kazim ,

Haider Ali, and John Panneerselvam
Abstract-Microblogging networks have gained popularity in recent years as a platform enabling expressions of human emotions, through which users can conveniently produce contents on public events, breaking news, and/or products.Subsequently, microblogging networks generate massive amounts of data that carry opinions and mass sentiment on various topics.Herein, microblogging is regarded as a useful platform for detecting and propagating new hot events.It is also a useful channel for identifying high-quality posts, popular topics, key interests, and high-influence users.The existence of noisy data in the traditional social media data streams enforces to focus on human-centric computing.This paper proposes a human-centric social computing (HCSC) model for hot-event detection and propagation in microblogging networks.In the proposed HCSC model, all posts and users are preprocessed through hypertext induced topic search (HITS) for determining high-quality subsets of the users, topics, and posts.Then, a latent Dirichlet allocation (LDA)-based multiprototype user topic detection method is used for identifying users with high influence in the network.Furthermore, an influence maximization is used for final determination of influential users based on the user subsets.Finally, the users mined by influence maximization process are generated as the influential user sets for specific topics.Experimental results prove the superiority of our HCSC model against similar models of hot-event detection and information propagation.
Index Terms-Event detection, event propagation, human centric, social computing.

I. INTRODUCTION
C YBER technology with its wide range of computing and communication devices has a significant impact on our daily life.Cyber-enabled online services [1]- [7] have become an integral part of daily activities.Microblogging is one of the popular cyber-enabled online services, which aid sharing and disseminating information anytime and anywhere [8]- [11].Being an information sharing platform [9], [12]- [14], microblogging also attracts many users on social media to establish friendships [15], exchange ideas, and to promote products.As a consequence, such activities in microblogging generate an enormous amount of data [16] with rich semantic content and structure.The availability of such data has created an opportunity for deeper research investigations in the fields of topic detection [1], [2], influence maximization [17], and information diffusion [18].Hotevent detection and propagation in social media are one of the recently emerging research areas, which predominantly focus on identifying highly influential users in microblogging networks.High-influence users can be exploited to achieve wider and quicker dissemination of hot events.Hot-event detection can be utilized in a wide range of applications such as product promotion [3], friend recommendation [19], and rumor control [20].
Specifically, microblogging hot-event detection and propagation emphasize finding most relevant influential speakers in order to propagate popular topics to enhance the effectiveness of hot-event dissemination.An efficient hot-event detection and propagation should necessarily incorporate the following three steps.

A. Choice of Posts
Microblogging generates a large number of posts daily.The ability to find high-quality posts is one of the most important requirements in the hot-event dissemination process.Selection of these high-quality posts can exert better influence during the hot-event detection and propagation.

B. Choice of Topics
Usually, the information generated in microblogging can be clustered under different topics.Determination of a suitable topic for every piece of information is another significant requirement in the process of hot-event dissemination.Selecting the most appropriate topic areas for information can drastically improve the dissemination effect.

C. Spreaders of Topics
Furthermore, spreaders in the network usually play a significant role during the dissemination of hot events.Usually, the most influential users with direct access to relevant target receivers are regarded as the best spreaders in the network.Highly influential spreaders utilizing the inherent power of diffusion of social networks can potentially improve the efficiency of the hot-event detection and propagation.
With this in mind, this paper proposes a human-centric social computing (HCSC) model for hot-event detection and propagation to mine the most influential spreader sets for popular topics.The proposed model carefully incorporates the aforementioned three integral requirements to achieve the maximum possible dissemination effect.The proposed HCSC model contains the following four decisive stages.First, all posts and users in the microblogging network are preprocessed using hypertext induced topic search (HITS) to identify highquality subsets of users, topics, and posts.Second, a latent Dirichlet allocation (LDA)-based multiprototype user topic detection method is applied to recognize the users with high influence under popular topics.Such identified users are used as the initial set of core users for hot-event detection and propagation.Third, a two-stage topic popularity-based influence maximization model is deployed for determining the most influential users based on the user subsets.The first stage mines users with the highest topic hub value.The second stage uses this input data to spread event information by the topic popularity-based information propagation model and then mines users with the greatest influence under popular topics.Finally, the users mined by the two-stage process are used as the influential user sets for specific topics.Important contributions of this paper are summarized as follows.
1) We propose a probabilistic latent semantic analysis (PLSA)-based hot-event detection model named HEE-PLSA that detects hot events based on popular topics.HEE-PLSA filters topic, post, and user information to guarantee high-quality social media big data.HEE-PLSA applies influence maximization based on topic popularity to improve the accuracy and quality of the event detection process for efficient soft computing.
The proposed approach creates a smaller, high-quality training dataset by selecting high-quality posts and/or users for popular user interests to reduce the impact of general user interests, garbage posts, and ordinary users.An LDA-based multiprototype user topic detection method called multiprototype-LDA (MP-LDA) is developed.The posts and users in each interest carry various weights to describe their degree of representation.The proposed MP-LDA model enables each topic to be represented by more than one post and a single user.The centrality value of posts and users is used for calculating prototype weights, while user topic similarity is utilized to help in partitioning the user's topics in the microblogging network.2) Based on the above user topic discovery, a two-stage user topic popularity-based hot-event propagation model is proposed to further reduce the time required by the classic greedy algorithm to complete rapid mining of the proper number of k key users.The first stage mines users with the highest user topic influence statically, and the second stage uses the users identified from the first stage to spread information based on the user topic popularity-based information propagation model, and then mines users with the largest influence increment greedily.The number of key users is initially allocated in proportion to the size of the topic while mining the key users for each topic in parallel.During the mining process for key users, a high-influence greedy strategy is adopted to form an alternative set of key users and from which an optimization based on submodularity is used to select the key users whom can exert the largest effect.3) Experiments conducted demonstrate the effectiveness of the proposed HCSC model in significantly improving the efficiency of hot-event detection and propagation when compared to the existing state-of-the-art methods presented in [1], [8]- [10], and [21]- [23].The rest of this paper is organized as follows.Section II reviews the existing works in event detection and propagation.Section III presents our proposed HCSC model.Experimental results are discussed in Section IV, and Section V concludes this paper.

II. RELATED WORK
The problem of efficient hot-event detection and propagation in social networks is gaining considerable attention in recent years.In general, existing works of event detection and propagation in the social network can be classified into the threshold and independent cascade (IC) models [17], [24], [25].These models simulated the NP-hard problem of influence diffusion as a discrete optimization problem and used a greedy algorithm to determine an approximate solution.In these approaches, the probability of each user being influenced usually increases with their corresponding number of influenced neighbors.A greedy algorithm [24] has been shown to perform considerably better than most of the generous algorithms, random algorithms, and the degree of the centricity-based algorithms.The performance of the former greedy algorithm has been claimed to achieve a near optimal solution within the range of 1 − 1/e.However, the deployment of the greedy algorithm identified three important problems [17].First, the algorithm runs longer, as it purely adopts a greedy guiding approach in each step in the global network.Furthermore, this method needs the implementation of the Monte Carlo method in every simulation graph of the global network based on the size of the sphere of influence set.It is worth noting that modestly sized networks with hundreds of nodes would require a lengthy computation time, so that the implementation of the greedy algorithm is not practically feasible for networks with thousands or millions of nodes.Second, it does not exploit the network topology properties such as the power-law distribution of nodes or community structure.Moreover, it does not incorporate the interactive information features such as social factors which differentiate between nodes and inconsistencies between node relationships.Third, not considering the influence of overlapping nodes is a noted drawback of the greedy approach.Each step in the greedy algorithm selects a local optimal node, which may not be necessarily the global optimal node.Moreover, this approach of the greedy algorithm might lead to a final node selection, consisting only of locally influential nodes and not globally influential nodes.Such drawbacks of the greedy algorithm necessitate further research to improve the quality of event detection [25].
Sohn et al. [26] developed an optimized algorithm to improve operational efficiency in the event detection process.This algorithm reduced the duration of the core-node mining process; however, the influence of the core node has been lower.Nguyen et al. [24] presented a new heuristic algorithm, which individually considered the propagation probabilities of nodes in the network along with the effect of multihop neighbors.Subsequently, it achieved a higher influence spread.Moreover, this work proposed a realistic network model with nonuniform propagation probabilities between nodes.However, both these methods do not explicitly consider the popularity and quality of users, which makes it hard to propagate the hot events to suitable users.
Liu et al. [17] investigated the influence of maximization and proposed a probabilistic solution.Sepehr and Beigy [25] proved that selecting the most influential nodes is an NP-hard problem.This work developed an efficient viral cascade probability estimation (VICE) method that leverages a special importance sampling approximation to achieve high accuracy even in the case of a small probability of influence.Aslay et al. [27] formulated the problem of selecting influential individuals as an optimization problem and developed an algorithm for a diffusion model and improved the efficiency and scalability of the influence maximization algorithms.Zhou et al. [18] used node topic distribution to estimate the node activation probability at the topic level and proposed an influence maximization algorithm called TopicRank to mine the influential node set under specific topic based on event probability.Some other researchers [1], [20], [22] have also developed algorithms based on this work and improved the time efficiency of mining influential users.
However, in these studies, the influence of a single user on all other users is the same, i.e., the activation probability of a user to activate other nodes always stays constant.This is not applicable to the real-world scenario where an individual may be considered highly influential among other peers when discussing the subject economy but completely unknown by peers in the area of sport.In other words, it is rare for any individual to be considered as an expert in multiple fields.Thus, the influence of a user in a social network is likewise related to both the user and the topic, while the influence of a given user is different for different topics.A few other works proposed in [22], [28], and [29] suffer common limitation, such that they only considered the topic influence on user activation probability and ignored the popularity degree of the topic, links between posts, and the diffusion power of users.Consequently, such drawbacks reflected in their lower efficiency and the improper number of final mining core users.
Few research studies integrated topic popularity degree scoring, user topic detection, and influence maximization to enhance the efficiency of event detection, while enlarging the sphere of influence of hot-event propagation.Each of the aforementioned studies has not considered the importance of hot-event topic information content and user preferences.The literature in the context of event detection demonstrates that the influence of users in the social network is related to the relationship between users and topic, while the influence of the same user is dissimilar under different topics [30].Furthermore, these studies in [11], [23], [31], and [32] focus on the effect of topic influence on user activation probability and ignored utilizing the popularity of topics, links between posts, and diffusion power of the users.This leads to the waste of users influence resulting in lower efficiency and the improper number of final mining core users.
To address this, we propose an efficient human-centric soft computing model for hot-event detection called HCSC that includes a PLSA-based hot-event detection model HEE-PLSA, an LDA-based multiprototype user topic detection method, and a two-stage topic popularity-based influence maximization for final determination of influential users based on user subsets.We efficiently and accurately mine an appropriate number of most influential spreader sets for hot-event propagation, and ordinary users are removed during the selection process.Moreover, there is no need to predefine a proper number of topics manually, as our approach automatically detects hot events and identifies influential spreaders under popular topics.

III. HUMAN-CENTRIC CYBER SOCIAL COMPUTING MODEL
This section details our proposed HCSC model for event detection and propagation.

A. HEE-PLSA Model
The proposed HEE-PLSA model selects the number of popular topics using the TD-HITS [8] approach, while hot events are detected according to the PLSA [28] model ensuring high-quality posts and high-influence users under popular topics.
HITS creates a smaller high-quality training dataset by extracting high-quality posts and influential users from a large pool of posts and users with high efficiency and accuracy.Then, we use a topic-decision method to determine the appropriate number of topics and to discover key posts from a large number of posts.Our proposed HEE-PLSA model exploits the TD-HITS method to identify high-influence users during hotevent detection.Fig. 1 demonstrates the process of hot-event detection and identification of high-influence users.

B. LDA-Based Multiprototype User Topic Detection Model
Existing event detection models hardly distilled the popular topics, which results in low quality of posts and users being  discovered under popular topics.Therefore, topic filtering method [8] is essential for determining the importance of users under popular topics.
The MP-LDA model is composed of two modules.After identifying the topic scores using the TD-HITS method, the topics are clustered by using the LDA topic model in the postnetwork and then the Gibbs sampling [8], [9] method is used for user topics detection.Then, the initial influential spreaders in influence maximization are discovered based on their hub value in the microblogging network and their local features in the user-user network using TS-LDA [8].Fig. 2 illustrates the process involved in user topic detection and discovery of initial influential spreaders.

C. Two-Stage User Topic Influence Maximization Model
The two-stage user topic influence maximization model consists of the following two stages.First, a topic popularitybased information propagation based on IC model [17].Second, a topic popularity-based influence maximization method to determine the proper number of spreaders set with the size of K for popular topics from the microblogging network through which the hot events can be propagated to more users.Fig. 3 depicts the process involved in user topic detection and discovery of influential spreaders in user topics.
Definition 1 (Microblogging Network G): Let V be the set of users in a topic popularity-based microblogging network, G, E denotes the set of edges in G, and T shows the set of topics in G.Then, the topic popularity-based microblogging network can be expressed as G(V,E,T), where V = {V 1 , V 2, …V n }, |V | = n, and E = {< u, v >|u, v ∈ V }, a link between u and v; it shows an edge between them; otherwise, T = {1, 2, …, z} and |T | = z.Definition 2 (Sphere of Influenceδt) : If S is the set of users with K size, then δ(S|t) denotes the sphere of influence of users set S for topic t, i.e., the number of users activated after S propagates information about topic t.

D. Mining High-Influence Users Under Popular Topics
In the initial stage, we choose the initial influential spreaders while not considering the information propagation characteristics of the microblogging network.The second stage uses the spreaders obtained from the first stage to spread information, and then iteratively mine top-k spreaders with the largest topic influence increment among the remaining influential users in a greedy way.In other words, an increment in the sphere of influence value of the spreader set after adding a spreader u is ensured to achieve the maximum effect.Therefore, the topic popularity-based influence maximization can be described as follows: given a microblogging network G(V,E,T), a parameter K , and a topic t; then, the goal is to find a user set S(|S| = K ) in G that can result in a maximum value of δ(S|t), i.e., maxS ( Topic popularity-based influence maximization tends to find a high-influence user set with K and t from G(V,E,T) through which the information can be propagated to more high-influence users.In order to discover these users, we first model the social network G(V,E) using the topic model HEE-PLSA and obtain the topics T and topic distribution of the users ψ T Then, according to the obtained topic popularity-based microblogging network G(V,E,T), a topic popularity-based influence maximization algorithm is proposed.
In the influence maximization, the selection of the initial propagation users depends on the information propagation trend and the sphere of influence.High-quality posts can draw more attention from high-influence users spreading and/or broadcasting such posts in the microblogging network.Developing the ability to identify influential spreaders effectively and efficiently in the microblogging network is a major challenge.A large number of important evaluation methods have been proposed to address this problem such as degree centrality [33], clustering coefficient centrality [34], and betweenness centrality [8], [26].However, degree centrality and clustering coefficient centrality of spreaders can only characterize the local information of networks.The complexity of computing betweenness centrality is very high due to the need to calculate the shortest path.Tweet composition between users and posts in the microblogging network cannot be directly analyzed by the application of these centrality methods alone.Other related theories describe that high-influence users are responsible for a large number of high-quality posts.We propose a novel solution to this problem.We believe that the initial influential spreader should satisfy a global condition and a network topology condition within a certain period, i.e., a high global feature and a high local feature.Here, we expect that high local feature users will trigger an early and rapid accumulation of contagious transmissions among a large number of candidate users.Finally, the global importance of users can be acquired by the hub value of users along with their posts.Subsequently, high global feature users will spread information, ideas, or rumors much faster than ordinary users.
In general, a user with more neighbor connections is highly likely to be an initial influential spreader in microblogging networks.Using this as inspiration, we propose a novel influence measure by considering the hub value and the degree value of user interaction.Moreover, the method proposed in this paper improves the accuracy of the identification of initial influential spreaders that can further improve the accuracy of the identification of final influential spreaders and the influential scope of influence maximization.

E. Topic Popularity-Based Hot-Event Information Propagation Model
This section explains the traditional information propagation, i.e., IC model [17] and our proposed topic popularitybased information propagation model.
IC is the most widely adopted information propagation model used to simulate information propagation in social networks.In the IC model, an active node can activate and/or inactive neighbor nodes in accordance with the activation probability between them.A higher activation probability depicts the level of ease in activating the nodes.
The activation probability in the IC model is generated randomly.However, the activation probability of nodes is related to the social relationships among the nodes and topics in the process of information propagation.It is worth noticing that nodes have a different activation probability for different topics.Therefore, we use a topic popularity-based information propagation model to calculate the node activation probability P t u,v for specific topics to simulate the information propagation in social networks in a more realistic way.P t u,v is influenced by social connections between the nodes.Closely related to the social connections between nodes, the greater connection times imply a more intimate relationship between nodes and have a higher activation probability.Therefore, user intimacy can be used to represent the degree of intimacy between nodes.Moreover, P t u,v is also significantly affected by the user's topic popularity.Information is usually quickly propagated between two highly popular users.Therefore, the topic popularity can influence P t u,v of two users.Definition 3 (User Intimacy):User intimacy, C u,v , denotes the frequency of connection between nodes u and v.It can be obtained from the ratio of the connection times of u and v to the connection times of u, and other nodes are given as follows: where R u,V i shows the connection times of nodes u and V i , and R u,v represents the connection times of nodes u and v. Definition 4 (Topic Popularity): Topic popularity TP T u,v shows the popularity degree of the two user's topic calculated as follows: where Authority T u,v denotes the authority of a key post T , Authority max represents the biggest authority of key posts, Concisely, P t u,v is influenced by the user intimacy C u,v and TP T u,v .Thus, P t u,v of a user u to v for specific t can be calculated as follows: The propagation process of the topic popularity-based information propagation (TPIP) model is similar to the IC model.Each user has only a chance to activate their neighbor users, while the user's activation process is independent of each other.The difference is that the user's activation probability of TPIP is dissimilar under different topics, which is more in line with the information propagation of microblogging networks.

IV. EXPERIMENTS AND RESULTS
We evaluate the efficiency of our proposed HCSC model against four existing topic models, including PLSA [28], LDA [8], bursty event dEtection (BEE) [23], and efficient eVent dEtection (EVE) [10] to compare the efficiency of the HCSC model.We generated our data sets from Twitter (http://twitter.com/)via TwitterAPI consisting of 1 500 000 posts and 36 052 users.

A. Popular Topics Detection
Fig. 4 shows the authority value and minimum distance of each post.Fig. 4 can be used to identify the proper number of topics, located in the upper right quadrant, which plays a key role in the spreading of hot events under a chosen specific topic community.Table I shows high-quality posts ranked by their minimum distance.It is evident from Table I that our proposed HCSC model can detect the top-k high-quality posts according to their authority value efficiently and effectively.The parameter k is set to 10 according to the number of posts, with the authority value being far greater than others in the right upper quadrant in Fig. 4. At the same time, it can be seen from Table II that the key users under popular topic communities can also be detected by our model, which further improves the efficiency of hot-event detection and guarantees the high quality of data for efficient soft computing.

B. Initial Influential Spreaders Analysis
The degree and hub values of users for topics according to the results achieved from the HEE-PLSA model can help distinguish the importance of users under each popular topic, which is better than the BEE and EVE models, as listed in Table III.We can also determine the number of influential users for each popular topic from Table IV by setting a different number of initial influential users.The proper number of initial influential spreaders can also be discovered in Fig. 5, which is better than the baseline approaches.With an increase in the number of initial influential users, the sphere of influence reaches 3266 when the number of initial influential users is 8 and saturates then after.Hence, the appropriate

C. Final Sphere of Influence
The contrast of the final sphere of influence aids influential spreader discovery.It can be noticed from Table VI that the final sphere of influence of the proposed HCSC model is approximately the same or even better than the IC model.Thus, the influential spreaders discovered by our HCSC model is more appropriate compared to the IC model.This is because of the fact that the activation probability of the nodes considers the topic popularity and intimacy between the nodes in our proposed model.

D. Influence Maximization Precision and Efficiency Analysis
In order to verify the sphere of influence and operation efficiency of the proposed HCSC model, we run all the baseline algorithms under the same experimental setup and computing platform.Each experiment is repeated five times, and average values are considered for evaluation.We then compare the similarity of top K users with those discovered by the other two models along with the run time of the studied algorithms under specific topics.The experimental results are summarized in Table VII; for K = 10, all the top ten users detected by two models are almost the same.The HCSC model considers the impact of the popularity subsequently; the number of influential users under a specific topic is more accurate than the IC model.The HCSC model selects enough nodes with high topic diffusion power as the influential nodes, such that their sphere of influence nearly covers the entire topic area.However, the IC model does not consider these parameters and further produces an improper number of selected nodes under specific topics.In addition, it can be observed from Table VIII that the HCSC model characterizes the shortest running time with an appropriate number of influential nodes.The HCSC model prepossesses data to remove users who might not affect the results and to directly mine users characterizing the highest topic diffusion power.Thus, the HCSC performs faster than the IC model.In summary, the proposed HCSC model is superior compared to the IC model in terms of time efficiency while possessing the same sphere of influence.

E. Event-Detection Precision and Efficiency
To compare the precision and efficiency of our HCSC model with PLSA, LDA, BEE, and EVE models, we performed experiments and calculated the precision as follows: where a represents the number of detected events matching real-life events and b shows the total number of the events detected by the same algorithm, while K is the number of possible events.
Table IX shows the precision results of the five methods (HCSC, PLSA, LDA, BEE, and EVE) and Table X summarizes their time efficiency.The HCSC model can find six events when K = 8.However, the performance of the PLSA, LDA, BEE, and EVE models are very similar when K > 6 such that these models can detect all the hot events only by artificial selection.If K = 1 − 4, some hot events would remain undetected.If the K = 4 is greater than 8, e.g., 10, these models can detect all of the events but their time complexity quite high.Thus, our proposed HCSC model is accurate and efficient and outperforms the baseline state-of-the-art models.

V. CONCLUSION
This paper proposed an efficient HCSC model for hotevent detection and propagation.HCSC integrated with the PLSA-based hot-event detection model can create a highquality training dataset from a large collection of users and posts containing noise.Moreover, our approach automatically detects the correct number of topics while efficiently identifying event-related key posts with high precision.Furthermore, it also detects critical events by analyzing the number of popular user topics and determines the influential spreaders linked to them using the LDA-based multiprototype user topic detection method.The HCSC approach utilizes both posts' and users' information which potentially enables a better understanding in a timely and accurate manner of the users involved in critical incidents by deploying the twostage user topic popularity-based hot-event propagation model.In addition, noisy posts and ordinary users are effectively removed from the selection, thus eliminating the need to predefine the proper number of topics manually, as our proposed model can effectively detect hot events while identifying the influential spreaders under popular topics.Experimental results demonstrated the effectiveness and efficiency of our HCSC model against five related approaches developed for guarantying the quality of posts, users, and popular topics discovery.Particularly, our proposed model excels in filtering out noisy data existing in the social network during hotevent detection and propagation.Our proposed model suffers limitations when there are changes in the user topic and behavior evolution during the event propagation.Therefore, future work will focus on user behavior analysis and predictive analytics for further improving the efficiencies of our proposed hot-event detection model.

Fig. 4 .
Fig. 4. Number of topics from the HCSC model.

TABLE I MINIMUM
DISTANCE AND AUTHORITY OF POSTS

TABLE III DEGREE
AND HUB VALUES OF TOP TEN INFLUENTIAL USERS UNDER TOPICS

TABLE IV NUMBER
OF USERS IN POPULAR TOPICS Fig. 5. Proper number initial influential spreaders.

TABLE V TOP
TEN INITIAL INFLUENTIAL SPREADERS MINING AND RELATION WITH POPULAR TOPICS number of popular topics is set to 4, i.e., (K = 4).It verifies the correctness of the appropriate number of popular topics identified in Table IV based on the key posts.The top ten initial influential spreaders and the popular topics they belong to are listed in Table V, which plays an important role in a specific topic.

TABLE VIII COMPARISON
OF TIME EFFICIENCY IN INFLUENCE MAXIMIZATION METHOD TIME (MINUTES)

TABLE IX COMPARISON
OF PRECISION IN HOT-EVENT DETECTION

TABLE X COMPARISON
OF TIME EFFICIENCY IN HOT-EVENT DETECTION METHOD TIME (MINUTES)