An agent-based model for emotion contagion and competition in online social media

An agent-based model for emotion contagion and competition in online social media

Abstract

Recent studies suggest that human emotions diffuse in not only real-world communities but also online social media. More and more mechanisms beyond emotion contagion are revealed, including emotion correlations which indicate their influence and the coupling of emotion diffusion and network structure such as tie strength. Besides, different emotions might even compete in shaping the public opinion. However, a comprehensive model that considers up-to-date findings to replicate the patterns of emotion contagion in online social media is still missing. In this paper, to bridge this vital gap, we propose an agent-based emotion contagion model which combines features of emotion influence and tie strength preference in the dissemination process. The simulation results indicate that anger-dominated users have higher vitality than joy-dominated ones, and anger prefers weaker ties than joy in diffusion, which could make it easier to spread between online groups. Moreover, anger’s high influence makes it competitive and easily to dominate the community, especially when negative public events occur. It is also surprisingly revealed that as the ratio of anger approaches joy with a gap less than 10%, angry tweets and users will eventually dominate the online social media and arrives the collective outrage in the cyber space. The critical gap disclosed here can be indeed warning signals at early stages for outrage controlling in online social media. All the parameters of the presented model can be easily estimated from the empirical observations and their values from historical data could help reproduce the emotion contagion of different circumstances. Our model would shed lights on the study of multiple issues like forecasting of emotion contagion in terms of computer simulations.

Emotion correlation, Emotion contagion, Tie strength, Agent-based model, Emotion competition

I Introduction

Recent years have witnessed the tremendous development of online social media. For example, Twitter attracts more than 310 million monthly active users and produces 500 million tweets every day. In online social media, the high-dimensional user-contributed content, which is a “big data” window (6), are easily to be collected and analyzed. Because of this, the growth of online social media provides an unprecedented opportunity to study human behavior with unparalleled richness and granularity.

Emotion contagion on real-world social networks is a traditional research area that has been studied by psychologist and sociologist for many years (27); (2); (22); (36). After the development of Internet, especially the explosion of online social media, researchers found that users’ emotions can also transfer to others via virtual online connections (7); (26); (12); (31); (14); (10). Kramer et al. found that sentimental states can be transferred to others via emotion contagion through a massive experiment on Facebook (32). Ferrara et al. analyzed the emotion valence of messages that Twitter users receive before they post positive or negative tweets to highlight the effect of emotion contagion (21). However, a model that simulate the emotion contagion and investigate the diffusion patterns in online social media is still missing. Aiming at filling this gap, in this paper, we propose an agent-based model to study the mechanisms of online emotion diffusion. In previous studies, users’ emotions are simply classified to positive and negative or just a score of happiness, neglecting the detailed aspect of human sentiments. However, fine-gained emotion states such as anger or sadness play essential roles in the bursts of social events like earthquake or terrorist attack. In this paper, we categorize human emotions into four categories which are anger, disgust, joy and sadness (46); (19) to systematically investigate and model the detailed mechanisms of different emotion contagion patterns.

On social networks, emotions flow with the diffusion of information, while information diffusion procedure is also influenced by user emotions behind the contents. Previous studies found that emotional tweets are easier to be retweeted and thus spread more quickly than neutral ones (41) and both positive and negative emotions might diffuse in the subsequent comments to corresponding blog entries (15). In psychology view, each emotion has three dimensions which are valence (positive–negative), arousal (passive/calm–active/excited), and tension (tense–relaxed) (45), among which both valence and arousal impact the diffusion procedure. High arousal emotions such as anger or anxiety boost diffusion more than low arousal emotions such as sadness or contentment (4); (5). For valence, researchers also found that bad is stronger than good, which implies that negative emotions have more impact than positive ones (3). Kim and Salehan also suggested that negative emotions have higher retweet performance than positive ones (30). Fan et al. found that anger, which is negative and of high arousal, is more influential than joy while sadness has extremely low correlation between connected users (19), which are consistent with the findings on the influence of valence and arousal in diffusion procedure.

Meanwhile, the dynamics beyond information diffusion can be influenced by the structure of underlying network such as the characteristics of nodes and ties. Since emotion spread is tightly coupled with the information flow in terms of retweets or comments, its contagion patterns are accordingly affected by the network features. Robert et al. found that nearly all transmission occurred between close friends, suggesting that strong ties are instrumental for spreading both online and real-world behavior in human social networks (8). Beyond that, the theory of weak ties, which has been long established, suggests that weak ties enhance diffusion by exchanging novel information across groups (25). Onnela et al. analyzed mobile communication networks and found that removal of the weak ties results in network collapse (35). In online social media, weak ties still play a crucial role in the dissemination of information (1). However, Zhao et al. suggested that weak ties can be categorized into positive ones and negative ones which can enhance and prevent the diffusion respectively (48); (47). For the coupling of emotion and tie strength, different emotions would have different preference on propagation ties, e.g., Fan et al. suggested that anger emotion prefers weaker ties compared to joy (20). In other words, joyful messages tend to be transferred between close friends while angry ones are more likely to evoke resonances of strangers. The existing findings indeed offer insightful mechanisms to establish competent contagion models for emotion in social networks.

In fact, traditional controlled experiments on emotion diffusion are extremely difficult because of the high costs and spatio-temporal limitations. While with a reliable diffusion model, researchers could study various emotion-related issues such as emotion competition and emotion dynamics in hot-event through computer simulation. However, few existing diffusion models take the influence of emotion and tie-strength preferences into accounts concurrently. In this paper, based on previous works, we proposed an agent-based model which combines the two critical features that function profoundly in the contagion process. Based on the model, we study the tie-strength preference of different emotions, emotion diffusion and how emotions compete in social network. Our model perfectly reproduces the vitality difference and the tie-strength preference of different emotions and offers an comprehensive understanding of emotion contagion and competition, and the emergence of dominated emotion that widely disseminates in the network. Our model would shed lights on study of emotion issues in terms of computer simulations.

Ii Background

In this section, we introduce our dataset and several up-to-date technologies and findings, which are the foundations of the present work. Our data set is public available and it can be downloaded freely through https://doi.org/10.6084/m9.figshare.4311920.v2.

ii.1 Dataset

Although we concentrate on the emotion contagion model, several experimental results are still necessary to define and estimate the parameters of the model. We collect over 11 million tweets posted by approximately 100,000 users over half a year in Weibo (Chinese Twitter) to perform the experiment. The following edges connected these users and construct a directed network denoted as , where and represent all the users and following edges respectively. is the set of edges’ strengths derived from common friends definition (which can be seen in II.3). To investigate the contagion mechanisms of fine-gained emotions, we employ a Bayesian classifier introduced in (46) to category tweets to four emotions which are anger, disgust, joy and sadness. The proportion of the four emotions in these tweets are illustrated in Table 1, which are important parameters in our model.

ii.2 Emotion correlation

From the view of conventional social theory, homophily is widely disclosed in social networks (33). After the explosion of online social media, researchers found that users’ physiological states are also assortative in online social networks (6); (7). Moreover, anger and joy are revealed to be significantly correlated and anger’s correlation is even higher than that of joy (20). However, disgust and sadness possess extremely low correlations. The emotion with higher correlation indicates that it is easier to affect others, which makes it an important feature of the diffusion process. In this paper, we apply the correlation calculation method on the connected users in our data and achieve similar results as in (20), as can be seen in Table 1. In our model, we apply the correlation results of four emotions to construct the model. While in the simulations, we only consider anger and joy and treat the diffusion of disgust and sadness as random reposts due to their extremely low influence.

Anger Joy Disgust Sadness
Proportion 19.2% 39.1% 13.7% 28.0%
Correlation 0.41 0.35 0.04 0.03
Table 1: Statistical results of emotion proportions and emotion correlations of anger, joy, disgust and sadness. The number of joyful messages are larger than that of others. Anger and Joy have large correlations which indicate that they are easily to influence other users.

ii.3 Tie-strength preference

Besides emotion, tie strength also has notably influence on diffusion. For example, users tend to communicate via strong ties (35), but weak ties are also irreplaceable in diffusion due to its “bridge” function between different groups. In this paper, we use the following three tie-strength definitions as in (20):

  • Common friends. More common friends of two users indicate that their relationship is stronger. Because of this, the first metric we apply is the proportion of common friends (35); (48), which is defined as for the tie between user and , where denotes the number of common friends of and . And and represent the the degree of and respectively.

  • Reciprocity. A higher ratio of reciprocity indicates more trust social connections and more significant homophily (49). Therefore, we define the proportion of reciprocal links, through which messages of a specific emotion disseminate, as a measurement of tie-strength of this emotion. Note that the result of reciprocity strength is a proportion rather than an averaged strength as that of the other two definitions.

  • Retweets. The idea of retweet strength is that larger number of retweets represents more frequent interactions. Note that, different from the previous two metrics, retweet strength evolves over time. Therefore, we count only the retweets that occurred before the relevant emotional retweet. Moreover, to smooth the comparison between anger and joy, the retweet strength (denoted as ) is normalized by , in which and separately represent the minimum and maximum values of all observations.

Based on these three metrics, researchers found that different emotions have different edge preferences in dissemination procedure (20), i.e., anger prefers weaker ties than joy. People tend to share joyful messages with their close friends, but can be stimulated by strangers’ anger in online social media. In this paper, we study the tie-strength preference by the simulation of our model and achieve similar results. Moreover, we reveal the underlying reason of different tie-strength preference of the two emotions.

ii.4 Information diffusion and emotion competition

Information diffusion is a traditional research area and several classical information diffusion models, such as linear threshold model (LT) and independent cascade model (IC) (24); (9); (29) are well-developed in previous literatures. LT model considers the weight of connection between and and generates threshold for node . becomes active if , which means the sum of weights between and its active neighbors exceeds a threshold. With the development of online social media, researchers found that information diffusion presented quite different mechanisms on this platform because of its unique pushing and republishing features (42); (48). Meanwhile, network structure such as tie strength makes significant effect on the diffusion procedure (35). Zhao. et al. proposed an information diffusion model which considers the pushing and republishing features and the tie-strength to investigate the relationship between the tie strength and information propagation in online social networks (47). Our model absorbs the advantages of previous literatures such as the pushing and republishing mechanisms of online social media, the preference of tie strength in diffusion procedure, and thresholds which are defined to determine the condition of diffusion.

As the theory of economics of attention (38); (13); (16); (18), users’ attention is a scarce resource that multiple information competes for. With the development of Internet, especially the explosion of online social media, netizens receive redundant messages from various online channels such as Facebook, Twitter and online forums, which results in extremely severe information overload problem. Because of this, information competes for users’ finite attention and the winners go vital in cyber space (23); (44). To investigate the competition mechanisms, researchers also proposed competition models (34); (43); (39); (28), e.g., Weng et al. proposed an agent-based model to study memes competition in Twitter. However, except for hashtags, topics and memes, emotions also compete for users’ limited attention and the competitive emotion would pervasively diffuse in the network. Based on this idea, we introduce a structure in our model to represent users’ limited attention and analyze the diffusion results from the perspective of emotion competition.

Iii Agent-based model

Agent-based model is widely employed in previous works to study network community, discussion process, emotion expression patterns and meme competition (37); (17); (40); (11); (43). Inspired by the model introduced in (43), we develop an emotion contagion model that combines emotion influence and tie strengths, to study emotion contagion mechanisms in social network. We simulate the diffusion process on the real-world following network and diffusion behaviors are reflected in terms of retweets which occur on connected edges. Each user has a screen which contains time-ordered messages that he/she has received. If users he/she follows post or repost messages, these messages will be pushed into the screen. In one circulation, a randomly selected user posts new messages with probability , or reposts several ones that in the screen. If the number of messages in the screen exceeds , the old messages will be removed. This process stems from the pushing and republishing mechanisms in online social media that users either post new messages or read messages he/she received and repost some of them. The critical principle of the dissemination is that if one message with emotion of significant influence and the tie strength between the two users are strong, the message will be preferentially reposted. The model is defined as follows:

  • Step 1: Randomly select a user from .

  • Step 2: posts a new message with the probability . The emotion of is decided by the statistical results of empirical data on which the probabilities (denoted as and ) of anger, disgust, joy and sadness are 19.2%, 13.7%, 39.1% and 28.0%, respectively, as can be seen in Table 1. At last, is pushed to the screens of all followers of . If ’s screen contains more than tweets, oldest tweets will be removed to narrow the screen until its size is . The procedure of pushing new messages is illustrated in Fig. 1(a).

  • Step 3: If does not post new message, he/she will read all the messages in his/her screen. When he/she reads a message that has emotion , and the tie strength of and his/her followee who posts or reposts this message is , the retweet tendency is defined as , where is the correlation of emotion . If the tendency exceeds a threshold , the message will be reposted and pushed to the screens of all followers of . We also remove oldest tweets that in ’s screen if the screen overloads. We call the above procedure republishing, which is shown in Fig. 1(b).

  • Repeat the above three steps for times.

Figure 1: Pushing and republishing procedures in our model. In (a), and follow . When posts a new tweet with probability , is put into the top of and ’s screens. (b) illustrates the republishing process. reads that in his/her screen, if the retweet tendency ( is the correlation of the emotion delivered by ) is higher than , will be reposted by and enter the screen of , who is the follower of , implying the emotion initially posted by transfers from to .

Table 2 illustrates important parameters in our model. such as and is the probability of four newly posted emotions, which can be adjusted to reflect the variation of new messages with one dominant emotion in social events. is achieved from the common friends strength definition, representing the strength of users and . represent users’ screen size, which is set to 20 in this paper because Weibo presents 10 newest tweets per page in their App. It can further be adjusted to investigate emotion contagion in different attention spans. After removing isolated nodes, we achieve 88,532 connected users which are used to construct . Because of this, the simulation steps is arbitrarily set to 8,853,200, which is 10 times of the network size and supposed to be sufficient in simulating the diffusion. We tune the threshold in the following simulation and further estimate the threshold in real online social media.

Symbol Meaning Value
The probability of posting new tweet 0.45
The probability of posting tweet with one emotion, such as See Table 1
The correlations of the four emotions See Table 1
The tie-strength between and
The size of the screen 20
The total number of steps 8,853,200
The retweet threshold
Table 2: The description of parameters of the model.

Our model considers two important features: emotion correlation and the tie-strength. Emotion correlation represents emotions’ influence. Hence emotion with high correlation such as anger is easier to be reposted. Meanwhile, as described in (35), users with stronger connections have more communications. Our model simulates the diffusion based on these two findings as described in Step 3: if the product of tie strength and is larger than a threshold, the message will be reposted. This definition combines the influence of content that is presented by emotion and the the network structure which is presented by tie-strength. The code of the model is also publicly available and it can be freely downloaded through https://doi.org/10.6084/m9.figshare.5092522.v1.

Iv Results

In this section, we first report the results of our emotion contagion model, including tie-strength preferences and user vitality difference of different emotions, estimation of parameters by comparing the simulation with empirical results, emotion diffusion and competition mechanisms.

iv.1 Tie-strength preference

Based on the three strength definitions, we calculate the average strength of ties on which anger and joy emotion transfer. Note that given the extremely low influence of disgust and sadness, which can be learnt from their poor corelations in the social network as can be seen in Table 1, only anger and joy are explored in the following simulations of emotion contagion. As can be seen in Fig. 2, all the results of the three strength definitions indicate that anger prefers transferring through weaker ties than joy. When the diffusion threshold is very small, e.g., , all messages can be retweeted regardless of the emotion and tie-strength. As a result, both of anger and joy diffuse on edges that have similar average strength. With the growth of , edges with small strength prevent some messages spreading on them, especially the ones with small correlations. On some edges, joyful messages cannot spread while angry ones could still pass. As a result, the average tie-strength for contagion of both anger and joy grows up with the growth of while their difference increases accordingly. Classical social theory suggests that weak ties play a key role in diffusion process because they always connect different communities and transfer information from one group to another. Because of this, the preference of weaker ties could make anger easier to spread widely. The preferences of tie strengths unraveled by our model are consistent with the empirical results in (20), which also suggested that anger prefers weaker ties than joy. Moreover, our model reveals where the different preferences come from. The diffusion in the network depends on the tie-strength and the influence of the contents’ emotion simultaneously. More influential emotions such as anger do not rely on ties with large strengths too much, which makes it easily spread via weaker ties than joy.

Moreover, we estimate the value of based on empirical analysis and the results are illustrated in Fig. 3. The empirical results of strength preference comes from (20), which are calculated from six months tweets posted by the network users based on the three strength definitions. As can be seen, we can draw the conclusion that the value of in real online social media ranges from 0.05 to 0.08. Our model results are assortative with observations from the real-world data and we can further estimate the threshold in Weibo, which are strong evidence of the reliability of our model.

Figure 2: Tie-strength preferences of anger and joy. In all the three strength definitions, including common friends (a), reciprocity (b) and retweets (c), anger prefers weaker ties than joy, especially when is large.
Figure 3: The tie-strength difference of joy and anger. The difference is defined as , where and are the average diffusion strengths of anger and joy respectively. The horizontal lines represent the difference calculated from realistic Weibo data (20).

iv.2 User Vitality

The user vitality, which reflects the extent of a user’s activeness in online social media, can be an excellent indicator of users’ intention in participating online activities. Here the user vitality is intuitively defined as the number of messages posted per day by a user after the registration in the empirical data. And for the simulations in the model, since all agents share the same age, the user-activity can be directly valued by the total number of posted messages by an agent. Then besides strength preferences, it is surprisingly found that in both simulation and empirical data, anger-dominated users demonstrate higher vitality than their joy-dominated counterparts, as can be seen in Fig. 4, in which anger (joy)-dominated users are those posting more angry (joyful) tweets than that of other emotions. The similarly long-tailed distribution of user vitality reproduced by our model, as shown in Fig. 4, again indicates its effectiveness in replicating emotion contagion in social media. And in fact, the consistence between the empirical data and our model further implies that being a sentiment with high arousal, anger can indeed enhance users’ activeness in online activities and make itself competent in capturing attention during the contagion.

Figure 4: The distributions of user vitality for respectively anger and joy-dominated users. (a) User vitality from the empirical data. (b) User vitality in simulations and here we set , which is close to empirical circumstance, as can be seen in Fig. 3. It is worth noting that different from the empirical data, there is an increment in vitality distribution of the model as the user vitality is very low. It comes from the configuration of the model that agents are evenly selected to publish or republish tweets at each step, which results in relatively less low-vitality users as compared to the empirical data.

iv.3 Emotion diffusion and competition

In this section, we discuss the emotion contagion patterns based on the simulation results. Fig. 5(a) illustrates the retweets proportions of anger and joy. When , tweets with all of the four emotions can be transferred via all edges and accordingly the retweets proportions of anger and joy are assortative with and , respectively. While with the growth of , angry tweets can spread on some weak ties on which joyful tweets can’t transfer and as a result, the proportion of anger retweets exceeds that of joy when . However, the proportions of all tweets decline because the number of retweets will decline when the threshold is large, as can be seen in Fig. 5(b). And the proportions of anger and joy are closest when but the number of angry tweets is still below joy because of the smaller value of . Anger-dominated users also reach a peak when . Because of the higher influence of anger, its number of retweets increases with the rising of and the total angry tweets and anger-dominated users also reach peaks at specific . However, as is much lower than , the total numbers of angry tweets and the anger dominated users are smaller than that of joy for all . From the view of emotion competition, people’s finite attention are filled with messages with different emotions. When the number of messages with one emotion increases, it becomes more competitive. Two factors that influence the amount of messages with one emotion are the probability of newly posted emotion messages and the amount of republishing messages with this emotion because both of pushing and republishing tweets will appear in followers’ screens. According to Weibo data, is much larger than , while anger has higher influence and easy to be republished which produces more angry retweets. In conclusion, anger’s higher influence can enhance its competition and makes it spread extensively from both the perspectives of messages and users.

Figure 5: The simulation results of anger and joy, in which emotion probabilities of new messages are 19.2%, 13.7%, 39.1% and 28.0% for anger, disgust, joy and sadness, respectively. These proportions are calculated from real Weibo data. In all of the three figures, x-axises represent the threshold and y-axises are the percents of retweets, tweets and users after the simulation. (a) and (b) separately illustrate the percentages of anger and joy in reposted messages and all messages including new tweets and reposted tweets. (c) is the proportion of anger- and joy-dominated users. Users who post or repost more angry (joyful) tweets than other emotions will be similarly labeled as anger(joy)-dominated users. And the number of a specific emotion dominated users reflects the diffusion scope of this emotion.

Moreover, in some circumstances, especially when negative societal events occur, people will post many angry messages in online social media and the possibility of publishing angry tweets will grow significantly. Because of this, we tune the emotion probabilities of newly posted tweets to 25% for all the four emotions to study the emotion competition from a equitable start. As can be seen in Fig. 6(b) and Fig. 6(c), the proportions of angry tweets and anger-dominated users are much more than that of joy messages. When , the proportions of angry and joyful tweets and users are similar. With grows up, the proportions of anger surpass joy and their difference in coverage reach a maximum at specific , which is about 0.04 for tweets and 0.01 for users respectively. However, when is too large, tweets with both anger and joy are hard to spread in the network, leading to a decreasing gap between them. In this simulation, joy has no superiority in the newly posted messages, and anger still has more influence. As a result, anger could dominate users’ attention, which makes it more competitive than joy. In conclusion, when negative events occur, people will post many angry tweets which are easily to influence others. As a result, angry users will dominate the whole network. It can well explain the bursty rise of negative events, in which anger dominates the public opinion (20). Because of anger’s high influence, website managers should try to prevent its extensively and unreasonably spread from the very beginning.

Figure 6: Simulation results of emotion competition. New tweets have the same probabilities (25%) to be labeled to the four emotions. (a) illustrates the percentages of angry and joyful retweets. The proportion of angry retweets increases with the growth of while of joy, the value first increases to a peak and then declines. (b) is the percentages of total tweets including new tweets and reposted tweets. The proportions of both of anger and joy increase first and then begin to drop. And the values of anger far exceed that of joy. (c) represents the anger and joy dominated users and the values of anger are still much more than that of joy.

To find the critical point that anger surpasses joy in the network, we tune the probability of newly posted anger and joy messages, denoted as and , to investigate when anger could dominate the network as approaches . As can be seen in Fig. 7, when the gap is large, e.g., , both of joyful tweets and joy-dominated users are more than angry ones. With shrinking of the gap, the number of both angry messages and anger-dominated users increases, and when the gap is smaller than 0.16 and 0.10, the proportion of angry messages and anger-dominated users exceeds that of joy messages and joy-dominated users respectively. The different critical gaps for tweets and users comes from the vitality difference, i.e., angry users demonstrate higher activeness in online activities. As a result, the number of angry messages exceeds that of joy although the number of anger-dominated users are still smaller than that of joy-dominated users. This finding demonstrate that when the proportion of newly posted angry messages is 10% less than that of joy, the network could be dominated by angry users. Network managers should be aware of that and adopt corresponding techniques to prevent the collective outrage in cyber space.

Figure 7: The proportions of angry and joy messages and anger and joy-dominated users with the reduction of . In this figure, we fix and to 0.25 and adjust . For example, when and , the gap of them is 0.10. (a) illustrate that when the gap is smaller than 0.16, the proportion of angry messages surpasses that of joy messages. (b) illustrate the proportion of anger and joy-dominated users. When the gap is smaller than 0.10, the proportion of anger-dominated users exceeds that of joy-dominated users.

V Conclusion

The explosive development of online social network produces massive data which provides an unprecedented opportunity to investigate human behavior. In this paper, we propose an agent-based model to well simulate the emotion contagion in online social media. Our model absorbs advantages by considering the pushing and republishing of online social media, the threshold which is the diffusion condition and the competition of users’ limited attention. Meanwhile, the most innovative point of our model is that we combine the influence of content which is presented by emotion and the network feature which is tie-strength. The primary idea of this combination is that higher emotion influence and stronger ties enhance diffusion in social network. By analyzing the simulation result, we find that as compared to joy, angry users possess higher vitalities and anger can spread on weaker ties which means it could widely diffuse across communities. These findings are consistent with previous empirical observations on Weibo. But our model further disclose the origin of this phenomenon which is high influence makes anger easily transfer via weak ties. Our model can also be used to study the emotion contagion, e.g., from the simulation results, we find that higher influence of anger can enhance its competition. Moreover, when the probability of newly posted angry tweet is high, i.e., in circumstances of negative societal events, anger will influence massive users and even dominate the network due to its higher influence. We also give a guide value that when the proportion of newly posted anger messages is 10% less than that of the joy, the network could be dominated by negative emotion. This is a warning signal in early stages, i.e., in some societal events, anger emotion could widely propagate and lead to the collective online rage.

Our model demonstrate that anger’s high influence can enhance its dissemination, and in some conditions, it can even dominate the network. Network managers should realize anger’s capability of being highly contagious and apply techniques to try to prevent the unreasonable prevalence of anger. Our emotion contagion model grabs multiple critical features of online diffusion and can be used to study multiple issues of emotion diffusion and even forecasting by adjusting parameters. For example, we simulate the diffusion of collective social event by changing the probabilities of newly posted emotions. Researchers can also study the influence of intention by adjusting the size of the screen.

Acknowledgements.
This work was supported by NSFC (Grant No. 71501005) and the fund of the State Key Lab of Software Development Environment (Grant Nos. SKLSDE-2015ZX-05 and SKLSDE-2015ZX-28). R. F. also thanks the Innovation Foundation of BUAA for PhD Graduates.

References

  1. Eytan Bakshy, Itamar Rosenn, Cameron Marlow, and Lada Adamic. The role of social networks in information diffusion. In Proceedings of the 21st International Conference on World Wide Web, WWW ’12, pages 519–528, New York, NY, USA, 2012. ACM.
  2. Sigal G. Barsade. The ripple effect: Emotional contagion and its influence on group behavior. Administrative Science Quarterly, 47(4):644–675, 2002.
  3. Roy F Baumeister, Ellen Bratslavsky, Catrin Finkenauer, and Kathleen D Vohs. Bad is stronger than good. Review of general psychology, 5(4):323–370, 2001.
  4. Jonah Berger. Arousal increases social transmission of information. Psychological Science, 22(7):891–893, 2011.
  5. Jonah Berger and Katherine L. Milkman. What makes online content viral? Journal of Marketing Research, 49(2):192–205, 2012.
  6. Catherine A. Bliss, Isabel M. Kloumann, Kameron Decker Harris, Christopher M. Danforth, and Peter Sheridan Dodds. Twitter reciprocal reply networks exhibit assortativity with respect to happiness. Journal of Computational Science, 3(5):388 – 397, 2012. Advanced Computing Solutions for Health Care and Medicine.
  7. Johan Bollen, Bruno Gonçalves, Guangchen Ruan, and Huina Mao. Happiness is assortative in online social networks. Artif. Life, 17(3):237–251, 2011.
  8. Robert M Bond, Christopher J Fariss, Jason J Jones, Adam DI Kramer, Cameron Marlow, Jaime E Settle, and James H Fowler. A 61-million-person experiment in social influence and political mobilization. Nature, 489(7415):295–298, 2012.
  9. Wei Chen, Yajun Wang, and Siyu Yang. Efficient influence maximization in social networks. In Proceedings of the 15th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD ’09, pages 199–208, New York, NY, USA, 2009. ACM.
  10. Anna Chmiel, Julian Sienkiewicz, Mike Thelwall, Georgios Paltoglou, Kevan Buckley, Arvid Kappas, and Janusz A Hołyst. Collective emotions online and their influence on community life. PloS ONE, 6(7):e22207, 2011.
  11. Anna Chmiel, Pawel Sobkowicz, Julian Sienkiewicz, Georgios Paltoglou, Kevan Buckley, Mike Thelwall, and Janusz A. Holyst. Negative emotions boost user activity at BBC forum. Physica A: Statistical Mechanics and its Applications, 390:2936–2944, 2011.
  12. Lorenzo Coviello, Yunkyu Sohn, Adam DI Kramer, Cameron Marlow, Massimo Franceschetti, Nicholas A Christakis, and James H Fowler. Detecting emotional contagion in massive social networks. PloS ONE, 9(3):e90315, 2014.
  13. Matthew B Crawford. The world beyond your head: On becoming an individual in an age of distraction. Macmillan, 2015.
  14. Linh Dang-Xuan and Stefan Stieglitz. Impact and diffusion of sentiment in political communication-an empirical analysis of political weblogs. In ICWSM, 2012.
  15. Linh Dang-Xuan and Stefan Stieglitz. Impact and diffusion of sentiment in political communication – an empirical analysis of political weblogs, 2012.
  16. Thomas H Davenport and John C Beck. The attention economy: Understanding the new currency of business. Harvard Business Press, 2013.
  17. F. Ding and Y. Liu. Modeling opinion interactions in a bbs community. The European Physical Journal B, 78(2):245–252, 2010.
  18. Josef Falkinger. Attention economies. Journal of Economic Theory, 133(1):266 – 294, 2007.
  19. Rui Fan, Jichang Zhao, Yan Chen, and Ke Xu. Anger is more influential than joy: Sentiment correlation in weibo. PloS ONE, 9(10):e110184, 2014.
  20. Rui Fan, Jichang Zhao, and Ke Xu. Higher contagion and weaker ties mean anger spreads faster than joy in social media. ArXiv e-prints, August 2016.
  21. Emilio Ferrara and Zeyao Yang. Measuring emotional contagion in social media. PloS ONE, 10(11):e0142390, 2015.
  22. James H Fowler and Nicholas A Christakis. Dynamic spread of happiness in a large social network: longitudinal analysis over 20 years in the framingham heart study. BMJ, 337, 2008.
  23. Michael H. Goldhaber. The attention economy and the net. First Monday, 2(4), 1997.
  24. Mark Granovetter. Threshold models of collective behavior. American Journal of Sociology, 83(6):1420–1443, 1978.
  25. Mark S Granovetter. The strength of weak ties. American journal of sociology, pages 1360–1380, 1973.
  26. Jamie Guillory, Jason Spiegel, Molly Drislane, Benjamin Weiss, Walter Donner, and Jeffrey Hancock. Upset now?: Emotion contagion in distributed groups. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, CHI ’11, pages 745–748, New York, NY, USA, 2011. ACM.
  27. Elaine Hatfield, John T Cacioppo, and Richard L Rapson. Emotional contagion. Current directions in psychological science, 2(3):96–100, 1993.
  28. Brian Karrer and M. E. J. Newman. Competing epidemics on complex networks. Phys. Rev. E, 84:036106, Sep 2011.
  29. David Kempe, Jon Kleinberg, and Éva Tardos. Maximizing the spread of influence through a social network. In Proceedings of the Ninth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD ’03, pages 137–146, New York, NY, USA, 2003. ACM.
  30. Dan Kim and Mohammad Salehan. The effect of sentiment on information diffusion in social media. 2015.
  31. Adam D.I. Kramer. The spread of emotion via facebook. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, CHI ’12, pages 767–770, New York, NY, USA, 2012. ACM.
  32. Adam D.I. Kramer, Jamie E. Guillory, and Jeffrey T. Hancock. Experimental evidence of massive-scale emotional contagion through social networks. Proceedings of the National Academy of Sciences, 111(24):8788–8790, 2014.
  33. McPherson Miller, Smith-Lovin Lynn, and Cook James, M. Birds of a feather: Homophily in social networks. Annual Review of Sociology, 27:415–444, 2001.
  34. Mehdi Moussaid, Dirk Helbing, and Guy Theraulaz. An individual-based model of collective attention. Proceedings of the European Conference on Complex Systems, 2009.
  35. J-P Onnela, Jari Saramäki, Jorkki Hyvönen, György Szabó, David Lazer, Kimmo Kaski, János Kertész, and A-L Barabási. Structure and tie strengths in mobile communication networks. Proceedings of the National Academy of Sciences, 104(18):7332–7336, 2007.
  36. J Niels Rosenquist, James H Fowler, and Nicholas A Christakis. Social network determinants of depression. Molecular psychiatry, 16(3), 2011.
  37. F. Schweitzer and D. Garcia. An agent-based model of collective emotions in onlinecommunities. The European Physical Journal B, 77(4):533–545, 2010.
  38. Herbert A Simon. Designing organizations for an information-rich world. 1971.
  39. Kim Sneppen, Ala Trusina, Mogens H. Jensen, and Stefan Bornholdt. A minimal model for multiple epidemics and immunity spreading. PLOS ONE, 5:1–7, 10 2010.
  40. P. Sobkowicz and A. Sobkowicz. Dynamics of hate based internet user networks. The European Physical Journal B, 73(4):633–643, 2010.
  41. Stefan Stieglitz and Linh Dang-Xuan. Emotions and information diffusion in social media—sentiment of microblogs and sharing behavior. Journal of Management Information Systems, 29(4):217–248, 2013.
  42. Eric Sun, Itamar Rosenn, Cameron Marlow, and Thomas Lento. Gesundheit! modeling contagion through facebook news feed. In International AAAI Conference on Web and Social Media, 2009.
  43. Lillian Weng, Alessandro Flammini, Alessandro Vespignani, and Fillipo Menczer. Competition among memes in a world with limited attention. Scientific reports, 2(335), 2012.
  44. Fang Wu and Bernardo A. Huberman. Novelty and collective attention. Proceedings of the National Academy of Sciences, 104(45):17599–17601, 2007.
  45. Wilhelm Max Wundt. Outlines of psychology. W. Engelmann, 1907.
  46. Jichang Zhao, Li Dong, Junjie Wu, and Ke Xu. Moodlens: an emoticon-based sentiment analysis system for Chinese tweets. In Proceedings of the 18th ACM SIGKDD international conference on knowledge discovery and data mining, pages 1528–1531. ACM, 2012.
  47. Jichang Zhao, Junjie Wu, Xu Feng, Hui Xiong, and Ke Xu. Information propagation in online social networks: a tie-strength perspective. Knowledge and Information Systems, 32(3):589–608, 2012.
  48. Jichang Zhao, Junjie Wu, and Ke Xu. Weak ties: Subtle role of information diffusion in online social networks. Phys. Rev. E, 82:016105, Jul 2010.
  49. Yu-Xiao Zhu, Xiao-Guang Zhang, Gui-Quan Sun, Ming Tang, Tao Zhou, and Zi-Ke Zhang. Influence of reciprocal links in social networks. PloS ONE, 9(7):e103007, 2014.
Comments 0
Request Comment
You are adding the first comment!
How to quickly get a good reply:
  • Give credit where it’s due by listing out the positive aspects of a paper before getting into which changes should be made.
  • Be specific in your critique, and provide supporting evidence with appropriate references to substantiate general statements.
  • Your comment should inspire ideas to flow and help the author improves the paper.

The better we are at sharing our knowledge with each other, the faster we move forward.
""
The feedback must be of minimum 40 characters and the title a minimum of 5 characters
   
Add comment
Cancel
Loading ...
122561
This is a comment super asjknd jkasnjk adsnkj
Upvote
Downvote
""
The feedback must be of minumum 40 characters
The feedback must be of minumum 40 characters
Submit
Cancel

You are asking your first question!
How to quickly get a good answer:
  • Keep your question short and to the point
  • Check for grammar or spelling errors.
  • Phrase it like a question
Test
Test description