Social media, regarded as two-layer networks consisting of users and items, turn out to be the most important channels for access to massive information in the era of Web 2.0. The dynamics of human activity and item popularity is a crucial issue in social media networks. In this paper, by analyzing the growth of user activity and item popularity in four empirical social media networks, i.e., Amazon, Flickr, Delicious and Wikipedia, it is found that cross links between users and items are more likely to be created by active users and to be acquired by popular items, where user activity and item popularity are measured by the number of cross links associated with users and items. This indicates that users generally trace popular items, overall. However, it is found that the inactive users more severely trace popular items than the active users. Inspired by empirical analysis, we propose an evolving model for such networks, in which the evolution is driven only by two-step random walk. Numerical experiments verified that the model can qualitatively reproduce the distributions of user activity and item popularity observed in empirical networks. These results might shed light on the understandings of micro dynamics of activity and popularity in social media networks.
Citation: Zhang P, Li M, Gao L, Fan Y, Di Z (2014) Characterizing and Modeling the Dynamics of Activity and Popularity. PLoS ONE 9(2): e89192. doi:10.1371/journal.pone.0089192
Editor: Santo Fortunato, Aalto University, Finland
Received: October 22, 2013; Accepted: January 16, 2014; Published: February 25, 2014
Copyright: © 2014 Zhang et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This work is supported by the Fundamental Research Funds for the Central Universities under Grant Nos. 2012RC0707 and 2012JBM067, National Natural Science Foundation of China under Grant Nos. 61174150, 71101009 and 61374175, NCET-09-0228 and the Doctoral Fund of the Ministry of Education (20110003110027). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
In recent years, social media networks, vital platforms for sharing contents with others in the era of Web 2.0, such as YouTube, Facebook, Delicious, Amazon, Flickr and Wikipedia, to name just a few, have experienced explosive growth , . These systems record the fingerprints of every user's activity and every item's popularity, providing “a wealth of data” to study the dynamics of human activity and item popularity at the global system scale. In particular, it is found that the probability distributions of the activity degree of users, e.g., editing in Wikipedia , voting in News2  as well as favorite marking in Flickr , and the popularity degree of items, e.g., the number of fans a photo has in Flickr , follow a power law. The power law distributions are explained by the rich-get-richer mechanism , , which is also called preferential attachment in the field of complex networks –. However, how these two distributions arise simultaneously due to human activity has yet to be determined.
The activity dynamics , – and popularity dynamics – have been investigated in the literatures, respectively. However, human activity and item popularity, two perspectives of the cross links between users and items, are interdependent; therefore, we can not study the dynamics of one aspect alone. In addition, individuals are always embedded in a social network. It is widely believed that information can spread quickly along social links using user-to-user exchanges, also known as “word-of-mouth” exchanges; moreover, the users' behaviors are strongly influenced by their neighbors –. In particular, the social degree and the activity degree depend on each other . Hence, it is considered worthwhile studying social networks to obtain deeper insights into the dynamics of human activity and item popularity. Until now, there has been no clear picture as to how online human activity and item popularity coevolve, so it is crucial to investigate the evolution of empirical human activity and item popularity as well as the theoretical model to obtain a better understanding of the possible generic laws governing the formation of activity distribution and popularity distribution.
In this paper, we first characterize the evolution of human activity and item popularity in the Amazon, Flickr, Delicious and Wikipedia networks. It is found that in such social media networks, both relative probabilities of users creating cross links and items acquiring cross links are proportional to the degree of activity and degree of popularity, respectively. In particular, the inactive users are more likely to trace popular items than the active users. Based on empirical observations, we then propose an evolving model based on two-step random walk. Finally, we justify the validity of our model by comparing the results of model with that of empirical networks. This work could shed light on the understanding of evolution of user activity and item popularity in social media networks, and it also could be helpful in certain applications, such as designing efficient strategies for virtual marketing and network marketing, etc.
Materials and Methods
Data description and notations
The Delicious data set was downloaded from http://data.dai-labor.de/corpus/delicious/, and consists of 132,500,391 bookmarks, 50,221,626 URLs (books), and 947,835 users between September, 2003 and December 31, 2007 . The Amazon user-movie rating data set was obtained from Stanford Large Network Dataset Collection (http://snap.stanford.edu/data/web-Amazon.html) . The data consists of 7,911,684 ratings, 267,320 movies and 759,899 users between August 1997 and October 2012. The Flickr data set was collected by daily crawling Flickr over 2.5 million users from Nov 2, 2006 to Dec 3, 2006, and again daily from February 3, 2007 to May 18, 2007 (http://socialnetworks.mpi-sws.org/datasets.html). Here, we only considered the users who had at least one favorite photo. With this constraint, there are 497,937 users, 11,232,836 photos and 34,734,221 favorite-markings in the data. The Wikipedia (English) data set was download from http://konect.uni-koblenz.de/networks/edit-enwiki. The data set consists of 21,416,395 articles written collaboratively by 3,819,691 volunteers around the world before September, 2010. The four datasets consist of individuals and items, such as movies in Amazon, URLs (books) in Delicious, photos in Flickr, and articles in Wikipedia. Moreover, users are able to show interest in these items using the network feature of rating in Amazon, bookmarking in Delicious, favorite-marking in Flickr, and editing in Wikipedia. Therefore, these systems are topologically equivalent. For analysis purposes, the user-item data can be mapped into a two-layer network, as shown schematically in Fig. 1. This network has two types of nodes: users and items totally. In principle, the individuals are embedded in a social network. For example, Flickr and Delicious allow users to make friends. Therefore, there should be two types of links: the cross links between users and items as well as the social links among users.
For these two types of links, we define three types of degrees. For example, U4 in the network has social degree (), and the activity degree (); I4 has the popularity degree (). Please note that there is no social links in some cases, such as Wikipedia.
Mathematically, the topology shown in Fig. 1 can be characterized by two matrices. , an adjacency matrix, represents the social links among users, with element if user declares user as his friend, otherwise . Similarly, , an adjacency matrix, characterizes the cross links, with element if user is interested in the item , otherwise . To be specific, we defined the following types of degrees to characterize the multi-relational connections. Two degrees are related to the cross links: (1) the activity degree: , i.e., the number of items interested by user ; (2) the popularity degree , i.e., the number of users who are interested in the item , which reasonably represents the popular extent in the network; and (3) the social degree , i.e., the number of friends for a given user. Note, and are two different perspectives of the cross links connecting users and items.
Measuring preferential attachment
Here, we explain the method for measuring the phenomenon of preferential attachment on temporal data , . The basic idea is to investigate whether new links are likely to attach to nodes with larger degree (size). We calculate the empirical value of the relative probability that a new cross link formed within a short period connects to a user (item), which has a degree of at the time , as follows, (1)
Here, is the degree at time . is the number of nodes with exact degree at , but creating (acquiring) new cross links within next small interval (e.g., one day in this article). is the number of users (items) with degree at . The preferential attachment hypothesis states that the rate with which a node with links acquires new links is a monotonically increasing function of , namely . To obtain a smooth curve from noisy data, we take the cumulative function form instead of : (2)
In our measurement, can be either degree of activity or degree of popularity . This method has been successfully used to verify the preferential attachment mechanism of BA model  in empirical evolving networks , , ,  and theoretical models .
Measuring relative contributions ratio
To measure the relative contribution ratio within a small interval (e.g., one day in this article), we extend the method proposed in the reference  as follows. Absolute contribution from the users with degree is measured simply by a percentage of new cross links created by these users within a short period out of the total number of cross links at time , (3)where is the number of new links, which are created by the users with degree and attached to the items with degree , within the period (e.g, one day in this paper). is the number of cross links attached to the items with degree at time . In order to observe the differences of users' activity with different degree , we present a more detailed breakdown of this absolute contribution by calculating the percentage of new cross links, which are from the users with degree and attach to the items with degree , out of all links attached to these items at time , and then normalized by the absolute contribution of these users, namely relative contribution ratio, (4)
Empirical analysis to temporal data
As shown in Fig. 1, Amazon, Flickr, Delicious and Wikipedia are typical social media networks consisting of users and items such as movies, photos, books, articles, etc (see Materials and Methods for data description and notations). Social media networks are more complicated than the networks with one type of links in previously studies –, including single node networks and bipartite networks, due to their multiplex nodes and multi-relations. Basic statistical properties for each data set are shown in Table 1. The degree of activity and degree of popularity follow an approximately power law distribution , , . In particular, the social degree follows a power law distribution in Flickr . In the following, we report the main findings of our empirical analysis of the Amazon, Flickr, Delicious and Wikipedia networks. We pay particular attention to the evolution of activity degree and popularity degree in these four networks.
Like many other complex networks, the growths of these four networks involve two major factors: adding new nodes and generating new links. Here we pay particular attention to the formation of new links during the evolution of networks, because this is the central process in which users can exchange information with each other. In the following, we focus on how the existing states of users and items affect the formation of new links and what encompasses the differences between various users' interests.
First, we examine the phenomenon of preferential creation on the existing users and preferential attachment on the existing items in these four data sets. To this end, we employ a numerical method, proposed to test preferential attachment (see Materials and Methods for more details), to investigate how the generation of new cross links depends on the existing degrees in the temporal data sets.
Figure 2 shows the cumulative function with respect to the degree of activity and degree of popularity. We see that the relative cumulative probability () for users (items) to create (acquire) cross links is proportional to the existing degree of activity (popularity). In particular, the cumulative functions approximately follow a straight line on the log-log scale, indicating that the relative cumulative probability of generating new degrees satisfies a power law with respect to the existing degrees, which can be characterized by the positive exponent where with denoting the degree. In Table 2, we list the characteristic exponents and determined by least-square fitting the functions for small as the curves deviate from the straight line for large due to low statistics. The positive exponents and indicate that the active users (with a higher degree of activity) have greater chance to create new cross links than the inactive users (with a lower degree of activity), while the popular items (with a higher degree of popularity) have greater chance to attract new cross links.
The cumulative functions of relative probability and , characterizing the influence of current activity degree and popularity degree on the formation of cross links between existing users and existing items respectively, where the time interval is one day. (a) Amazon for the snapshot of 01/06/2012, (b) Flickr for the snapshot of 01/03/2007, (c) Delicious for the snapshot of 01/12/2007 and (d) Wikipedia for the snapshot of 01/09/2010. In the insets, are the relative probability of existing users creating cross links to new items, and are the relative probability of existing items acquiring cross links from new users. Through this paper, the results of empirical analysis are for the same snapshot, and the similar patterns are also observed in other snapshots of the data. The straight lines are guide to the eye.
As these four systems expand rapidly, we then investigate the formation of cross links between new users (items) and existing items (users). In the insets of Fig. 2, characterizes the relative probability that the existing users are interested in new items with respect to the users' degree of activity, whereas characterizes the relative probability that the existing items attract the attentions of new users with respect to the items' degree of popularity. Interestingly, as seen in the insets of Fig. 2, these cumulative functions also follow a power law. The positive exponents and indicate that the newly created items are more likely to attract the attentions of active users, while the new users are more likely to be interested in popular items. The above results suggest that the users are likely to trace popular items overall, and that the active users are more likely to create new cross links than the inactive users.
What is the influence of activity (popularity) degree on the intensity of users tracing popular items (items attracting attentions of users)? To attack this problem, we classify the users and items into different groups according to their activity degree and popularity degree. Then, we investigate the cumulative functions of relative probability and for different group of users and items, respectively. As seen from Fig. 3, the slops of inactive users (with smaller ) tracking items look qualitatively larger than those of active users. For instance, in Wikipedia, the slope is 2.2 for , while it is 1.75 for . This indicates that the inactive users more severely trace popular items than the active users. Moreover, as seen in the insets of Fig. 3, the slops of unpopular items (with smaller ) attracting users look slightly larger than those of popular items, indicating that the unpopular items attract a greater interest amongst active users. Please note that the differences between different groups of users or items are smaller for Amazon than for the other three networks. This may be due to the different spreading modes of items such as movies. For instance, a popular movie is similar to the well-known global information in the Amazon user-movie network, but there is no such counterpart in the Flickr user-photo, Delicious user-book and Wikipedia author-article networks.
The cumulative function of relative probability for different groups of users and for different groups of items (in the insets), in Amazon(a), Flickr (b), Delicious (c) and Wikipedia (d). The users and items are classified into different groups according to the degree of activity and the degree of popularity.
To provide an additional evidence for the different intensity of users tracing popular items, we also calculate the relative contribution of users with activity degree , who create some new cross links to items with popularity degree within one day (see Materials and Methods for the detail). Ideally, if the intensities of users tracing popular items are identical, the relative contribution ratios should be always equal to 1 for all group of users. As seen in Fig. 4, the relative contribution of active users (with larger ) to unpopular items (with smaller ) is larger than 1 but is smaller than 1 for popular items (with larger ), indicating that the active users make higher contributions to unpopular items than average but lower to popular items than average. Meanwhile, the inactive users exhibit the opposite behavior with the exception of Amazon. Especially in Flickr and Wikipedia, it is obviously found that increases for the most inactive users, while decreases for the most active users with respect to popularity degree. Based on Eq. (5), the cumulative function for the inactive users will increase more faster than that for the active users, indicating that the slop for inactive users is larger than that for active users (as shown in Fig. 3). Furthermore, it is found that medium active users within Wikipedia, e.g., , make almost equivalent contributions to articles having different degree of popularity (as shown in Fig.4 (d)), indicating that they may not care about the article's popularity when they edit them. These results also indicate that the inactive users are more likely to trace popular items than the active users, in agreement with previous observations.
(a) Amazon, (b) Flickr, (c) Delicious and (d) Wikipedia. The users are classified into following types according to different degrees of activity, i.e., , , , and . The values of fluctuate significantly for large due to low statistics.
To further understand the mechanisms governing the evolution of real networks, we attempt to set up a theoretical model for user-item networks. Our primary goal is to qualitatively reproduce the human activity and the item popularity observed in the four empirical networks previously mentioned. In the above numerical analysis, the rich-get-richer phenomenon has been observed in the growth of the user's degree of activity and the item's degree of popularity, i.e., the active users (the popular items) have a higher probability of creating (acquiring) new cross links.The mechanism of preferential attachment has successfully explained the rich-get-richer phenomenon in previous works –, but it implicitly requires global information, e.g. degree of all nodes. However, it is impossible for individuals to collect global information in real social systems. Therefore, this only gives a macroscopic explanation of how a user's degree of activity and an item's degree of popularity evolve. Moreover, the formation of cross links will simultaneously affect the activity degree and popularity degree. This poses an interesting question: What is the microscopic mechanism governing the growth of a user's activity degree and an item's popularity degree while giving rise to the various distribution observed?
There are two crucial questions to be considered. First, how are the users activated to create new cross links? It is very difficult to formulate the users' behaviors because human dynamics are very complicated due to the inherent diversity of real world circumstances. In the empirical analysis, the users with a larger degree of activity are more likely to create cross links. Moreover, it is found that the degree of activity is positively correlated to the social degrees . In addition, the users' activities are influenced by their neighbors' activities as the information can spread along social links by user-to-user exchanges –. Therefore, we believe that the users with more friends are frequently activated as receiving more information from neighbors, and the users interested in more items are easily activated because they are very sensitive to stimuli. Hence, we employ the random walk, starting from one user and via either social or cross links, to select users, who will actively create cross links. The users with more friends and more items have a greater chance of being reached by random walk. It is found that random walk might be one possible micro mechanism governing the evolution of social networks, , and is equivalent to preferential attachment from a macro perspective , .
The second question to consider is: How do the activated users access the items? In Flickr, it has been found that over 80% of new social links are formed between friends' friends and over 50% of new cross links are formed between one user and his friends' favorite photos . Moreover, the probability of a user favorite-marking one photo increases with the number of his friends who have favorite-marked the photo , indicating that the user is influenced by his friends and reaches the photo via his friends. We therefore assume that the users access the items via their friends by two-step random walk process. It is also worth noting that the popular items are exposed to more users, so they have a higher probability of being reached by the random walk process than the unpopular items.
For simplicity, we made the following assumptions in our model: (1) the users can befriend other users (see the example of Fig. 1), (2) the activated users are selected by random walk either via social or cross links, and (3) new links (except the first links attached to newly added users and items) are formed between one user and one of his second neighbors (either users or items). In this way, the link growth process can be understand as two-step random walks via the social links or via the cross links. One way to model this network is to select the activated users who will then actively create new links. Another is to select the target nodes (either users or items) that will passively acquire new links. In our numerical simulation, we employ a two-step random walk for simplicity.
Numerically, the topology evolves according to the following rules: (1) the initial network consists of a few users () and items (). The users form a small random social network, while the items are randomly rated by the users. (2) At each time step, one new user is added into the system and randomly connected to one user and one of items rated by the user that is being connected to. Meanwhile, new items are added into the system, each of which is rated by one activated user selected by two-step random walk. (3) At each time step, users are activated by using a two-step random walk, and each of them connects to his second neighbors by a two-step random walk via common friends or via common items if they are not previously connected. (4) At each time step, users are activated by a two-step random walk, and each of them connects to one of items rated by his friends by a two-step random walk if he has not previously rated it.
We carried out numerical simulations to validate the model. We set the parameters as follows: and . We then ran the simulation up to . Based on the simple assumptions of random walk, the model can reproduce the distributions of activity degree and popularity degree observed in the above four empirical networks. In Fig. 5, we compare the degree distributions of the model with those of the empirical networks. It is found that the distributions of the model networks are qualitatively consistent with their counterparts of the empirical networks. Though the distributions of popularity degree in the model do not quantitatively match those of empirical networks as shown in Figs. 5 (c) and (d), the slops are consistent with each other. In the insets of Figs. 5 (c) and (d), we compare the distributions of larger degree of popularity of the model with those of the empirical networks. It is observed that they are approximately consistent with each other. These results indicate that our assumptions are reasonable that the users are activated by a two-step random walk, and subsequently find items of interest to them in the same manner.
Distributions of activity degree and popularity degree . (a) Amazon, where , and . (b) Flickr, where , and . (c) Delicious, where , and . The inset is the distribution of degree of popularity for . (d) Wikipedia, where , and . The inset is the distribution of degree of popularity for .
Figure 6 displays the influences of the parameters on the distributions. From Fig. 6 (a), we can see that the ratio between the number of new cross links and the number of new items has an obvious influence on the distribution of popularity degree; however, the distributions of activity degree are greatly affected by the parameter as seen in Fig. 6 (b). Furthermore, the distributions of activity degree depend on the number of new social links to some extent, while the distributions of social degree are almost independent of the number of new cross links and the number of new items . In particular, the distributions of social degree follow a power law, which is qualitatively consistent with that of Flickr networks.
In this study, we first carried out an empirical analysis to four empirical networks: Amazon, Flickr, Delicious and Wikipedia. Our study revealed the growth patterns of the users' degrees of activity and items' degrees of popularity within these networks, both of which follow the law of the rich-get-richer. It was found that the users are likely to trace popular items, but the intensities are different for the users with different activity degrees. For example, the active users make a greater than average contributions to the unpopular items whereas the inactive users make a greater than average contributions to the popular items. Motivated by the empirical findings in these four networks, we proposed an evolving model based on a two-step random walk, which is able to qualitatively reproduce the activity and popularity distributions observed in empirical networks. Based on both the empirical analysis and the theoretical model, we believe that the information spreading amongst individuals, which is simplified as a two-step random walk in the model, could represent one possible micro mechanism governing the dynamic evolution of human activity and item popularity. Of course, the dynamics of human activity and item popularity are very complicated due to the inherent diversity of human behaviors and the varying nature of items. Hence, there may be other microscopic mechanisms governing the dynamics of human activity and item popularity.
It should be noted that the results of our model are only qualitatively consistent with the empirical results. The quantitative mismatch is due to the simplifications in our model. For example, the users are only activated by a two-step random walk, and then reach the items by another two-step random walk via friends. In reality, the situation could be much more complicated. In the empirical networks, for example, besides the stimulus of neighbors, the occasional events can inspire users to participate in network-relating events. For instance, the users can access photos through various other channels in Flickr, such as the list of interesting photos provided by the web site, the search engine, the links between similar photos and so on. Furthermore, an item's particular attributes, such as being an award-winning picture, may disproportionally affect how quickly an item's popularity changes. These factors have remarkable influence on the growth of human activity and item popularity. We believe that if we consider more realistic factors in the model, we can improve the performance of our model and obtain more helpful insights in understanding the dynamics of human activity and item popularity. These problems deserve further investigations in the future.
Conceived and designed the experiments: PZ MHL LG YF ZRD. Performed the experiments: PZ MHL LG. Analyzed the data: PZ MHL LG. Contributed reagents/materials/analysis tools: PZ MHL LG. Wrote the paper: PZ MHL.
- 1. Kaplan AM, Haenlein M (2010) Users of the world, unite! The challenges and opportunities of Social Media. Business Horizons 53: 59–68. doi: 10.1016/j.bushor.2009.09.003
- 2. Kietzmann JH, Hermkens K, McCarthy IP, Silvestre BS (2011) Social Media? Get serious! Understanding the functional building blocks of Social Media. it Business Horizons 54: 241–251. doi: 10.1016/j.bushor.2011.01.005
- 3. Muchnik L, Pei S, Parra LC, Reis SDS, Andrade JS Jr, et al. (2013) Origins of power-law degree distribution in the heterogeneity of human activity in social networks,. Sci. Rep 3: 1783. doi: 10.1038/srep01783
- 4. Cha M, Mislove A, Gummadi KP (2009) A Measurement-driven Analysis of Information Propagation in the Flickr Social Network. WWW '09 Proceedings of the 18th international conference on World wide web, pp 721–730.
- 5. Simon HA (1955) On a class of skew distribution functions. Biometrika 42: 425. doi: 10.1093/biomet/42.3-4.425
- 6. Price DS (1976) A general theory of bibliometric and other cumulative advantage processe. J. Am. Soc. Inf. Sci 27: 292. doi: 10.1002/asi.4630270505
- 7. Albert R, Barabási A-L (2002) Statistical mechanics of complex networks. Rev. Mod. Phys 74: 47–97. doi: 10.1103/revmodphys.74.47
- 8. Newman MEJ (2003) The Structure and Function of Complex Networks. SIAM Rev 45: 167–256. doi: 10.1137/s003614450342480
- 9. Boccaletti S, Latora V, Moreno Y, Chavez M, Hwang D-U (2006) Complex networks: Structure and dynamics. Phys. Rep 424: 175–308. doi: 10.1016/j.physrep.2005.10.009
- 10. Barabási A-L (2005) The origin of bursts and heavy tails in human dynamics. Nature 435: 207. doi: 10.1038/nature03459
- 11. Castellano C, Fortunato S, Loreto V (2009) Statistical physics of social dynamics. Rev. Mod. Phys 81: 591–646. doi: 10.1103/revmodphys.81.591
- 12. Yasseri T, Sumi R, Kertész J (2012) Circadian Patterns of Wikipedia Editorial Activity: A Demographic Analysis. PLoS ONE 7(1): e30091. doi: 10.1371/journal.pone.0030091
- 13. Zhou T, Kiet HAT, Kim BJ, Wang B-H, Holme P (2008) Role of activity in human dynamics. EPL 82: 28002. doi: 10.1209/0295-5075/82/28002
- 14. Zhao Z, Cai S, Huang J, Fu Y, Zhou T (2012) Scaling behavior of online human activity. EPL 100: 48004. doi: 10.1209/0295-5075/100/48004
- 15. Ratkiewicz J, Fortunato S, Flammini A, Menczer F, Vespignani A (2010) Characterizing and Modeling the Dynamics of Online Popularity. Phys. Rev. Lett 105: 158701. doi: 10.1103/physrevlett.105.158701
- 16. Wu F, Huberman BA (2007) Novelty and collective attention. Proc. Natl. Acad. Sci. U.S.A 104: 17599. doi: 10.1073/pnas.0704916104
- 17. Dezso Z, Almaas E, Lukács A, Rácz B, Szakadát I, et al. (2006) Dynamics of information access on the web. Phys. Rev. E 73: 066132. doi: 10.1103/physreve.73.066132
- 18. Szabo G, Huberman BA (2010) Predicting the popularity of online content. Communications of the ACM 53: 80–88. doi: 10.1145/1787234.1787254
- 19. Crane R, Sornette D (2008) Robust dynamic classes revealed by measuring the response function of a social system. Proc. Natl. Acad. Sci. U.S.A 105: 15649. doi: 10.1073/pnas.0803685105
- 20. Eom Y-H, Fortunato S (2011) Characterizing and Modeling Citation Dynamics. PLoS ONE 6(9): e24926. doi: 10.1371/journal.pone.0024926
- 21. Weng L, Flammini A, Vespignani A, Menczer F (2012) Competition among memes in a world with limited attention. Sci. Rep 2: 335. doi: 10.1038/srep00335
- 22. Yan Q, Wu L (2012) Impact of Bursty Human Activity Patterns on the Popularity of Online Content. Discrete Dyn. Nat. Soc 2012: 872908. doi: 10.1155/2012/872908
- 23. Salganik MJ, Dodds PS, Watt DJ (2006) Experimental study of inequality and unpredictability in an artificial cultural market. Science 311: 854–856. doi: 10.1126/science.1121066
- 24. Onnela J-P, Reed-Tsochas F (2010) Spontaneous emergence of social influence in online systems. Proc. Natl. Acad. Sci. USA 107: 18375–18380. doi: 10.1073/pnas.0914572107
- 25. Lorenz J, Rauhut H, Schweitzer F, Helbing D (2011) How social influence can undermine the wisdom of crowd effect. Proc. Natl. Acad. Sci. USA 108: 9020–9025. doi: 10.1073/pnas.1008636108
- 26. Centola D (2010) The Spread of Behavior in an Online Social Network Experiment. Science 329: 1194–1197. doi: 10.1126/science.1185231
- 27. Bond RM, Fariss CJ, Jones JJ, Kramer ADI, Marlow C, et al. (2012) A 61-million-person experiment in social influence and political mobilization. Nture 489: 295–298. doi: 10.1038/nature11421
- 28. Wetzker R, Zimmermann C, Bauckhage C (2008) Analyzing social bookmarking systems: A del.icio.us cookbook. In Proceedings of Mining Social Data Workshop, collocated with ECAI 2008, pp 26–30.
- 29. McAuley J, Leskovec J (2013) From Amateurs to Connoisseurs: Modeling the Evolution of User Expertise through Online Reviews. WWW '13 Proceedings of the 22nd international conference on World Wide Web, pp 897–908.
- 30. Newman MEJ (2001) Clustering and preferential attachment in growing networks. Phys. Rev. E 64: 025102 (R).. doi: 10.1103/physreve.64.025102
- 31. Jeong H, Néda Z, Barabási AL (2003) Measuring preferential attachment in evolving networks. Europhys. Lett 61: 567–572. doi: 10.1209/epl/i2003-00166-9
- 32. Barabási AL, Albert R (1999) Emergence of Scaling in Random Networks. Science 286: 509–512. doi: 10.1126/science.286.5439.509
- 33. Wei T, Li M, Wu C, Yan X, Fan Y, et al. (2013) Do scientists trace hot topics? . Sci. Rep 3: 2207. doi: 10.1038/srep02207
- 34. Li M, Zou H, Guan S, Gong X, Li K, et al. (2013) A coevolving model based on preferential triadic closure for social media networks. Sci. Rep 3: 2512. doi: 10.1038/srep02512
- 35. Li M, Gao L, Fan Y, Wu J, Di Z (2010) Emergence of global preferential attachment from local interaction. New Journal of Physics 12: 043029. doi: 10.1088/1367-2630/12/4/043029
- 36. Saramaki J, Kaski K (2004) Scale-free networks generated by random walkers. Physica A 341: 80–86. doi: 10.1016/j.physa.2004.04.110