Patterns of users/communities/media in the Web: list of papers

Hi,

i want to share a list of papers that are observing user traces in the Web and extract patterns from these. You can add your/other papers by commenting + i am updating the list. The citation style is not consistent and i'm not guaranteeing that citations are 100% correct (pages, proceedings, etc). If you use them please check and let me know if anything is wrong.

I add as well some comments about the papers.

The last but not least, good luck with your research.

• 2006
• Danyel Fisher, Marc Smith, and Howard T. Welser. 2006. You Are Who You Talk To: Detecting Roles in Usenet Newsgroups. In Proceedings of the 39th Annual Hawaii International Conference on System Sciences - Volume 03 (HICSS '06), Vol. 3. IEEE Computer Society, Washington, DC, USA, 59.2-. DOI=10.1109/HICSS.2006.536 http://dx.doi.org/10.1109/HICSS.2006.536
  
  The authors define a question-answer person  that is characterized by responding to those with low out-degree and infrequently responding to those with higher degree. Depending on a forum topic, the behavior of its users differentiate. Forums devoted to hobby topics admit outsiders though they have cores. While in flame forums people are replying only to well-connected users.
• Ralf Klamma, Marc Spaniol, and Dimitar Denev. PALADIN: A Pattern Based Approach to Knowledge Discovery in Digital Social Networks. In Proceedings of I-KNOW ’06, 6th International Conference on Knowledge Management, Graz, Austria, September 6 - 8, 2006, J.UCS (Journal of Universal Computer Science) Proceedings, pages 457–464. Springer, 2006.
  
  Focus of this paper is to create a pattern language that describe patterns of digital traces in Web. After analyzing mailing lists, authors find patterns like no answering person, spammer and troll based on patterns they described with the language.
• 2009
• R. Dean Malmgren, Jake M. Hofman, Luis A.N. Amaral, and Duncan J. Watts. 2009. Characterizing individual communication patterns. In Proceedings of the 15th ACM SIGKDD international conference on Knowledge discovery and data mining (KDD '09). ACM, New York, NY, USA, 607-616. DOI=10.1145/1557019.1557088 http://doi.acm.org/10.1145/1557019.1557088
  
  The authors define a model that describe communication patterns (e-mail communication of students); find that users do not change their behaviors over semesters (their belonging to initial clusters are stable) while they differentiate between each other. 
• Meeyoung Cha, Juan Antonio Navarro Perez, and Hamed Haddadi Flash Floods and Ripples: The Spread of Media Content through the Blogosphere. In Proc. of the AAAI Conference on Weblogs and Social Media (ICWSM) Data Challenge Workshop, San Jose, May 2009
  
  The paper investigates patterns of content diffusion by investigating blog networks. Most of blogs (70%) produces only 30 % of the content. While other blogs serve like mass media. These are community-edited blogs, spammed blogs, microblogging blogs and content aggregators. Two types of content spreading was defined: flash floods like news which quickly distribute but did not appear later and ripples which can be discovered a long time after appearing, like songs.
  
• Fabrício Benevenuto, Tiago Rodrigues, Meeyoung Cha, and Virgílio Almeida. 2009. Characterizing user behavior in online social networks. In Proceedings of the 9th ACM SIGCOMM conference on Internet measurement conference (IMC '09). ACM, New York, NY, USA, 49-62. DOI=10.1145/1644893.1644900 http://doi.acm.org/10.1145/1644893.1644900
  
  User activities with social networks reveal more often user activities in SN. Moreover sequence of activities were investigated as well to find out when and why users perform some actions, like viewing pictures. 
• 2012
• Claudia Wagner, Matthew Rowe, Markus Strohmaier, and Harith Alani What Catches Your Attention? An Empirical Study of Attention Patterns in Community Forums. . ICWSM, The AAAI Press,
  
  The study focuses on LDA (Latent Dirichlet Allocation) application on investigating forum texts. The authors consider user, focus, post, title and community features to find the most relevant features for each community. They discover differences in features' significance depending on communities. For example, in communities with no specific topics a post should be short to increase a probability to be answered. While in communities with specific topics newbies get replies quiet often.
• Hillmann, R.; Trier, M. Dissemination Patterns and Associated Network Effects of Sentiments in Social Networks 2012 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM) 
  
  The authors consider Web communities according to dynamic network motif analysis. They find network patterns like reciprocity (bi-directional connection) and transitivity (a friend of my friend becomes my friend) in BBC and Digg forum discussions. But emotions, positive and negative, expressed in textual messages have no impact on these classical network patterns.
• Cha, M., Benevenuto, F., Haddadi, H., and Gummadi, K.  The World of Connections and Information Flow in Twitter. Systems, Man and Cybernetics, Part A: Systems and Humans, IEEE Transactions on 42, no. 4 (2012): 991–998. 
  
  The authors divide Twitter users onto three groups: grassroots, evangelists and mass media. The division was done according to the indegree of the users. After that, these users were described, e.g. reciprocity, appearance in events, etc. One of the statement of the paper is that evangelists extend the reach of mass media up to 25 %. 
• 2013
• Kizilcec, René F., Piech, Chris, and Schneider, Emily. Deconstructing disengagement: analyzing learner subpopulations in massive open online courses. 2013. http://doi.acm.org/10.1145/2460296.2460330.
  
  The main focus of the paper to find patterns of learner engagement in Massive Open Online Courses. Authors use demographic, forum activity, overall experience and some other information for detecting clusters and explain user disengagement by consideration of particular classes.
• 2014
• Ashton Anderson, Daniel Huttenlocher, Jon Kleinberg, and Jure Leskovec Engaging with massive online courses. In Proceedings of the 23rd international conference on World wide web (WWW '14). International World Wide Web Conferences Steering Committee, Republic and Canton of Geneva, Switzerland, 687-698. DOI=10.1145/2566486.2568042 http://dx.doi.org/10.1145/2566486.2568042 
  
  Here activities of users in MOOCs were analyzed. The authors define patterns of users trivially by a number of lecture viewing. Later grades of users were compared with viewing of lectures and participating in forums. Two types of users are submitting assignments: hard working students that are engaged in a course all the time and students that have a knowledge of the course and just submit its assignments. Forum participants are mostly consist of forum users. In forum threads users who answer threads are those who are more active in the forum than initiators of the threads. Or are those who have higher grades. 

A Datawarehouse for Social Media for Learning

Recently i started to redesign our Mediabase, the database that includes data crawled from social media. A data warehouse model is more convenient for analyzing social media. Here i'm going to explain why.

All in all, the Mediabase design is not wrong. Behind it, its creators put the actor network theory (ANT) that claims all items in a system as actors. The following picture will help to understand how this idea applied to the Mediabase.

The Mediabase was an empty database with set of defined tables. Now it is filled on a regular basis with the help of watcher scripts. A watcher crawls one of Media: blogs, feeds, podctasts, e-mail archives. The traces left in a medium by its users consist of texts, tags or labels, urls and many other components. We collect these Artifacts in the Mediabase to be able later to apply them for different analytic tasks. Moreover, we examine users, e.g. e-mail senders, blog owners, forum posts authors and others. We can apply different analysis investigating changes of behaviors of uniquely defined users. Such an analysis shows activity patterns of users, their roles in communities, their interests and their closed peers in networks.

The Artifact and Medium are actors according to the ANT. The Community, the Agent (or User) and the Process are actors as well. All they have an influence on activities in online networks. That is why the Mediabase aim is to capture these changes.

But why do we need to collect all these data? Let us follow a teacher that aims to understand what problems do students have by solving an exercise. He can try to find an answer by observing online student communities and finding answers to the following questions:

• Are there any questions about the exercise?
• How many people have viewed the topic?
• How many people have took part in the discussions?
• What resources were used? How active are the students that talk about the exercise?
• What are the sentiments and intents expressed in the discussions?

To find the answers can be time-consuming and impossible manually (depending on the number of students and the number of involved online communities). 1) Communities have unique structures: the positions of users in structures specify roles that represent community experts, communicative members and brokers which are members connecting several groups. 2) Huge amount of texts include relevant and irrelevant for learning information. It may include interesting explanations and links but can be overlooked by a person.

So the help of machines is needed not only because the communities are big. We are talking about situations when teachers, managers or stakeholders of online learning communities want to get answers to different questions about different time intervals of their communities.

Therefore, we need a database model that provides the right information in the right place at the right time with the right cost in order to support the right decision. In the traditional databases we get a data item by specifying two parameters, e.g. specifying the time and the medium parameters we get posts appeared in the medium in the given time period. Or we get users that were active at the given time period in the given medium. But more sophisticated queries require some time to execute because of joins, etc. Moreover, analyzing communities researchers have to mine data and afterwards make some experiments to find patterns.

To simplify the life cf researchers, I propose a data cube that is a multidimensional model with a set of measures - facts. We can navigate over the data cube by specifying dimensions that correlate with Actors of ANT model. Particularly, we can choose a User that posted at a Time in a Medium so that we get the output that characterize the User in the given Time point in the given Medium. But in this case the output includes different measures captured by mining social networks, texts or user activities.

On the following picture you can see the hierarchy of dimensions to the proposed data cube. The construction of the Mediabase cube is still not ready, as it should reflect all actors we find important in the online networks. Thus I still need to elaborate Process, Agent(types), Community(types), Artifacts dimensions.

Users editing articles in several Wikipedia

I was wondering about contributors who are posting in several Wikipedia, Who are they? What country are they from?

Surely English Wikipedia is a leader in the number of contributors, articles, articles edits and number of users who contribute both in English and any other Wikipedia is high. I never doubt it. What about other Wikipedia? Looking at Wikipedia we can find which nations interact more often with each other than others.

A bit more than 1% of all contributors from our data set editing/creating articles in 2 Wikipedia. Only 12 of 4,919,026 users have ever worked in 5 Wikipedia.

Most of the cross-Wikipedia authors are registered; therefore, we cannot identify the place where the authors come from. A majority of the cross-Wikipedia anonymous authors are coming from Germany (15%) and the United States (11%)

The cross-Wikipedia authors contributing to more than 3 Wikipedia are summarized in following table. 

We find 1,718 contributors that did 618,584 edits. In the table we summarize the contributors work: the number of edits and the number of edits per cross-Wikipedia user. Cross-Wikipedia users contributing in 3 Wikipedia, are participating often in Russian, Japanese and any other Wikipedia from the set. The highest activity (more than 100 edits per cross-Wikipedia user) was shown in Macedonian, Russian, Catalan and Arabic Wikipedia instances. Generally cross-Wikipedia users are more active than contributors working only in one Wikipedia.

Data set:

We analyzed the Wikipedia data starting from June, 30th 2001 till January, 1 2009 and divide this time into 16 equal time intervals to keep visualizations simple. The period of the 1st interval is from June 30th, 2001 till December 31st, 2001, the period of the 2nd interval is from January 1st, 2002 till June, 29, 2002 and so on. Because of hardware limitations, we did not consider all revisions done in the period for constructing author networks. As Wikipedia instances have different number of articles, we choose a number of revisions depending on the number of all revisions in the instances. The revisions are picked up along an article timeline. The overall structure of the Wikipedia network and activities of Wikipedians are not impaired by this data reduction. Finally, we got datasets with number of revisions comparable to each other.

We chose both European and Asian Wikipedia. The instances were selected according to their size: large European Wikipedia (Spanish and Russian), large Asian Wikipedia (Japanese and Turkish), small European Wikipedia (Bulgarian, Catalan, Danish, Greek, Macedonian, and Ukrainian) and small Asian Wikipedia (Arabic, Hindi, and Korean). The list of small European Wikipedia list includes Wikipedia of different Slavic languages (Bulgarian, Macedonian, Ukrainian) and the Catalan Wikipedia, the Wikipedia of a minority language group in Spain.

some notes during reading "Networks, Crowds and Markets" by David Easley and Jon Kleinberg - Introduction

a network is a pattern of interconnections among a set of things

adding resources in a network can undermine its efficiency - Braess's paradox

a notion of equilibrium - a state that is self-reinforcing in that it provides no individual with an incentive to unilaterally change his or her strategy, even knowing how others will behave

we have a fundamental inclination to behave as we see other to behave: 1) the behavior of others convey the information or it is a direct benefit from aligning your behavior with that of others

in many cases you care more about aligning your own behavior with the behavior of your immediate neigbors  in the social network, rather than with the population as a whole.

a new behavior starts with a small set of initial adopters and then spread radially outward through the network.

a diffusion of technologies can be blocked by the boundary of a densely connected cluster in the network- a "closed community" of individuals who have a high amount of linkage among themselves, and hence are resistant to outside influences,

How long do communities live?

The question I'm solving  in my dissertation is "Do any signs exist that indicate that community will die or will continue to exist successfully?". I don't mean any Shaman signs and i'm not trying to read coffee fortune. it is pure science.

I'm following the life of communities and community knowledge. A community is impossible to survive without users thus first of all the users need to be analyzed. I'm examining what are users behaviors, where a behavior is a broad term in Internet world.

One of the examples of such an examination was done in the paper Activity Span. The researchers analyzed survival patterns of users in web community. They were considering Yahoo Answers, Baidu Knows (Chinese) and Naver Knowledge-IN (Korean) knowledge sharing systems. The outcomes the researchers come to: for all three communities the users need to be trained to ask questions otherwise the questions remain without answers. As it was expected, people don't leave community if they got answers to their questions. The experts of communities stay longer there if their answers were awarded (best answers).

The users lifespan implies the lifespan of their community. Surely if there are clear indications of users that are going to finish their life in a community, it can be used as a sign that community will die. Let say if there are questions that nobody answers and answers nobody thanks for produce an alarm for community managers that something have to be done about their community or it disappear.

Things about communities are not so simple to figure out. A community consists not only of askers, answerers. Other types of users exist. Their behaviors are dependent on the activity in posting answers and questions, in participating in discussions, in sentiments they use in posts, in their intents. I state here that it is a complex task to describe a community as it has many participants with different behaviors, i.e different patterns. Thus it makes hard to imagine what happens with the community afterwards: is it die or survive?

Identity is not important for some communities?!

I've just watched very interesting talk by Michael S. Bernstein about 4chan and /b/ discussion boards where more than 90% of users are anonymous. The threads in the boards are not archived but deleted after sometime (~3,9 minutes). Although the anonymity of users and very short life of threads, communities have their culture and policies.For example, if somebody posted without using a settled slang, he gets a "Lurk further" reply.

Such kind of communities haven't been researched a lot. It is interesting that in communities like 4chan nobody looks for a reputation of a user. Even so some posts have "long" life (~ 39 minutes) when other users reply. In the boards it is going only about posted content and not about who has posted.


 
 
4chan and /b/: An Analysis of Anonymity and Ephemerality in a Large Online Community

My presentation at the doctoral consortium of EC-TEL 2010

Self-modeling and self-reflection of E-learning communities

View more presentations from petrushyna.

I would like to thank for the feedback i got from EC-TEL Doctoral Consortium participants. It helps me to narrow the domain of my research.

Reflective patterns in online communities

The rate of return (ROI) answers are the money gained or lost. It measures a percentage of the interest to the investment, e.g. you gained 5 € on your 10€ investment - the ROI is 5/10=50%. The ROI can be as well negative.

How can we analyse the interest of the investment in online communities? What events cause positive or negative activities in networks? There are some papers trying to explain the matter of things basing on communities contributions:

In Burke et al. the scholars calculated the probabilities of getting a response in discussion boards. The probabilities highly depends on the type of introduction("i'm lurking here for several months" - group introduction or "working as a professional in this area" - topic introduction) and sometimes on the type of group.

The wiki case is observed in "Wikis as social networks: dynamics and evolution", where the vandalism on a wiki page was detected if the page regulates the information flow within Wikis. Particularly, if the page is a hub between many other pages, the probability of trashing is high.

Anyway, it is more complex to find out the ROI for human interactions as to calculate the finance ROI, where monetary values are used. The investements in human interactions and the interests of these are not uniform and hardly comparable. What you know from the childhood: if you use rude words in the conversation, your listener will reply coarsely as well (in most cases). The rule "if ... then ... " is on the stage. If we have the pattern "rude question", will we get the pattern "rude answer"? Why yes and why now? What are dependencies? Shouldn't the activities/events/actions in communities be modeled? Then will modeling be useful to find the provoked profit?

In the informatics there is a bag of artificial intelligence problems like reasoning, learning, decision making. Those are aware or at least try to be aware of all the inconsistencies and uncertanties human interactions have. Those should be used to find the reflective patterns.