Web-Scale Interaction I: Open Source & Social Computing

From CS260Wiki
Jump to: navigation, search

Bjoern's Slides

Extra Materials

Dan Pink's TED Talk on the effects of incentivizing intellectual work

A 10-minute video highlighting Dan Pink's views on incentives.

Academic work on making sense of Wikipedia editing behaviors.

Discussant's Slides and Materials


Reading Responses

Kurtis Heimerl - 9/21/2010 14:09:46

Coase's Penguin, or Linux and the Nature of the Firm, Leave it to economists to make anything boring. The open-source movement initially confused economists, and this paper attempts to explain the incentive structures of these communities. It’s horribly dated, which is surprising as it’s not all that old. It’s also really LOOONNGGG so this writeup is likely of limited quality.

This seems to be of some value to someone, but not me. So much of ICTD is viewing and understanding incentives that I feel I have a good grasp of the structures inherent in open-source.

The article definitely sidestepped a lot of the commercial dedication to things like Linux, but that’s probably because they are out of scope. Those incentives are old and understood, but the questions of crowdsourcing and participation by users without any monetary benefit are not.

To close this out, I think this analysis leads to an interesting discussion of Turk. Can adding a payment structure to crowdsourcing make it better, or should we be doing volunteer work?

Predicting Tie Strength With Social Media This paper details a method for evaluating the strength of social network ties through social network metrics such as length of relationship, number of messages, and so on.

Good idea, and concise paper. I’m not surprised, either about the fact that it works or the factors they chose. That leaves me with little discussion material, unfortunately. They did a pretty good job explaining potential uses of this information, reducing the noise in my facebook noise feed. In fact, it seems as though facebook took this research seriously, as the feed have improved pretty dramatically since 2009.

I guess I’m mostly surprised that this research wasn’t done before this point. It’s not like facebook is the first (or even third) social network. I’m going to spend a few hours trying to find some more low-hanging fruit. Maybe we can predict the current status of a relationship from the social network?

matthew chan - 9/21/2010 15:26:34

> Predicting Tie Strength With Social Media

IN Gilbert and Karahalios' paper on explorign tie strength, this paper is fairly important because of today's dominated life style of social media and social networks--in particular, the authors used Facebook for their research. The results and findings were a model that could correctly classify friends as strong or weak more than 85% of the time. The technique/method used was to recruit ~35 university students/staff and had them go onto their Facebook account. Using a Firefox extension/add-on and GreaseMonkey, they had the participants answer questions to determine the strength of their relationship. I found the choice of questions very intersting such as "How would you feel asking this friend to loan you $100 or more?" since the questions were more complicated and hypothetical. This paper strongly relates to today's technologies bc it's abolutely true that on FB, you're either a stranger or a friend, when there are lots of in btwns. Moreover, by classifying our strong/close friends, we might be able to get better streams of info on them (ie. NewsFeed) This paper does not relate to my field of work; in addition, there are some blind spots (which the paper even admitted), such as explorign videos on FB, and every other social networking tool, ie. Twitter, out there.

> Coase's Penguin, or Linux and the Nature of the Firm In part I, we learn about something that sounds very close to crowd-sourcing. With examples such as NASA's clickster, Wikipedia and Amazon.com, the author(s) describe a system where peers contribute and dictate the relevance of material or even extending software, ie. open source. The author then explains the accredation (a 15 yr old giving legal advice learned from CourtTV) by exploring Slashdot where users and their comments play a critical role. Furthmore, distribution is an essential part. As shown by the author, Napster's biggest success was using the hard drives of every user to form a decentralized network for easy access. In part III, the author seeks to figure out what makes the contributors ticked, to figure out the motivation behind the users. Monetary, hedonic, and social-psychological rewards are explored to help clarify the diversity of motivation.

Overall, this paper is very important because it highlights the growing trend of our society. As mentioned, many organizations are beginning to try out crowd-sourcing, especially Wikipedia. This paper doesn't relate to my own work. One potential blind spot I'd like to explore is the role of social networks, especially since Facebook and Google are beginning to perform real-time search that even includes Twitter posts, etc.

Airi Lampinen - 9/21/2010 15:54:18

Benkler's text "Coase's Penguin, or Linux and the Nature of the Firm" is an overview of peer-production. Furthermore, Benkler discusses the motivations and potential incentive problems related to participation.

The first section stresses how peer-production is a wider phenomenon than the mere issue of free software. Benkler gives examples from NASA to Project Gutenberg to illustrate the general emerging phenomenon in the organization of information production. The text has merit in showing how in the case of peer-production, collaborative technologies are allowing new types of behaviors and activities emerge. What is revolutionary are not so much the technological platforms but the change in collective action they are making possible.

The third section is a more economics-oriented analysis of the motivations and potential incentive problems related participation. He chooses to look at motivations through a very limited lens, focusing on monetary rewards, intrinsic hedonic rewards and social-psychological rewards. Due to this decision many interesting questions are beyond his scope.

Benkler discusses the difficulty of integrating small contributions to a meaningful and states that this problem is a central factor limiting the viability of peer production. He summarizes technology, iterative peer production, social norms and market or hierarchical mechanisms as approaches to integration.

Gilbert and Karahalios' article "Predicting Tie Strength With Social Media" addresses the problem that social media often treats all users the same: trusted friend or total stranger, with little or nothing in between.

The central concept of the text is tie strength. The authors present a predicte model for mapping social media data to infere tie strength from explicit information such as amount of communication, friends in common and the like. Next to explaining the model, the paper includes an interesting outlyer analysis - excerpts from interviews where users explain why certain inferences of tie strength went badly wrong.

Gilbert and Karahalios present interesting ideas on how modeling tie strength can improve social media design elements, such as privacy controls, message routing, friend introductions and information prioritization. While the model seems to work fairly well in what it was designed to do, predict tie strength, it does not answer how to take into account the issue that tie strength is not a sole and sufficient characteristic of a relationships. A user may have two Facebook friends with whom she/he has an equally strong tie. That, however, does not mean that the user would want those two people to have access to the same components of his/her profile etc. All in all, however, the paper is one of the more convincing studies of modeling social variables that I have seen.

Charlie Hsu - 9/21/2010 18:02:01

Coase's Penguin

In this reading, Yochai Benkler describes the peer production of information, basing some of his insights on observations of the open-source software development community. Benkler examines the sort of information produced on the Internet via peer networks: content such as Wikipedia, and NASA's Clickworkers. He then examines the validation and accreditation of such peer-generated content, and the distribution models. Benkler also discusses the problems with motivation and organization in a peer production model.

I found Benkler's analysis of peer-produced content and communication not only dry and long-winded, but also not particularly insightful, especially in the field of human-computer interaction. Much of the content section was simply about how different websites/systems facilitate content generation by users. The accreditation section has one mildly interesting insight (you can use peers to not only generate your content, but also verify it! What an idea!), and the distribution section repeats this exact same finding (users can also distribute your content!). Perhaps I am being overly critical, but the peer production summary of Benkler's reading was more of a tedious overview and selected examples of how peer production systems looked at the time of his analysis, information that is pretty readily available and well understood.

The analysis of the problems with peer production, motivation and organization, was much more interesting. I especially took note of the idea of monetary rewards interfering with social-psychological rewards. I feel that since monetary rewards are useful universally, in so many ways, they lose the uniqueness inherent in the rewards gained by working for a non-profit/open-source peer-production project. It is exactly the uniqueness of the reward that increases the value of the hedonic or social-psychological reward motivating participation in the peer-production project. You can't buy StackOverflow karma, you can't gain reputation on Wikipedia by doing anything but work on Wikipedia.

I also felt that the problems with organization and integration of work, though interesting and certainly applicable to peer production, actually applied to just about all engineering projects, solo or group, personal or office-based work. Modularity is key to scaling projects, leaving the big-scale picture open for redesign, allowing appropriate 3rd party contributions to be easily integrated into the project, and much more. Granularity is important to keep a module's scope understandable and reduce overhead when onboarding new developers. Integration is a more difficult problem with peer production than with traditional systems, but Benkler seems to simply hand-wave it away by saying "the integration function must either be low-cost or itself sufficiently modular to be peer-produced." Perhaps this works with information generation models like Wikipedia, but what about with open-source code projects? Who manages the project and integration then? Modularity, granularity, and a clean system for peer submission, review, and checkin are required then for successful integration.

Predicting Tie Strength With Social Media

This paper described research into attempting to predict friendship tie strength with social media such as Facebook. The authors developed a predictive model using Facebook variables to determine strong vs. weak ties among participants' Facebook friendships. The authors then interviewed participants about relationship anomalies, and provided practical and theoretical implications for their research results.

I felt that the authors did well in their analysis of tie strength dimensions, the execution of their experiment, and analysis of their results. The base idea of the research seems sound and relevant to me; I and many of the people I converse about Facebook with share similar sentiments about the problems with binary friend/not-friend relationships in social media. What exactly does friendship mean in this sort of binary social context? Shouldn't there be an option to more easily facilitate sharing information with other friends on the basis of how well you want them to know you, just like in real life? The tie strength dimensions chosen by the authors seemed relevant and were based off of existing literature. I felt that their variables were not only well chosen and categorized, but also relatively easy to retrieve via social media. The direction these researchers chose by using digitized social media was one filled with concrete evidence, instead of one with users mistakenly recalling communications with friends. The analysis of relationship anomalies was also a nice touch; it very strongly emphasized the complexity of relationships and offered directions in future research.

I felt the paper did an excellent job of explicitly stating the contributions it made to HCI research. Their findings on the relative importance of tie strength dimensions (intimacy, intensity, and educational difference in social differences being some of the stronger ones) are important not only in steering the direction for future social media tie strength research, but also in general social science as well. Their decision to calculate tie strength as a continuous value was not only successful, but strongly highlights the idea that relationship strength is complex and perhaps better represented through a continuum rather than a discrete scale. The practical implications of their research are also highly relevant to the problems Facebook faces today: calls for better controls over privacy, attempts to stream the most relevant content to the news feed, and recommendation engines that provide appropriate suggestions for new friends and reconnections.

I felt that my personal experience with Facebook also lent some opinions on their actual results as well. I felt that the anomaly with inbox thread depth might be partially explained by a strange phenomenon I observed with private messages on Facebook. When I want to communicate privately with a friend, often times Facebook private message is the last option that pops into my head; usually the medium is email or instant message. Facebook private message, in my case, is most commonly used by irritating campus groups that "invite" you to an event and spam you with private messages reminding you about the event. This would explain the inbox thread depth's negative correlation to tie strength in my case; often times the people who are spamming me via mass event private message are those who don't know me well enough to personally ask.

Pablo Paredes - 9/21/2010 18:13:46

Summary for Benkley, Y. – Coase’s Penguin, or, Linux and The nature of the Firm

The fist part of this paper is related to the definition of what are the key components of a collaborative development approach as the fundamental process underlying open source development. The author describes peer production as a modern alternative to market-based or firm-based production and defines three main components of commons-based (as in the tragedy of the commons) peer production: Content, Relevance/Accreditation and Value-Added Distribution.

Many participants refer content as the creation of new information in a distributed fashion created through small inputs. One example of this is the NASA clickworkers, where hundreds of volunteers help solve complex problems by adding small value whenever and wherever they are available.

Relevance/Accreditation is the process that ensures there is compensated outcome, where high-quality and low-quality input is leveled out. A good example is Slashdot.org, where comments made by devoted users, which are measured by karma (defined as a meritocracy in which the number or positive quotations that a user has, as well as the quality of the quotations, i.e. if other karma authors are quoting the posting) generate a well refined method to get valuable and accredited content that are relevant to the audience of the site.

Finally, Value-Added Distribution is the notion that the Internet is not only a cheap transport/distribution medium, but it also enables quality control and increased reliability distribution methods. Examples of this shown are Project Gutenberg, Akamai and Napster, where hundreds of distributed peers provide proofreading as well as reliability (redundancy) to the distribution process.

The author closes by signaling that the overall process of creation and collaboration is based always on individual (therefore diverse) motivations. He expands the basic hedonic creative approach to our lonely existence and the inherent indirect appropriation opportunity that rises from the huge potential to have mundane individual gains through professional services based on open frameworks platforms to a clearer definition of what are the types of activities/tasks that can better benefit from peer production. In short he describes that any production method is better than the others when it can motivate better behavior in humans. He lays the fundaments of motivations into three categories, (M) Monetary rewards, (H) Intrinsic Hedonistic Rewards and (SP) Socio-psychological rewards, which could be independent or combined.

The Rewards equation R=Ms + H + SPp,jalt (where s = satiation, i.e. rate at which M decreases; p = positive or negative interpretation of having money; jalt = jealously or altruism). However, the author completes the analysis introducing a cost/benefit view, which is described by some transactional costs Cm and Csp, related to M and SP respectively. Defining V as the marginal value of an agent’s action, opportunities where there is potential for collaboration without monetary rewards can be described by the equations: Cm > V > Csp and H + SP – Csp > 0.

The author shows how M has a different effect on R depending on the values of s and p, meaning that there are some activities where people that are satiated economically will prefer to undertake activities with high SP component rather than M. Complimentary it is evident that the (positive or negative) value of p is dependent on cultural dynamics and that can change across cultures or with time.

Overall a couple of interesting conclusions of the paper are that (1) peer production can be embraced when the number of contributions is large enough to make monetary rewards less valuable to a group of people with equivalent motivations. However a corollary to this conclusion is that open source and any other peer production systems generates a social segmentation, and therefore does not reflect a true transversal expression of the community, and (2), peer production projects suitable to fulfill the motivational structure should have the right combination of modularity (extent to which a project can be broken up into independent and asynchronous modules), granularity (size of modules in terms of time and effort – the number or people willing to participate is inversely proportional to this value, and also the smaller the module the more prone to more SP driven motivations), and cost/opportunity of integration (composed of a quality control, i.e. the possibility of success of the project based on the quality of its inputs, and a summarization elements). To embrace a peer production process, integration must be either low-cost or modular itself.

I believe the foundation of this analysis is eye opening to go beyond the view of simple open source, and to see the underlying phenomena of commons-based peer production. From this perspective, I would like to have seen some deeper analysis on the definition of integration in reference to the tolerance to failure that a culture has, and how this could impact the undertaking of peer production based projects. I assume that higher tolerance to failure could trigger less well integrated projects, and therefore steam up new ideas to be explored in a peer oriented process, while lower tolerance to failure (whether it be in the realms of M, H or SP) generates opposing results.

Summary for Gilbert, E. and Karahalios K. – Predicting Tie Strength with Social Media

The paper describes a method to bridge the gap between tie strength theory (from sociology and psychology research) and social media. An initial definition of tie strength is defined as a (probably linear) combination of four dimensions: amount of time, the emotional intensity, the intimacy (mutual confiding), and the reciprocal services, which characterize the tie [Granovetter, M. S.]. Combinations of these dimensions characterize a tie as weak or strong. Additional research introduces other dimensions such as topological structure, emotional support and social distance. Weak and strong ties can render more positive or negative outcomes of human group-centered tasks, and it is evident strong does NOT mean better.

Many of the many variables assessed in the study to measure the previously cited tie strength ties do provide relevant content that could show eventually the value of the underlying tie strength. However some limitations inherent of social media, like the lack of goods exchange, and the potential lack of accountability for “real” life social interaction are not explored as additional error factors that could reduce the efficiency of using social networks as predictors of tie strength. Additionally, a generalization of this model assumes there is a genuine mapping of a “real” persona with a “virtual” one. This mapping can clearly be altered by either a transformation of the “virtual” persona, in many ways, from psychological to socio-cultural aspect, as well as the possibility of a one to many mapping.

I have some remarks regarding its approach: a) it is not clear to me the influence of cliques in the analysis, and how this is accounted for; b) the sample selected to perform the test does not represent clearly a good random sample of society; c) there is not paired validation of the meaning of friendship among common friends, i.e. the notion of tie strength can be different to different people, and I do not see how this is accounted for in the model; d) it is not clear to me if there is any measure that incorporates context or specific personal situations (emotional, economic, social, etc.), which could be aspects that can alter the perception of the relationship with an specific person.

However, despite my remarks, I really appreciate this paper as an indicator that social networks are much more complex and rich in dynamic processes that actually may require a completely new assessment of a more effective network (or topology) to describe these systems. I believe the current simple “friending” approach followed by Facebook and the likes is not close to map well these models and therefore the novelty of the application could be a higher factor of success rather than its ability to map social relations.

Drew Fisher - 9/21/2010 18:23:12

Coase's Penguin:

This article discusses (in great detail) the existence, motivation, and value behind peer production systems, including GNU/Linux, Slashdot, and Wikipedia. It makes a case for how this sort of system is different from traditional systems of production, why such systems exist, how they can be sustainable, and where they might outperform traditional systems.

I liked the psychological analysis, and the attention to the fact that people work for things other than money. It reminded me of Dan Pink's TED talk - http://www.ted.com/talks/dan_pink_on_motivation.html to watch. The discussion on jealousy/altruism makes me think about Google's "Don't be evil" and how that plays into their persistant relative good standing in the open-source community.

I didn't get that much out of this paper, since I'm a contributor to the open-source community, intimately familiar with several of the example peer production systems mentioned, and fairly well-read in the study of human motivation. I did, however, find it to be a thorough, well-reasoned exploration of the topic. I also appreciated his conclusion that "For regulators, the implications are quite significant. In particular, the current heavy focus on strengthening intellectual property rights is exactly the wrong approach to increasing growth through innovation and information production if having a robust peer production sector is important to an economy’s capacity to tap its human capital efficiently," a problem that has continued misguidance of our legislators has continued to worsen.

Predicting Tie Strength With Social Media

This paper makes an attempt to model how close two people are, based on their interactions on a social network - in this case, Facebook. If successful and accurate, such a model would suggest that data on social networks tells a consistent (if limited) story about the interactions between persons.

I found it interesting (but not surprising) that in Table 4, Job had the weakest correlation to other questions. This is the source of my major criticism of the paper: it has been shown that persons in lower socioeconomic classes tend to seek jobs through strong links, whereas those in higher socioeconomic classes do so through weak links. I question whether the "Job" question was a good choice for analyzing strength of social ties.

Shaon Barman - 9/21/2010 18:27:43

Coase's Penguin

The authors discuss why certain peer produced environments, such as Wikipedia are successful. Peer production has typically faced two critiques: there is little motivation and bad organization. They decompose peer production enterprises into a model, and analyze how the different variables relate to each other.

The paper discusses several projects, including Clickworkers, Wikipedia, Kuro5hin, Amazon, Google, Slashdot etc... These projects represent a wide range of interests and organizational hierarchies. But all projects use a large group of users in order to improve the quality of their product. One of the overarching themes is in order to get the masses involved, the amount of one a single person has to do must be negligible. By breaking a task into bite size pieces, a single user can contribute without sacrificing more important things. This allows the enterprise to draw from a large and varied crowd. Such systems can also by manipulated so certain checks and balances must be put into place. Slashdot accomplishes this by limited the votes a single user has and has moderators and meta-moderators. A system with such checks allows a large number of users to participate while maintaining quality.

In the third section, the author discusses the motivations in these peer production system, balancing money, intrinsic hedonic rewards and social rewards. The model he proposes seems too simplistic and has to be arbitrarily appended in many situations. There are also interactions between money and social rewards which are not captured in this mode (one example of this is the jalt factor mentioned by the author). While this model is a start to predict how a "crowd" would react to a new system, it seems to be over-fit to the different peer production systems analyzed and shows little beyond these systems. The final paragraphs key needs that should be met in order to motivate people to contribute to such products, such modularity, heterogeneity, motivation (including integration and allocation), freeloading and defection.

Overall, I thought the paper made some good points about how peer productions systems worked. The author analyzed many current systems and found how the checks and balances provided the user base a system which allows many users to contribute without over taxing them. I thought the model of reward they provided was too brittle, and their analysis of the systems using this model was too long (and tedious).

Predicting Tie Strength with Social Media

This paper discuses how to use Facebook in order to predict relationships between individuals. They extract statistics from a user's profile in order to predict how they would react to a series of questions. The model uses the statistics, interactions between different statics and the network structure in its predictions.

Overall, I thought this paper made good use of the enormous amount of social data found in the internet. It also exposed me to the pitfalls of user studies. The order of the questions and the observations within each participant not being independent affected the results. It also seems when they looked for "intimacy" words within a text, the context of the word can drastically affect the meaning.

One aspect they do not deal with is how users would react to the system auto-classifying their friends. In the past, whenever a large company tries to automatically create such interactions, many users get upset. When dealing with privacy issues, large companies must tread carefully when predicting the behavior of their users. It would also be interesting to use this data to correlate advertisements/shopping habits in order to create more profit.

Aaron Hong - 9/21/2010 18:43:14

In "Predicting Tie Strength with Social Media" by Gilbert, et al. we see how they use some variables from Facebook, data they collected from a lab, and social theory to construct links that are more meaningful then exits/does not exist.

Overall, it was an interesting article however I do see the limitations of applying social theory to computer science. They did do a significant amount of research, but I still think social interactions are more complicated. The complexity is noted by showing the negative correlation of inbox depth and tie strength, which was counter-intuitive. Also relationships that are "strong" can quickly change in dynamics such as a break-up or fall-out. It is interesting enough though, however with the fuzzy results it is better to let the user control their virtual relationships manually and not automatically.

In "Coase’s Penguin" by Benkler is about the formation of the strange phenomena that is the open source movement. It talks about why they would work, especially in a networked environment. What's particularly interesting is that I'm taking a class on Psychology of Creativity and we've talked about intrinsic interests as being a big reason why they are creative and motivated to work. In a similar sense part of the open source movement is due to people wanting to write software, not because of rewards like money, social pressure, etc.

Anand Kulkarni - 9/21/2010 18:50:56

Coase's Penguin

The author argues that open-source development represents a third model of work (as opposed to Coase's division of work between employees working to satisfy managers, or managers working to satisfy market signals), and discusses implications and advantages of this model.

The core contribution is an attempt to explain the success of massively collaborative volunteer efforts, with a special emphasis on understanding open-source software development models. This contribution can have implications in HCI for understanding how to motivate crowds to perform work in certain ways, as well as understanding how to prevent open-source projects from failing. Two issues of values that the author addresses include how collaborative work can be modularized and how users can be incentivized, both of which are critical for collaborative work. I would like to see more specific application of these models.

I found the author's argument to be rather weakly supported. Most of the models suggested in the paper are developed only with intuitive and at times philosophical arguments presented for their adoption. At the same time, the absence of good models means that the author's explanation for open-source collaboration's success is perhaps the only good one. It may also be unreasonable to expect quantitative or more precise models when considering an inherently social phenomenon.

Predicting Tie Strength The authors discuss the problem of predicting the strength of links within a social network and present a model that can distinguish between strong and weak ties with high accuracy.

The core contribution of the paper is a model for distinguishing between strong and weak ties, as well as some secondary (unstated) contributions in terms of the design of experiments for analyzing tie strength. This work may have greater value in providing a foundation for other researchers to carry out precise quantitative analyses of the underlying impact of tie strength in more interesting settings rather than the model itself. The problem of determining tie strength has implications in HCI in analyzing the spread and impact of information within a network - strong ties imply more impact when one node is affected. All together, the paper does fill a missing gap between the theory of tie strength and its practice in social networks; it seems to be one of the first to test such a model with Facebook. The authors' suggestion that their results may be useful in adding features to Facebook seriously undervalue their results.

Because this is an experimental paper, the authors justify their model and approach thoroughly throughout, and provide excellent justification for their approach. Close attention is paid to the way that friends of users are chosen so as not to let large-networked individuals dominate the results. There are some problems in the authors' definitions of tie strength in terms of a set of five questions; they contend that this reflects and incorporates several views on what tie strength means, and have an extended discussion based on the existing literature, but I'd argue that self-reported tie strength as in their first question is really the most important factor. The other main difficulty with the author's approach is that Facebook does not always reflect tie strength between two closely tied users. The authors acknowledge this issue as a weakness in their experiment and even use it to explain some of the inaccuracy in their model.

Brandon Liu - 9/21/2010 18:57:06

“Predicting Tie Strength with Social Media”

The contribution of this paper is that it confirms our expectations about how people act on a social network. We can see from interactions with a website that these interactions mirror the strength of real-life friendships. A result of the paper was a model that could classify Facebook friends as strong or weak.

One problem with the paper is that it assumes that Facebook is prototypical of all social media sites. It describes how certain features of Facebook tie in to determining the strength of bonds, but doesn’t elaborate on how Facebook’s mechanisms play in. For example, the “Gifts” part of Facebook may factor into the strength of a tie, but Gifts were only really popular when it was just released. Another criticism of the paper is that it inherits ambiguity from its use of sentiment analysis. In my experience, a lot of casual communication on walls couldn’t be correctly classified as positive or negative by these techniques.

There were also some problems in the survey technique used, for example, asking a participant how much a facebook friend could help them find a job. Since tech industry workers seem more likely to have a career-related acquaintance on facebook than other industries, it would have been better if the author included statistics to make sure this was a valid question.

The paper mentioned how Facebook closeness may be negatively correlated with the strength of a bond, due to friends preferring IM, text messaging, or phone calls over facebook messaging. It would be an interesting research direction to see each mode of Facebook communication plotted against other methods on some kind of closeness spectrum.

Other mechanisms of social networking may yield different results. For example, all of the connections on Facebook are positive, while on a site like Slashdot, negative relationships can be inferred (see Leskovec, Hottenlocher and Kleinberg, “Predicting Positive and Negative Links in Online Social Networks”). Also, unidirectional relationships, such as “Followers” on Twitter, may have a different model.

“Coase’s Penguin and the Nature of the Firm”

In Part I of the paper, the author discussed the free software culture as compared to other “public knowledge” communities, such as science and academia. He specifically addresses the problems of relevance and accreditation, which are dealt with in peer-reviewed journals, but not so much by online communities. The contribution of this part is to clarify what a mode of peer production needs to succeed. This is relevant to HCI since it describes specific mechanisms (such as Slashdot karma) that when implemented affect how the users interact with and judge information.

One important point I liked in Part 1 was how it described the rise of Google. The author’s opinion is that a ‘willingness to pay to be seen’ model for search engines is detrimental both to the anonymous users and the search engine itself. The author describes how Google is successful since it provides satisfaction by giving users the most useful result by considering relevance and accreditation.

I found Part III of the paper personally relevant. As a contributor to open source software, I definitely believed the author’s point that it can drive away contributors when a project ‘Sells out’ to commercial interests. I would extend this analysis by saying it is even more emotional than that: for example, when Sun, a proponent of open source software, was bought by Oracle, a number of developers in the community suggested that the project be abandoned. Thus, the perception of an open source project as something done on leisure time and purely for artistic purposes (i.e. negatively correlated monetary and socio-psychological reward) is really a result of marketing effort on the part of the project. Another example is the open source datastore Redis, which is exploding in popularity, even though it was recently ‘bought out’ - the lead developer was hired by VMWare. Organizations must cultivate their image if they want open source communities to work in their favor. Debian is another good example of a project that eschews corporate interests.

In my personal experience, I’ve found that monetary and sociopsychological rewards often go hand in hand in the open source community, especially with well known hackers (for example, a Rails core contributor) who makes a lucrative career out of consulting while being revered in the community.

Luke Segars - 9/21/2010 18:57:47

Coase's Penguin, or Linux and the Nature of the Firm

This paper identifies an interesting and world-shaking phenomenon that has recently been the subject of a decent amount of attention from economists, behavioral psychologists, and businesspeople. Benkler identifies the emergence of a third “mode of production” in the open source software movement that is based more around our intrinsic motivations to produce something useful instead of the traditional extrinsic motivator of money.

There has been a substantial amount of attention drawn to this topic recently due to its deep conflict with traditional economics. The idea, made clear by Benkler and other authors, has tremendous effects for understanding human behavior and therefore how technology might be able to aid humans in group interactions. For instance, it is possible to determine what areas of a project consistently turn out weaker in distributed open source projects than in paid commercial ventures. Many parts may have similar quality – this suggests that fancy collaboration tools are not necessary for a team to be productive on this particular component. Other parts may be significantly behind, potentially indicating difficulty in collaboration among the distributed team members.

Open source software development is, in ways, simply the far end of the spectrum that traditional businesses are moving in as well: geographically distributed teamwork. Tools that are helpful for open source developers (email, version control software, instant messaging) are also very important in the commercial space as well. From an evaluation perspective, using open source contributors as a testbed for a particular tool or technique could prove to be a very accurate and simple process.

I am personally very excited by the principals that the success of the open source software movement proposes about human nature. Technology has provided us with digital data that can be copied, shared and distributed for free – this seems to have changed so much about that our society that I am comfortable suggesting that it may shake the very core of the financial markets as well. A number of branches are already straining under pressure, but while the old system struggles we are seeing the emergence of a new people-driven approach to productivity, and that means big things for almost every sector of society.

Matthew Can - 9/21/2010 18:57:48

Coase’s Penguin

In traditional economics, productive activities can be viewed as responses to either market prices or managerial commands. In this paper, Benkler characterizes a new mode of economic production dubbed “commons-based peer-production” in which groups of people collaborate on large projects, driven by diverse motivational factors and social signals, not by prices or employers. The paper provides examples of this phenomenon, its component characteristics, and a framework for understanding how it operates.

Although this paper is from outside the HCI community, it is still relevant to HCI research initiatives because in HCI, we are interested in how large, distributed networks of people use electronic media to produce content and services. It is useful to us to understand the dynamics at play in such networks and what contributes to the success of the peer production effort. This paper makes a stride toward achieving those goals because it presents a framework for analyzing peer-production networks.

Specifically, I liked the way the author decomposed motivation into three components: monetary rewards, intrinsic hedonic rewards, and social-psychological rewards. Such a model can explain why peer-production is sustainable even though it lacks a monetary component (in fact, monetary rewards may be inversely related to social-psychological rewards such as community recognition). Moreover, the author did well to tie the issue of motivation into the analysis of content integration. The way in which the peer-produced content is collected, refined, and then distributed creates a feedback loop with the motivational mechanism.

Predicting Tie Strength with Social Media

In this paper, the authors present a predictive model of tie strength. The model takes as input a user’s social media data and maps that to a numerical value, tie strength, for each of that user’s social connections. The paper demonstrates that tie strength is a much more useful measure of social connection than a mere binary connection because it can be used to inform the user’s privacy controls and to filter information presented to the user.

As an initial attempt at bridging the theory behind tie strength and the reality of relationships (as they are manifested in social media), this paper does a wonderful job of presenting a predictive model that validates theory against reality. However, the model suffers from several drawbacks that severely limit its applicability.

The authors acknowledge that there may be a problem with the model being too specific to Facebook because the coefficients of the predictive variables may not generalize. It seems that the problem is even worse. The predictive variables themselves were chosen with Facebook in mind. For example, “appearances together in a photo” is only relevant if the social networking application in question has image sharing features. How would this extend to Twitter?

More importantly, this kind of model completely fails to capture the tie strength (or incorrectly predicts very low tie strength) for individuals that communicate through a medium other than Facebook. The authors touched on the idea that some social interactions, perhaps those of high tie strength, bypass online social networking entirely. For example, I don’t write on the walls of my best friends. In fact, I don’t even check their profiles regularly. I use synchronous forms of communication, chat and telephone, to interact with them. Perhaps the model would be strengthened by incorporating variables from a number of different and varying communication channels to predict tie strength.

David Wong - 9/21/2010 18:57:49

1) The "Coase's Penguin" paper discusses the phenomenon of distributed peer projects that go against the traditional model of market or firm based production. Section 1 discusses examples of using peer production and its benefits in regards to content, accreditation, and distribution. Section 3 discusses a way of representing rewards obtained from money, hedonism, and social/psychological rewards. Given the model, the section discusses how to structure rewards in peer production models so that the projects are sustainable.

The "Predicting Tie Strength With Social Media" paper proposes a model to indicate tie strength in relationships and evaluates whether that model fits into social networks. Their experiments showed that their model was able to distinguish between strong and weak ties with over 85% accuracy.

2) The "Coase's Penguin" paper offers an interesting perspective on analyzing the successes of various distributed peer production systems. I think it adds value to the HCI research community as a way to structure large, distributed, systems for a large volume of users. Given the advances in networking and ubiquitous computing since 2002, when the paper was written, I think the paper's analysis is even more pertinent.

The "Predicting Tie Strength With Social Media" paper offers interesting results to the HCI community, especially for the field of social networks. The binary model of friendship is severely limited and has room for improvement. With a more sophisticated friendship model, users of social networks can be shielded from malicious strangers and can have their data better protected. As such, the results described in the paper can inspire more research into creating a sophisticated model for analyzing the strength of relationships.

3) The "Coase's Penguin" paper addresses a problem that is well motivated. Given the success of some systems, such as Slashdot and Wikipedia, and given the increased connectivity of the world, the ability to understand the models of these sites and to reproduce them applied to a different concept is invaluable. Although there were no experiments and all of the analysis was solely empirical, the paper clearly illustrated its point. While the models described for rewards may lack enough granularity to accurately model the real world, they give a solid conceptual basis for analysis.

The "Predicting Tie Strength With Social Media" paper has a relatively strong argument. The problem is well motivated as there have been recent issues with privacy and safety when using social networks, for instance the situation where a girl was killed by a stalker on facebook or the situation where a young girl committed suicide after her online boyfriend, who was really a middle-aged man who posed himself as a teenager, broke up with her. Also, this problem extends the literature on defining social ties. While the model itself was quite comprehensive, it isn't totally clear on how this can be extended to other domains.

Bryan Trinh - 9/21/2010 18:58:30

Coase's Penguin or Linux, the Nature of the Firm

In this paper Yochai Benkler discusses the implications that crowd sourced information and open source communities have on our world. He gives several examples of how services and information provided through online peer organizations can surpass those created by an organized firm.

This paper is important because it highlights the opportunities of human creation that were not ever possible without the internet and the connections that can be created by them. Especially in the domain of information gathering; the internet provides an extremely vast pool of information that can be analyzed in order to drive all types of beneficial output. By better understanding the mechanisms that create relevant and valuable information and services using peer production, we can extend this process to tackle ever larger problems.

Today there are a whole bunch of start ups that are trying to organize and make use of all the data that we have on the internet. The amount of unstructured data on the internet is ever increasing, but by using peer production ideas, people are organizing this information into more useful forms.

Peer production provides a very effective means of creating certain types of services and information, but there are probably many instances where they just cannot compete with a larger firm. It would be interesting to compare and contrast similar products produced by peer production and an organized firm in order to classify the set of problems that are better suited for organized development.

Predicting Tie Strength With Social Media

In this paper, Eric Gilbert and Karrie Karahalios implement a predictive model to find the tie strength between individuals in an online social network. Tie strengths are a set of parameters that attempt to place a numerical value on relationships. By analyzing data from Facebook, they attempted to corollate numerical data such as number of comments to a set of questions that determined how close the two individuals were. Their predictive model is created by using a well known strategy in design called design of experiments. The underlying assumption is that the value at hand, strength, is effected by the set of parameters decided by the experimenter.

This would be a useful model for generating news feeds or content for social network viewers. Features that seem like a black box to the user is a pretty good place to apply this formula. Actually I was surprised that facebook did not already use something like this for their news feeds.

This model has its limitations though. I wouldn't trust such a simple model, let alone any model, to control my privacy settings or information feeds. Basically in any situation where the user has complete control, I think that these sorts of analytics only confuses users. Instead of figuring out for the user, figure out a better way to aid the user. Models such as these just cannot predict the intricacies of human behavior and social networks.

The authors of this paper essentially used techniques in experiment designs to model social behavior. I don't think this paper does anything more than provide a simple heuristic for facebook to decided news feed order. Any other automated smarts would put facebook users in an outrage.

Linsey Hansen - 9/21/2010 18:59:06

In his article Benklar describes “peer productions” and how they can be more efficient in certain situations as opposed to market or hierarchy based productions. He then goes on to explain why people might be interested in doing free work, and presents a function of rewards being a combination of intrinsic and monetary values.

Considering that this is a recent article, I feel that Benklar hit most topics dead on, and the fact that he did admit that “peer production” is only good for certain cases, as opposed to being either extremely for it or against it, managed to cover up any potential blind spots. I definitely feel like forms of peer production can be used to cover things that were once considered more “professional” tasks, especially when presented in a fun way. For example, the click workers seems like a really efficient way to free up scientists and grad students, but then there is always the probability of the majority getting something terribly wrong (though there could always be professionals checking median results occasionally).

More research can definitely be done in the future on peer production-like tasks, which can maybe help get certain tasks done even faster, while sparing those who once did the task plenty of boredom (though on the other hand, this could do damage to the currently delicate economy if it completely opened certain positions).

Gilbert and Karahalios on the other hand discuss current relationship branding in social media, and how it can be improved with a new method of differentiating strong from weak tie strength.

In their method, they look at variables reflecting intimacy, duration, reciprocal services, emotional support, social distance, demographic and usage, and intensity, in order to better predict the ties in a relationship. Participants would then be asked questions about certain friends relating to these things, and eventually relating to the bond the participant felt towards the person. Aside from a little error, the model was mostly able to accurately predict the strength of a tie. As the author's suggest, the user could then use this data to limit the privacy of certain friends- ie there could be an automatic “friend ranking” ai based on interactions that the user has with a particular person.

Thomas Schluchter - 9/21/2010 19:00:34

    • Coase's penguin

Benkler's article describes a framework for understanding what makes large-scale collaborations in a digital environment work despite the absence of what has traditionally been understood to motivate it: market incentives or managerial structures. The framework makes it possible to evaluate whether certain environments/constellations foster or smother peer production.

I found the perspective of the paper very useful: It shows that peer production has certain advantages over other forms of organized labor, but that the conditions to make it 'tick' are very specific. It thus shows that peer production will not supplant markets or firms due to the fairly volatile structure of commitment, but rather that it can complement both. The key strenghts, discovering and allocating human talent to projects, seems indeed to be one of the more challenging aspects of running a business. Peer production has the advantage that people self-select which virtually guarantees motivation. The trade-off is that there needs to be a balancing act of moderation. Not all contributors are equally qualified, and ensuring quality of the end product is one of the primary challenges for peer-produced works.

This leads to the very interesting question how labor is distributed in communities of volunteers. The things that one is allowed to do, or the parts of a product that one can influence through one's contributions become determinants of reputation. In this regard, peer production is not so different from more traditional forms of labor. The interesting aspect here is that due to the structural weakness of other traditional incentives (like money), the role of reputation becomes far more important.

The model of the three components (M, H, SP) to motivation can probably be extended to a generalized theory of what makes people collaborate. In the light of our discussion from last time regarding the adoption of groupware in organizations, this angle seems very promising as a design complement to technical solutions.

    • Predicting tie strength

This paper presents research on a method to qualify the notion of a tie (or a connection) beyond the binary state that most online social networks currently use by looking at logged relationship data. The prediction method is validated against self-reported evaluations of participants' relationships.

The prospect of having more intelligent social networks that adapt their behavior to users' behavior is not particularly new. Facebook's business model relies on this in some areas, such as targeted advertising, but the approach seems promising when it comes to making sense of relationship data to actually say something about the relationship itself. Particularly the adaptive privacy settings would be a quantum leap in usability of these systems. Facebook has time and again been criticized for the complexity of their privacy settings. The problem with this criticism, partly, is that the relationship structures we ourselves build on Facebook are blown out of proportion. The average number of friends being at 300, as the paper indicates, illustrates that the indiscriminate adding of people to one's networks is what materially contributes to privacy problems.

One of the key methodical elements of the paper is to validate the predicitions against self-reported evaluations. The authors do not talk about where the scale for this particular instrument came from. If they've developed it themselves, it would be nice to have some pre-test data on the performance of that instrument. At first sight, it seems like for five questions, this should better be very highly optimized for validity and reliability.

Richard Shin - 9/21/2010 19:01:56

Coase's Penguin, or Linux and the Nature of the Firm

This paper explores the recently-arisen trend of peer production of information, or what we might today call "crowdsourcing". The authors describe how inexpensive computers and the internet have led to unprecedentedly-low costs of capturing information and disseminating it, making human creativity itself the most salient part of creating and distributing information. The paper argues that peer production, unlike existing structures such as firms or markets, use human creativity the most efficiently, by avoiding transaction costs. Starting from the example of open-source software, the paper describes how the peer production model has been applied to various fields such as locating craters from satellite images of Mars, building an encyclopedia, and moderating internet forum comments.

Unlike other papers that we have read, the topic of this paper didn't seem particularly related to human-computer interaction; perhaps it could be better described as how computers mediate human creativity, allowing large numbers of people to effectively work together. Nevertheless, I thought that this new mode of collaboration identified and explored in the paper was interesting and relevant. One would need to look no further than the massive success that Wikipedia (in the paper, described as a fledgling project to build an encyclopedia) has become, to determine the potential of a system that allows a large number of people to contribute a little bit each. Similarly, I thought that the authors clearly identified the value provided by traditional mechanisms for producing information (producing actual 'utterance', distributing it widely, and identifying it as credible and relevant), and how peer production can also address these needs.

However, I didn't find part 3, specifically where the author analyzes what might motivate people to take part in peer production, quite as convincing. While the paper does discuss the issue somewhat formally, by dividing motivation into three parts (monetary, hedonistic, and social/psychological), it still seemed firmly grounded in intuition rather than data. It seemed to me that this part of the paper would especially have been apt for peer production, by asking those who partake in it about why they do so and summarizing the results, rather than building a model in a vacuum.

Arpad Kovacs - 9/21/2010 19:02:21

The Gilbert paper explores how accurately the traditional 7 dimensions of Intensity, Intimacy, Duration, Reciprocal Services, Structural, Emotional Support and Social Distance can predict tie strength between individuals on social networking websites such as Facebook. The authors' approach was to ask each participant 5 survey questions about the strength of their relation with each of their friends, and in the meantime datamining various indicators of each connection such as communication reciprocity, mutual friends, recency of communication, and interaction frequency.

I am not quite sure what the contribution of this paper to HCI is; it seems to be more of a social-science/psychology-oriented publication. Nevertheless, I found it interesting that it is indeed possible to gauge the strength of a relationship with a high degree of accuracy using only 15 predictive factors. The most interesting finding was that inbox thread depth was inversely correlated to tie strength; apparently close friends on facebook like to discss a wide variety of random topics, while presumably business/professional relationships are more focused and involve conversations of longer. I also think that the importance of wall postings in the predictive factors is quite an important finding; it would be wise for facebook to expend the most effort on this feature rather than applications, since it appears that the wall is where the most significant interactions occur.

I thought that the authors' model, which is just a linear combination of the predictive factors, was quite simplistic, but as shown by their results, it appears to work quite well. I am quite interested in how well these findings generalize to other websites, such as LinkedIn, whose focus would be professional relationships, and involves less intimate, longer-term interactions; I would predict that structural and social distance would play a much greater role.

The Benkler paper explores the issues of motivation in a commons-based peer production environment. The author surmises that the open-source/free software movement is a new socio-economic model, which is unlike the traditional property and contract-based economics that drive firms and markets. Instead, individuals who contribute to these large, collaborative projects have their own unique, and often intrinsic motivations, such as recognition, a need to learn, etc. This process has been catalyzed by the declining price of physical capital and communications, which emphasize that human capital is the primary asset, and make network-based collaboration.

The problem with the open-source movement is that due to the heterogeneous motivations of the contributors, and the lack of a hierarchical power structure, individuals make small contributions, but rarely is there a unified vision behind the whole. As a result, the open-source community consists of a patchwork of perpetually unfinished, unpolished projects, as well as a vast wasteland of the detritus of abandoned projects. As a result, Benkler advocates that it should be companies who embed information into these projects, and leverage the power of volunteers to reduce their own costs of production, while providing a vision and guidance towards a successful, but open product.

Kenzan boo - 9/21/2010 19:17:15

Predicting Tie Strength With Social Media This is an excellent article pointing out one of the finer details in relationships to other people which I feel face book has missed. The article points out how many people are connected to each other by many measurable data points gathered from their interactions on facebook such as messages sent and their high correlation to the users explicit statement of how strong their tie is with the other person. To a large part, I feel most networking sites have missed this portion of social interactions. Their ways of measuring correlation between connections is effective, however like they said, they have missed a lot of other aspects of human interactions like interacting through a child’s facebook. While this is a great way of mining data to get more information about human data, it does miss a few points. However, given that we already have almost all this data on servers on facebook’s end, with the user’s permission and cooperation with facebook we could design much better advertising or social recommendations.

Coase’s Penguin, or, Linux and The Nature of the Firm The article discusses the merits of open source and general human contribution to a knowledge base without requiring payment or ownership of a product. This reminds me of a speech I once heard that the internet is a place full of kindness. Everyone everywhere is coming together to help each other. Think about forums where someone new to a product comes to ask a question about something. For example how to create an adhoc computer. Many people will come and answer the question. Another example is stackoverflow.com where people come to share knowledge about programming. The biggest example of this I wikipedia.com which has grown to the most diverse knowledge base in the world.

Richard Shin - 9/21/2010 19:44:56

I only discussed the first reading in the response I submitted earlier, so here is my response to the second reading:

Predicting Tie Strength With Social Media

This paper describes a system that statistically predicts 'tie strength' between two connected people by measuring their engagement with each other in the Facebook service. The authors constructed a system that asks participants in their study to self-report the strength of their tie with a random selection of Facebook friends, and then collects a variety of data such as the frequency and intensity of their communication, the number of mutual friends, their social distance (difference in age, education level, political views, etc.). Then, using this data, the authors predicted the tie strength, compared it to the self-reported tie strength, and determined the correlation between parts of data and the tie strength.

I thought that the idea of automatically determining which social ties are important, and which are not, is the fundamental contribution of this paper; as the paper notes, knowing this information could enable a variety of useful functions, such as delivering only the status updates from the users that are deemed most relevant to the user. This idea has been applied fairly widely to social networking services today, such as Facebook which recommends 'people you may know' and sorts the News Feed by relevance, and Twitter which suggests new people to follow based on similar data.

However, while the work in this paper seemed promising, it didn't seem to me that the predictions were necessary accurate enough to be useful for some of the cases presented in the paper, such as automatically deciding to whom to make your private profile visible. Perhaps, the data could be more useful if, instead of optimizing for general correlation between the predicted tie strength and the self-reported tie strength, eliminating false positives or false negatives were instead emphasized; then these two forms of data could be used for different purposes, such as friend recommendations and automatically revealing data to predicted ties, for example.