http://www.hpl.hp.com/research/scl/ Recent papers from HP Labs' Social Computing Lab en-us Wed, 17 Apr 2013 21:14:21 GMT Wed, 17 Apr 2013 21:14:21 GMT http://www.hpl.hp.com/research/scl/papers/adaptive/adaptive.pdf How to increase the cost of access without loosing the audience When should a necessary inconvenience be introduced gradually, and when should it be imposed all at once? The question is crucial to web content providers, who in order to generate revenue must sooner or later introduce advertisements, subscription fees, or other inconveniences. In a setting where people eventually fully adapt to changes, the answer depends on the shape of the "survivor curve" S(x), which represents the fraction of a user population willing to tolerate inconveniences of size x (Aperjis and Huberman 2011). We report a new laboratory experiment that, for the first time, estimates the shape of survivor curves in several different settings. We engage laboratory subjects in a series of six desirable activities, e.g., playing a video game, viewing a chosen video clip, or earning money by answering questions. For each activity we introduce a chosen level of a particular inconvenience, and each subject chooses whether to tolerate the inconvenience or to switch to a bland activity for the remaining time. Our key finding is that the survivor curve is log-concave in all six activities. Theory suggests that web content providers therefore will generally find it profitable to introduce inconveniences gradually over time, with the timing chosen to balance the number of long-term users against more rapid revenue acquisition. Christina Aperjis, Ciril Bosh-Rosa, Daniel Friedman and Bernardo A. Huberman 19 Feb 2013 15:44:56 -0800 adaptation ecommerce social http://www.hpl.hp.com/research/scl/papers/completion/completion.pdf Making others truthful Both in online labor markets and within enterprises, a worker often has better information than his employer about the likelihood of completing a task or project on time. This information asymmetry prevents managers from having accurate estimates of project completion times. We design and experimentally test an incentive mechanism which induces workers to reveal their true likelihoods of completion times while minimizing the expected payments by the employers. Our results show that our mechanism performs very well at the aggregate level. Jing Wang, Christina Aperjis and Bernardo A. Huberman 13 Feb 2013 10:38:10 -0800 truthful reporting social http://www.hpl.hp.com/research/scl/papers/socialmedia/IntegrateSources.pdf Dirichlet Topic models for integrating content from heterogeneous web sources In recent years, the phenomenal growth and popularity of social media, news and discussion websites has led to a vast number of information sources available online. These sources generate massive amounts of real-time content on a daily basis making it increasingly difficult to glean true and relevant information from them. In this paper, we discuss how content can be integrated from heterogeneous web sources such as news, blogs and social media to reduce noise and redundancy, and extract relevant contextual information in the form of latent topics, essential for tasks such as web document classification. First, we describe a probabilistic extension of LDA (Probabilistic Source LDA) that can handle heterogeneous sources. We then proceed to introduce a novel discriminative topic model, Discriminative Dirichlet Allocation (DDA) which is naturally designed to work with heterogeneous sources and is more attuned to the words of the documents and their relationships. We compare and contrast these methods using real data on the US elections 2012. Our results demonstrate that the Discriminative Dirichlet Allocation method can extract highly relevant and discriminative topics that outperform other LDA-based methods. Rumi Ghosh and Sitaram Asur 27 Nov 2012 13:25:22 -0700 topic models heterogeneous sources data integration latent topics http://www.hpl.hp.com/research/scl/papers/socialmedia/tweet_summary.pdf Generating automatic short summaries of an event from social media Social media services such as Twitter generate phenomenal volume of content for most real-world events on a daily basis. Digging through the noise and redundancy to understand the important aspects of the content is a very challenging task. We propose a search and summarization framework to extract relevant representative tweets from an unfiltered tweet stream in order to generate a coherent and concise summary of an event. We introduce two topic models that take advantage of temporal correlation in the data to extract relevant tweets for summarization. The summarization framework has been evaluated using Twitter data on four real-world events. Evaluations are performed using Wikipedia articles on the events as well as using Amazon Mechanical Turk (MTurk) with human readers (MTurkers). Both experiments show that the proposed models outperform traditional LDA and lead to informative summaries. Freddy Chong Tat Chua and Sitaram Asur 27 Nov 2012 13:20:22 -0700 summarization social media topic models temporal correlation http://www.hpl.hp.com/research/scl/papers/echo/echo.pdf Using the wisdom the crowd to solve attention problems The explosive growth of user generated content, along with the continuous increase in the amount of traditional sources of content, has made it extremely hard for users to digest the relevant pieces of information that they need to pay attention to in order to make sense of their needs. Thus, solutions are needed to help both professionals (e.g lawyers, analysts, economists) and ordinary users navigate this flood of information. We present a novel interaction model and system called Echo which uses machine learning techniques to traverse a corpus of documents and distill crucial opinions from the collective intelligence of the crowd. Based on this analysis, Echo creates an intuitive and elegant interface, as though constructed by an editor, that allows users to quickly find salient documents and opinions, all powered by the wisdom of the crowd. The Echo UI directs the user's attention to critical opinions using a natural magazine style metaphor, with visual call outs and other typographic changes. This technique allows a user to read as normal, while focusing her attention on the important opinions within documents, and showing how these opinions relate to those of the crowd. Joshua Hailpern and Bernardo A. Huberman 19 Sep 2012 11:25:22 -0700 attention collective social computing http://www.youtube.com/watch?v=MGLKFE_eEWg Video of Echo The explosive growth of user generated content, along with the continuous increase in the amount of traditional sources of content, has made it extremely hard for users to digest the relevant pieces of information that they need to pay attention to in order to make sense of their needs. Thus, solutions are needed to help both professionals (e.g lawyers, analysts, economists) and ordinary users navigate this flood of information. We present a novel interaction model and system called Echo which uses machine learning techniques to traverse a corpus of documents and distill crucial opinions from the collective intelligence of the crowd. Based on this analysis, Echo creates an intuitive and elegant interface, as though constructed by an editor, that allows users to quickly find salient documents and opinions, all powered by the wisdom of the crowd. The Echo UI directs the user's attention to critical opinions using a natural magazine style metaphor, with visual call outs and other typographic changes. This technique allows a user to read as normal, while focusing her attention on the important opinions within documents, and showing how these opinions relate to those of the crowd. Joshua Hailpern and Bernardo A. Huberman 19 Sep 2012 15:25:22 -0700 attention collective social computing video http://www.hpl.hp.com/research/scl/papers/privatedata/PricingPrivateData.pdf Taking advantage of risk aversion to decrease the price We consider a market where buyers can access unbiased samples of private data by appropriately compensating the individuals to whom the data corresponds (the sellers) according to their privacy attitudes. We show how bundling the buyers' demand can decrease the price that buyers have to pay per data point, while ensuring that sellers are willing to participate. Our approach leverages the inherently randomized nature of sampling, along with the risk-averse attitude of sellers in order to discover the minimum price at which buyers can obtain unbiased samples. We take a prior-free approach and introduce a mechanism that incentivizes each individual to truthfully report his preferences in terms of different payment schemes. We then show that our mechanism provides optimal price guarantees in several settings. Vasilis Gkatzelis, Christina Aperjis, and Bernardo A. Huberman 14 Sep 2012 15:54:36 -0700 privacy private data pricing markets for data social computing economics electronic commerce http://www.hpl.hp.com/research/scl/papers/random/random.pdf Online interactions are far more predictable than it appears The massive amounts of data that social media generates has facilitated the study of online human behavior on a scale unimaginable a few years ago. At the same time, the much discussed apparent randomness with which people interact online makes it appear as if these studies cannot reveal predictive social behaviors that could be used for developing better platforms and services. We use two large social databases to measure the mutual information entropy that both individual and group actions generate as they evolve over time. We show that user's interaction sequences have strong deterministic components, in contrast with existing assumptions and models. In addition, we show that individual interactions are more predictable when users act on their own rather than when attending group activities. Chunyan Wang and Bernardo A. Huberman 30 Jul 2012 15:22:10 -0700 social computingonline interactions http://firstmonday.org/htbin/cgiwrap/bin/ojs/index.php/fm/article/view/4013/3209 How to get paid when they use your data Since there is, in principle, no reason why third parties should not pay individuals for the use of their data, we introduce a realistic market that would allow these payments to be made while taking into account the privacy attitude of the participants. And since it is usually important to use unbiased samples to obtain credible statistical results, we examine the properties that such a market should have and suggest a mechanism that compensates those individuals that participate according to their risk attitudes. Equally important, we show that this mechanism also benefits buyers, as they pay less for the data than they would if they compensated all individuals with the same maximum fee that the most concerned ones expect. Christina Aperjis and Bernardo A. Huberman 25 Apr 2012 16:45:02 -0700 social computing private data markets for data http://www.hpl.hp.com/research/scl/papers/comments/comments.pdf Predicting the growth dynamics and structural properties of conversation threads We present an analysis of user conversations in on-line social media and their evolution over time. We propose a dynamic model that accurately predicts the growth dynamics and structural properties of conversation threads. The model successfully reconciles the differing observations that have been reported in existing studies. By separating artificial factors from user behaviors, we show that there are actually underlying rules in common for on-line conversations in different social media websites. Results of our model are supported by empirical measurements throughout a number of different social media websites. Chunyan Wang, Mao Ye and Bernardo A. Huberman 27 Feb 2012 15:23:43 -0800 social computing social media blogs.