Groups of firms often achieve a competitive advantage through the formation of geo-industrial clusters. Although many exemplary clusters, such as Hollywood or Silicon Valley, have been frequently studied, systematic approaches to identify and analyze the hierarchical structure of the geo-industrial clusters at the global scale are rare. In this work, we use LinkedIn's employment histories of more than 500 million users over 25 years to construct a labor flow network of over 4 million firms across the world and apply a recursive network community detection algorithm to reveal the hierarchical structure of geo-industrial clusters. We show that the resulting geo-industrial clusters exhibit a stronger association between the influx of educated-workers and financial performance, compared to existing aggregation units. Furthermore, our additional analysis of the skill sets of educated-workers supplements the relationship between the labor flow of educated-workers and productivity growth. We argue that geo-industrial clusters defined by labor flow provide better insights into the growth and the decline of the economy than other common economic units.

Word embedding spaces are powerful tools for capturing latent semantic relationships between terms in corpora, and have become widely popular for building state-of-the-art natural language processing algorithms. However, studies have shown that societal biases present in text corpora may be incorporated into the word embedding spaces learned from them. Thus, there is an ethical concern that human-like biases contained in the corpora and their derived embedding spaces might be propagated, or even amplified with the usage of the biased embedding spaces in downstream applications. In an attempt to quantify these biases so that they may be better understood and studied, several bias metrics have been proposed. We explore the statistical properties of these proposed measures in the context of their cited applications as well as their supposed utilities. We find that there are caveats to the simple interpretation of these metrics as proposed. We find that the bias metric proposed by Bolukbasi et al. 2016 is highly sensitive to embedding hyper-parameter selection, and that in many cases, the variance due to the selection of some hyper-parameters is greater than the variance in the metric due to corpus selection, while in fewer cases the bias rankings of corpora vary with hyper-parameter selection. In light of these observations, it may be the case that bias estimates should not be thought to directly measure the properties of the underlying corpus, but rather the properties of the specific embedding spaces in question, particularly in the context of hyper-parameter selections used to generate them. Hence, bias metrics of spaces generated with differing hyper-parameters should be compared only with explicit consideration of the embedding-learning algorithms particular configurations.

Entity linking is the task of mapping potentially ambiguous terms in text to their constituent entities in a knowledge base like Wikipedia. This is useful for organizing content, extracting structured data from textual documents, and in machine learning relevance applications like semantic search, knowledge graph construction, and question answering. Traditionally, this work has focused on text that has been well-formed, like news articles, but in common real world datasets such as messaging, resumes, or short-form social media, non-grammatical, loosely-structured text adds a new dimension to this problem. This paper presents Pangloss, a production system for entity disambiguation on noisy text. Pangloss combines a probabilistic lineartime key phrase identification algorithm with a semantic similarity engine based on context-dependent document embeddings to achieve better than state-of-the-art results (>5% in F1) compared to other research or commercially available systems. In addition, Pangloss leverages a local embedded database with a tiered architecture to house its statistics and metadata, which allows rapid disambiguation in streaming contexts and on-device disambiguation in low-memory environments such as mobile phones.

The study of social phenomena is becoming increasingly reliant on big data from online social networks. Broad access to social media data, however, requires software development skills that not all researchers possess. Here we present the IUNI Observatory on Social Media , an open analytics platform designed to facilitate computational social science. The system leverages a historical, ongoing collection of over 70 billion public messages from Twitter. We illustrate a number of interactive open-source tools to retrieve, visualize, and analyze derived data from this collection. The Observatory, now available at osome.iuni.iu.edu , is the result of a large, six-year collaborative effort coordinated by the Indiana University Network Science Institute.

With the exploding popularity of online social networks and microblogging platforms, social media have become the turf on which battles of opinion are fought. This section discusses a particularly insidious type of abuse of social media, aimed at manipulation of political discourse online. Grassroots campaigns can be simulated using techniques of what has come to be known as astroturf with the goal of promoting a certain view or candidate, slandering an opponent, or simply inciting or suppressing the vote. Such deception threatens the democratic process. We describe various attacks of this kind and a system designed to detect them. 6.3.1 The Rise of Online Grassroots Political Movements The 2008 presidential election will go down in history as the first to be dominated by grassroots movements organized and coordinated online. The ultimate success of Senator Obama’s campaign was due in no small part to its pioneering use of social media. An approach of direct dialog with his grassroots supporters captivated and connected with untapped layers of society and initiated a new era of political participation in American politics. On the other side of the aisle, the aftermath of the election brought about a reaction 232 Chapter 6 The Internet and the Physical World culminating in the Tea Party movement [397]. In both cases, it was clear that citizens would no longer be content as passive targets of political messages. They demanded an increased role in defining political discourse. As individuals gradually turn to the Internet in search of political and economic information, they naturally use existing social networks and platforms to discuss their views and ideals with their peers. Microblogging tools such as Twitter play an important role in this movement by allowing individuals to act as media aggregators and curators who are able to influence their followers and who are, in turn, influenced by the people that they elect to follow. Over time, trust develops between followers and followees making the latter more likely to accept content and information provided by the former. Perhaps the most striking demonstration of the relevance of this type of discourse, and of how aligned it is with public opinion at large, can be found in a 2010 paper by Tumasjan et al. [470]. By analyzing over 100, 000 tweets containing direct political references to parties or politicians in the ramp-up to the German Federal Election in 2009, they found that the fraction of activity within Twitter corresponding to each party closely matched the vote shares in the final election results. If this result could be generalized, this would imply that Twitter can be used as a real-time public sentiment monitoring tool. Based on this finding, Tumasjan et al. proposed that Twitter be used as a distributed real-time campaign monitoring tool. If it is true that Twitter truly mirrors public perception, then perhaps it is also true that by manipulating perception within Twitter one is also able to manipulate it in the real world. This inference has not escaped the attention of institutions and groups interested in promoting specific topics or actions. Mustafaraj and Metaxas [364] studied in detail one such case that occurred during the 2010 special Massachusetts senate election. They observed how a network of fake, computer controlled, accounts produced almost 1000 tweets in just over 2 h, containing a specific URL smearing one of the candidates. The goal of the perpetrators was to generate as much traffic as possible to reach a wide audience and thus influence the outcome of the election. To achieve this goal, specific users perceived as influential were they targeted in hopes that they would retweet the URL, thus bestowing upon it an added layer of credibility. Blunt as it was, this attempt was extremely successful, generating such a large number of retweets to briefly place the URL in the first page of Google results for the query Martha Coakley—the name of the smeared candidate. Coordinated deception of this sort, where a single agent forges the appearance of widespread support for an idea or position, is known as astroturfing, a name that stems from the parallel between fake grassroots movement and the common type of artificial grass used in sports stadiums. In this section, we look in detail at several tactics being used to promote misinformation in a covert way and a system that aims to automatically detect and track such attempts. 6.3.2 Spam and Astroturfing As anyone with an email inbox is well aware, spammers have decades of experience in reaching huge audiences. Their techniques range from simple mass email campaigns to sophisticated techniques that automatically customize each message to avoid detection by automated countermeasures. As with other communication media in the past, spammers have descended upon Twitter and adapted their toolbox to this new medium. Many of these techniques and potential countermeasures have been analyzed in detail [113,223,479,499]. Although there seems to be limited amounts of collusion between spammer accounts [223] in the form of spam campaigns designed to make users click a specific URL, there are specific characteristics that can identify spammer accounts. Defining features include the frequency of tweets, the age of the accounts, 6.3 Abuse of Social Media and Political Manipulation 233 and their periphery in the social graph [499]. The combination of content and user behavior attributes makes it possible to train machine learning algorithms to automatically detect spam accounts with a large accuracy [113]. This is likely due to the fact that spam relies on large numbers of accounts controlled by a small number of spammers. At first glance the goals of spammers and astroturfers might seem similar. They both want to communicate a message to a large audience of users, and both want to effect action (clicks, votes, changes of opinion) in the targeted users. However, there are several fundamental differences between the two types of attacks. Astroturfers, to create the illusion of widespread autonomous support, must retain some degree of credibility and appear independent with respect to commercial or political interests. Likewise, while spammers can use a single account to target many users, astroturfers rely on the fact that users are more receptive to messages they perceive as coming from multiple independent sources. These different techniques necessitate distinct approaches to the detection problem. Spam detection systems often focus on the content of messages—for instance, determining whether the message contains a certain link or set of tags. In detecting astroturf, the focus must be on how the message is delivered rather than its content. The fact that the message is delivered in the guise of legitimate online chatter instead of an organized campaign is more relevant than its veracity. Content may be a legitimate opinion or information resource; the fraud is not a product of the content but rather the distribution mechanism. Further, many of the users involved in propagating a successful astroturf message may in fact be legitimate users who are unwittingly complicit in the deception, having been deceived themselves. Thus, methods for detecting spam that focus on properties of user accounts, such as the number of URLs in tweets originating from an account or the interval between successive tweets, are likely to be unsuccessful in the detection of astroturf. A normal user may come to believe and disseminate a piece of information that had its origins on a campaign of this type. As more and more normal users join the dissemination of this message, any information that could potentially be extracted from analyzing the properties of the accounts spreading it will become increasingly muddled. 6.3.3 Deceptive Tactics Anyone trying to increase their visibility on Twitter has an obvious strategy: create an account, start tweeting and gradually accumulate followers. However, the egalitarian nature of the platform means that they are just one voice in a crowd of millions. When the goal is to have your voice heard no matter what the cost, several deceptive tactics can be used to quickly gain a large number of followers and obtain an aura of influence or importance within the community [141]. 6.3.3.1 Centrally and computer controlled accounts The old tenet, “nothing attracts a crowd like a crowd” holds true online. Astroturfers take advantage of this fact to catalyze faux grassroots activity by creating the illusion that a large number of people are behind a message or movement. The simplest way to achieve this effect is the creation of multiple centrally controlled accounts, known as sockpuppets, which are used to simulate several independent actors promoting a coherent message. These accounts can then be used to broadcast a message seemingly independent of one another, or be manipulated to appear as though they are engaged socially with one another. One advantage of the first approach is that it creates the appearance of independent actors responding to an exogenous influence at the expense of the credibility that comes with a rich social circle. The second approach relies on social 234 Chapter 6 The Internet and the Physical World expectations to create the appearance of authenticity at the expense of appearing independent. Common to both of these approaches is a reliance on a large number of centrally coordinated accounts. To effectively astroturf at scale requires automation, and Chu et al. [153] studied the behavioral differences between real users and bots on Twitter. They distinguish between two types of bots: “benign” bots, which often self-identify as automated processes and simply relay information from RSS feeds or other automated sources; and “malicious” bots, which spread spam or malicious content while acting as real users. One of the key distinguishing features between humans and bots is that bots ten

We examine the temporal evolution of digital communication activity relating to the American anti-capitalist movement Occupy Wall Street. Using a high-volume sample from the microblogging site Twitter, we investigate changes in Occupy participant engagement, interests, and social connectivity over a fifteen month period starting three months prior to the movement's first protest action. The results of this analysis indicate that, on Twitter, the Occupy movement tended to elicit participation from a set of highly interconnected users with pre-existing interests in domestic politics and foreign social movements. These users, while highly vocal in the months immediately following the birth of the movement, appear to have lost interest in Occupy related communication over the remainder of the study period.

Social movements rely in large measure on networked communication technologies to organize and disseminate information relating to the movements' objectives. In this work we seek to understand how the goals and needs of a protest movement are reflected in the geographic patterns of its communication network, and how these patterns differ from those of stable political communication. To this end, we examine an online communication network reconstructed from over 600,000 tweets from a thirty-six week period covering the birth and maturation of the American anticapitalist movement, Occupy Wall Street. We find that, compared to a network of stable domestic political communication, the Occupy Wall Street network exhibits higher levels of locality and a hub and spoke structure, in which the majority of non-local attention is allocated to high-profile locations such as New York, California, and Washington D.C. Moreover, we observe that information flows across state boundaries are more likely to contain framing language and references to the media, while communication among individuals in the same state is more likely to reference protest action and specific places and times. Tying these results to social movement theory, we propose that these features reflect the movement's efforts to mobilize resources at the local level and to develop narrative frames that reinforce collective purpose at the national level.

The broad adoption of the web as a communication medium has made it possible to study social behavior at a new scale. With social media networks such as Twitter, we can collect large data sets of online discourse. Social science researchers and journalists, however, may not have tools available to make sense of large amounts of data or of the structure of large social networks. In this paper, we describe our recent extensions to Truthy, a system for collecting and analyzing political discourse on Twitter. We introduce several new analytical perspectives on online discourse with the goal of facilitating collaboration between individuals in the computational and social sciences. The design decisions described in this article are motivated by real-world use cases developed in collaboration with colleagues at the Indiana University School of Journalism. Author Keywords

Correction to A. Lourenço, M. Conover, A. Wong, A. Nematzadeh, F. Pan, H. Shatkay, and L.M. Rocha."A Linear Classifier Based on Entity Recognition Tools and a Statistical Approach to Method Extraction in the Protein-Protein Interaction Literature". BMC Bioinformatics 2011, 12(Suppl 8):S12. doi:http://10.1186/1471-2105-12-S8-S12.

Social computing technologies are pervasive in our work, relationships, and culture. Despite their promise for transforming the structure of communication and human interaction, the complex social dimensions of these technological systems often reproduce offline social ills or create entirely novel forms of conflict and deviance. This panel brings together scholars who study deviance and failure in diverse social computing systems to examine four design-related themes that contribute to and support these problematic uses: theft, anonymity, deviance, and polarization.

We examine partisan differences in the behavior, communication patterns and social interactions of more than 18,000 politically-active Twitter users to produce evidence that points to changing levels of partisan engagement with the American online political landscape. Analysis of a network defined by the communication activity of these users in proximity to the 2010 midterm congressional elections reveals a highly segregated, well clustered partisan community structure. Using cluster membership as a high-fidelity (87% accuracy) proxy for political affiliation, we characterize a wide range of differences in the behavior, communication and social connectivity of left- and right-leaning Twitter users. We find that in contrast to the online political dynamics of the 2008 campaign, right-leaning Twitter users exhibit greater levels of political activity, a more tightly interconnected social structure, and a communication network topology that facilitates the rapid and broad dissemination of political information.

The widespread adoption of social media for political communication creates unprecedented opportunities to monitor the opinions of large numbers of politically active individuals in real time. However, without a way to distinguish between users of opposing political alignments, conflicting signals at the individual level may, in the aggregate, obscure partisan differences in opinion that are important to political strategy. In this article we describe several methods for predicting the political alignment of Twitter users based on the content and structure of their political communication in the run-up to the 2010 U.S. midterm elections. Using a data set of 1,000 manually-annotated individuals, we find that a support vector machine (SVM) trained on hash tag metadata outperforms an SVM trained on the full text of users' tweets, yielding predictions of political affiliations with 91% accuracy. Applying latent semantic analysis to the content of users' tweets we identify hidden structure in the data strongly associated with political affiliation, but do not find that topic detection improves prediction performance. All of these content-based methods are outperformed by a classifier based on the segregated community structure of political information diffusion networks (95% accuracy). We conclude with a practical application of this machinery to web-based political advertising, and outline several approaches to public opinion monitoring based on the techniques developed herein.

In this study we investigate how social media shape the networked public sphere and facilitate communication between communities with different political orientations. We examine two networks of political communication on Twitter, comprised of more than 250,000 tweets from the six weeks leading up to the 2010 U.S. congressional midterm elections. Using a combination of network clustering algorithms and manually-annotated data we demonstrate that the network of political retweets exhibits a highly segregated partisan structure, with extremely limited connectivity between left- and right-leaning users. Surprisingly this is not the case for the user-to-user mention network, which is dominated by a single politically heterogeneous cluster of users in which ideologically-opposed individuals interact at a much higher rate compared to the network of retweets. To explain the distinct topologies of the retweet and mention networks we conjecture that politically motivated individuals provoke interaction by injecting partisan content into information streams whose primary audience consists of ideologically-opposed users. We conclude with statistical evidence in support of this hypothesis.

Online social media are complementing and in some cases replacing person-to-person social interaction and redefining the diffusion of information. In particular, microblogs have become crucial grounds on which public relations, marketing, and political battles are fought. We demonstrate a web service that tracks political memes in Twitter and helps detect astroturfing, smear campaigns, and other misinformation in the context of U.S. political elections. We also present some cases of abusive behaviors uncovered by our service. Our web service is based on an extensible framework that will enable the real-time analysis of meme diffusion in social media by mining, visualizing, mapping, classifying, and modeling massive streams of public microblogging events.

This is the training data used to produce the results shown in the paper listed below. Source: Sampled public tweets from Twitter streaming API. Date range: September 14 to October 27, 2010. Contains: data.arff: holds the un-resampled training data. data_balanced.arff: holds the resampled training data. data.instance_to_id.pickle: holds a Python pickle relating instance IDs in the data.arff file with Meme IDs in the Truthy database. To view the page for a particular meme ID, go to http://truthy.indiana.edu/m?id= Please cite: Jacob Ratkiewicz, Michael Conover, Mark Meiss, Bruno Goncalves, Alessandro Flammini, and Filippo Menczer. Detecting and Tracking Political Abuse in Social Media. Proc. 5th International AAAI Conference on Weblogs and Social Media ICWSM, 2011.

We study astroturf political campaigns on microblogging platforms: politically-motivated individuals and organizations that use multiple centrally-controlled accounts to create the appearance of widespread support for a candidate or opinion. We describe a machine learning framework that combines topological, content-based and crowdsourced features of information diffusion networks on Twitter to detect the early stages of viral spreading of political misinformation. We present promising preliminary results with better than 96% accuracy in the detection of astroturf content in the run-up to the 2010 U.S. midterm elections.