Search

The machine learning (ML) research community has landed on automated hate speech detection as the vital tool in the mitigation of bad behavior online. However, it is not clear that this is a widely supported view outside of the ML world. Such a disconnect can have implications for whether automated detection tools are accepted or adopted. Here we lend insight into how other key stakeholders understand the challenge of addressing hate speech and the role automated detection plays in solving it. To do so, we develop and apply a structured approach to dissecting the discourses used by online platform companies, governments, and not-for-profit organizations when discussing hate speech. We find that, where hate speech mitigation is concerned, there is a profound disconnect between the computer science research community and other stakeholder groups—which puts progress on this important problem at serious risk. We identify urgent steps that need to be taken to incorporate computational researchers into a single, coherent, multistakeholder community that is working towards civil discourse online.

We explore how platforms influence user-generated content by comparing reviews made on the retail platform Amazon, with those on the (non-retail) community platform Goodreads. We find the retail setting gives rise to shorter, more declamatory, and persuasive reviews, while the non-retail community generates longer, more reflective, tentative reviews with more diverse punctuation. These differences are pronounced enough to enable automatic inference of the platform from which reviews were taken with over 90% F1. Significant differences in star-ratings appear to parallel differences in review content. Both platforms allow users to give feedback on reviews. Experiments show that a subtle difference in the review feedback features influences review-promotion behavior, which may in part explain the differences in review content. Our results show that the context and design of a platform has a strong but subtle effect on how users write and engage with content.

Platforms like Reddit have attracted large and vibrant communities, but the individuals in those communities are free to migrate to other platforms at any time. History has borne this out with the mass migration from Slashdot to Digg. The underlying motivations of individuals who migrate between platforms, and the conditions that favor migration online are not well-understood. We examine Reddit during a period of community unrest affecting millions of users in the summer of 2015, and analyze large-scale changes in user behavior and migration patterns to Reddit-like alternative platforms. Using self-reported statements from user comments, surveys, and a computational analysis of the activity of users with accounts on multiple platforms, we identify the primary motivations driving user migration. While a notable number of Reddit users left for other platforms, we found that an important pull factor that enabled Reddit to retain users was its long tail of niche content. Other platforms may reach critical mass to support popular or “mainstream” topics, but Reddit’s large userbase provides a key advantage in supporting niche topics.

Exercise plays a central role in many peoples' fitness goals. While prior work has examined how individuals pursue these health and fitness goals on general purpose platforms such as Twitter, the lack of precise activity recording has limited detailed analyses of individual and group behaviors. In this study, we explore a recent social media platform dedicated to exercise and use nearly four years of longitudinal exercising history of over 188,000 users to discover large-scale exercising patterns corresponding to different motivations such as sports or general fitness.

Much work on the demographics of social media platforms such as Twitter has focused on the properties of individuals, such as gender or age. However, because credible detectors for organization accounts do not exist, these and future large-scale studies of human behavior on social media can be contaminated by the presence of accounts belonging to organizations. We analyze organizations on Twitter to assess their distinct behavioral characteristics and determine what types of organizations are active. We first create a dataset of manually classified accounts from a representative sample of Twitter and then introduce a classifier to distinguish between organizational and personal accounts. In addition, we find that although organizations make up less than 10% of the accounts, they are significantly more connected, with an order of magnitude more friends and followers.

Geolocated social media data provides a powerful source of information about place and regional human behavior. Because little social media data is geolocation-annotated, inference techniques serve an essential role for increasing the volume of annotated data. One major class of inference approaches has relied on the social network of Twitter, where the locations of a user's friends serve as evidence for that user's location. While many such inference techniques have been recently proposed, we actually know little about their relative performance, with the amount of ground truth data varying between 5% and 100% of the network, the size of the social network varying by four orders of magnitude, and little standardization in evaluation metrics. We conduct a systematic comparative analysis of nine state-of-the-art network-based methods for performing geolocation inference at the global scale, controlling for the source of ground truth data, dataset size, and temporal recency in test data. Furthermore, we identify a comprehensive set of evaluation metrics that clarify performance differences. Our analysis identifies a large performance disparity between that reported in the literature and that seen in real-world conditions. To aid reproducibility and future comparison, all implementations have been released in an open source geoinference package.

Social media data mining and analytics has stimulated a wide array of computational research. Traditionally, individual researchers are responsible for acquiring and managing their own datasets. However, the temporal nature of social data, the challenges involved in correctly preparing a dataset, the sheer scale of many datasets, and the proprietary nature of many data sources can make extending and comparing computational methods difficult and often impossible. In light of this, because replicability is a fundamental pillar of the scientific process and because method comparison is essential to characterizing computational advancements, we require an alternative to the traditional model of researcher-owned datasets. In this paper we propose FREESR, a framework that gives researchers the ability to develop and test method performance without requiring direct access to “shared” datasets. As a case study and first community resource, we have implemented the FREESR paradigm around the task of Tweet geolocation. The implementation showcases the clear suitability of this framework for the social media research context. Beyond the implementation, we see the FREESR paradigm as being an important step towards making study reproducibility and method comparison more principled and ubiquitous in the social media research community.

Book reviewing is a commonplace activity on many ecommerce sites. However, it is nested within the broader context of book buying and selling. Goodreads, an online platform for social curation of book collections, provides an opportunity to observe on-line book reviewing in an environment that is not (at least overtly) focused on commercialization. In this study, we perform a careful comparative study of reviewer behavior and engagement in Goodreads and Amazon.com, constrained to a single genre (biography), including 21,394 books and 2.5 million reviews. We discover marked differences between the platforms that suggest disparate population composition and objectives of review-writing across the two platforms. Our findings suggest an important and generalizable principle: that two platforms engaging users on the same task (e.g., book review writing) may elicit quite different behavior depending on the implicit or explicit context and motivation present.

In order for a municipality to effectively service and engage its constituency, it must understand the composition of the communities within it. Up to the present, such demographic estimates for target populations have been obtained largely from census data or expensive, time-intensive surveys. In this paper, we use Twitter microblog content to estimate the gender makeup of commuting populations using different modes of transportation (cars, public transportation, and bikes) in Toronto, Canada. We apply a demographic inference algorithm to 33,215 public Twitter accounts that follow one of three popular transportation-related Twitter-based news feeds (one for traffic, one for public transportation updates, and one for bicycling). Recent census data provides ground truth against which to compare the estimates we derive from Twitter. We find that, for all three communities (car drivers, public transport users, and bicyclists), the estimates obtained from Twitter reflect the majority-minority relationships between genders reported in census data. This provides preliminary, but compelling evidence that Twitter may be a platform that can go beyond simply signaling the presence of physical communities to actually measure their compositions.