COPCI

2.2.1 Memes and Trends

The meme concept was first proposed by Richard Dawkins in his influential book “The Selfish Gene" [95]. Dawkins defines the meme as “a unit of cultural transmission, or a unit of imitation". Nowadays we have adopted this concept to represent hashtags, keywords, and URLs on the internet.

Tracking and grouping similar concepts is easier when they are presented as quantifiable units. Memes serve this purpose. Most of the studies that analyze the content generated online isolate memes as starting points for their initial datasets.

There has been a large body of work in the area of information diffusion through net- works. Several early models for information diffusion were inspired from classical disease propagation models in epidemiology, such as SIR and SIS [16]. There has also been exten- sive work on modeling the adoption or spread of an idea, content or product in a social network. Well known classes of models in this domain include Threshold [142] and Cascade models [135], that specify how a node adopts a particular idea or product based on the adoption pattern prevalent in its neighborhood. The concept of diffusion was initially intro- duced by social scientists and theory was developed to study how innovations and novelties spread [244]. Studies also define different categories for adopters such as innovators, early adopters, majority, and laggards based on their rank in involvement. Other related diffusion models for product marketing included the Bass [22] model that is based on an S-shaped adoption curve [122].

In recent work, Goel et.al proposes a formal measure, structural virality, of the degree to which a cascade reaches its audience through broadcast-like mechanisms vs. viral mecha- nisms [131]. The authors conduct a large scale empirical study of a billion diffusion events for news, videos, images and petitions on Twitter and observe a wide range of diverse cascading structures with varying structural virality, and show a low correlation between popularity and structural virality. The authors then show how a simple SIR model can capture several of the empirically-observed properties of the cascades. However, they note that their model could not explain the large variance in structural virality that they observed empirically.

Trendsrepresent interesting collective communication phenomena: they are user-generated, continually changing and mostly ungoverned (although orchestrated hijacking attempts have been observed [52,240,241]). Different information diffusion mechanisms may determine the trending dynamics of hashtags and other memes on social media. Exogenous and endogenous dynamics produce memes with distinctive characteristics [116, 119, 181, 216, 263]: external

events occurring in the real world (e.g., a natural disaster or a terrorist attack) can generate chatter on the platform and therefore trigger the trending of a new, unforeseen hashtag; other topics (e.g., politics or entertainment) are continuously discussed and sometimes a particular conversation can garner lots of attention and generate trending memes. So far, trends have been studied as a proxy to detect exogenous real-world events discussed in social media [5,23,87,246], emerging topics, or news of interest for the online community [60,183]. Recent work analyzes emerging topics, memes, and conversations triggered by real world events [5,23,60]. Studies of information dissemination reveal mechanisms governing content production and consumption [73] as well as prediction of future content popularity. Chenget al. study the prediction of photo-sharing cascade size [65] and recurrence [66] on Facebook.

2.2.2 Geography of Information Diffusion

It has been suggested that social media may overcome the spatio-temporal limitations of traditional communication: technologically-mediated systems make it possible to ignore physical and geographic distances [75, 217]. This, however, does not imply that commu- nication patterns on social media are not affected by physical distances and geographic borders [209,227].

Trends are also strongly localized in space and time: the temporal and geographic di- mensions play a crucial role to determine the success of a trend in terms of spreading and longevity. Unveiling the spatio-temporal dynamics that drive trending conversations on social media is instrumental for many purposes: from designing successful advertising campaigns, to understanding virality and popularity that characterize some topics. Recent studies took advantage of platforms such as Yelp and Foursquare, which provide customized services to their users based on their physical location (e.g., recommendations of events or places), to study geographic user activity patterns [221, 247–249]. Others have used plat-

forms such as Twitter and Facebook, that enrich user profiles with geographic information and accompany user generated content with location-based data, to map user demograph- ics [174,206].

Onnela et al. [227] noted that, although the probability of observing a tie between two individuals in a social network (in that case, a mobile phone call network) decreases as a power law with physical distance, the geographic spread of social groups quickly increases with the size of the group; even groups of modest dimensions (≈ 30 members) are able to span hundreds of kilometers, suggesting that, in technologically-mediated social systems, there exist distinctive social dynamics that govern the communication among individuals. Geographic locations and physical distances have been found to be correlated to friendship behaviors in online social networks [187], to determine patterns in human mobility networks [51,137], and to affect collaboration schemes in science networks [228].

Geographic factors have also been recently found to be crucial in the adoption of lan- guages and dialects [209], and in the expression of sentiment [207, 236, 237] in online social media. Mocanuet al. [209] showed how social media data can be used to characterize lan- guage geography at different levels of granularity, to highlight patterns such as linguistic homogeneity and linguistic mixture in multilingual regions.

2.2.3 Proxy to Analyze Human Behaviors

Studies by Mitchell et al. suggests that the adoption of online social media content can be instrumental to describe emotional, demographic, and geographic characteristics of users of these socio-technical systems; in particular, they investigated Twitter users active in the US in terms of happiness and individual satisfaction [125,207]. A study of happiness on Twitter led to a hedonometer project, in which the authors study temporal changes of global happiness and the relation between local low and high points with real-world events [107].

People use social media to reflect their emotions and events affecting their lives through social media. The mismatch between the social representation and real state of the user can pose challenges for research that leverages social media data because many individual worries about their online representations and conform online norms [109, 151]. However, our behaviors on social networks still carry a lot of information about personality, cultural, political and sexual preferences [133,238,239].

The use of social media also shows strong correlation with public health measures [98, 230]. Researchers have been studying several health related topics using social media data [85, 96,97,99,225,226] and search logs [229,231,301].

Similarly, services for online shopping have rich information about our preferences and tastes. We use health monitoring devices to track our work-out routines and sleep qual- ity [143, 200]. Location based services like Foursquare, AirBnB and Yelp track our eating habits and navigation history [1,167,201,257].

2.2.4 Detection of Emerging Topics

Another recent research line related to our work is that of the detection of emerging trends, topics, memes, and events in online social networks and social media [5, 23, 60, 87, 114, 183, 195,246].

Social media data can be used to make educated guesses on the outcome of real-word events, such as elections or competitions [104]. Ciulla et al. [75] combined trends and geographic information of Twitter data to demonstrate that online social media can be exploited to predict social events in the real-world. They collected trending hashtags and phrases related to contestants of the popular TV show American Idol, mapping the fan base of each candidate to different geographic regions inside and outside the US, to identify spatial patterns in attention allocation and preferences expressed on the online platform.

These signals were then combined and used to predict voting behaviors of fans achieving good accuracy.