=Paper=
{{Paper
|id=Vol-1914/SIDEWAYS17-9
|storemode=property
|title=Event-driven TV Programs Web Community Exploration
|pdfUrl=https://ceur-ws.org/Vol-1914/SIDEWAYS17-9.pdf
|volume=Vol-1914
|authors=Ruggero G. Pensa,Claudio Schifanella,Luca Vignaroli
|dblpUrl=https://dblp.org/rec/conf/ht/PensaSV17
}}
==Event-driven TV Programs Web Community Exploration==
https://ceur-ws.org/Vol-1914/SIDEWAYS17-9.pdf
Event-driven TV Programs Web Community Exploration
Ruggero G. Pensa Claudio Schifanella Luca Vignaroli
Department of Computer Science, Department of Computer Science, RAI CRIT - Turin
University of Turin University of Turin Via Cavalli 6
Corso Svizzera 185 Corso Svizzera 185 Turin, Italy 10149
Turin, Italy 10149 Turin, Italy 10149 luca.vignaroli@rai.it
pensa@di.unito.it schi@di.unito.it
ABSTRACT the University of Turin led in the last years to the design and devel-
With the goal of understanding how major public interest events opment of a general framework1 for the collection and analysis of
are perceived by the TV public and how the users’ interests evolve heterogeneous data coming from standard TV sources (EPGs, edito-
in time, we introduce a data collection, integration and analysis rial metadata, audience metrics) and social networks conversations
framework that allows to compute, characterize and explore dy- [1]. Within our general purpose social data collection, integration
namic social web communities. We focus on communities of Twitter and analysis framework [9], in this paper, with the goal of under-
users that interact each-other around specific TV events and track standing how major public interest events are perceived by the TV
their public cybersocial activities to analyze and visualize the in- public and how the users’ interests evolve in time, we present a
formation diffusion processes and understand how they affect the new application that allows to compute, characterize and explore
community structure of the social network. dynamics social web communities.
Our application enables both guided and automatic extraction
KEYWORDS of emergent topics and concepts from Twitter public conversations.
Based on such concepts, it builds the social network of users by
social network analysis, social TV, community detection
leveraging their interactions. Two users are considered as interact-
Reference format: ing each other if they mention each-other explicitly (in a retweet or
Ruggero G. Pensa, Claudio Schifanella, and Luca Vignaroli. 2017. Event- a reply) or whenever they refer to similar concepts in their conver-
driven TV Programs Web Community Exploration. In Proceedings of Inter- sations. Moreover, our application enables the execution of multiple
national Workshop on Social Media World Sensors (SIDEWAYS), Prague Czech community detection algorithms [2–4] to uncover the underlying
Republic, July 2017 (Sideways2017), 4 pages, CEUR-WS.org. community structure of the network at different time points. Those
communities are then described in terms of social influence of their
nodes using several metrics (pagerank, betweenness and eigen-
1 INTRODUCTION vector centrality) [7]) as well as in terms of topics and concepts
People on the Web talk about television. Trending topics and con- characterizing their interactions (by using summarization). [8]).
cepts arise from TV users’ “cybersocial” activities thus influencing Finally, the application enables the observation, analysis and char-
the community structure of the social network. Users aggregate acterization of the diffusion of topics and concepts in the overall
themselves around topics and concepts that emerge from major network and among the different communities [10]. It could then
events. The U.S. presidential primaries have a clear effect on the support the activity of researchers, analysts and practitioners inter-
community structure of users around the world, but such groups ested in social media analysis and network dynamics. Our use case
are subject to changes as the candidates pronounce some speech in relies on real public data gathered from Twitter, related to one of
favor or against some policies (e.g., on immigration, protectionism, the most important TV event broadcasted by RAI.
welfare) that concern them, directly or indirectly. Capturing and
understanding the evolution of such communities promptly is of 2 RELATED WORK
great value for a number of stockholders: advertisers, broadcast In the last three years, a number of tools have been proposed to
programmers, market and society analysts. address the problem of complexity and dynamicity in network anal-
Within the RAI research project on the study of the integration ysis and visualization (eg., [6, 11, 13]). GalaxyExplorer [6] is an
between social networks and the TV world, the collaboration with influence-driven visual analysis system for exploring how users
influence each other in a social network. The authors use a galaxy-
based visual metaphor to simplify the visual complexity of large
graphs. In [12], the authors propose a spatial visualization system
to detect geo-social event from Twitter conversations. Gazouille [5]
is another location-based system for discovering local events in geo-
localized social media streams. Insight4News [11], instead, connects
news articles to social conversations, by using topic detection and
tweet summarization, and performs hashtag recommendation. Fi-
Sideways2017, July 2017, Prague Czech Republic
Copyright held by the author(s). nally, in [13] the authors adopts semi-supervised machine learning
1 http://hdm.di.unito.it/mesoontv.html
Sideways2017, July 2017, Prague Czech Republic R. Pensa et al.
Figure 1: Architecture of the MeSoOnTV framework.
to perform community detection via a label propagation approach to a label propagation algorithm. A ego-minus-ego network is de-
that leverages an epidemic spreading model. Differently from them, fined as the ego-network of a node v without the node v itself [3].
the focus of our work is the community. Our framework, in fact, The analysis module implementing Louvain and the metrics that
aims at detecting and characterizing the evolution of the commu- measure user influence is based on Gephi4 , while, for DEMON, we
nities gravitating around TV events. Furthermore, we analyze and use our own adaptation of the authors’ Java implementation.
visualize the information diffusion dynamics at both the single node The visualization layer is constituted of the web application, that
level and the community level. adopt modern web technologies to enable responsiveness and high-
level interaction patterns: node.js as web page server, a REST API
3 MESOONTV SYSTEM OVERVIEW for database querying, D3.js as graph visualization and browsing
Our framework consists of three internal layers, covering all the library. To cope with the potential huge amount of data, we adopt
phases from data collection, representation and integration to data a caching mechanism on the layout, nodes, links and communities.
analysis, and a visualization layer (see Figure 1). More specifically,
a Source processing layer contains the different modules for col- 4 DEMONSTRATION SCENARIO
lecting all the data from web sources to be conveyed in the social In our use case scenario, we gathered Twitter data from February
graph and other data representation technologies. It accesses a 9th to 13th 2016. The data are related to the five episodes of the 2016
number of predefined web/social/media sources (Twitter, official edition of the Sanremo Music Festival, the most popular Italian song
web sites, TV channels, ontological information sources) and con- contest and awards. Overall, more than 2.52M tweets of 176, 760
tinuously processes the information collected in real-time to detect users have been processed. In these conversations, the relevant
the named entities (people, places and events) trough the use of a concepts are mentioned more than 1.80M times, while the overall
Named-Entity Recognition module and a topic extraction module. number of hashtags is 3.60M. Our goal is to recognize and char-
The collected users, topics, concepts and relationships among them acterize those communities that gravitates around some specific
are then stored in the social graph layer based on Neo4j2 and events that happened during this major national event. In partic-
MySQL3 , which contains all the modules needed to store and man- ular, during the 2016 edition, apart from the song contest itself,
age the social graph. The social query and analysis layer offers several events attracted the public interest and were extensively
functionalities for querying, browsing and analyzing the graph. discussed in newspapers, news broadcasting, blogs and social me-
More specifically the analysis module provides a set of community dia. For instance, in all episodes, some artists and guests promoted
detection and social network analysis components. the upcoming law on civil rights for the LGBT community. During
Community detection if performed by adopting two well-known the second episode, there was a touching exhibition by Ezio Bosso,
algorithms: Louvain [2] and DEMON [3]. Louvain is a greedy op- an Italian composer known worldwide affected by an autoimmune
timization method that attempts to optimize the “modularity” of neurological disorder. In the final episode, Elio e le Storie Tese, a
a partition of the network. Modularity measures the density of well-known comedy rock band, performed their song dressed as
links inside communities compared to links between communities Kiss, the famous metal band; this exhibition was appreciated by Kiss
[2]. DEMON, instead, is an algorithm that detect hierarchical and frontman Gene Simmons, thus gaining visibility at international
overlapping communities in networks. It enables the discovery of level
global communities from multiple ego-minus-ego networks thanks A user has the possibility of selecting a set of concepts of interest
(e.g., the artists exhibiting during the event, “LGBT”, “Ezio Bosso”,
2 http://www.neo4j.com
3 http://www.mysql.com 4 https://gephi.org
Event-driven TV Programs Web Community Exploration Sideways2017, July 2017, Prague Czech Republic
Figure 2: Visualization of community details.
Figure 3: Visualization of concept diffusion.
“Kiss”) or a set of “trending/emerging concepts” according to some interest community dealing with Ezio Bosso and the civil rights of
basic statistics. Then, a community detection algorithm is executed LGBT people (#SanremoArcobaleno).
on the Twitter interaction network of the users that mentioned The application supports also the analysis of the traces of the
at least one of the selected concepts in their conversations (also information diffusion processes involving individual users and com-
including replies and retweets). The network and their communities munities (see Figure 3). By retracing the timeline of each episode,
are visualized as shown in Figure 2. Each community can be then the application allows the user to visualize and compare the spread
inspected individually and the application returns some related of a number of selected concepts in the interaction network. Dur-
statistics: the list of most influential users (according to pagerank, ing each time interval, Twitter users that tweet, retweet or reply
betweenness centrality or other network metrics) the list of the in conversations mentioning a specific concept c are “turned on”
most relevant concepts and hashtags (identified by a summarization and highlighted with the color identifying c. In this visualization
technique) including both selected concepts and other co-occurrent modality, the information diffusion process can be compared with
concepts. In Fig. 2, for instance, Community 2 is characterized by the overall Twitter activity represented by the curve on bottom of
the fact that the involved users talk about Dear Jack (an Italian rock Figure 3. This facility supports the discovery of diffusion patterns,
band), while the hashtag #MezzoRespiro refers to the title of the as well as the visual comparison of diffusion trends. Notice that ba-
song presented during the contest. Community 5, instead, is a broad sic complex network statistics (number of nodes/edges, number of
connected components, clustering coefficient) are always available
Sideways2017, July 2017, Prague Czech Republic R. Pensa et al.
on screen. Moreover, the user can observe the time evolution of [6] Xiaotong Liu, Srinivasan Parthasarathy, Han-Wei Shen, and Yifan Hu. 2015.
such metrics referring only to the highlighted subgraph. Though GalaxyExplorer: Influence-Driven Visual Exploration of Context-Specific Social
Media Interactions. In Proceedings of WWW 2015 - Companion Volume. 215–218.
usable on mobile devices, the application is optimized for 4K UHD [7] Mark E. J. Newman. 2010. Networks: An Introduction. Oxford University Press,
displays. New York, NY, USA.
[8] Brendan O’Connor, Michel Krieger, and David Ahn. 2010. TweetMotif: Ex-
Through the described platform, it is possible to identify and ploratory Search and Topic Summarization for Twitter. In Proceedings of ICWSM
visualize phenomena that occur on social networks with reference 2010.
to the television world such as emerging phenomena, their temporal [9] R. G. Pensa, M. L. Sapino, C. Schifanella, and L. Vignaroli. 2016. Leveraging
Cross-Domain Social Media Analytics to Understand TV Topics Popularity. IEEE
evolution and the relationships between the entities involved. The Computational Intelligence Magazine 11, 3 (Aug 2016), 10–21. https://doi.org/10.
conceptual navigation network allows the user to walk through 1109/MCI.2016.2572518
the concepts and involved social users mentioned within television [10] Daniel M. Romero, Brendan Meeder, and Jon M. Kleinberg. 2011. Differences in
the mechanics of information diffusion across topics: idioms, political hashtags,
events in a particular timeline based on mentions of mentions and complex contagion on twitter. In Proceedings of WWW 2011. 695–704.
during television events. [11] Bichen Shi, Georgiana Ifrim, and Neil Hurley. 2014. Insight4News: Connecting
News to Relevant Social Conversations. In Proceedings of ECML PKDD 2014, Part
The flow of data coming from social networks is now seen as III. 473–476.
a true additional information channel for the end user, for some [12] Shoko Wakamiya, Adam Jatowt, Yukiko Kawai, and Toyokazu Akiyama. 2016.
television programs it is a fruition line for some verse independent Analyzing Global and Pairwise Collective Spatial Attention for Geo-social Event
Detection in Microblogs. In Proceedings of WWW 2016, Companion Volume. 263–
of the television program it is generated. The platform described, 266.
in addition to being used as a support to marketing choices, can [13] Ying Wen, Yuanhao Chen, and Xiaolong Deng. 2014. Epdemic spreading model
be used as a data source to tell television events from the point of based overlapping community detection. In Proceedings of ASONAM 2014. 954–
959.
view of users who have lived and commented on them, creating
an unseen story that sees as protagonists the viewers themselves
with their perception and emotions tried during the viewing of a
television program. In fact, Digital Storytelling, or Narrative made
with digital instruments, consists of organizing selected content
from the web in a coherent system based on a narrative structure,
in order to obtain a tale composed of multiple elements of various
formats (video, audio, images, texts, maps, etc.). Digital Storytelling
is therefore a form of narrative particularly suitable for communica-
tive forms such as journalism, politics, marketing, autobiography,
didactics and it could became a new frontiers of the television world
to provide information and entertainment experiences that can be
visualized through different tools of fruition in synergy with each
other.
5 CONCLUSIONS
In this paper we presented an application that allows to find, char-
acterize and explore community of users in social networks con-
versations. We demonstrate the effectiveness of the approach by
exploiting a use case in the TV setting related to a well known
Italian song festival. The collaboration between University of Turin
and RAI, the Italian broadcaster, is constantly evolving: in the next
months, we will integrate other sources that we are already moni-
toring (e.g., Facebook), as well as new features for a more in-depth
characterization and comparison of users and communities.
REFERENCES
[1] Alessio Antonini, Luca Vignaroli, Claudio Schifanella, Ruggero G. Pensa, and
Maria Luisa Sapino. 2013. MeSoOnTV: a media and social-driven ontology-based
TV knowledge management system. In Proc. of ACM HT ’13. 208–213.
[2] Vincent D Blondel, Jean-Loup Guillaume, Renaud Lambiotte, and Etienne Lefeb-
vre. 2008. Fast unfolding of communities in large networks. J. Stat. Mech. Theor.
Exp. 2008, 10 (2008), P10008.
[3] Michele Coscia, Giulio Rossetti, Fosca Giannotti, and Dino Pedreschi. 2014. Un-
covering Hierarchical and Overlapping Communities with a Local-First Ap-
proach. TKDD 9, 1 (2014), 6:1–6:27.
[4] Santo Fortunato. 2010. Community detection in graphs. Physics Reports 486,
3âĂŞ5 (2010), 75 – 174. https://doi.org/10.1016/j.physrep.2009.11.002
[5] Pierre Houdyer, Albrecht Zimmermann, Mehdi Kaytoue, Marc Plantevit, Joseph
Mitchell, and Céline Robardet. 2015. Gazouille: Detecting and Illustrating Local
Events from Geolocalized Social Media Streams. In Proceedings of ECML PKDD
2015, Part III. 276–280.