Final Report Summary - ISOCIAL (iSocial: Decentralized Online Social Networks)
1. Introduction: The Online Social Network (OSN) industry is currently dominated by global behemoths, such as Facebook, Google and Twitter, that built their businesses by monetizing private data of their customers. Not surprisingly, their services come with a wide spectrum of serious concerns, including issues of ownership of private information, the protection of privacy, service interoperability, etc. Furthermore, such centralized services require heavy investment in infrastructure (e.g. building power-hungry data centers), preventing easy entrance to any new service provider, and thus effectively monopolizing the market. However, emerging technological trends are bringing a transformational change into the market, by enabling the shift from centralized services to totally decentralized systems that bring promise to address the aforementioned issues.
The main objective of iSocial is to develop such decentralized technologies for OSNs and provide world class training for a next generation of researchers and computer scientists. iSocial emphasizes on a strong combination of advanced understanding, in both theoretical and experimental approaches, of methodologies and tools that are required to develop decentralized platforms.
iSocial’s consortium is made of 7 full academic and industrial partners from different European countries, and 6 associate business partners. iSocial training network funded and trained 11 ESRs and 5 ERs, who collectively carried out 32 secondments at the academic and industrial partners of iSocial. Furthermore, the consortium organized two intensive post-graduate summer schools, 7 online courses, three thematic workshops, and a final comprehensive workshop. Social and outreach activities have also been undertaken. The consortium organized two international video challenge programs targeting high school students, and published periodic public newsletters.
The research output of iSocial contributed to the fields of P2P systems, big data analytics, decentralized machine learning, decentralized security and privacy, as well as modelling and simulation of complex social networks. The corresponding academic results within the duration of the project are 47 publications at renowned journals and conferences, 8 submissions under review, and 3 defended Ph.D. theses (7 more Ph.D. candidates are expected to finish next year).
2. Comprehensive Summary Overview of Results: Throughout the lifetime of iSocial project, research on the following tracks were pursued: Decentralized Infrastructure, Big Data Analytics and Machine Learning, Security and Privacy, and Simulation and Modelling.
*Decentralized Infrastructure: iSocial fellows have designed and built a Peer-to-Peer (P2P) architecture for Distributed Notification Systems over DOSNs. The proposed architecture provides a novel topology of the P2P network and exploits the social graph to establish connections between peers. By this, the latency required for the communication between two users in the social network is significantly reduced. iSocial fellows have also investigated real-time P2P video streaming using WebRTC, a new W3C standard for browser to browser communication. This study resulted in a new commercial project at Peerialism (an industrial partner of iSocial), and is being deployed at multiple customers. The load-balancing problems were also addressed, resulting in the development of novel algorithms to reduce the load imbalance across distributed systems. The algorithms were integrated into Apache Storm, an open source stream processing framework.
*Big Data Analytics and Machine Learning: iSocial fellows developed novel massively parallel graph-based algorithms that suitably fit DOSNs and eliminate the need of centralized aggregation points. New efficient algorithms were developed for multiple graph mining problems, such as top-k densest subgraph problem in a fully dynamic setting, and streaming graph partitioning. This enabled new approaches for efficient extraction of knowledge from the streams generated within the online social media, including topic detection, entity disambiguation and location prediction. iSocial fellows also worked on the integration of graph analytics with machine learning, to analyze autonomous data sources as well as users interactions in decentralized settings. Moreover, iSocial fellows worked towards studying the Twitter social network, and performed a comprehensive study of the 2009 and 2015 social graph snapshots. To the best of the team’s knowledge, this work is the first quantitative study on the entire Twittersphere, which compares the evolution of the network in such a large scale.
*Security and Privacy: iSocial fellows developed novel fine-grained and efficient algorithms for the management of privacy and risk in DOSNs. Alternative algorithms for more flexible and more expressive access control mechanisms in fully decentralized settings were developed. Social authentication mechanisms, that are used by one of the most popular OSNs, were also investigated. iSocial fellows demonstrated that an adversary can easily defeat that mechanism with a simple attack. As a result, a new more robust system for social authentication was designed by exploiting more advanced image analysis techniques. iSocial fellows also worked on the problem of Sybil (fake) accounts detection in DOSNs, and privacy leakage from collective resources (e.g. group photos). Fully decentralized algorithms were developed for the estimation of risk scores associated with each node in the network based on a multi-level clustering approach, effectively enhancing the accuracy of fake accounts detection compared to the state of the art.
*Modelling and Simulation: iSocial fellows discovered a quantitative two-parameter model which reproduces the entire topological evolution of a quasi-isolated OSN with unprecedented precision from the birth of the network. It allowed to precisely gauge the fundamental macroscopic and microscopic mechanisms involved. iSocial fellows introduced an ecological theory of the digital world, which exhibits a stable coexistence of several networks as well as the domination of a single one, in contrast to the principle of competitive exclusion. Investigation of how multiple coexisting networks, which form a so called multiplex system, was also carried out. This allowed to perform accurate trans-layer link prediction, where connections in one layer can be predicted by observing the hidden geometric space of another layer.
3. ITN training activities: iSocial consortium organized a total of 4 International Workshops, 2 Summer Schools, and Annual Research Meetings, where iSocial fellows had the opportunity to present their work through different media (talks, posters, discussion panels). iSocial consortium has also offered 7 online courses covering knowledge topics in distributed systems, complex networks, security, and data privacy preserving computing. Some events (both Summer Schools) were jointly organised together with other EU initiatives, in particular with Erasmus Mundus Joint Doctorate in Distributed Computing Programme, as well as with EIT Digital to bring in entrepreneurial aspects to the fellow training.
4. Dissemination: Research within iSocial resulted in 37 conference and 10 journal publications, including a groundbreaking publication in Nature Physics (Impact Factor: 20), and the best student paper award in IEEE Big Data Conference, 2016. All the research was summarized in a periodic iSocial newsletter disseminated to all the stakeholders as well as available online.
5. Socio-economic impacts and conclusions: iSocial project trained 16 fellows who will be joining the ranks of expert data scientists with expertise in security & privacy, decentralized technologies, and modelling of complex systems. It also significantly advanced the field of Decentralized Online Social Networks, by making available completely new technologies that will pave the way for more decentralized, sharing-based digital economy of the future. The groundbreaking research on Social Networks modelling was published in Physics Nature. Some of the technologies developed within iSocial are adopted by European SMEs, such as Hive Streaming, and Gavagai.
iSocial webpage: http://isocial-itn.eu/
The main objective of iSocial is to develop such decentralized technologies for OSNs and provide world class training for a next generation of researchers and computer scientists. iSocial emphasizes on a strong combination of advanced understanding, in both theoretical and experimental approaches, of methodologies and tools that are required to develop decentralized platforms.
iSocial’s consortium is made of 7 full academic and industrial partners from different European countries, and 6 associate business partners. iSocial training network funded and trained 11 ESRs and 5 ERs, who collectively carried out 32 secondments at the academic and industrial partners of iSocial. Furthermore, the consortium organized two intensive post-graduate summer schools, 7 online courses, three thematic workshops, and a final comprehensive workshop. Social and outreach activities have also been undertaken. The consortium organized two international video challenge programs targeting high school students, and published periodic public newsletters.
The research output of iSocial contributed to the fields of P2P systems, big data analytics, decentralized machine learning, decentralized security and privacy, as well as modelling and simulation of complex social networks. The corresponding academic results within the duration of the project are 47 publications at renowned journals and conferences, 8 submissions under review, and 3 defended Ph.D. theses (7 more Ph.D. candidates are expected to finish next year).
2. Comprehensive Summary Overview of Results: Throughout the lifetime of iSocial project, research on the following tracks were pursued: Decentralized Infrastructure, Big Data Analytics and Machine Learning, Security and Privacy, and Simulation and Modelling.
*Decentralized Infrastructure: iSocial fellows have designed and built a Peer-to-Peer (P2P) architecture for Distributed Notification Systems over DOSNs. The proposed architecture provides a novel topology of the P2P network and exploits the social graph to establish connections between peers. By this, the latency required for the communication between two users in the social network is significantly reduced. iSocial fellows have also investigated real-time P2P video streaming using WebRTC, a new W3C standard for browser to browser communication. This study resulted in a new commercial project at Peerialism (an industrial partner of iSocial), and is being deployed at multiple customers. The load-balancing problems were also addressed, resulting in the development of novel algorithms to reduce the load imbalance across distributed systems. The algorithms were integrated into Apache Storm, an open source stream processing framework.
*Big Data Analytics and Machine Learning: iSocial fellows developed novel massively parallel graph-based algorithms that suitably fit DOSNs and eliminate the need of centralized aggregation points. New efficient algorithms were developed for multiple graph mining problems, such as top-k densest subgraph problem in a fully dynamic setting, and streaming graph partitioning. This enabled new approaches for efficient extraction of knowledge from the streams generated within the online social media, including topic detection, entity disambiguation and location prediction. iSocial fellows also worked on the integration of graph analytics with machine learning, to analyze autonomous data sources as well as users interactions in decentralized settings. Moreover, iSocial fellows worked towards studying the Twitter social network, and performed a comprehensive study of the 2009 and 2015 social graph snapshots. To the best of the team’s knowledge, this work is the first quantitative study on the entire Twittersphere, which compares the evolution of the network in such a large scale.
*Security and Privacy: iSocial fellows developed novel fine-grained and efficient algorithms for the management of privacy and risk in DOSNs. Alternative algorithms for more flexible and more expressive access control mechanisms in fully decentralized settings were developed. Social authentication mechanisms, that are used by one of the most popular OSNs, were also investigated. iSocial fellows demonstrated that an adversary can easily defeat that mechanism with a simple attack. As a result, a new more robust system for social authentication was designed by exploiting more advanced image analysis techniques. iSocial fellows also worked on the problem of Sybil (fake) accounts detection in DOSNs, and privacy leakage from collective resources (e.g. group photos). Fully decentralized algorithms were developed for the estimation of risk scores associated with each node in the network based on a multi-level clustering approach, effectively enhancing the accuracy of fake accounts detection compared to the state of the art.
*Modelling and Simulation: iSocial fellows discovered a quantitative two-parameter model which reproduces the entire topological evolution of a quasi-isolated OSN with unprecedented precision from the birth of the network. It allowed to precisely gauge the fundamental macroscopic and microscopic mechanisms involved. iSocial fellows introduced an ecological theory of the digital world, which exhibits a stable coexistence of several networks as well as the domination of a single one, in contrast to the principle of competitive exclusion. Investigation of how multiple coexisting networks, which form a so called multiplex system, was also carried out. This allowed to perform accurate trans-layer link prediction, where connections in one layer can be predicted by observing the hidden geometric space of another layer.
3. ITN training activities: iSocial consortium organized a total of 4 International Workshops, 2 Summer Schools, and Annual Research Meetings, where iSocial fellows had the opportunity to present their work through different media (talks, posters, discussion panels). iSocial consortium has also offered 7 online courses covering knowledge topics in distributed systems, complex networks, security, and data privacy preserving computing. Some events (both Summer Schools) were jointly organised together with other EU initiatives, in particular with Erasmus Mundus Joint Doctorate in Distributed Computing Programme, as well as with EIT Digital to bring in entrepreneurial aspects to the fellow training.
4. Dissemination: Research within iSocial resulted in 37 conference and 10 journal publications, including a groundbreaking publication in Nature Physics (Impact Factor: 20), and the best student paper award in IEEE Big Data Conference, 2016. All the research was summarized in a periodic iSocial newsletter disseminated to all the stakeholders as well as available online.
5. Socio-economic impacts and conclusions: iSocial project trained 16 fellows who will be joining the ranks of expert data scientists with expertise in security & privacy, decentralized technologies, and modelling of complex systems. It also significantly advanced the field of Decentralized Online Social Networks, by making available completely new technologies that will pave the way for more decentralized, sharing-based digital economy of the future. The groundbreaking research on Social Networks modelling was published in Physics Nature. Some of the technologies developed within iSocial are adopted by European SMEs, such as Hive Streaming, and Gavagai.
iSocial webpage: http://isocial-itn.eu/