|D4.1 – Heterogeneous Social Network Graph topology and lifecycle||Download||This deliverable presents the description of the Heterogeneous Social Graph topology and the lifecycle of the topology. The document describes what the Heterogeneous Social Network Graph is and the local knowledge of the network each node has. The document introduces the local structure of each user, namely Contextual Ego Network, which is a multi-layer network where each layer represents a specific context of the use, and - is implemented by exploiting the ego network social model. The document provides an overview of how a user is identified in the HELIOS social network by introducing the main characteristics of the user profile, and how trust affects the lifecycle of the Contextual Ego Network structure of each user.|
|D4.2 – Time-dependent Social Graph||Download||This deliverable presents the modelling of the Heterogeneous Social Graph using a temporal network and describes what the Heterogeneous Social Network Graph is and the importance to model it is using a graph formalism that includes the temporal complexity. Furthermore, introduces the Contextual Ego Network as the local view of each user of the whole Social Overlay. Thanks to the Contextual Ego Network, each user has to maintain information that is local instead of the whole graph, and the Social Overlay is maintained by the means of the union of each local view. The Contextual Ego Network is a multi-layer network where each layer represents a specific context of the user, and each layer is modelled by exploiting the ego network social model. The document moreover provides an overview of some possible studies that can be carried out on each layer, with a special focus on the usage of Graph Neural Networks.|
|D4.3 – Mining the social graph||Download|
In this deliverable, we explore the state-of-the-art and new approaches for mining heterogeneous social graphs in decentralized time-evolving systems. Such approaches can discover patterns of relations and interactions driven by the underlying preferences of HELIOS users, which in turn can be used to improve the quality of HELIOS applications. In particular, we investigate existing graph mining practices, such as graph clustering and graph neural networks, which can mine the time-evolving organization and preferences of social media users and propose novel protocols and adaptations that help adopt them in the decentralized setting of HELIOS.
To this end, we propose a novel community detection protocol for discovering the community structure of decentralized networks, as well as a novel decentralized adaptation of temporal graph neural networks of mining user preferences. The latter can recommend future interactions with similarly high efficacy to their centralized counterpart in social network datasets with similar properties as those of the envisioned HELIOS networks and are deployed in a graph recommendation module that facilitates social recommendation and analysis tasks. To support the integration of that module with the rest of the HELIOS platform, we also improve the Contextual Ego Network management library.
|D4.4 – Define rewarding methodologies ||Download||The purpose of this deliverable is on the one side to analyze which is the best blockchain platform where to build the Helios Rewarding system and on the other side, to define the actions to be rewardable as well as the algorithm to calculate the number of tokens to be distributed for each of the users who contributes on top of HELIOS. In order to do so, an analysis of different rewarding systems used by some of the most popular decentralized social networks has been conducted. Furthermore, different types of the blockchain (permissionless vs permissioned) are investigated. Last but not least, the HELIOS Reward System is defined taking into account use case applicability.|
|D4.5 – Computational Trust||Download|
In this deliverable, we first show how the problem of trust is framed by state-of-the-art research, by giving a brief survey of the studies carried out in this field; based on this survey, we then outline which are the main properties of trust in social environments. Thereafter, we introduce the HELIOS trust model we theorized, by defining and describing the features that are exploited, and we provide the actual formula that we use in the module for trust computation.
Finally, we give an overview of the APIs that can be employed by the other HELIOS modules or the client application to instantiate the trust module, update it in case of new events happening in the Conceptual Ego Network (CEN), or retrieve the last trust value that has been computed for an alter in a specific context.
|D4.6 – Neuro-based services||Download|
In this deliverable, we describe the Neuro Behavioural module of HELIOS. It aims to characterise the relationship between two HELIOS users, based on the analysis of their communications. A set of models have been developed that analyse text messages and images shared between users on mobile phones.
Once the trust manager needs to compute the trust to modulate the contextual ego network, it calls the neuro-behaviour module that provides a set of cognitive and emotional metrics using the previously analysed communications: attention, which characterizes the number of communications; arousal, the intensity of the emotions in terms of activation; and valence, which represents whether an emotion is perceived as positive or negative.
The module has been validated through an experimental study involving 30 participants. They were asked to do several tasks using HELIOS, such as inviting their partner to breakfast the next day. Neurophysiological responses
of the ego were recorded, including electrodermal activity, electrocardiogram and electroencephalogram. Statistical analysis was performed considering the user's emotional self-assessment, the HELIOS module and the neurophysiological responses.
The results support the algorithms and models developed to track behaviour among users in social media networks.
|D4.7 – Organic social graph creation||Download|
This report will define the properties required to establish organic social graphs that evolve based on the users' environment. It serves as a basis for implementation and clearly states what
elements are required to establish contextual and ad-hoc (organic) social networks.
The main concept of Organic Social Graph creation is the concept of context. This concept is discussed throughout the document and specifically analysed in the state of the art section. In relation to
this, an Organic Social Graph is created based on proximity within a certain context which strongly relates to the place, time and purpose of the social interaction.
Specifically, this deliverable describes researching tasks and development towards establishing organic social networks in a non-intrusive and privacy-sensitive manner that require little to non-user intervention.
The deliverable has two interconnected parts. The first is the creation of an Organic Social Graph in a smart environment. For this, a Raspberry Pi is utilised. The second part serves as an extension and focused on creating Organic Social Graphs in different contexts via a mobile phone. The overall result of this deliverable is a research-based discussion and subsequent implementation of two Proof of Concepts illustrating how to create an Organic Social Graph in different settings and contexts. It serves as the foundation for further research towards several key innovations such as context detection, automatic social graph management and automatic social organic graph creation.
|D4.8 – Development of content-aware social graphs||Download|
In this deliverable, we explore the development of content-based features to enrich the HELIOS Heterogeneous Social Graph (HSG), and in that way provide support for content-aware user profiling and matching services. To this end, we aggregate the image collections of the HELIOS users and extract semantic information in three complementary ways; the first focuses on analyzing the user interests, the second construct a user representation in a self-supervised way and the third relies on processing pieces of text extracted from the images.