A Graph Neural Network For Fuzzy Twitter Graphs Georgios Drakopoulos 1 , Eleanna Kafeza 2 , Phivos Mylonas 1 and Spyros Sioutas 3 1 Department of Informatics, Ionian University, Tsirigoti Sq. 7, Kerkyra 49100, Hellas 1 College of Technological Innovation, Dubai Academic City, E-L1-108, UAE 3 Computer Engineering and Informatics Department, University of Patras, Patras 26504, Hellas Abstract Social graphs abound with information which can be harnessed for numerous behavioral purposes including online political campaigns, digital marketing operations such as brand loyalty assessment and opinion mining, and determining public sentiment regarding an event. In such scenarios the efciency of the deployed methods depends critically on three factors, namely the account behavioral model, the social graph topology, and the nature of the information collected. A prime example is Twitter which is especially known for the lively activity and the intense conversations. Here an extensible computational methodology is proposed based on a graph neural network operating on an edge fuzzy graph constructed by a combination of structural, functional, and emotional Twitter attributes. These graphs constitute a strong algorithmic cornerstone for engineering cases where a properly formulated potential or uncertainty functional is linked to each edge. Starting from the ground truth in each individual vertex, the graph neural network progressively computes in an unsupervised manner a global graph state which can in turn be subject to further processing. The results, obtained using as a benchmark a recent similar graph neural network architecture along with two Twitter graphs, are promising. Keywords Fuzzy graphs, graph mining, graph neural networks, behavioral analytics, emotional polarity, Twitter 1. Introduction Graph mining is an integral part of the interconnected era since it lays the groundwork for numerous applica- tions across a wide array of fnancial and technological felds including among others social network analysis, database query optimization, graph signal processing (GSP), supply chain and logistics networks, and brain cir- cuit analysis. In this context modeling a graph in terms of vertices, connectivity patterns, and associated features is tantamount to data model selection. Edge fuzzy graphs extend classical graphs as probabilities drawn from a sin- gle distribution which may well have unknown param- eters to be estimated. Said distribution is closely linked to the semantics and functional nature of the underlying graph. For instance, in a transportation network edge ex- istence probabilities can show how likely a specifc road is to be blocked from snow in winter months, whereas in a computer network they may model the chance of a virus being propagated along a given link. In order to compute an estimation of the global graph state which allows not only a higher level overview but also subsequent processing, in this work will be used a CIKM’21: 30th ACM International Conference on Information and Knowledge Management, November 01–05, 2021, Virtual Event, QLD, Australia  c16drak@ionio.gr (G. Drakopoulos); eleana.kafeza@zu.ac.ae (E. Kafeza); fmylonas@ionian.gr (P. Mylonas); sioutas@ceid.upatras.gr (S. Sioutas)  0000-0002-0975-1877 (G. Drakopoulos); 0000-0001-9565-2375 (E. Kafeza); 0000-0002-6916-3129 (P. Mylonas) © 2021 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). CEUR Workshop Proceedings http://ceur-ws.org ISSN1613-0073 CEUR Workshop Proceedings (CEUR-WS.org) graph neural network (GNN) architecture. GNNs con- stitute a class of unsupervised neural networks where each vertex, representing a processing node, starts with a local ground truth information vector and iteratively a global status is derived based on the fundamental fact that graphs contain inherently higher order information in a distributed manner. The resulting graph global state can be subsequently further processed in order to derive global properties such as community discovery. The primary research objective of this conference pa- per is the development of a GNN architecture designed for edge fuzzy Twitter graphs constructed from incor- porating structural, functional, and behavioral features. The proposed methodology can be inherently extended to other possible attribute types, making it thus appropriate for mining graphs originating from social media or evolv- ing computational ecosystems for that matter. This work diferentiates itself from previous ones in two aspects, namely the fusion of various heterogeneous attributes and the induced edge fuzzy topology. The remaining of this work is structured as follows. In section 2 the recent scientifc literature regarding GNNs, graph mining, and computational behavioral science is briefy reviewed. The proposed methodology along with the relevant intuition are given in 3. The results obtained from the experiments are the focus of section 4. Future research directions are given in 5. Technical acronyms are defned the frst time they are encountered in the text. Finally, the notation of this conference paper is summarized in table 1. 1