High-Performance Computing and ABMS for High-Resolution COVID-19 Spreading Simulation

This paper presents an approach for the modeling and the simulation of the spreading of COVID-19 based on agent-based modeling and simulation (ABMS). Our goal is not only to support large-scale simulations but also to increase the simulation resolution. Moreover, we do not assume an underlying network of contacts, and the person-to-person contacts responsible for the spreading are modeled as a function of the geographical distance among the individuals. In particular, we defined a commuting mechanism combining radiation-based and gravity-based models and we exploited the commuting properties at different resolution levels (municipalities and provinces). Finally, we exploited the high-performance computing (HPC) facilities to simulate millions of concurrent agents, each mapping the individual’s behavior. To do such simulations, we developed a spreading simulator and validated it through the simulation of the spreading in two of the most populated Italian regions: Lombardy and Emilia-Romagna. Our main achievement consists of the effective modeling of 10 million of concurrent agents, each one mapping an individual behavior with a high-resolution in terms of social contacts, mobility and contribution to the virus spreading. Moreover, we analyzed the forecasting ability of our framework to predict the number of infections being initialized with only a few days of real data. We validated our model with the statistical data coming from the serological analysis conducted in Lombardy, and our model makes a smaller error than other state of the art models with a final root mean squared error equal to 56,009 simulating the entire first pandemic wave in spring 2020. On the other hand, for the Emilia-Romagna region, we simulated the second pandemic wave during autumn 2020, and we reached a final RMSE equal to 10,730.11.

Detection of fickle trolls in large-scale online social networks

Online social networks have attracted billions of active users over the past decade. These systems play an integral role in the everyday life of many people around the world. As such, these platforms are also attractive for misinformation, hoaxes, and fake news campaigns which usually utilize social trolls and/or social bots for propagation. Detection of so-called social trolls in these platforms is challenging due to their large scale and dynamic nature where users' data are generated and collected at the scale of multi-billion records per hour. In this paper, we focus on fickle trolls, i.e., a special type of trolling activity in which the trolls change their identity frequently to maximize their social relations. This kind of trolling activity may become irritating for the users and also may pose a serious threat to their privacy. To the best of our knowledge, this is the first work that introduces mechanisms to detect these trolls. In particular, we discuss and analyze troll detection mechanisms on different scales. We prove that the order of centralized single-machine detection algorithm is O ( n 3 ) which is slow and impractical for early troll detection in large-scale social platforms comprising of billions of users. We also prove that the streaming approach where data is gradually fed to the system is not practical in many real-world scenarios. In light of such shortcomings, we then propose a massively parallel detection approach. Rigorous evaluations confirm that our proposed method is at least six times faster compared to conventional parallel approaches.

Mobility in Unsupervised Word Embeddings for Knowledge Extraction—The Scholars’ Trajectories across Research Topics

In the knowledge discovery field of the Big Data domain the analysis of geographic positioning and mobility information plays a key role. At the same time, in the Natural Language Processing (NLP) domain pre-trained models such as BERT and word embedding algorithms such as Word2Vec enabled a rich encoding of words that allows mapping textual data into points of an arbitrary multi-dimensional space, in which the notion of proximity reflects an association among terms or topics. The main contribution of this paper is to show how analytical tools, traditionally adopted to deal with geographic data to measure the mobility of an agent in a time interval, can also be effectively applied to extract knowledge in a semantic realm, such as a semantic space of words and topics, looking for latent trajectories that can benefit the properties of neural network latent representations. As a case study, the Scopus database was queried about works of highly cited researchers in recent years. On this basis, we performed a dynamic analysis, for measuring the Radius of Gyration as an index of the mobility of researchers across scientific topics. The semantic space is built from the automatic analysis of the paper abstracts of each author. In particular, we evaluated two different methodologies to build the semantic space and we found that Word2Vec embeddings perform better than the BERT ones for this task. Finally, The scholars’ trajectories show some latent properties of this model, which also represent new scientific contributions of this work. These properties include (i) the correlation between the scientific mobility and the achievement of scientific results, measured through the H-index; (ii) differences in the behavior of researchers working in different countries and subjects; and (iii) some interesting similarities between mobility patterns in this semantic realm and those typically observed in the case of human mobility.

ActorNode2Vec: An Actor-based solution for Node Embedding over large networks

 The application of Machine Learning techniques over networks, such as prediction tasks over nodes and edges, is becoming often crucial in the analysis of Complex systems in a wide range of research fields. One of the enabling technologies in that sense is represented by Node Embedding, which enables us to learn features automatically over the network. Among the different approaches proposed in the literature, the most promising are DeepWalk and Node2Vec, where the embedding is computed by combining random walks and neural language models. However, characteristic limitations with these techniques are related to memory requirements and time complexity. In this paper, we propose a distributed and scalable solution, named ActorNode2vec, that keeps the best advantages of Node2Vec and overcomes the limitations with the adoption of the actor model to distribute the computational load. We demonstrate the efficacy of this approach with a large network by analyzing the sensitivity of walk length and number of walks parameters and make a comparison also with Deep walk and an Apache Spark distributed implementation of Node2Vec. Results show that with ActorNode2vec computational times are drastically reduced without losing embedding quality and overcoming memory issues.