Laura Sani

GPU-Based Parallel Search of Relevant Variable Sets in Complex Systems

Various methods have been proposed to identify emergent dynamical structures in complex systems. In this paper, we focus on the Dynamical Cluster Index (DCI), a measure based on information theory which allows one to detect relevant sets, i.e. sets of variables that behave in a coherent and coordinated way while loosely interacting with the rest of the system. The method associates a score to each subset of system variables; therefore, for a thorough analysis of the system, it requires an exhaustive enumeration of all possible subsets. For large systems, the curse of dimensionality makes the problem solvable only using metaheuristics. Even within such approaches, however, DCI computation has to be performed for a huge number of times; thus, an efficient implementation becomes a mandatory requirement. Considering that a candidate relevant set’s DCI can be computed independently of the others, we propose a GPU-based massively parallel implementation of DCI computation. We describe the algorithm’s structure and validate it by assessing the speedup in comparison with a single-thread sequential CPU implementation when analyzing a set of dynamical systems of different sizes.

Dynamics of Emotions and Relations in a Facebook Group of Patients with Hidradenitis Suppurativa

Hidradenitis suppurativa (HS) is an orphan, underdiagnosed and painful disease of the skin that has a considerable negative impact on quality of life and on emotional well-being. As reported by the italian HS patients’ association (Inversa Onlus), this condition brings patients to develop an emotional closure with the consequence that they often don’t talk about their condition with anybody. In this paper we discuss some results obtained by applying automatic emotion detection and social network analysis techniques on the Facebook group of the Inversa Onlus association. In particular, we analyze the patients’ emotional states, as expressed by the post published from 2010 to 2016, and how these emotions are influenced by friendships in the group, during the years.

Social Relevance Index for Studying Communities in a Facebook Group of Patients

Identifying Relevant Sets, i.e., variable subsets that exhibit a coordinated behavior, in complex systems is a very relevant research topic. Systems that exhibit complex dynamics are, for example, social networks, which are characterized by complex and dynamic relationships among users. A challenging topic within this context regards the identification of communities or subsets of users, both within the whole network and within specific groups. We applied the Relevance Index method, which has been shown to be effective in many situations, to the study of communities of users in the Facebook group of the Italian association of patients affected by Hidradenitis Suppurativa. Since the need for computing the Relevance Index for each possible variable subset of users makes the exhaustive computation unfeasible, we resorted to the help of an efficient niching evolutionary metaheuristic, hybridized with local searches. The communities detected through the aforementioned method have been studied to search similarities in terms of number of posts, sentiments, number of contacts, roles, behaviors, etc. The results demonstrate that it is possible to detect such subsets of users in the particular Facebook group we analyzed.

Searching Relevant Variable Subsets in Complex Systems Using K-Means PSO

The Relevance Index method has been shown to be effective in identifying Relevant Sets in complex systems, i.e., variable sub-sets that exhibit a coordinated behavior, along with a clear independence from the remaining variables. The need for computing the Relevance Index for each possible variable sub-set makes such a computation unfeasible, as the size of the system increases. Because of this, smart search methods are needed to analyze large-size systems using such an approach. Niching metaheuristics provide an effective solution to this problem, as they join search capabilities to good exploration properties, which allow them to explore different regions of the search space in parallel and converge onto several local/global minima.

A combined approach for the analysis of support groups on Facebook - the case of patients of hidradenitis suppurativa

Hidradenitis Suppurativa (HS), also known as Acne Inversa, is a chronic, underdiagnosed, often debilitating and painful disease that affects the folds of the skin. It has a considerable negative impact on the quality of life and on the emotional well-being. In this paper we discuss some results obtained by applying automatic Emotion Detection and Social Network Analysis techniques on the Facebook group of the Italian patients’ association (Inversa Onlus). In particular, we analyze the patients’ emotional states, as expressed by the posts and comments published from 2009 to 2017, and how these emotions are influenced by different social network factors, such as interactions and friendships in the group, during the observed years.

Can the Relevance Index be Used to Evolve Relevant Feature Sets

The Relevance Index (RI) is an information theory-based measure that was originally defined to detect groups of functionally similar neurons, based on their dynamic behavior. More in general, considering the dynamical analysis of a generic complex system, the larger the RI value associated with a subset of variables, the more those variables are strongly correlated with one another and independent from the other variables describing the system status. We describe some early experiments to evaluate whether such an index can be used to extract relevant feature subsets in binary pattern classification problems. In particular, we used a PSO variant to efficiently explore the RI search space, whose size equals the number of possible variable subsets (in this case \(2^{104}\)) and find the most relevant and discriminating feature subsets with respect to pattern representation. We then turned such relevant subsets into a new smaller set of richer features, whose values depend on the values of the binary features they include. The paper reports some exploratory results we obtained in a simple character recognition task, comparing the performance of RI-based feature extraction and selection with other classical feature selection/extraction approaches.

A holistic system for troll detection on Twitter

Abstract Various techniques based on artificial intelligence have been proposed for the automatic detection of online anti-social behaviors, both in existing systems and in the scientific literature. In this article, we describe TrollPacifier, a holistic system for troll detection, which analyses many different features of trolls and legitimate users on the popular Twitter platform. In this system, the most known and promising approaches and research lines are applied, along with original new ideas, in a form that fits such a large public platform. In particular, we have identified six groups of features, based respectively on the analysis of writing style, sentiment, behaviors, social interactions, linked media, and publication time. As its main scientific contributions, this work provides: (i) an up-to-date analysis of the state of the art for the problem of troll detection; (ii) the systematic collection and grouping of features, on Twitter; (iii) the description of a working holistic system for troll detection, with a very high accuracy (95.5%); and (iv) a comparison among the different features, with a machine learning approach. Our results demonstrate that automatic classification can be useful in the whole process of identification and management of online anti-social behaviors. However, a multi-faceted approach is required, in order to obtain an adequate accuracy.

A Relevance Index Method to Infer Global Properties of Biological Networks

Many complex systems, both natural and artificial, may be represented by networks of interacting nodes. Nevertheless, it is often difficult to find meaningful correspondences between the dynamics expressed by these systems and the topological description of their networks. In contrast, many of these systems may be well described in terms of coordinated behavior of their dynamically relevant parts. In this paper we use the recently proposed Relevance Index approach, based on information-theoretic measures. Starting from the observation of the dynamical states of any system, the Relevance Index is able to provide information about its organization. Moreover, we show how the application of the proposed approach leads to novel and effective interpretations in the T helper network case study.