Abstract The use of social media has grown exponentially in recent years up to become a reflection of human social attitudes and to represent today the main channel for conducting discussions and sharing opinions. For this reason, the vast amount of information generated is often used for predicting outcomes of real-world events in different fields, including business, politics, and health, as well as in the entertainment industry. In this paper, we focus on how data from Twitter can be used to predict ratings of a large set of TV shows regardless of their specific genre. Given a show, the idea is to exploit features concerning the pre-release hype on Twitter for rating predictions. We propose a novel machine learning-based approach to the genre-independent TV show popularity prediction problem. We compared the performance of several well-known predictive methods, and as a result, we discovered that LSTM and Random Forest can predict the ratings in the USA entertainment market, with a low mean squared error of 0.058. Furthermore, we tested our model by using data of “never seen” shows, by deriving interesting results in terms of error rates. Finally, we compared performance against relevant solutions available in the literature, with discussions about challenges arousing from the analysis of shows in different languages.

Abstract Optimal portfolio selection—composing a set of stocks/assets that provide high yields/returns with a reasonable risk—has attracted investors and researchers for a long time. As a consequence, a variety of methods and techniques have been developed, spanning from purely mathematics ones to computational intelligence ones. In this paper, we introduce a method for optimal portfolio selection based on multi-objective evolutionary algorithms, specificallyNondominated Sorting Genetic Algorithm-II (NSGA-II), which tries tomaximizethe yield andminimizethe risk, simultaneously. The system, namedEvoFolio, has been experimented on stock datasets in a three-years time-frame and varying the configurations/specifics of NSGA-II operators.EvoFoliois aninteractivegenetic algorithm, i.e., users can provide their own insights and suggestions to the algorithm such that it takes into account users’ preferences for some stocks. We have performed tests with optimizations occurring quarterly and monthly. The results show howEvoFoliocan significantly reduce the risk of portfolios consisting only of stocks and obtain very high performance (in terms of return). Furthermore, considering the investor’s preferences has proved to be very effective in the portfolio’s composition and made it more attractive for end-users. We argue thatEvoFoliocan be effectively used by investors as a support tool for portfolio formation.

Visual Knowledge Discovery (VKD), the emerging research field exploring the integration between Artificial Intelligence and Visual Analytics is impacting a growing number of contexts. This paper dwells on the potential intersections between the VKD perspective and the legal world, a scenario for many reasons drawn by the idea of combining empirical insights offered by computation with intuitive, visually-enhanced forms of data mining. We will focus on criminal justice, taking the cue from an ongoing experimental research project that investigates novel applications of computational social science methods - complex network analysis - to analyze criminal organizations for both scientific and investigative purposes. The work will describe the VKD solutions devised to allow public prosecutors to visually browse the content of phone calls and environmental tapping gathered during preliminary inquiries. We will discuss how given visualizations - graph-based visualizations in the first place - can enhance text mining techniques helping to capture in-a-glance communication flows, discussions' topics and both structural and functional features of criminal networks at hand.

Mobile devices are nowadays ubiquitous. They are equipped with a variety of sensors, each designed for capturing specific signals which can be exploited to acquire discriminating user’s traits, thus allowing to recognize, for instance, the authorized users. In this regard, we focus on capturing soft biometric traits from smartphones. Soft biometric information extracted from a human body (e.g., gender and age) is ancillary information proved to improve the performance of biometric authentication systems, and it has drawn a great deal of attention for its applications in healthcare, smart spaces and digital world as a whole. This paper presents an approach to gender and age-group recognition, namely TGSB, leveraging a transfer learning strategy applied to state-of-the-art Convolutional Neural Networks (CNNs) fed with image-based representations of touch gestures performed by users on mobile devices. We perform experiments considering, one at a time, touch gestures of the same kind, and combinations thereof, with intermediate and late fusion learning strategies. Experiments prove that TGSB is a promising approach, with up to 94% accuracy for the gender recognition and up to 99% for the age-group recognition. We highlight the most useful touch gesture for gender and age-group recognition, that is Scroll with 81% and 96% accuracy, respectively. We show that combining multiple touch gestures (intermediate fusion) with a joint latent subspace learning mechanism in CNNs improves the TGSB performance, up to 99% accuracy when considering a combination of two Scroll. Compared to previous works, TGSB exhibits much better performance in both gender and age-group recognition.

Abstract Automatic short answer grading (ASAG), a hot field of natural language understanding, is a research area within learning analytics. ASAG solutions are conceived to offload teachers and instructors, especially those in higher education, where classes with hundreds of students are the norm and the task of grading (short)answers to open-ended questionnaires becomes tougher. Their outcomes are precious both for the very grading and for providing students with “ ad hoc ” feedback. ASAG proposals have also enabled different intelligent tutoring systems. Over the years, a variety of ASAG solutions have been proposed, still there are a series of gaps in the literature that we fill in this paper. The present work proposes GradeAid , a framework for ASAG. It is based on the joint analysis of lexical and semantic features of the students’ answers through state-of-the-art regressors; differently from any other previous work, (i) it copes with non-English datasets, (ii) it has undergone a robust validation and benchmarking phase, and (iii) it has been tested on every dataset publicly available and on a new dataset (now available for researchers). GradeAid obtains performance comparable to the systems presented in the literature (root-mean-squared errors down to 0.25 based on the specific tuple $$\langle $$ ⟨ dataset-question $$\rangle $$ ⟩ ). We argue it represents a strong baseline for further developments in the field.

The digital evolution of economies and markets brings changes that go largely beyond growth and efficiency. In the gig economy, also fueled by algorithmic management solutions, digital labour platforms (DPLs) raise significant issues that include power asymmetries, new forms of workers' abuse and discrimination, algorithms' opacity and over-control. In such a scenario, while normative frameworks evolve novel safeguards, a crucial challenge is that of feeding public (social, institutional) oversight on the dynamics that, at various levels, affect the gig work world. In this paper, we show how the combination of visual analytics and data-driven heuristics can be used to offer new insights on and promote higher levels of transparency and awareness about the fairness of DPLs' activity seen, in the first place, from the workers' perspective. Solutions will be presented as part of GigAdvisor, an experimental cross-platform application developed within an ongoing research that draws on computational social science methods to enable new critical approaches to the digital economy.

Plagiarism is a debated and controversial topic in different fields. For example, in Law, where the subjectivity of the judges that have to pronounce a suspicious case usually lead to long and often unsolved cases, and in Music, where huge amounts of money are invested every year to face and try to solve suspicious cases. In this scenario, the automatic detection of music plagiarism is fundamental by representing useful support for judges during their pronouncements and an important result to avoid musicians spending more time in court than on composing music. This paper shows how the combination of visual analytics and the employment of adaptive meta-heuristics can assist domain experts in judging suspicious cases. Solutions will be presented as part of PlagiarismDetection, a cross-platform tool that leverages text-similarity algorithms, computational intelligence, optimization methods, and visualization techniques to enable new critical approaches to music plagiarism analysis.

Abstract Financial bubbles represent a severe problem for investors. In particular, the cryptocurrency market has witnessed the bursting of different bubbles in the last decade, which in turn have had spillovers on all the markets and real economies of countries. These kinds of markets and their unique characteristics are of great interest to researchers. Generally, investors and financial operators study market trends to understand when bubbles might occur using technical analysis tools. Such tools, which have been historically used, resulted in being precious allies at the basis of more advanced systems. In this regard, different autonomous, adaptive and automated trading agents have been introduced in the literature to study several kinds of markets. Among these, we can distinguish between agents withZero/Minimal Intelligence (ZI/MI)andComputational Intelligence (CI)-based agents. The first ones typically trade on the market without resorting to complex learning strategies; the second ones usually use (deep) reinforcement learning mechanisms. However, these trading agents have never been tested on the cryptocurrencies market and related financial bubbles, which are still mostly overlooked in the literature. It is unclear how these agents can make profits/losses before, during, and after a bubble to adjust their strategy and avoid critical situations. This paper compares a broad set of trading agents (betweenZI/MIandCIones) and evaluates them with well-known financial indicators (e.g., volatility, returnsSharpe ratio, drawdown,SortinoandOmega ratio). Among the experiment’s outcomes,ZI/MIagents were more explainable thanCIones. Based on the results obtained above, we introduceGGSMZ, a trading agent relying on a neuro-fuzzy mechanism. The neuro-fuzzy system is able to learn from the trades performed by the agents adopted in the previous stage.GGSMZ’s performances overcome those of other tested agents. We argue thatGGSMZcould be used by investors as a decision support tool.

Abstract Gender classification of mobile devices’ users has drawn a great deal of attention for its applications in healthcare, smart spaces, biometric-based access control systems and customization of user interface (UI). Previous works have shown that authentication systems can be more effective when considering soft biometric traits such as the gender, while others highlighted the significance of this trait for enhancing UIs. This paper presents a novel machine learning-based approach to gender classification leveraging the only touch gestures information derived from smartphones’ APIs. To identify the most useful gesture and combination thereof for gender classification, we have considered two strategies:single-viewlearning, analyzing, one at a time, datasets relating to a single type of gesture, andmulti-viewlearning, analyzing together datasets describing different types of gestures. This is one of the first works to apply such a strategy for gender recognition via gestures analysis on mobile devices. The methods have been evaluated on a large dataset of gestures collected through a mobile application, which includes not only scrolls, swipes, and taps but also pinch-to-zooms and drag-and-drops which are mostly overlooked in the literature. Conversely to the previous literature, we have also provided experiments of the solution in different scenarios, thus proposing a more comprehensive evaluation. The experimental results show thatscroll downis the most useful gesture andrandom forestis the most convenient classifier for gender classification. Based on the (combination of) gestures taken into account, we have obtained F1-score up to 0.89 in validation and 0.85 in testing phase. Furthermore, the multi-view approach is recommended when dealing with unknown devices and combinations of gestures can be effectively adopted, building on the requirements of the system our solution is built-into. Solutions proposed turn out to be both an opportunity for gender-aware technologies and a potential risk deriving from unwanted gender classification.

One of the challenges in the emerging field of computational crime analysis is that of extracting actionable knowledge from heterogeneous (both legal and empirical) information hidden into criminal proceedings. Public prosecutors generally deal with information systems that do not provide advanced information extraction functionalities and that boil down to databases containing complaints, criminal records or police reports. In this paper, we dwell on how information visualization can support knowledge mining in criminal investigations by playing a three-fold role: (a) depicting the structural and qualitative features of both criminal organizations and their members; (b) showing the evolution of criminal networks over time; (c) enhance the interaction between the domain expert and computational heuristics in the knowledge construction process. We present three visualizations designed to support knowledge mining in criminal investigations that have been tested with real data and evaluated by legal scholars and public prosecutors within a computational crime analysis project.

Several works on representation learning for graph-structured data have been proposed in recent literature. However, most of such techniques have several downsides. On the one hand, graph kernels which use handcrafted features (e.g., shortest paths) are hampered by poor generalization problems. On the other hand, methods for learning representations of whole graphs deal with unattributed or single-attributed graphs.In this work, we propose a novel technique for graph embedding learning able to take into account multi-attribute graphs (from 1 to an arbitrary number). Given a multi-attribute graph, the proposed method generates an embedding vector as follows: (i) the graph is split into several single-attribute graphs; for each of these, one numeric vector is generated by using state-of-the-art graph embedding techniques; (ii) the obtained vectors are concatenated in one representative vector using a multi-view learning integration technique; (iii) the size of such a vector is reduced through deep autoencoders.Experiments have been conducted on the music style recognition problem. We focus on the corpus of 4-voice J. S. Bach’ compositions. First, such a corpus has been decomposed and translated into graph-based structures corresponding to the music scores. Then, the proposed method is applied to generate the embedding vectors from the obtained graphs. Finally, a Random Forest model trained on such obtained vectors is used for generating novels music compositions in the learned style. Results obtained show the effectiveness of the proposed approach.

Abstract The growth of Internet and the pervasiveness of ICT have led to a radical change in social relationships. One of the drawbacks of this change is the exposure of individuals to threats during online activities. In this context, thetechno-regulationparadigm is inspiring new ways to safeguard legally interests by means of tools allowing to hamper breaches of law. In this paper, we focus on the exposure of individuals to specific online threats when interacting with smartphones. We propose a novel techno-regulatory approach exploiting machine learning techniques to provide safeguards against threats online. Specifically, we study a set of touch-based gestures to distinguish between underages or adults who is accessing a smartphone, and so to guarantee protection. To evaluate the proposed approach’s effectiveness, we developed an Android app to build a dataset consisting of more than 9000 touch-gestures from 147 participants. We experimented bothsingle-viewandmulti-viewlearning techniques to find the best combination of touch-gestures able of distinguishing between adults and underages. Results show that the multi-view learning combining scrolls, swipes, and pinch-to-zoom gestures, achieves the best ROC AUC (0.92) and accuracy (88%) scores.

Abstract Terms of Service (ToS) are fundamental factors in the creation of physical as well as online legally relevant relationships. They not only define mutual rights and obligations but also inform users about contract key issues that, in online settings, span from liability limitations to data management and processing conditions. Despite their crucial role, however, ToS are often neglected by users that frequently accept without even reading what they agree upon, representing a critical issue when there exist potentially unfair clauses. To enhance users’ awareness and uphold legal safeguards, we first propose a definition of ToS unfairness based on a novel unfairness measure computed counting the unfair clauses contained in a ToS, and therefore, weighted according to their direct impact on the customers concrete interests. Secondly, we introduce a novel machine learning-based approach to classify ToS clauses, represented by using sentence embedding, in different categories classes and fairness levels. Results of a test involving well-known machine learning models show that Support Vector Machine is able to classify clauses into categories with a F1-score of 86% outperforming state-of-the-art methods, while Random Forest is able to classify clauses into fairness levels with a F1-score of 81%. With the final goal of making terms of service more readable and understandable, we embedded this approach into ToSware, a prototype of a Google Chrome extension. An evaluation study was performed to measure ToSware effectiveness, efficiency, and the overall users’ satisfaction when interacting with it.

During the traditional music analysis process, stylistic rules usually have to be deduced directly from examples of compositions or past performance. In such cases, musicians create external representations of a music style domain as source for reflection, inspiration and collaboration. However, due to the large number of music examples, creating such representations can be essential, but at the same time, slow and costly.In this paper, we show that interactive visualization and machine learning could aid in supporting and enhancing musician cognition and team-based collaboration. Specifically, we propose an approach to this problem which: (1) allows musicians to visually externalize their evolving mental models of a music domain, in the form of thematically organized anchored pairs. i.e., (narrative, graph), each one corresponding to a specific music pattern, and (2) uses such pairs to develop a music style classification system based on machine learning, as support for musicians during their activities (composition, performance). To this end, we introduce a novel graph representation of music stylistic patterns and discuss the advantages of linking such a representation to machine learning. Results of a preliminary study involving 10 musicians provided us with overall positive feedback about the effectiveness of our approach as well as further directions to explore.

Over the last few years, biometrics has emerged as an increasingly reliable solution to recognize people using their physiological or behavioural characteristics. Despite their advantages, biometric systems raise many practical, ethical and legal issues. While, understandably, main concerns involve privacy and the risk of covert surveillance, profiling, and social control, another relevant question is the potential exclusion of individuals that, due to injuries, disability or genetic defects, may not meet the physical requirements used for the identification. In such situations, the risk comes out from the limits of current biometrics systems, which could exclude entire classes of individuals with negative spillovers on the possibility of access services and even exercise rights. In this paper, we focus on the recognition of iris suffering from Coloboma, a congenital abnormality of membranes of the eye. We first show how this pathological state impacts on the performance of the Daugman's algorithm, which represents the most widespread method used for the iris localization step in eye-based biometrics. Second, we designed and tested a classifier based on Convolutional Neural Network able to detect the presence of Coloboma with 95.45% accuracy. This result opens up new perspectives towards the definition of more sophisticated "diversity-aware" biometric systems.

Sifting through the flood of documents today available and making sense of them is an ever-growing challenge of our time. Visualization is offering new ways to navigate through large sets of documents, mainly through graphs that intuitively depict textual entities (documents or parts thereof) and relations tying them. A relevant issue is that graphs can be smoothly used, especially in desktop environments. The small size of tablets' and smartphones' screens makes it difficult to simply port graph visualizations from desktop to mobile devices, just as the use of the latter is growing for browsing and reading texts. In the light of the above, a challenge is seamlessly merging abstraction power offered by visualization, information retrieval, and the access to texts. This paper presents a Human-Computer Interaction strategy integrating mobile devices (for visualization-based, high-level and intuitive interaction with data and their relations) and desktop computers (for full-text reading). We focus on Sliding Treemap, a visualization designed to render graph-like (hierarchical, relational) structures on mobile touch devices. The solution, suitable to be applied also in other domains, has been designed based on the needs of the legal domain, which is "highly textual" and still rather unfamiliar with visualization techniques. Sliding Treemap has been developed in the prototype of a cross-platform app using as testbed the catalogue of the European Court of Justice Library, and sets of heterogeneous legal sources from the Italian legal system.

The paper presents ongoing research aiming to ease the interaction with large amounts of public sector data. The project focuses on the development of a modular online platform for both desktop and mobile devices exploiting Visual Analytics to offer citizens, researchers and policymakers crosscutting reading of heterogeneous information. We experiment new ways to integrate, analyze, and visualize administrative, legal, and economic data as they appear to be a powerful ally not only to increase the transparency of government activities but also to enhance evidence-based policy making and agenda setting.

Visual languages are studied in many different disciplines including Formal Logic. Several diagram methods have been proposed for the visual representation of the logical relations and in particular for First Order Predicate Logic (FOPL) formulas. Among these, logical symbolism, Euler diagrams, semantic networks, conceptual grids, conceptual spaces and so on. It is shown that these representations are formally equivalent and can be inter-translated algorithmically, but provide different and complementary visualizations such that the use of multiple representations may provide greater insight than any alone. We present a new visual language, V-Logic, which support different visual representation schemes: spatial, graphical and symbolic notations. It specifies rules for mapping FOPL formulas in special semantically equivalent diagrams, named V-diagrams. Logical inference based on the interpretation of a V-diagram is essentially a translation of the diagram into a logical formalism. Such a translation is natural and could be used to teach FOPL. We performed a preliminary study with 10 students. Results provided us with overall positive feedback about the effectiveness of our approach as well as further directions to explore.

The paper presents three experimental platforms for legal analytics, online environments integrating heterogeneous computational heuristics, information processing, and visualization techniques to extract actionable knowledge from legal data. Our goal is to explore innovative approaches to issues spanning from information retrieval to the quantitative analysis of legal corpora or to the study of criminal organizations for research and investigative purposes. After a brief introduction to the e-science paradigm and to the role played in it by research platforms, we focus on visual analytics as a viable way to interact with legal data. We then present the tools, their main features and the results so far obtained. The paper ends up with some considerations about the computational turn of science and its role in promoting a much needed interdisciplinary and empirical evolution of legal research.