In the recent digitization era, capturing, representing, and understanding knowledge is essential in countless real-world scenarios. Knowledge graphs emerged as a powerful tool for representing information through an adequately interconnected and interpretable structure in such a context. Nevertheless, generating proper knowledge graphs usually requires significant manual effort and domain expertise, resulting in graphs often affected by human subjectivity, limited scalability, or inability to capture implicit knowledge or handle heterogeneity. This paper proposes an innovative zero-shot strategy tailored to uncover reliable knowledge from text leveraging the recent highly effective generative large language models, with a particular focus on the GPT-3.5 model. Our proposal aims to create a suitable knowledge graph or improve existing ones by discovering missing qualitative triples. To assess the effectiveness of our methodology, we performed experiments on domain-specific datasets, confirming its potential for scalable and versatile knowledge discovery.

Abstract Ultrasound is a readily available, non-invasive and low-cost screening for the identification of endometriosis lesions, but its diagnostic specificity strongly depends on the experience of the operator. For this reason, computer-aided diagnosis tools based on Artificial Intelligence techniques can provide significant help to the clinical staff, both in terms of workload reduction and in increasing the overall accuracy of this type of examination and its outcome. However, although these techniques are spreading rapidly in a variety of domains, their application to endometriosis is still very limited. To fill this gap, we propose and evaluate a novel multi-scale ensemble approach for the automatic segmentation of endometriosis lesions from transvaginal ultrasounds. The peculiarity of the method lies in its high discrimination capability, obtained by combining, in a fusion fashion, multiple Convolutional Neural Networks trained on data at different granularity. The experimental validation carried out shows that: (i) the proposed method allows to significantly improve the performance of the individual neural networks, even in the presence of a limited training set; (ii) with a Dice coefficient of 82%, it represents a valid solution to increase the diagnostic efficacy of the ultrasound examination against such a pathology.

This work is focused on the sailing domain, for which several innovative technologies are being adopted to improve sailing efficiency, performance, and safety. In this context a knowledge graph could be used, for example, to represent information about different types of boats, sailing techniques, maritime safety, or weather conditions. Although numerous construction methods or ready-to-go knowledge graphs have been proposed in many fields, the sailing domain still needs to be explored. As the most effective methods rely on domain-specific datasets, the absence of suitable and available sailing datasets is one of the main challenges. Although several Open Information Extraction (OpenIE) methods may generate relevant triplets (the elementary units composing a knowledge graph) from arbitrary text without any additional information about its topic, such methods usually generate many incorrect triplets. In this paper, we aim (i) to address the aforementioned problem by proposing an innovative method that combines in an improved and strengthened way different OpenIE tools to generate proper triplets from domain-specific sources and, in particular, (ii) to build and release a suitable dataset for the sailing domain. Results confirm that our proposal can maximize the extracted information and infer unique information irretrievable by the classical OpenIE tools and, furthermore, that the generated dataset is significantly valuable for the sailing scenario.

Abstract In the last years, scientific and industrial research has experienced a growing interest in acquiring large annotated data sets to train artificial intelligence algorithms for tackling problems in different domains. In this context, we have observed that even the market for football data has substantially grown. The analysis of football matches relies on the annotation of both individual players’ and team actions, as well as the athletic performance of players. Consequently, annotating football events at a fine-grained level is a very expensive and error-prone task. Most existing semi-automatic tools for football match annotation rely on cameras and computer vision. However, those tools fall short in capturing team dynamics and in extracting data of players who are not visible in the camera frame. To address these issues, in this manuscript we present FootApp, an AI-based system for football match annotation. First, our system relies on an advanced and mixed user interface that exploits both vocal and touch interaction. Second, the motor performance of players is captured and processed by applying machine learning algorithms to data collected from inertial sensors worn by players. Artificial intelligence techniques are then used to check the consistency of generated labels, including those regarding the physical activity of players, to automatically recognize annotation errors. Notably, we implemented a full prototype of the proposed system, performing experiments to show its effectiveness in a real-world adoption scenario.

Machine learning techniques have recently become the norm for detecting patterns in financial markets. However, relying solely on machine learning algorithms for decision-making can have negative consequences, especially in a critical domain such as the financial one. On the other hand, it is well-known that transforming data into actionable insights can pose a challenge even for seasoned practitioners, particularly in the financial world. Given these compelling reasons, this work proposes a machine learning approach powered by eXplainable Artificial Intelligence techniques integrated into a statistical arbitrage trading pipeline. Specifically, we propose three methods to discard irrelevant features for the prediction task. We evaluate the approaches on historical data of component stocks of the S&P500 index and aim at improving not only the prediction performance at the stock level but also overall at the stock set level. Our analysis shows that our trading strategies that include such feature selection methods improve the portfolio performances by providing predictive signals whose information content suffices and is less noisy than the one embedded in the whole feature set. By performing an in-depth risk-return analysis, we show that the proposed trading strategies powered by explainable AI outperform highly competitive trading strategies considered as baselines.

Manual event tagging may be a very long and stressful activity, due the monotonous operations involved. This is particularly true when dealing with online video tagging, as for football matches, in which the burden of events to tag can consist of many thousands of actions, according to the desired level of granularity. In this work we describe an actual solution, developed for an existing football match tagging application, in which the GUI has been enhanced and integrated with a Voice User Interface, aiming at reducing tagging time and error rate. Empirical tests have revealed the efficiency and the benefits brought by the developed solution.