HERE'S OUR BEST

The "Atlas of Data Science Research" (DS-Atlas), an interactive visualization tool designed to study the landscape of the DS field. The DS-Atlas is built on a dataset of approximately 1.3 million scientific publications from the Elsevier Scopus database, leveraging Natural Language Processing, Large Language Models, and dimensionality reduction techniques to generate a semantic representation of the DS research. The DS-Atlas provides interactive operations to explore the dataset by allowing users to focus on specific areas, filter by keywords and/or time periods, and uncover thematic connections and research trends.

LIA (Law Innovation Assistant) is an advanced AI tool for legislators designed to prevent contradictions when amending complex laws. By analyzing legal codes, it identifies all relevant articles related to a specific topic or proposes the necessary amendments to achieve a legislative goal (e.g., repealing a regulation). This ensures new legislation is consistent and well-integrated with the existing legal framework. LIA's core function is to reduce the risk of legislative conflicts, thereby improving the quality and effectiveness of parliamentary work and increasing the transparency of the legislative process.

Redcapp is a web app for interactive exploration of the Little Red Riding Hood fairy tale. The app has been developed in the framework of the EU H2020 VAST project with the aim to foster a discussion around the moral values that the fairy tale can communicate. The Redcapp app can be used in two modes. One mode is "individual" and will allow a single person to explore the fairy tale by making choices according to personal preferences. A different mode is "collective" and will allow to explore the fairy tale by making choices collegially in case the goal is to enforce a shared experience with other participants.

Canvas Conversation Tales is a web application to create stories in the form of imaginary conversations, based on the visual content of a selected artwork. This creation process turns out to be an example of crowdsourcing, more specifically of collaborative story writing, a term we use to identify an activity carried out online, by a group of people not supposed to know each other, producing a narrative as output. Canvas Conversation Tales is intended as an edutainment tool that can be adopted in a museum or exhibition area to engage both onsite and online visitors with a creative, collective activity. Further details in publications.

This research investigates how text-to-image diffusion models internally represent artistic concepts like content and style when generating artworks. Using cross-attention analysis, we examine how these models separate content-describing and style-describing elements in prompts. Our findings reveal that diffusion models show varying degrees of content-style separation, with content tokens typically influencing object regions and style tokens affecting backgrounds and textures. Explore the complete set of generated images here!

TimeFrame allows to explore the whole textual 1948 edition of Corriere della Sera (10,418 newspaper articles) and to visualize on a timeline over 146,000 event mentions and their participants. TimeFrame makes it possible to build event-based queries at different granularity levels.