Guido Rocchietti

Korbinian Randl, Guido Rocchietti, Aron Henriksson et al.

Retrieval-Augmented Generation (RAG) systems combine dense retrievers and language models to ground LLM outputs in retrieved documents. However, the opacity of how these components interact creates challenges for deployment in high-stakes domains. We present RAG-E, an end-to-end explainability framework that quantifies retriever-generator alignment through mathematically grounded attribution methods. Our approach adapts Integrated Gradients for retriever analysis, introduces PMCSHAP, a Monte Carlo-stabilized Shapley Value approximation, for generator attribution, and introduces the Weighted Attribution-Relevance Gap (WARG) metric to measure how well a generator's document usage aligns with a retriever's ranking. Empirical analysis on TREC CAsT and FoodSafeSum reveals critical misalignments: for 47.4% to 66.7% of queries, generators ignore the retriever's top-ranked documents, while 48.1% to 65.9% rely on documents ranked as less relevant. These failure modes demonstrate that RAG output quality depends not solely on individual component performance but on their interplay, which can be audited via RAG-E.

Chiara Pugliese, Francesco Lettich, Guido Rocchietti et al.

In this resource paper, we present two publicly available datasets of semantically enriched human trajectories, together with the pipeline to build them. The trajectories are publicly available GPS traces retrieved from OpenStreetMap. Each dataset includes contextual layers such as stops, moves, points of interest (POIs), inferred transportation modes, and weather data. A novel semantic feature is the inclusion of synthetic, realistic social media posts generated by Large Language Models (LLMs), enabling multimodal and semantic mobility analysis. The datasets are available in both tabular and Resource Description Framework (RDF) formats, supporting semantic reasoning and FAIR data practices. They cover two structurally distinct, large cities: Paris and New York. Our open source reproducible pipeline allows for dataset customization, while the datasets support research tasks such as behavior modeling, mobility prediction, knowledge graph construction, and LLM-based applications. To our knowledge, our resource is the first to combine real-world movement, structured semantic enrichment, LLM-generated text, and semantic web compatibility in a reusable framework.