arXiv subjects

Gabriella Pasi

Publications and source records attributed to Gabriella Pasi.

At least 19 recordsLinked to original sources

arXiv2026

ExpertLens: Visualizing Embedding Spaces for Post-Hoc Explainability in MoE Enhanced Retrievers

Neural models, including dense retrievers, have been widely adopted in Information Retrieval (IR), often delivering state-of-the-art performance. Despite their effectiveness, these models operate as black boxes, limiting the interpretability of their ranking decisions. Existing post-hoc explainability methods for neural rankers primarily focus on feature-level attributions, which can be insufficient to capture the complexity of learned embedding spaces. In this work, we propose ExpertLens, a post-hoc explainability framework for Mixture-of-Experts (MoE)-enhanced dense retrievers that shifts focus from local scalar feature importance to representation-level global interpretability. ExpertLens leverages discriminative embedding space visualizations jointly with automatically extracted Concept Activation Vectors to reveal how expert routing drives embedding space formulation and retrieval effectiveness. Experiments across five IR benchmarks and two MoE-enhanced dense retrievers show that expert routing consistently improves embedding space structure, positioning queries and their relevant documents into better-defined geometric neighborhoods. Analysis of expert subspaces further reveals general-purpose dominant experts, along with minority experts exhibiting distinct linguistic specialization, with subspaces arranged according to multi-semantic concept similarity. Our code is publicly available.

cs.IR↗

arXiv2026

DRAMA: Domain Retrieval using Adaptive Module Allocation

Neural models are increasingly used in Web-scale Information Retrieval (IR). However, relying on these models introduces substantial computational and energy requirements, leading to increasing attention toward their environmental cost and the sustainability of large-scale deployments. While neural IR models deliver high retrieval effectiveness, their scalability is constrained in multi-domain scenarios, where training and maintaining domain-specific models is inefficient and achieving robust cross-domain generalisation within a unified model remains difficult. This paper introduces DRAMA (Domain Retrieval using Adaptive Module Allocation), an energy- and parameter-efficient framework designed to reduce the environmental footprint of neural retrieval. DRAMA integrates domain-specific adapter modules with a dynamic gating mechanism that selects the most relevant domain knowledge for each query. New domains can be added efficiently through lightweight adapter training, avoiding full model retraining. We evaluate DRAMA on multiple Web retrieval benchmarks covering different domains. Our extensive evaluation shows that DRAMA achieves comparable effectiveness to domain-specific models while using only a fraction of their parameters and computational resources. These findings show that energy-aware model design can significantly improve scalability and sustainability in neural IR.

cs.IR↗

arXiv2025

Mixture of Experts Approaches in Dense Retrieval Tasks

Dense Retrieval Models (DRMs) are a prominent development in Information Retrieval (IR). A key challenge with these neural Transformer-based models is that they often struggle to generalize beyond the specific tasks and domains they were trained on. To address this challenge, prior research in IR incorporated the Mixture-of-Experts (MoE) framework within each Transformer layer of a DRM, which, though effective, substantially increased the number of additional parameters. In this paper, we propose a more efficient design, which introduces a single MoE block (SB-MoE) after the final Transformer layer. To assess the retrieval effectiveness of SB-MoE, we perform an empirical evaluation across three IR tasks. Our experiments involve two evaluation setups, aiming to assess both in-domain effectiveness and the model's zero-shot generalizability. In the first setup, we fine-tune SB-MoE with four different underlying DRMs on seven IR benchmarks and evaluate them on their respective test sets. In the second setup, we fine-tune SB-MoE on MSMARCO and perform zero-shot evaluation on thirteen BEIR datasets. Additionally, we perform further experiments to analyze the model's dependency on its hyperparameters (i.e., the number of employed and activated experts) and investigate how this variation affects SB-MoE's performance. The obtained results show that SB-MoE is particularly effective for DRMs with lightweight base models, such as TinyBERT and BERT-Small, consistently exceeding standard model fine-tuning across benchmarks. For DRMs with more parameters, such as BERT-Base and Contriever, our model requires a larger number of training samples to achieve improved retrieval performance. Our code is available online at: https://github.com/FaySokli/SB-MoE.

cs.IR↗

arXiv2025

Investigating Large Language Models' Linguistic Abilities for Text Preprocessing

Text preprocessing is a fundamental component of Natural Language Processing, involving techniques such as stopword removal, stemming, and lemmatization to prepare text as input for further processing and analysis. Despite the context-dependent nature of the above techniques, traditional methods usually ignore contextual information. In this paper, we investigate the idea of using Large Language Models (LLMs) to perform various preprocessing tasks, due to their ability to take context into account without requiring extensive language-specific annotated resources. Through a comprehensive evaluation on web-sourced data, we compare LLM-based preprocessing (specifically stopword removal, lemmatization and stemming) to traditional algorithms across multiple text classification tasks in six European languages. Our analysis indicates that LLMs are capable of replicating traditional stopword removal, lemmatization, and stemming methods with accuracies reaching 97%, 82%, and 74%, respectively. Additionally, we show that ML algorithms trained on texts preprocessed by LLMs achieve an improvement of up to 6% with respect to the $F_1$ measure compared to traditional techniques. Our code, prompts, and results are publicly available at https://github.com/GianCarloMilanese/llm_pipeline_wi-iat.

cs.CL↗

arXiv2025

PARK: Personalized academic retrieval with knowledge-graphs

Academic Search is a search task aimed to manage and retrieve scientific documents like journal articles and conference papers. Personalization in this context meets individual researchers' needs by leveraging, through user profiles, the user related information (e.g. documents authored by a researcher), to improve search effectiveness and to reduce the information overload. While citation graphs are a valuable means to support the outcome of recommender systems, their use in personalized academic search (with, e.g. nodes as papers and edges as citations) is still under-explored. Existing personalized models for academic search often struggle to fully capture users' academic interests. To address this, we propose a two-step approach: first, training a neural language model for retrieval, then converting the academic graph into a knowledge graph and embedding it into a shared semantic space with the language model using translational embedding techniques. This allows user models to capture both explicit relationships and hidden structures in citation graphs and paper content. We evaluate our approach in four academic search domains, outperforming traditional graph-based and personalized models in three out of four, with up to a 10\% improvement in MAP@100 over the second-best model. This highlights the potential of knowledge graph-based user models to enhance retrieval effectiveness.

cs.IR↗

arXiv2025

Reasoning Capabilities and Invariability of Large Language Models

Large Language Models (LLMs) have shown remarkable capabilities in manipulating natural language across multiple applications, but their ability to handle simple reasoning tasks is often questioned. In this work, we aim to provide a comprehensive analysis of LLMs' reasoning competence, specifically focusing on their prompt dependency. In particular, we introduce a new benchmark dataset with a series of simple reasoning questions demanding shallow logical reasoning. Aligned with cognitive psychology standards, the questions are confined to a basic domain revolving around geometric figures, ensuring that responses are independent of any pre-existing intuition about the world and rely solely on deduction. An empirical analysis involving zero-shot and few-shot prompting across 24 LLMs of different sizes reveals that, while LLMs with over 70 billion parameters perform better in the zero-shot setting, there is still a large room for improvement. An additional test with chain-of-thought prompting over 22 LLMs shows that this additional prompt can aid or damage the performance of models, depending on whether the rationale is required before or after the answer.

cs.CL↗

arXiv2025

Investigating Task Arithmetic for Zero-Shot Information Retrieval

Large Language Models (LLMs) have shown impressive zero-shot performance across a variety of Natural Language Processing tasks, including document re-ranking. However, their effectiveness degrades on unseen tasks and domains, largely due to shifts in vocabulary and word distributions. In this paper, we investigate Task Arithmetic, a technique that combines the weights of LLMs pre-trained on different tasks or domains via simple mathematical operations, such as addition or subtraction, to adapt retrieval models without requiring additional fine-tuning. Our method is able to synthesize diverse tasks and domain knowledge into a single model, enabling effective zero-shot adaptation in different retrieval contexts. Extensive experiments on publicly available scientific, biomedical, and multilingual datasets show that our method improves state-of-the-art re-ranking performance by up to 18% in NDCG@10 and 15% in P@10. In addition to these empirical gains, our analysis provides insights into the strengths and limitations of Task Arithmetic as a practical strategy for zero-shot learning and model adaptation. We make our code publicly available at https://github.com/DetectiveMB/Task-Arithmetic-for-ZS-IR.

cs.IR↗

arXiv2024

Investigating Mixture of Experts in Dense Retrieval

While Dense Retrieval Models (DRMs) have advanced Information Retrieval (IR), one limitation of these neural models is their narrow generalizability and robustness. To cope with this issue, one can leverage the Mixture-of-Experts (MoE) architecture. While previous IR studies have incorporated MoE architectures within the Transformer layers of DRMs, our work investigates an architecture that integrates a single MoE block (SB-MoE) after the output of the final Transformer layer. Our empirical evaluation investigates how SB-MoE compares, in terms of retrieval effectiveness, to standard fine-tuning. In detail, we fine-tune three DRMs (TinyBERT, BERT, and Contriever) across four benchmark collections with and without adding the MoE block. Moreover, since MoE showcases performance variations with respect to its parameters (i.e., the number of experts), we conduct additional experiments to investigate this aspect further. The findings show the effectiveness of SB-MoE especially for DRMs with a low number of parameters (i.e., TinyBERT), as it consistently outperforms the fine-tuned underlying model on all four benchmarks. For DRMs with a higher number of parameters (i.e., BERT and Contriever), SB-MoE requires larger numbers of training samples to yield better retrieval performance.

cs.IR↗

arXiv2024

Aligning Generalisation Between Humans and Machines

Recent advances in AI -- including generative approaches -- have resulted in technology that can support humans in scientific discovery and forming decisions, but may also disrupt democracies and target individuals. The responsible use of AI and its participation in human-AI teams increasingly shows the need for AI alignment, that is, to make AI systems act according to our preferences. A crucial yet often overlooked aspect of these interactions is the different ways in which humans and machines generalise. In cognitive science, human generalisation commonly involves abstraction and concept learning. In contrast, AI generalisation encompasses out-of-domain generalisation in machine learning, rule-based reasoning in symbolic AI, and abstraction in neurosymbolic AI. In this perspective paper, we combine insights from AI and cognitive science to identify key commonalities and differences across three dimensions: notions of, methods for, and evaluation of generalisation. We map the different conceptualisations of generalisation in AI and cognitive science along these three dimensions and consider their role for alignment in human-AI teaming. This results in interdisciplinary challenges across AI and cognitive science that must be tackled to provide a foundation for effective and cognitively supported alignment in human-AI teaming scenarios.

cs.AI↗

arXiv2024

Leveraging Large Language Models for Medical Information Extraction and Query Generation

This paper introduces a system that integrates large language models (LLMs) into the clinical trial retrieval process, enhancing the effectiveness of matching patients with eligible trials while maintaining information privacy and allowing expert oversight. We evaluate six LLMs for query generation, focusing on open-source and relatively small models that require minimal computational resources. Our evaluation includes two closed-source and four open-source models, with one specifically trained in the medical field and five general-purpose models. We compare the retrieval effectiveness achieved by LLM-generated queries against those created by medical experts and state-of-the-art methods from the literature. Our findings indicate that the evaluated models reach retrieval effectiveness on par with or greater than expert-created queries. The LLMs consistently outperform standard baselines and other approaches in the literature. The best performing LLMs exhibit fast response times, ranging from 1.7 to 8 seconds, and generate a manageable number of query terms (15-63 on average), making them suitable for practical implementation. Our overall findings suggest that leveraging small, open-source LLMs for clinical trials retrieval can balance performance, computational efficiency, and real-world applicability in medical settings.

cs.IR↗

arXiv2024

Synthetic Data Generation with Large Language Models for Personalized Community Question Answering

Personalization in Information Retrieval (IR) is a topic studied by the research community since a long time. However, there is still a lack of datasets to conduct large-scale evaluations of personalized IR; this is mainly due to the fact that collecting and curating high-quality user-related information requires significant costs and time investment. Furthermore, the creation of datasets for Personalized IR (PIR) tasks is affected by both privacy concerns and the need for accurate user-related data, which are often not publicly available. Recently, researchers have started to explore the use of Large Language Models (LLMs) to generate synthetic datasets, which is a possible solution to generate data for low-resource tasks. In this paper, we investigate the potential of Large Language Models (LLMs) for generating synthetic documents to train an IR system for a Personalized Community Question Answering task. To study the effectiveness of IR models fine-tuned on LLM-generated data, we introduce a new dataset, named Sy-SE-PQA. We build Sy-SE-PQA based on an existing dataset, SE-PQA, which consists of questions and answers posted on the popular StackExchange communities. Starting from questions in SE-PQA, we generate synthetic answers using different prompt techniques and LLMs. Our findings suggest that LLMs have high potential in generating data tailored to users' needs. The synthetic data can replace human-written training data, even if the generated data may contain incorrect information.

cs.IR↗

arXiv2024

Report on the Workshop on Simulations for Information Access (Sim4IA 2024) at SIGIR 2024

This paper is a report of the Workshop on Simulations for Information Access (Sim4IA) workshop at SIGIR 2024. The workshop had two keynotes, a panel discussion, nine lightning talks, and two breakout sessions. Key takeaways were user simulation's importance in academia and industry, the possible bridging of online and offline evaluation, and the issues of organizing a companion shared task around user simulations for information access. We report on how we organized the workshop, provide a brief overview of what happened at the workshop, and summarize the main topics and findings of the workshop and future work.

cs.IR↗

arXiv2024

Health Misinformation Detection in Web Content via Web2Vec: A Structural-, Content-based, and Context-aware Approach based on Web2Vec

In recent years, we have witnessed the proliferation of large amounts of online content generated directly by users with virtually no form of external control, leading to the possible spread of misinformation. The search for effective solutions to this problem is still ongoing, and covers different areas of application, from opinion spam to fake news detection. A more recently investigated scenario, despite the serious risks that incurring disinformation could entail, is that of the online dissemination of health information. Early approaches in this area focused primarily on user-based studies applied to Web page content. More recently, automated approaches have been developed for both Web pages and social media content, particularly with the advent of the COVID-19 pandemic. These approaches are primarily based on handcrafted features extracted from online content in association with Machine Learning. In this scenario, we focus on Web page content, where there is still room for research to study structural-, content- and context-based features to assess the credibility of Web pages. Therefore, this work aims to study the effectiveness of such features in association with a deep learning model, starting from an embedded representation of Web pages that has been recently proposed in the context of phishing Web page detection, i.e., Web2Vec.

cs.IR↗

arXiv2024

DESIRE-ME: Domain-Enhanced Supervised Information REtrieval using Mixture-of-Experts

Open-domain question answering requires retrieval systems able to cope with the diverse and varied nature of questions, providing accurate answers across a broad spectrum of query types and topics. To deal with such topic heterogeneity through a unique model, we propose DESIRE-ME, a neural information retrieval model that leverages the Mixture-of-Experts framework to combine multiple specialized neural models. We rely on Wikipedia data to train an effective neural gating mechanism that classifies the incoming query and that weighs the predictions of the different domain-specific experts correspondingly. This allows DESIRE-ME to specialize adaptively in multiple domains. Through extensive experiments on publicly available datasets, we show that our proposal can effectively generalize domain-enhanced neural models. DESIRE-ME excels in handling open-domain questions adaptively, boosting by up to 12% in NDCG@10 and 22% in P@1, the underlying state-of-the-art dense retrieval model.

cs.IR↗

arXiv2023

SE-PEF: a Resource for Personalized Expert Finding

The problem of personalization in Information Retrieval has been under study for a long time. A well-known issue related to this task is the lack of publicly available datasets that can support a comparative evaluation of personalized search systems. To contribute in this respect, this paper introduces SE-PEF (StackExchange - Personalized Expert Finding), a resource useful for designing and evaluating personalized models related to the task of Expert Finding (EF). The contributed dataset includes more than 250k queries and 565k answers from 3 306 experts, which are annotated with a rich set of features modeling the social interactions among the users of a popular cQA platform. The results of the preliminary experiments conducted show the appropriateness of SE-PEF to evaluate and to train effective EF models.

cs.IR↗

arXiv2023

Denoising Attention for Query-aware User Modeling in Personalized Search

The personalization of search results has gained increasing attention in the past few years, thanks to the development of Neural Networks-based approaches for Information Retrieval and the importance of personalization in many search scenarios. Recent works have proposed to build user models at query time by leveraging the Attention mechanism, which allows weighing the contribution of the user-related information w.r.t. the current query. This approach allows taking into account the diversity of the user's interests by giving more importance to those related to the current search performed by the user. In this paper, we first discuss some shortcomings of the standard Attention formulation when employed for personalization. In particular, we focus on issues related to its normalization mechanism and its inability to entirely filter out noisy user-related information. Then, we introduce the Denoising Attention mechanism: an Attention variant that directly tackles the above shortcomings by adopting a robust normalization scheme and introducing a filtering mechanism. The reported experimental evaluation shows the benefits of the proposed approach over other Attention-based variants.

cs.IR↗

arXiv2023

Fuzzy order-sorted feature logic

Order-Sorted Feature (OSF) logic is a knowledge representation and reasoning language based on function-denoting feature symbols and set-denoting sort symbols ordered in a subsumption lattice. OSF logic allows the construction of record-like terms that represent classes of entities and that are themselves ordered in a subsumption relation. The unification algorithm for such structures provides an efficient calculus of type subsumption, which has been applied in computational linguistics and implemented in constraint logic programming languages such as LOGIN and LIFE and automated reasoners such as CEDAR. This work generalizes OSF logic to a fuzzy setting. We give a flexible definition of a fuzzy subsumption relation which generalizes Zadeh's inclusion between fuzzy sets. Based on this definition we define a fuzzy semantics of OSF logic where sort symbols and OSF terms denote fuzzy sets. We extend the subsumption relation to OSF terms and prove that it constitutes a fuzzy partial order with the property that two OSF terms are subsumed by one another in the crisp sense if and only if their subsumption degree is greater than 0. We show how to find the greatest lower bound of two OSF terms by unifying them and how to compute the subsumption degree between two OSF terms, and we provide the complexity of these operations.

cs.AI↗

arXiv2023

Effective Matching of Patients to Clinical Trials using Entity Extraction and Neural Re-ranking

Clinical trials (CTs) often fail due to inadequate patient recruitment. This paper tackles the challenges of CT retrieval by presenting an approach that addresses the patient-to-trials paradigm. Our approach involves two key components in a pipeline-based model: (i) a data enrichment technique for enhancing both queries and documents during the first retrieval stage, and (ii) a novel re-ranking schema that uses a Transformer network in a setup adapted to this task by leveraging the structure of the CT documents. We use named entity recognition and negation detection in both patient description and the eligibility section of CTs. We further classify patient descriptions and CT eligibility criteria into current, past, and family medical conditions. This extracted information is used to boost the importance of disease and drug mentions in both query and index for lexical retrieval. Furthermore, we propose a two-step training schema for the Transformer network used to re-rank the results from the lexical retrieval. The first step focuses on matching patient information with the descriptive sections of trials, while the second step aims to determine eligibility by matching patient information with the criteria section. Our findings indicate that the inclusion criteria section of the CT has a great influence on the relevance score in lexical models, and that the enrichment techniques for queries and documents improve the retrieval of relevant trials. The re-ranking strategy, based on our training schema, consistently enhances CT retrieval and shows improved performance by 15\% in terms of precision at retrieving eligible trials. The results of our experiments suggest the benefit of making use of extracted entities. Moreover, our proposed re-ranking schema shows promising effectiveness compared to larger neural models, even with limited training data.

cs.IR↗