Davide Buscaldi

Mohamed Dhouib, Davide Buscaldi, Sonia Vanier et al.

Detecting tampered text in document images is a challenging task due to data scarcity. To address this, previous work has attempted to generate tampered documents using rule-based methods. However, the resulting documents often suffer from limited variety and poor visual quality, typically leaving highly visible artifacts that are rarely observed in real-world manipulations. This undermines the model's ability to learn robust, generalizable features and results in poor performance on real-world data. Motivated by this discrepancy, we propose a novel method for generating high-quality tampered document images. We first train an auxiliary network to compare text crops, leveraging contrastive learning with a novel strategy for defining positive pairs and their corresponding negatives. We also train a second auxiliary network to evaluate whether a crop tightly encloses the intended characters, without cutting off parts of characters or including parts of adjacent ones. Using a carefully designed generation pipeline that leverages both networks, we introduce a framework capable of producing diverse, high-quality tampered document images. We assess the effectiveness of our data generation pipeline by training multiple models on datasets derived from the same source images, generated using our method and existing approaches, under identical training protocols. Evaluating these models on various open-source datasets shows that our pipeline yields consistent performance improvements across architectures and datasets.

Hadi Abdine, Moussa Kamal Eddine, Michalis Vazirgiannis et al.

Word sense induction (WSI) is a difficult problem in natural language processing that involves the unsupervised automatic detection of a word's senses (i.e. meanings). Recent work achieves significant results on the WSI task by pre-training a language model that can exclusively disambiguate word senses, whereas others employ previously pre-trained language models in conjunction with additional strategies to induce senses. In this paper, we propose a novel unsupervised method based on hierarchical clustering and invariant information clustering (IIC). The IIC is used to train a small model to optimize the mutual information between two vector representations of a target word occurring in a pair of synthetic paraphrases. This model is later used in inference mode to extract a higher quality vector representation to be used in the hierarchical clustering. We evaluate our method on two WSI tasks and in two distinct clustering configurations (fixed and dynamic number of clusters). We empirically demonstrate that, in certain cases, our approach outperforms prior WSI state-of-the-art methods, while in others, it achieves a competitive performance.

Vanni Zavarella, Sergio Consoli, Diego Reforgiato Recupero et al.

Numerous methods and pipelines have recently emerged for the automatic extraction of knowledge graphs from documents such as scientific publications and patents. However, adapting these methods to incorporate alternative text sources like micro-blogging posts and news has proven challenging as they struggle to model open-domain entities and relations, typically found in these sources. In this paper, we propose an enhanced information extraction pipeline tailored to the extraction of a knowledge graph comprising open-domain entities from micro-blogging posts on social media platforms. Our pipeline leverages dependency parsing and classifies entity relations in an unsupervised manner through hierarchical clustering over word embeddings. We provide a use case on extracting semantic triples from a corpus of 100 thousand tweets about digital transformation and publicly release the generated knowledge graph. On the same dataset, we conduct two experimental evaluations, showing that the system produces triples with precision over 95% and outperforms similar pipelines of around 5% in terms of precision, while generating a comparatively higher number of triples.

Hugo Attali, Nathalie Pernelle, Davide Buscaldi et al.

Graph Neural Networks are powerful models for learning from graph-structured data, yet their effectiveness is often limited by two critical challenges: over-squashing, where information from distant nodes is excessively compressed, and over-smoothing, where repeated propagation makes node representations indistinguishable. Both phenomena stem from the interaction between message passing and the input topology, ultimately degrading information flow and limiting the performance of GNNs. In this survey, we examine graph rewiring techniques, a class of methods designed to modify the graph topology to enhance information propagation in GNNs. We provide a comprehensive review of state-of-the-art rewiring approaches, delving into their theoretical underpinnings, practical implementations, and performance trade-offs.

Vijini Liyanage, Davide Buscaldi

Thanks to the state-of-the-art Large Language Models (LLMs), language generation has reached outstanding levels. These models are capable of generating high quality content, thus making it a challenging task to detect generated text from human-written content. Despite the advantages provided by Natural Language Generation, the inability to distinguish automatically generated text can raise ethical concerns in terms of authenticity. Consequently, it is important to design and develop methodologies to detect artificial content. In our work, we present some classification models constructed by ensembling transformer models such as Sci-BERT, DeBERTa and XLNet, with Convolutional Neural Networks (CNNs). Our experiments demonstrate that the considered ensemble architectures surpass the performance of the individual transformer models for classification. Furthermore, the proposed SciBERT-CNN ensemble model produced an F1-score of 98.36% on the ALTA shared task 2023 data.

Mohamed Dhouib, Davide Buscaldi, Sonia Vanier et al.

Visual Language Models require substantial computational resources for inference due to the additional input tokens needed to represent visual information. However, these visual tokens often contain redundant and unimportant information, resulting in an unnecessarily high number of tokens. To address this, we introduce PACT, a method that reduces inference time and memory usage by pruning irrelevant tokens and merging visually redundant ones at an early layer of the language model. Our approach uses a novel importance metric to identify unimportant tokens without relying on attention scores, making it compatible with FlashAttention. We also propose a novel clustering algorithm, called Distance Bounded Density Peak Clustering, which efficiently clusters visual tokens while constraining the distances between elements within a cluster by a predefined threshold. We demonstrate the effectiveness of PACT through extensive experiments.

Hugo Attali, Davide Buscaldi, Nathalie Pernelle

Graph Neural Networks (GNNs) are powerful tools for learning from graph-structured data, but their effectiveness is often constrained by two critical challenges: oversquashing, where the excessive compression of information from distant nodes results in significant information loss, and oversmoothing, where repeated message-passing iterations homogenize node representations, obscuring meaningful distinctions. These issues, intrinsically linked to the underlying graph structure, hinder information flow and constrain the expressiveness of GNNs. In this survey, we examine graph rewiring techniques, a class of methods designed to address these structural bottlenecks by modifying graph topology to enhance information diffusion. We provide a comprehensive review of state-of-the-art rewiring approaches, delving into their theoretical underpinnings, practical implementations, and performance trade-offs.

Jérémie Dentan, Alexi Canesse, Davide Buscaldi et al.

Claim-level Uncertainty Quantification (UQ) is a promising approach to mitigate the lack of reliability in Large Language Models (LLMs). We introduce MUCH, the first claim-level UQ benchmark designed for fair and reproducible evaluation of future methods under realistic conditions. It includes 4,873 samples across four European languages (English, French, Spanish, and German) and four instruction-tuned open-weight LLMs. Unlike prior claim-level benchmarks, we release 24 generation logits per token, facilitating the development of future white-box methods without re-generating data. Moreover, in contrast to previous benchmarks that rely on manual or LLM-based segmentation, we propose a new deterministic algorithm capable of segmenting claims using as little as 0.2% of the LLM generation time. This makes our segmentation approach suitable for real-time monitoring of LLM outputs, ensuring that MUCH evaluates UQ methods under realistic deployment constraints. Finally, our evaluations show that current methods still have substantial room for improvement in both performance and efficiency.

Jérémie Dentan, Davide Buscaldi, Sonia Vanier

Verbatim memorization in Large Language Models (LLMs) is a multifaceted phenomenon involving distinct underlying mechanisms. We introduce a novel method to analyze the different forms of memorization described by the existing taxonomy. Specifically, we train Convolutional Neural Networks (CNNs) on the attention weights of the LLM and evaluate the alignment between this taxonomy and the attention weights involved in decoding. We find that the existing taxonomy performs poorly and fails to reflect distinct mechanisms within the attention blocks. We propose a new taxonomy that maximizes alignment with the attention weights, consisting of three categories: memorized samples that are guessed using language modeling abilities, memorized samples that are recalled due to high duplication in the training set, and non-memorized samples. Our results reveal that few-shot verbatim memorization does not correspond to a distinct attention mechanism. We also show that a significant proportion of extractable samples are in fact guessed by the model and should therefore be studied separately. Finally, we develop a custom visual interpretability technique to localize the regions of the attention weights involved in each form of memorization.

Mathis Le Bail, Jérémie Dentan, Davide Buscaldi et al.

Sparse Autoencoders (SAEs) have been successfully used to probe Large Language Models (LLMs) and extract interpretable concepts from their internal representations. These concepts are linear combinations of neuron activations that correspond to human-interpretable features. In this paper, we investigate the effectiveness of SAE-based explainability approaches for sentence classification, a domain where such methods have not been extensively explored. We present a novel SAE-based architecture tailored for text classification, leveraging a specialized classifier head and incorporating an activation rate sparsity loss. We benchmark this architecture against established methods such as ConceptShap, Independent Component Analysis, and other SAE-based concept extraction techniques. Our evaluation covers two classification benchmarks and four fine-tuned LLMs from the Pythia family. We further enrich our analysis with two novel metrics for measuring the precision of concept-based explanations, using an external sentence encoder. Our empirical results show that our architecture improves both the causality and interpretability of the extracted features.

Vijini Liyanage, Davide Buscaldi, Adeline Nazarenko

Automatic text generation based on neural language models has achieved performance levels that make the generated text almost indistinguishable from those written by humans. Despite the value that text generation can have in various applications, it can also be employed for malicious tasks. The diffusion of such practices represent a threat to the quality of academic publishing. To address these problems, we propose in this paper two datasets comprised of artificially generated research content: a completely synthetic dataset and a partial text substitution dataset. In the first case, the content is completely generated by the GPT-2 model after a short prompt extracted from original papers. The partial or hybrid dataset is created by replacing several sentences of abstracts with sentences that are generated by the Arxiv-NLP model. We evaluate the quality of the datasets comparing the generated texts to aligned original texts using fluency metrics such as BLEU and ROUGE. The more natural the artificial texts seem, the more difficult they are to detect and the better is the benchmark. We also evaluate the difficulty of the task of distinguishing original from generated text by using state-of-the-art classification models.

Danilo Dessì, Francesco Osborne, Diego Reforgiato Recupero et al.

The continuous growth of scientific literature brings innovations and, at the same time, raises new challenges. One of them is related to the fact that its analysis has become difficult due to the high volume of published papers for which manual effort for annotations and management is required. Novel technological infrastructures are needed to help researchers, research policy makers, and companies to time-efficiently browse, analyse, and forecast scientific research. Knowledge graphs i.e., large networks of entities and relationships, have proved to be effective solution in this space. Scientific knowledge graphs focus on the scholarly domain and typically contain metadata describing research publications such as authors, venues, organizations, research topics, and citations. However, the current generation of knowledge graphs lacks of an explicit representation of the knowledge presented in the research papers. As such, in this paper, we present a new architecture that takes advantage of Natural Language Processing and Machine Learning methods for extracting entities and relationships from research publications and integrates them in a large-scale knowledge graph. Within this research work, we i) tackle the challenge of knowledge extraction by employing several state-of-the-art Natural Language Processing and Text Mining tools, ii) describe an approach for integrating entities and relationships generated by these tools, iii) show the advantage of such an hybrid system over alternative approaches, and vi) as a chosen use case, we generated a scientific knowledge graph including 109,105 triples, extracted from 26,827 abstracts of papers within the Semantic Web domain. As our approach is general and can be applied to any domain, we expect that it can facilitate the management, analysis, dissemination, and processing of scientific knowledge.

Bilal Ghanem, Davide Buscaldi, Paolo Rosso

The online new emerging suspicious users, that usually are called trolls, are one of the main sources of hate, fake, and deceptive online messages. Some agendas are utilizing these harmful users to spread incitement tweets, and as a consequence, the audience get deceived. The challenge in detecting such accounts is that they conceal their identities which make them disguised in social media, adding more difficulty to identify them using just their social network information. Therefore, in this paper, we propose a text-based approach to detect the online trolls such as those that were discovered during the US 2016 presidential elections. Our approach is mainly based on textual features which utilize thematic information, and profiling features to identify the accounts from their way of writing tweets. We deduced the thematic information in a unsupervised way and we show that coupling them with the textual features enhanced the performance of the proposed model. In addition, we find that the proposed profiling features perform the best comparing to the textual features.