https://example.com/publication_types/journal-article/Journal-ArticleHugoBlox Kit (https://hugoblox.com)en-usThu, 01 Jan 2026 00:00:00 +0000https://example.com/media/icon_hu_521f220490f7a7b8.pnghttps://example.com/publication_types/journal-article/https://example.com/publications/2026-adapting-language-models-with-continual-learning-for-temporal-drifts/Thu, 01 Jan 2026 00:00:00 +0000https://example.com/publications/2026-adapting-language-models-with-continual-learning-for-temporal-drifts/https://example.com/publications/2026-an-experimental-comparison-of-the-most-popular-approaches-to-fake-news-detection/Thu, 01 Jan 2026 00:00:00 +0000https://example.com/publications/2026-an-experimental-comparison-of-the-most-popular-approaches-to-fake-news-detection/https://example.com/publications/2026-challenging-the-abilities-of-large-language-models-in-italian-a-community-initiative/Thu, 01 Jan 2026 00:00:00 +0000https://example.com/publications/2026-challenging-the-abilities-of-large-language-models-in-italian-a-community-initiative/https://example.com/publications/2025-continually-learn-to-map-visual-concepts-to-language-models-in-resource-constrained-e/Wed, 01 Jan 2025 00:00:00 +0000https://example.com/publications/2025-continually-learn-to-map-visual-concepts-to-language-models-in-resource-constrained-e/https://example.com/publications/2025-prompting-encoder-models-for-zero-shot-classification-a-cross-domain-study-in-italian/Wed, 01 Jan 2025 00:00:00 +0000https://example.com/publications/2025-prompting-encoder-models-for-zero-shot-classification-a-cross-domain-study-in-italian/https://example.com/publications/2024-characterizing-fashion-influencers-behavior-on-instagram/Mon, 01 Jan 2024 00:00:00 +0000https://example.com/publications/2024-characterizing-fashion-influencers-behavior-on-instagram/https://example.com/publications/2024-continual-pre-training-mitigates-forgetting-in-language-and-vision/Mon, 01 Jan 2024 00:00:00 +0000https://example.com/publications/2024-continual-pre-training-mitigates-forgetting-in-language-and-vision/<h2 id="abstract">Abstract</h2> <p>: Pre-trained models are commonly used in Continual Learning to initialize the model before training on the stream of non-stationary data. However, pre-training is rarely applied during Continual Learning. We investigate the characteristics of the Continual Pre-Training scenario, where a model is continually pre-trained on a stream of incoming data and only later fine-tuned to different downstream tasks. We introduce an evaluation protocol for Continual Pre-Training which monitors forgetting against a Forgetting Control dataset not present in the continual stream. We disentangle the impact on forgetting of 3 main factors: the input modality (NLP, Vision), the architecture type (Transformer, ResNet) and the pre-training protocol (supervised, self-supervised). Moreover, we propose a Sample-Efficient Pre-training method (SEP) that speeds up the pre-training phase. We show that the pre-training protocol is the most important factor accounting for forgetting. Surprisingly, we discovered that self-supervised continual pre-training in both NLP and Vision is sufficient to mitigate forgetting without the use of any Continual Learning strategy. Other factors, like model depth, input modality and architecture type are not as crucial.</p>https://example.com/publications/2024-dataset-for-multimodal-fake-news-detection-and-verification-tasks/Mon, 01 Jan 2024 00:00:00 +0000https://example.com/publications/2024-dataset-for-multimodal-fake-news-detection-and-verification-tasks/https://example.com/publications/2024-deep-continual-learning-for-medical-call-incidents-text-classification-under-the-pres/Mon, 01 Jan 2024 00:00:00 +0000https://example.com/publications/2024-deep-continual-learning-for-medical-call-incidents-text-classification-under-the-pres/<h2 id="abstract">Abstract</h2> <p>The aim of this work is to develop and evaluate a deep classifier that can effectively prioritize Emergency Medical Call Incidents (EMCI) according to their life-threatening level under the presence of dataset shifts. We utilized a dataset consisting of 1982746 independent EMCI instances obtained from the Health Services Department of the Region of Valencia (Spain), with a time span from 2009 to 2019 (excluding 2013). The dataset includes free text dispatcher observations recorded during the call, as well as a binary variable indicating whether the event was life-threatening. To evaluate the presence of dataset shifts, we examined prior probability shifts, covariate shifts, and concept shifts. Subsequently, we designed and implemented four deep Continual Learning (CL) strategies-cumulative learning, continual fine-tuning, experience replay, and synaptic intelligence-alongside three deep CL baselines-joint training, static approach, and single fine-tuning-based on DistilBERT models. Our results demonstrated evidence of prior probability shifts, covariate shifts, and concept shifts in the data. Applying CL techniques had a statistically significant ($\alpha$=0.05) positive impact on both backward and forward knowledge transfer, as measured by the F1-score, compared to non-continual approaches. We can argue that the utilization of CL techniques in the context of EMCI is effective in adapting deep learning classifiers to changes in data distributions, thereby maintaining the stability of model performance over time. To our knowledge, this study represents the first exploration of a CL approach using real EMCI data.</p>https://example.com/publications/2023-special-issue-nl4ai-2022-workshop-on-natural-language-for-artificial-intelligence/Sun, 01 Jan 2023 00:00:00 +0000https://example.com/publications/2023-special-issue-nl4ai-2022-workshop-on-natural-language-for-artificial-intelligence/https://example.com/publications/2022-in-context-annotation-of-topic-oriented-datasets-of-fake-news-a-case-study-on-the-not/Sat, 01 Jan 2022 00:00:00 +0000https://example.com/publications/2022-in-context-annotation-of-topic-oriented-datasets-of-fake-news-a-case-study-on-the-not/<h2 id="abstract">Abstract</h2> <p>The problem of fake news detection is becoming increasingly interesting for several research fields. Different approaches have been proposed, based on either the content of the news itself or the context and properties of its spread over time, specifically on social media. In the literature, it does not exist a widely accepted general-purpose dataset for fake news detection, due to the complexity of the task and the increasing ability to produce fake news appearing credible in particular moments. In this paper, we propose a methodology to collect and label news pertinent to specific topics and subjects. Our methodology focuses on collecting data from social media about real-world events which are known to trigger fake news. We propose a labelling method based on crowdsourcing that is fast, reliable, and able to approximate expert human annotation. The proposed method exploits both the content of the data (i.e., the texts) and contextual information about fake news for a particular real-world event. The methodology is applied to collect and annotate the Notre-Dame Fire Dataset and to annotate part of the PHEME dataset. Evaluation is performed with fake news classifiers based on Transformers and fine-tuning. Results show that context-based annotation outperforms traditional crowdsourcing out-of-context annotation.</p>https://example.com/publications/2020-lessons-learned-from-evalita-2020-and-thirteen-years-of-evaluation-of-italian-languag/Wed, 01 Jan 2020 00:00:00 +0000https://example.com/publications/2020-lessons-learned-from-evalita-2020-and-thirteen-years-of-evaluation-of-italian-languag/https://example.com/publications/2018-the-emotions-of-abstract-words-a-distributional-semantic-analysis/Mon, 01 Jan 2018 00:00:00 +0000https://example.com/publications/2018-the-emotions-of-abstract-words-a-distributional-semantic-analysis/https://example.com/publications/2017-how-to-harvest-word-combinations-from-corpora-methods-evaluation-and-perspectives/Sun, 01 Jan 2017 00:00:00 +0000https://example.com/publications/2017-how-to-harvest-word-combinations-from-corpora-methods-evaluation-and-perspectives/https://example.com/publications/2017-learning-affect-with-distributional-semantic-models/Sun, 01 Jan 2017 00:00:00 +0000https://example.com/publications/2017-learning-affect-with-distributional-semantic-models/