Social media platforms have enabled large-scale influence campaigns, impacting democratic processes. To fight against these threats, continuous training is needed. A typical training session is based on a fictive scenario describing key elements which are instantiated into a dedicated platform.Such a platform simulates social networks, which host a huge amount of content aligned with the training scenario. However, directly using Large Language Models to create appropriate content result in low content diversity due to coarse-grained and high-level scenario constraints, which compromises the trainees’ immersion.We address this issue with SocialForge, a system designed toenhance the diversity and realism of the generated content while ensuring its adherence to the original scenario.Specifically, SocialForge refines and augments the initial scenario constraints by generating detailed subnarratives, personas, and events.We assess diversity, realism, and adherence to the scenario through custom evaluation protocol. We also propose an automatic method to detect erroneous constraint generation, ensuring optimal alignment of the content with the scenario.SocialForge has been used in real trainings and in several showcases, with great end-user satisfaction. We release an open-source dataset generated with SocialForge for the research community.

Narratives are a new tool to propagate ideas that are sometimes well hidden in press articles. The SemEval-2025 Task 10 focuses on detecting and extracting such narratives in multiple languages. In this paper, we explore the capabilities of encoder-based language models to classify texts according to the narrative they contain. We show that multilingual encoders outperform monolingual models on this dataset, which is challenging due to the small number of samples per class per language. We perform additional experiments to measure the generalization of features in multilingual models to new languages.

We present a corpus of 100 documents, named OBSINFOX, selected from 17 sources of French press considered unreliable by expert agencies, annotated using 11 labels by 8 annotators. By collecting more labels than usual, by more annotators than is typically done, we can identify features that humans consider as characteristic of fake news, and compare them to the predictions of automated classifiers. We present a topic and genre analysis using Gate Cloud, indicative of the prevalence of satire-like text in the corpus. We then use the subjectivity analyzer VAGO, and a neural version of it, to clarify the link between ascriptions of the label Subjective and ascriptions of the label Fake News. The annotated dataset is available online at the following url: https://github.com/obs-info/obsinfox Keywords: Fake News, Multi-Labels, Subjectivity, Vagueness, Detail, Opinion, Exaggeration, French Press

This paper investigates the language of propaganda and its stylistic features. It presents the PPN dataset, standing for Propagandist Pseudo-News, a multisource, multilingual, multimodal dataset composed of news articles extracted from websites identified as propaganda sources by expert agencies. A limited sample from this set was randomly mixed with papers from the regular French press, and their URL masked, to conduct an annotation-experiment by humans, using 11 distinct labels. The results show that human annotators were able to reliably discriminate between the two types of press across each of the labels. We use different NLP techniques to identify the cues used by annotators, and to compare them with machine classification: first the analyzer VAGO to detect discourse vagueness and subjectivity, and then four different classifiers, two based on RoBERTa, one CATS using syntax, and one XGBoost combining syntactic and semantic features.

Maritime security requires full-time monitoring of the situation, mainly based on technical data (radar, AIS) but also from OSINT-like inputs (e.g., newspapers). Some threats to the operational reliability of this maritime surveillance, such as malicious actors, introduce discrepancies between hard and soft data (sensors and texts), either by tweaking their AIS emitters or by emitting false information on pseudo-newspapers. Many techniques exist to identify these pieces of false information, including using knowledge base population techniques to build a structured view of the information. This paper presents a use case for suspect data identification in a maritime setting. The proposed system UMBAR ingests data from sensors and texts, processing them through an information extraction step, in order to feed a Knowledge Base and finally perform coherence checks between the extracted facts.

Ces dernières années, les contributions majeures qui ont eu lieu en apprentissage automatique supervisé ont mis en evidence la nécessité de disposer de grands jeux de données annotés de haute qualité. Les recherches menées sur la tâche de reconnaissance d’entités nommées dans des textes en français font face à l’absence de jeux de données annotés “à grande échelle” et avec de nombreuses classes d’entités hiérarchisées. Dans cet article, nous proposons une approche pour obtenir un tel jeu de données qui s’appuie sur des étapes de traduction puis d’annotation des données textuelles en anglais vers une langue cible (ici au français). Nous évaluons la qualité de l’approche proposée et mesurons les performances de quelques modèles d’apprentissage automatique sur ces données.

Arabizi is a written form of spoken Arabic, relying on Latin characters and digits. It is informal and does not follow any conventional rules, raising many NLP challenges. In particular, Arabizi has recently emerged as the Arabic language in online social networks, becoming of great interest for opinion mining and sentiment analysis. Unfortunately, only few Arabizi resources exist and state-of-the-art language models such as BERT do not consider Arabizi. In this work, we construct and release two datasets: (i) LAD, a corpus of 7.7M tweets written in Arabizi and (ii) SALAD, a subset of LAD, manually annotated for sentiment analysis. Then, a BERT architecture is pre-trained on LAD, in order to create and distribute an Arabizi language model called BAERT. We show that a language model (BAERT) pre-trained on a large corpus (LAD) in the same language (Arabizi) as that of the fine-tuning dataset (SALAD), outperforms a state-of-the-art multi-lingual pretrained model (multilingual BERT) on a sentiment analysis task.