VerbaNexAI

Nowadays, persuasive messages are more and more frequent in social networks, which generates great concern in several communities, given that persuasion seeks to guide others towards the adoption of ideas, attitudes or actions that they consider to be beneficial to themselves. The efficient detection of news genre categories, detection of framing and detection of persuasion techniques requires several scientific disciplines, such as computational linguistics and sociology. Here we illustrate how we use lexical features given a news article, determine whether it is an opinion piece, aims to report factual news, or is satire. This paper presents a novel strategy for news based on Lexical Weirdness. The results are part of our participation in subtasks 1 and 2 in SemEval 2023 Task 3.

Our approach achieved first place in the CLEF 2025 JOKER Task 3 competition, outperforming all other participating teams and establishing new benchmarks for LLM-based creative translation systems. Testing five different models using advanced prompting strategies. Our methodology involved systematic prompt engineering with Chain-of-Thought reasoning and universe-specific translation patterns. ChatGPT-4o achieved the best performance with 29.5% exact matches and 30.6% accent-tolerant matches, resulting in an overall 60.1% success rate, demonstrating the potential of LLM-based approaches for creative multilingual wordplay translation.

This study presents an advanced approach for predicting human preferences and generating explanations in large language models (LLMs) within the context of the "Preference Prediction" task of ELOQUENT Lab 2025. We implemented techniques such as Few-Shot Learning, Auto Chain-of-Thought (Auto-CoT), and Retrieval-Augmented Generation (RAG), evaluating multiple pre-trained models, from LLaMA-3 to distilgpt2. The system developed by the VerbaNexAI team achieved first place in the competition, standing out for its high performance in both safety (94.15%) and truthfulness (75.16%) criteria.

The BioNNE-L 2025-1 challenge advances biomedical entity linking (BEL) by mapping textual mentions to UMLS concepts, which is crucial for clinical and research applications. This study addresses Subtask 1 (English) with a novel SapBERT-based system. It integrates a hybrid re-ranking strategy combining cosine, Jaccard, and Levenshtein similarities, optimizing weights via grid search. Evaluated on the BioNNE-L development set, our system achieved an Accuracy@1 of 0.718, Accuracy@5 of 0.802, and MRR of 0.750. In the official competition, the VerbaNex AI Lab team secured first place in Accuracy@1 (0.70).

This study presents a novel approach utilizing DistilBERT, a lightweight variant of BERT, to identify inconsistencies in piping and instrumentation diagrams (P&IDs) within SmartPID systems. A structured dataset was constructed by extracting engineering design data from a SQL-based SmartPID database, monitoring all modifications and updates made throughout the design phase. The DistilBERT model was fine-tuned on this dataset to recognize inconsistencies in real-time, achieving an impressive F1 score of 99% and a loss of 0.04%. The model’s performance was validated by domain experts, who confirmed the detected inconsistencies as highly accurate. Our approach significantly reduces the manual effort required for P&ID review and improves design consistency, demonstrating the potential for enhanced safety and efficiency in complex industrial projects. Future work will focus on refining the model’s parameters and expanding its application across different industries.