14 research outputs found
Streamlining event relation extraction: A pipeline leveraging pretrained and large language models for inference
No full text
Extraction sémantique de relations entre événements à partir de textes avec des graphes de connaissances
No full textThis research addresses the scientific challenge of accurately capturing event flows from textual data, crucial for informed decision-making, historical analysis, and predictive modeling. We introduce FARO, an ontology that structures 25 distinct relationships among events and facts, enabling richer semantic representations. To support robust event relation extraction, we leverage large language models (LLMs), common sense knowledge from the ATOMIC knowledge graph, and generative AI techniques to create a novel annotated dataset of over 500,000 sentences encompassing refined relations such as direct causality, enablement, prevention, and intention. Utilizing this comprehensive resource, we develop and comparatively evaluate an advanced extraction model capable of identifying fine-grained causal event relationships. The practical effectiveness of our approach is validated through two applications: enhanced narrative generation via structured and context-rich knowledge graphs, and automated fact-checking using causal reasoning, achieving a notable F1-score of 61.56% on the AVeriTeC dataset, thus establishing a strong foundation for future research in causal-aware methodologiesCette recherche traite du défi scientifique que représente l'extraction précise des flux d'événements à partir de données textuelles, essentielle pour la prise de décision, l'analyse historique et la modélisation prédictive. Nous proposons FARO, une ontologie qui structure 25 relations distinctes entre événements et faits, permettant des représentations sémantiques enrichies. Afin de soutenir l'extraction robuste des relations événementielles, nous exploitons des grands modèles de langage (LLMs), des connaissances du bon sense issues du graphe ATOMIC, et des techniques d'intelligence artificielle générative pour créer une nouvelle base de donnée annoté de plus de 500 000 phrases couvrant des relations sémantiques raffinées telles que la causalité directe, la facilitation, la prévention et l'intention. À partir de cette ressource, nous développons et évaluons comparativement un modèle performant, capable d'identifier avec précision ces relations causales fines. L'efficacité pratique de notre approche est validée par deux applications : la génération narrative enrichie grâce à des graphes de connaissances structurés et contextualisés, et la vérification automatique des faits utilisant un raisonnement causal, atteignant un score F1 notable de 61,56 % sur le jeu de données AVeriTeC, établissant ainsi une solide base pour les futures recherches en méthodes intégrant la causalit
Optimization approach for calcined clay and slag based geopolymer mortar – an experimental investigation for 3DCP
No full textAfter water, concrete is the most used substance on the planet. Approximately three tonnes of concrete is produced per person annually (Gagg, 2014). Traditionally, large amounts of Ordinary Portland Cement (OPC) are needed for the production of concrete. The production of OPC is very energy intensive and therefore a major generator of carbon dioxide, which is considered to be a potent greenhouse gas. High CO2-emissions are caused by calcination of limestone and combustion of fossil fuel during an energy intensive production process. To reduce the CO2-emissions caused by the concrete manufacturing, there is a growing interest in alternative and low CO2-binders. A great example of alternative binders are alkali-activated materials (AAMs), often referred to as geopolymers. Alkali-activated materials are inorganic polymers acting as the binder agent in concrete, often containing by-products from the industry such as fly ash and blast furnace slag (Davidovits, 1989). Unlike OPC, AAMs are synthesized by activation of an aluminosilicate source (fly ash, slag, metakaolin) with alkaline activators. AAMs have attracted a lot of attention in the industry because of its superior mechanical properties, excellent resistance to sulphate attack, low creep and low drying shrinkage compared to OPC. Besides sustainability, the total amount of costs is also a relevant factor for the construction industry. 3D-printing of concrete removes the need for formwork and enables the industry to create complex shapes as well as optimization of material use. It also provides an opportunity for an automated building process with a minimal amount of labour and material wastage. Combining the use of alkali-activated materials (AAMs) in an innovative and automated 3D-printing process may therefore offer many advantages for the construction industry...Civil Engineerin
