“…In 2021, we tackled these issues by introducing the information extraction approach described in Dessì et al [18], which is able to combine information from different tools according to a domain ontology and produce large-scale KGs. This approach inspired several further works in the field [19,20,21,22,23,24] and was used to produce the Artificial Intelligence Knowledge Graph (AI-KG) [25], a knowledge base describing 820K research entities in the field of AI. However, this first attempt also suffered from several limitations, such as: i) the entity extraction modules did not take advantage of the expert knowledge acquired from the analysis of the resulting knowledge graphs, ii) a limited ability to merge together multiple versions of the same entity (e.g., data cleaning algorithm, data cleaning automation, and preset data cleaning strategy), iii) a shallow and manual method for mapping verbal predicates to semantic relations, and iv) a limited methodology for assessing the validity of a triple, based on a very simple multilayer perceptron classifier.…”