The need of structured data: Introducing the OKgraph project

Although many computational problems can be approached using Deep Learning, in this position paper we argue that in the case of Information Retrieval tasks this is not mandatory and even detrimental whenever alternatives exist. Instead of learning (by training) how to solve the full problem, we suggest to split it into two sub-problems: A) producing structured data (specifically knowledge graphs) out of the corpora, and b) providing usable tools (including natural language) to querying such structured data. Motivated by this two-step approach and its need of structured data, we introduce the Open Knowledge Graph (OKgraph) project, an initiative recently funded by Regione Autonoma della Sardegna aiming at providing insights on the first part of the problem: A general way of generating knowledge graphs from text corpora, unsupervisedly.