Semantic Web technologies and deep learning share the goal of creating intelligent artifacts that emulate human capacities such as reasoning, validating, and predicting. There are notable examples of contributions leveraging either deep neural architectures or distributed representations learned via deep neural networks in the broad area of Semantic Web technologies. Knowledge Graphs (KG) are one of the most well-known outcomes from the Semantic Web community, with wide use in web search, text classification, entity linking etc. KGs are large networks of real-world entities described in terms of their semantic types and their relationships to each other.
A challenging but paramount task for problems ranging from entity classification to entity recommendation or entity linking is that of learning features representing entities in the knowledge graph (building “knowledge graph embeddings”) that can be fed into machine learning algorithms. The feature learning process ought to be able to effectively capture the relational structure of the graph (i.e. connectivity patterns) as well as the semantics of its properties and classes, either in an unsupervised way and/or in a supervised way to optimize a downstream prediction task. In the past years, Deep Learning (DL) algorithms have been used to learn features from knowledge graphs, resulting in enhancements of the state-of-the-art in entity relatedness measures, entity recommendation systems and entity classification. DL algorithms have equally been applied to classic problems in semantic applications, such as (semi-automated) ontology learning, ontology alignment, duplicate recognition, ontology prediction, relation extraction, and semantically grounded inference.