Contatto di riferimento: Aldo Gangemi
A number of studies have been conducted for measuring the semantic similarity between two entities (i.e., words, sentence, documents, graphs) which can further be used for broader and more practical applications Question Answering, Chatbots, Plagiarism Detection, Document Clustering etc… One of the most well known technique for measuring such kind of similarities belongs to the family of methods focused on generating Vector Space Models starting from textual data or Knowledge Graphs. This talk gives a broad know-how of the classical methods for generating vector spaces and an interesting and efficient evolution over these classical methods using Neural Network Models i.e., Word2Vec. Further advancements for dealing with Knowledge Graphs represented as RDF using a combination of graph mining methods and Word2Vec will also be discussed called as RDF2Vec. Several linguistic resources such as BabelNet and FrameNet use these methods for generating embeddings on the sense level (SEW-Embed) and event level (Frame2Vec). These embeddings are proved to perform well for specific applications such as word similarities, frame similarities, Knowledge Reconciliation etc..