An analysis of word embedding spaces and regularities

Word embeddings are widely use in several applications due to their ability to capture semantic relationships between words as relations between vectors in high dimensional spaces. One of the main problems to obtain the information is to deal with the phenomena known as the Curse of Dimensionality,...

Descripción completa

Detalles Bibliográficos
Autor: Gijón Agudo, Manuel
Tipo de recurso: tesis de maestría
Fecha de publicación:2019
País:España
Institución:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2117/166266
Acceso en línea:https://hdl.handle.net/2117/166266
Access Level:acceso abierto
Palabra clave:Artificial intelligence
Word embeddings
Embedding space
Distances
Semantic relations
WordNet
Intel·ligència artificial
Classificació AMS::68 Computer science::68T Artificial intelligence
Àrees temàtiques de la UPC::Informàtica::Intel·ligència artificial
Descripción
Sumario:Word embeddings are widely use in several applications due to their ability to capture semantic relationships between words as relations between vectors in high dimensional spaces. One of the main problems to obtain the information is to deal with the phenomena known as the Curse of Dimensionality, the fact that some intuitive results for well known distances are not valid in high dimensional contexts. In this thesis we explore the problem to distinguish between synonyms or antonyms pairs of words and non-related pairs of words attending just to the distance between the words of the pair. We considerer several norms and explore the problem in the two principal kinds of embeddings, GloVe and Word2Vec.