Flexible Similarity Search of Semantic Vectors Using Fulltext Search Engines

This publication doesn't include Faculty of Arts. It includes Faculty of Informatics. Official publication website can be found on muni.cz.



Type Article in Proceedings
Conference CEUR Workshop Proceedings, Vol. 1923
MU Faculty or unit

Faculty of Informatics

Field Informatics
Keywords vector space modelling; semantic vectors encodings; inverted-index; systems performance; document representations; Latent Semantic Analysis; doc2vec; GloVe; Elasticsearch; evaluation; performance optimization
Description Vector representations and vector space modeling (VSM) play a central role in modern machine learning. In our recent research we proposed a novel approach to ‘vector similarity searching’ over dense semantic vector representations. This approach can be deployed on top of traditional inverted-index-based fulltext engines, taking advantage of their robustness, stability, scalability and ubiquity. In this paper we validate our method using varied datasets ranging from text representations and embeddings (LSA, doc2vec, GloVe) to SIFT descriptors of image data. We show how our approach handles the indexing and querying in these domains, building a fast and scalable vector database with a tunable trade-off between vector search performance and quality, backed by a standard fulltext engine such as Elasticsearch.
Related projects: