Learning Latent Vector Spaces for Product Search

• DOI: 10.1145/2983323.2983702
• Publication Date: 2016-10-26T13:20:08Z
• Authors: Christophe Van Gysel; Maarten de Rijke; Evangelos Kanoulas
• Keywords: Word2vec; Discriminative model; Feature vector; Rank (graph theory); Latent semantic analysis; Vector space model

We introduce a novel latent vector space model that jointly learns the representations of words, e-commerce products and mapping between two without need for explicit annotations. The power lies in its ability to directly discriminative relation particular word. compare our method existing models (LSI, LDA word2vec) evaluate it as feature learning rank setting. Our achieves enhanced performance better product representations. Furthermore, from words benefit errors propagated back during parameter estimation. provide an in-depth analysis analyze structure learned