Attr2vec: A Neural Network Based Item Embedding Method

Abstract

Mathematically modeling discrete objects to represent them in the form of low-dimensional vectors, which enable math calculations between discrete objects, is a key component of many computer systems. In recent years, Word2vec model has been proposed in natural language processing, and the low-dimensional vector embedding of words has achieved excellent results. In this paper, we extend the embedding model of word vector and add the representation method of object attributes so that the model can be applied to general objects. Besides, this paper verifies the validity of this model on a large-scale data set, indicating that it is superior to the traditional method SVD.