user2Vec: Social Media User Representation Based on Distributed Document Embeddings

Abstract

Recent improvements in word representations (word embeddings) have improved a wide range of text-based information retrieval applications. Successfully representing many semantic characteristics of words in low dimensional vector spaces with Continuous Bag of Words (CBOW) and Skip-Gram models invoked new techniques for textual input representations. In this study we introduce a neural embedding model for representing social media users using document representation model (doc2vec). We propose a simple method for evaluating the quality of the user vectors. We also share our results on simply averaging user vectors of the same category as category vectors. The experiment results show that our user2vec model creates semantically meaningful representations of users and it is very open for new improvements. We also share the dataset used in this study.