Word Embeddings for User Profiling in Online Social Networks

User profiling in social networks can besignificantly augmented by using available full-text itemssuch as posts or statuses and ratings (in the form oflikes) that users give them. In this work, we applymodern natural language processing techniques basedon word embeddings to several problems related touser profiling in social networks. First, we present anapproach to create user profiles that measure a user’sinterest in various topics mined from the full texts of theitems. As a result, we get a user profile that can be used,e.g., for cold start recommendations for items, targetedadvertisement, and other purposes; our experimentsshow that the interests mining method performs on alevel comparable with collaborative algorithms while atthe same time being a cold start approach, i.e., itdoes not use the likes of an item being recommended.Second, we study the problem of predicting a user’sdemographic attributes such as age and gender basedon his or her full-text items. We evaluate theefficiency of various age prediction algorithms based onword2vec word embeddings and conduct an extensiveexperimental evaluation, comparing these algorithmswith each other and with classical baseline approaches.