Dissertation

Predicting Demographics and Affect in Social Networks

Volkova, Svitlana

2015

Overview

The recent explosion of social media services like Twitter, Facebook and Google+ has led to an interest in predicting hidden information from the large amounts of freely available public content. As compared to the earlier explosion of documents arising from the web, social media content is significantly more personalized – written in the first person, informal, and often revealing of latent attributes of users. The task of inferring latent user properties from social media data has become known as user modeling, personal analytics or user profiling task. Previous approaches treated the task of user attribute prediction as static super- vised classification, applied textual features extracted from user tweets and relied on an unrealistic amount of content per user (thousands of tweets). This dissertation relies mainly on Twitter data and focuses on several important but previously unex- plored aspects of the task of user attribute prediction: (1) developing novel models and practical techniques that reflect the dynamic streaming nature of social media; (2) studying predictive power and latent relationships between user demographics, emotions and interests in social media; and (3) showing that extra-linguistic features such as user demographics, personality and emotions can improve a variety of downstream applications, e.g., sentiment analysis and attribute-affect specific language modeling. (Abstract shortened by ProQuest.)

• Publisher: ProQuest Dissertations & Theses
• Subject: Artificial intelligence / Computer science / Information science
• ISBN: 1369476418, 9781369476415