Social media-driven news personalization

Abstract

While social media have achieved significant and widespread adoption as platforms for sharing information, their use as a source of data for predicting user interests has not yet been fully explored. In this paper, we present a content-based approach to modeling user interests based on Twitter. Our recommendation system uses information retrieval techniques to represent tweets and users as collections of news topics, including high-level categories (e.g., sports, politics, business) and detailed subtopics (e.g., Chicago Bulls, Mitt Romney, entrepreneurship). We discuss the design of a system that uses this information to deliver news recommendations in the form of a personalized newspaper. Finally, we describe a novel method for evaluating recommendation systems based on Twitter that involves mining Twitter data to identify explicit indicators of news interests and comparing these to retroactive system recommendations.