Learning Bayes Nets for Relational Data and Heterogenous Networks

点击翻译按钮获取中文摘要

Vast amounts of data are gathered daily for different purposes. Many, if not most, new datasets contain information about networks of linked entities. Sources of link data include social media, such as Twitter, social networks, computer networks such as the Internet, biological networks, and the relational databases that most organizations use to maintain their data. A crucial competitive edge comes with the ability to efficiently find informative patterns in network data and to learn from them to improve decision-making under uncertainty. The aim of this project is to extend and enhance computational tools from Machine Learning and Artificial Intelligence so that they can be applied to network data. Key statistical questions include the following. (1) Are there connections between different types of links? For example, if a user searches for a webpage that covers a topic, are they likely to search for a video that covers the same topic? (2) Which individuals are especially important in a network? For example, do some Twitter accounts influence their followers more than others? The ranking website Klout.com uses data from 7 networks such as Twitter, Google+, Facebook to rank individuals. (Out of a maximum influence score of 100, Justin Bieber scores 92 and Barack Obama 99). Klout.com produces scores for 100+ Million people and reports 30 Billion monthly calls to their API. (3) Can we explain the existence of links by finding similarities among individuals? Answers to these basic questions have applications in predicting links, predicting ratings, making recommendations, marketing, web search, and other areas, including major commercial applications. Rating and recommendation systems are used by many companies to recommend products to customers, such as Amazon and Netflix.