Modeling Time, Space, and Networks for Effective and Efficient Information Retrieval

点击翻译按钮获取中文摘要

Citizens of modern societies are inundated with increasing quantities of information with no relief in sight. Search remains one of the best solutions today for combating this deluge. In today's mobile world, search systems need to understand time, space, and networks as they pertain to user needs: for example, we demand the most up-to-date facts, prefer nearby locations, and ask friends for help. Similarly, search systems need to model these important aspects of context: documents are created at specific times and locations, often with well-defined relationships to other documents (e.g., hyperlinks on the web, reply to a tweet). My research program lies at the intersection of information retrieval and big data, and aims to build tools that help users find relevant information in large text collections while tackling fundamental challenges in computer and information science. My focus is to explore the rich interactions between time, space, and networks in building search systems that are both effective (i.e., return high quality results) and efficient (i.e., exhibit low query latencies, small memory footprints, etc.). I propose an integrated approach that will make contributions to efficient and scalable indexes for querying documents that have temporal, spatial, and network metadata; rich contextual models of relevance for document ranking and predictive user modeling; and cross-platform interfaces that support real-world information seeking. These capabilities will be demonstrated in two main applications: (1) Real-time information filtering on social media streams - a journalist might be interested in "tracking" a stream of social media posts for updates about a particular topic, e.g., the European refugee crisis. (2) Scholarly analysis of web archives - web archives contain temporal snapshots of web content and offer valuable resources for digital historians studying the recent past or social scientists exploring how political debates evolve. The impact of this research program is many-fold: A better understanding of the roles that time, space, and networks play in search will inform the design of future search engines for complex tasks. With substantial previous experience at Twitter building production "big data" analytics infrastructure and user-facing products, I have a strong record of bridging the divide between academia and industry; these experiences will assist in successful technology transfer. My former students and post-doctoral researchers possess highly-coveted skills, knowledge, and experience in working with big data, and my research program will continue training valuable contributors to modern information-centric economies in Canada and around the world.