PhD Defense: Cross-Platform Question Answering in Social Networking Services

Talk
Mossaab Bagdouri
Time: 
05.16.2017 09:30 to 11:30
Location: 
AVW 1146

The last two decades have made the Internet a major source for knowledge seeking. Several platforms have been developed to find answers to one’s questions such as search engines and online encyclopedias. The wide adoption of social networking services has pushed the possibilities even further by giving people the opportunity to stimulate the generation of answers that are not already present on the Internet. Some of these social media services are primarily community question answering (CQA) sites, while the others have a more general audience but can also be used to ask and answer questions.

The choice of a particular platform (e.g., a CQA site, a microblogging service, or a search engine) by some user depends on several factors such as awareness of available resources and expectations from different platforms, and thus will sometimes be suboptimal. Hence, we introduce cross-platform question answering, a framework that aims to improve our ability to satisfy complex information needs by returning answers from different platforms, including those where the question has not been originally asked.

We propose to build this core capability by defining a general architecture for designing and implementing real-time services for answering naturally occurring questions. This architecture consists of four key components: (1) real-time detection of questions, (2) a set of platforms from which answers can be returned, (3) question processing by the selected answering systems, which optionally involves question transformation when questions are answered by services that enforce differing conventions from the original source, and (4) answer presentation, including ranking, merging, and deciding whether to return the answer.

We demonstrate the feasibility of this general architecture by instantiating a restricted development version in which we collect the questions from one CQA website, one microblogging service or directly from the asker, and find answers from among some subset of those CQA and microblogging services. To enable the integration of new answering platforms in our architecture, we introduce a framework for automatic evaluation of their effectiveness.

Examining Committee:

Chair: Dr. Douglas W. Oard

Dean’s rep: Dr. Philip Resnik

Members: Dr. Héctor Corrada Bravo

Dr. Hal Daumé III

Dr. David D. Lewis