Undergrad Research Project - HCII Epistemo project

Spring 2017

Irene Lin
Jeffrey Bigham
Project description

Conversational assistants have the potential to help people find and make sense of personal information like email as they go about their everyday lives. Support for conversational search is still however limited, both by the brittleness of automated approaches and by privacy concerns introduced when considering private repositories like email. The Epistemo project in HCII department is a conversational search system that enables users to search, extract and manage information from emails in a privacy-preserving way by having crowds write programs on-the-fly to help the automated system work better.

When a user asks a question, Epistemo follows a three step process to get the answer. First, users question is processed into a query by crowd workers and is used to search the emails. Second, an information extractor is trained based on example answers from crowd workers. Third, the extractor is applied and the crowd workers help rank and filter through search results. Through all the steps, crowd workers never have direct access to emails.

My objective in the project is to enable Epistemo’s approach to work across a broader range of questions. Other objectives include enabling Epistemo to automatically acquire additional relevant information from search tasks and design conversational interactions around such information.

Return to project list