|Title||:||Query Relaxation and Ranking for Extended Knowledge-bases|
|Speaker||:||Maya Ramanath (IIT Delhi)|
|Details||:||Fri, 3 Feb, 2017 12:00 PM @ BSB 361|
|Abstract:||:||Large-scale knowledge-bases (KBs) store information in a structured manner and facilitate querying by users who have well-defined and precise information needs. However, knowing SPARQL, the declarative query language for KBs, does not guarantee a good user experience.
First, if the number of results are large there is a possibility of the user being overwhelmed with uninteresting information. Therefore, there is a need to rank the results returned by the KB. We have developed a framework for result ranking based on language models (LMs). I will describe this ranking model, which is one of the first to use LMs for structured data.
Second, in order to formulate a meaningful query, the user must be familiar with both the terminology of the KB as well as the structure of the information stored in the KB. Incomplete knowledge of either of these could result in query formulations which throw out errors or return empty answers. We have developed a framework to automatically relax the user-submitted query. I will describe this framework and how it integrates into the ranking model above.
Finally, flexibility is lost if query formulation is restricted to just the vocabulary of the KB. Instead, if the vocabulary can be expanded to include Open IE style relationships (that is, textual phrases denoting entities and relationships instead of their canonical equivalent), then more meaningful relaxations are possible. I will describe an extension of our relaxation framework which considers Open IE style relationships as a part of an extended knowledge-base.
I will conclude with some work in progress and open problems in this domain.
Bio: Maya Ramanath is an assistant professor in the Department of Computer Science and Engg. at IIT Delhi and is a member of the Data Analytics and Intelligence and Research in the department. She completed her PhD in IISc and subsequently held a post-doctoral researcher position at the Max-Planck Institute for Informatics in Saarbruecken, Germany. Her research interests lie in semantic web data management and in particular, making it easy for casual users to make effective use of the semantic web.