Entity-centric summarization: generating text summaries for graph snippets

In recent times, focus of information retrieval community has shifted from traditional keyword-based retrieval to techniques utilizing the semantics in the text. Since such techniques require the understanding of relationships between entities, efforts are ongoing to organize the Web into large entity-relationship graphs. These graphs can be leveraged to answer complex relationship queries. However, most of the research has focused upon extracting structural information between entities such as a path, Steiner tree, or subgraphs. Little attention has been paid to the comprehension of these structural results, which is necessary for the user to understand relationships encapsulated in these structures. In this doctoral proposal, we pursue the idea of entity-centric summarization and propose a novel framework to produce entity-centric summaries which describe the relationships among input entities. We discuss the inherent challenges associated with each module in the framework and present an evaluation plan. Results from our preliminary experiments are encouraging and substantiate the feasibility of summarization problem.