Title: From Graph Snippets to Summaries : Can We Read Knowledge Graphs?
Speaker: Srikanta Bedathur, IIIT Delhi
Abstract:
With the growth in size and coverage of automatically constructed
knowledge graphs, they are increasingly being used to derive deeper
understanding of how real-world entities are connected with each other.
One such application that has gathered lot of steam is that of
relationship mining -- identify long-range, multi-faceted relationships
between a pair or a group of entities. Typical solutions generate a
snippet of graph that ``best'' connects the queried entities in the
knowledge graph.
However, the non-trivial challenge of actually interpreting these
results rests with the user. Due to the lack of context (time, text and
space) of the presented relationships, and the size of resulting
snippets, users often struggle to make sense of these results. In order
to overcome this, we are working towards automatically generating an
interpretable presentation of these results in the form of text
summaries. In this talk, I plan outline our recent work and further
challenges in this direction.
Relevant paper:
Towards Generating Text Summaries for Entity Chains. Shruti Chhabra and Srikanta Bedathur. ECIR 2014.