Question Answering and Query Processing for Extended Knowledge Graphs

schemaless nature that enables them to grow seamlessly, allowing for new 
relationships and entities as needed. With this rapid growth, several issues 
arise: (i) how to allow users to query knowledge graphs in an expressive, yet 
user-friendly, manner, which shields them from all the underlying complexity, 
(ii) how, given a structured query, to return satisfactory answers to the user 
despite possible mismatches between the query vocabulary and structure and the 
knowledge graph, and (iii) how to automatically acquire new knowledge, which can 
be fed into a knowledge graph. In this dissertation, we make the following 
contributions to address the above issues:

  – We present DEANNA, a framework for question answering over knowledge
    graphs, allowing users to easily express complex information needs using
    natural language and obtain tuples of entities as answers thereby taking
    advantage of the structure in the knowledge graph.
  – We introduce TriniT, a framework that compensates for unsatisfactory results
    for structured queries over knowledge graphs either due to mismatches with 
    the knowledge graph or the knowledge graph's inevitable incompleteness. 
    TriniT tackles the two issues by extending the knowledge graph using 
    information extraction over textual corpora, and supporting query relaxation 
    where a user's query is rewritten in a manner transparent to the user to 
    compensate for any mismatches with the data.
  – We present ReNoun, an open information extraction framework for extracting 
    binary relations mediated by noun phrases and their instances from text. Our 
    scheme extends the state of the art in open information extraction which has 
    thus far focused on relations mediated by verbs.

Our experimental evaluations of each of the above contributions demonstrate the 
effectiveness of our methods in comparison to state-of-the-art approaches.