Relation Extraction and its Application to Question Answering

  • Author / Creator
    Xu, Ying
  • Information extraction, extracting structured information from text, is a vital component for many natural language tasks such as question answering. It generally consists of two components: (1) named entity recognition (NER), identifying noun phrases that are names of organizations, persons, or countries; and (2) relation extraction, extracting relations between entities. In this dissertation, we assume the entities are given, and concentrate on the relation extraction task. Traditional relation extraction task seeks to confirm a predefined set of relations in a text, such as the employment or family relation. These systems are difficult to extend by including additional relations. In contrast, the open information extraction (Open IE) task attempts to extract all relations, using words in sentences to represent the relations. My dissertation focuses on Open IE. We first proposed a tree kernel based-Open IE system that achieved state of the art performance. One advantage of the tree kernel model is that it exploits information in syntactic parse trees without feature engineering. After observing the importance of words in relation extraction, we then incorporated word embeddings into the tree kernel and improved the system’s performance. However, previous systems have not considered implicit relations, i.e., relations implied in noun phrase structures such as Germany’s people, Google Images, and Shakespeare’s book. We call this type of structure nested named entities. To study the implicit relation phenomenon, we automatically extracted thousands of instances of training data from Wikipedia. We demonstrate the feasibility of recovering implicit relations with a supervised classification model. Our data and model provides a baseline for future work on this task. Last but not least, to show the effect of our relation extraction systems, we built an Open IE-based question answering system and achieved promising results. Our analysis indicates the weakness of the current Open IE systems, the role of our information extraction results, and gives directions for improvement.

  • Subjects / Keywords
  • Graduation date
    Spring 2017
  • Type of Item
  • Degree
    Doctor of Philosophy
  • DOI
  • License
    This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.
  • Language
  • Institution
    University of Alberta
  • Degree level
  • Department
  • Supervisor / co-supervisor and their department(s)
  • Examining committee members and their departments
    • Schuurmans, Dale (Computing Science)
    • Ringlstetter, Christoph (Linguistic)
    • Sun, Maosong ( Computer Science)
    • Barbosa, Denilson (Computing Science)
    • Kondrak, Grzegorz (Computing Science)