Pitt Logo LinkContact Us

RELATIONSHIP ANALYSIS OF IMAGE DESCRIPTIONS: AN ONTOLOGICAL, CONTENT ANALYTIC APPROACH

Benson, Allen C (2011) RELATIONSHIP ANALYSIS OF IMAGE DESCRIPTIONS: AN ONTOLOGICAL, CONTENT ANALYTIC APPROACH. Doctoral Dissertation, University of Pittsburgh.

[img]
Preview
PDF - Primary Text
Download (4Mb) | Preview

    Abstract

    The relationships humans express when describing images have powerful, but poorly understood, effects on how visual information is represented, structured, and processed in information systems. This study evaluates the benefits and difficulties of using content analysis and ontological analysis as predictors of relationship instances and types occurring in image descriptions. A random sample of 36 documented reference transactions from the administrative files of the Pittsburgh Photographic Library is analyzed in light of three describing contexts: image searcher, curator, and cataloger. Through the qualitative and quantitative assessment of image descriptions, the research leads to several key findings and contributions. The most important findings vindicate the claim that recognition, capture, and classification of relationship instances can be empirically grounded utilizing content analysis and ontological tools and methods. Evidence comes in successfully ascertaining and capturing in a Corpus the existence of 1,655 relationship instances. Further, the analysis finds evidence of relationship types and subtypes of relationships whose members share certain recognizable properties in common. The study also brings useful, new insights to the capture of background information surrounding events using situation-templates, introduces methods for formulating case relations and image attributes as binary predicates, and it offers a new, finer-grained definition of relationship. Contributions of this study include a corpus of relationship instances, an ontology of relationship types, and a methodological framework that provides significantly better results than earlier studies in the prediction of relationships (the architecture of which is depicted in Figure 22 on page 102). There are a number of ways this research could be extended and corroborated. First, event analysis ought to be tied to a system of semantic frame analysis. Second, test the content analysis form against other texts, which will result in elaboration of the core ontology of relationship types. Third, expand image description analysis beyond the current domain to include image description in visual ethnography, art history and criticism, and photography practices. Fourth, test how inference engines reason over relationships in knowledge-based environments. Finally, to aid in the analysis of the meanings of relationships, more work is needed in formalizing the ontological concepts used in image descriptions.


    Share

    Citation/Export:
    Social Networking:

    Details

    Item Type: University of Pittsburgh ETD
    ETD Committee:
    ETD Committee TypeCommittee MemberEmailORCID
    Committee ChairTomer, Christingerctomer@sis.pitt.edu
    Committee MemberHe, Daqingdaqing@sis.pitt.edu
    Committee MemberBowker, Geoffrey Cgbowker@sis.pitt.edu
    Committee MemberFahlman, Scott Esef@cs.cmu.edu
    Title: RELATIONSHIP ANALYSIS OF IMAGE DESCRIPTIONS: AN ONTOLOGICAL, CONTENT ANALYTIC APPROACH
    Status: Unpublished
    Abstract: The relationships humans express when describing images have powerful, but poorly understood, effects on how visual information is represented, structured, and processed in information systems. This study evaluates the benefits and difficulties of using content analysis and ontological analysis as predictors of relationship instances and types occurring in image descriptions. A random sample of 36 documented reference transactions from the administrative files of the Pittsburgh Photographic Library is analyzed in light of three describing contexts: image searcher, curator, and cataloger. Through the qualitative and quantitative assessment of image descriptions, the research leads to several key findings and contributions. The most important findings vindicate the claim that recognition, capture, and classification of relationship instances can be empirically grounded utilizing content analysis and ontological tools and methods. Evidence comes in successfully ascertaining and capturing in a Corpus the existence of 1,655 relationship instances. Further, the analysis finds evidence of relationship types and subtypes of relationships whose members share certain recognizable properties in common. The study also brings useful, new insights to the capture of background information surrounding events using situation-templates, introduces methods for formulating case relations and image attributes as binary predicates, and it offers a new, finer-grained definition of relationship. Contributions of this study include a corpus of relationship instances, an ontology of relationship types, and a methodological framework that provides significantly better results than earlier studies in the prediction of relationships (the architecture of which is depicted in Figure 22 on page 102). There are a number of ways this research could be extended and corroborated. First, event analysis ought to be tied to a system of semantic frame analysis. Second, test the content analysis form against other texts, which will result in elaboration of the core ontology of relationship types. Third, expand image description analysis beyond the current domain to include image description in visual ethnography, art history and criticism, and photography practices. Fourth, test how inference engines reason over relationships in knowledge-based environments. Finally, to aid in the analysis of the meanings of relationships, more work is needed in formalizing the ontological concepts used in image descriptions.
    Date: 25 July 2011
    Date Type: Completion
    Defense Date: 23 June 2011
    Approval Date: 25 July 2011
    Submission Date: 21 July 2011
    Access Restriction: No restriction; The work is available for access worldwide immediately.
    Patent pending: No
    Institution: University of Pittsburgh
    Thesis Type: Doctoral Dissertation
    Refereed: Yes
    Degree: PhD - Doctor of Philosophy
    URN: etd-07212011-193416
    Uncontrolled Keywords: construction grammar; content analysis; image; image description; knowledge base; natural language processing; natural language understanding; ontological analysis; ontology; photograph; photograph description; relationships; Scone
    Schools and Programs: School of Information Sciences > Library and Information Science
    Date Deposited: 10 Nov 2011 14:52
    Last Modified: 22 Jun 2012 10:46
    Other ID: http://etd.library.pitt.edu/ETD/available/etd-07212011-193416/, etd-07212011-193416

    Actions (login required)

    View Item

    Document Downloads