Refinement-based OWL class induction with convex measures

David Ratcliffe*, Kerry Taylor

*Corresponding author for this work

    Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

    5 Citations (Scopus)

    Abstract

    Beam-search may be used to iteratively explore and evaluate refinements of candidate hypotheses expressed in logical formalisms such as description logic. In this paper, we analyse heuristics for beam search methods over OWL classes and present a novel search algorithm, OWL-Miner which leverages the properties of convex measure functions to deliver an improved memory-bounded beam search. We present performance results on the mutagenesis benchmark that demonstrate world-best 10-fold cross-validated accuracy. Our improvements to the space and time-based efficiency of refinement-based learning algorithms are significant for expanding the size of learning problems that can be feasibly addressed by refinement learning and the quality of solutions that can be found with limited resources.

    Original languageEnglish
    Title of host publicationSemantic Technology - 7th Joint International Conference, JIST 2017, Proceedings
    EditorsZhe Wang, Kewen Wang, Anni-Yasmin Turhan, Xiaowang Zhang
    PublisherSpringer Verlag
    Pages49-65
    Number of pages17
    ISBN (Print)9783319706818
    DOIs
    Publication statusPublished - 2017
    Event7th Joint International Conference on Semantic Technology, JIST 2017 - Gold Coast, Australia
    Duration: 10 Nov 201712 Nov 2017

    Publication series

    NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
    Volume10675 LNCS
    ISSN (Print)0302-9743
    ISSN (Electronic)1611-3349

    Conference

    Conference7th Joint International Conference on Semantic Technology, JIST 2017
    Country/TerritoryAustralia
    CityGold Coast
    Period10/11/1712/11/17

    Fingerprint

    Dive into the research topics of 'Refinement-based OWL class induction with convex measures'. Together they form a unique fingerprint.

    Cite this