Research Problems and Directions in Semantic Web


Winter 2019
Instructor: Raghava Mutharaju
IIIT-Delhi
IIIT Delhi

Ontology Modelling

  • Automatically building ontologies with complex relations
    • Given a dataset (text, tables, excel sheets), competency questions, and list of ODPs, can we (auto/semi-automatically) build ontologies that can answer the competency questions?
    • This is called as Ontology Learning
  • Automatic refactoring of ontologies with ODPs
    • Rebuild large (bulky) ontologies with ODPs
    • Can we make the bulky ontologies more modular by replacing pieces of the ontology with ODPs?
    • The "intent" or the behaviour of the modular ontology should be the same as the original ontology
    • The modular ontology should answer the same competency questions as the original one
  • Extracting ODPs from large ontologies
    • There are several large ontologies that may have similar patterns and can be extracted to form an ODP
    • Ontologies have to be analyzed and common structures have to be identified

Description Logic Reasoning


  • Approximate Reasoning
    • Reasoning over highly expressive and/or very large ontologies is time consuming
    • Approximate Reasoning is about sacrificing either soundness or completeness for better reasoning runtime

Ontology Alignment


  • Given two ontologies, what are the different entities that are similar to each other?
    • Concepts
    • Properties
    • Instances
  • Ontology Alignment Evaluation Initiative (OAEI) provides benchmarks for alignment

Knowledge Graph Construction


  • Extraction of triples (instance data) from text
  • There are some popular automatically constructed KGs
    • DBpedia
    • Never Ending Learning (NELL)
    • YAGO
    • Knowledge Graphs from Google, LinkedIn, Amazon, eBay
  • Open Information Extraction tools such as OpenIE, ClausIE, FRED, Stanford OpenIE etc. can be used for getting triples from text

Knowledge Graph Embedding


  • Entities and relations in a KG are represented as vectors
  • This enables us to perform vector operations on KGs
  • Similar to word embedding
  • KG embedding is useful in a variety of applications
    • Knowledge Graph Completion
    • Question Answering

Question Answering using Knowledge Graphs


Semantic Web and IoT


Making Semantic Technologies user friendly


References


  1. Collected Research Questions Concerning Ontology Design Patterns. Karl Hammar et. al. Ontology Engineering with Ontology Design Patterns. 2016