Instance Store Database Support for Reasoning over Individuals

  • Slides: 8
Download presentation
Instance Store Database Support for Reasoning over Individuals 1. S Bechhofer, I Horrocks, D

Instance Store Database Support for Reasoning over Individuals 1. S Bechhofer, I Horrocks, D Turi. Instance Store - Database Support for Reasoning over Individuals. http: //instancestore. man. ac. uk/instancestore. pdf Dumitru Roman 1

Outline • • Introduction Instance Store Implementation Comparisons & Evaluation 2

Outline • • Introduction Instance Store Implementation Comparisons & Evaluation 2

Introduction • • • Description Logic (DL): T-Box & A-Box. More expressive DL languages

Introduction • • • Description Logic (DL): T-Box & A-Box. More expressive DL languages - distinction between the T- and A-Box collapses. A full DL A-box allows us to make assertions about individuals of the form I € D and P(i, j) for individuals i and j, arbitrary concept descriptions D and properties (binary relationships) P. We can then query the A-box with queries such as: – – • Answering such queries requires full A-box reasoning, a task that is known to be difficult in the presence of expressive languages for the T-box – • Retrieval Which individuals occurring in the assertions are instances of some concept description Q? Realization Given an individual i, what are the most specific concepts C in the T-box that i is an instance of? an alternative approach – the instance store – providing a limited, yet effective form of A-box reasoning. The paper discusses implementation strategies for an instance store using a combination of relational database and DL reasoner. 3

Instance Store • well known technique for implementing an instance store - simply treat

Instance Store • well known technique for implementing an instance store - simply treat individuals as primitive concept definitions and add new primitive concepts “pseudo-individuals” for each individual. – – – Retrieval then simply involves classifying the query and returning all individuals such that their associated “pseudo-individuals” lie below the query in the concept hierarchy. Problem - with realistic knowledge bases likely to run into millions of instances, such an approach is unlikely to scale. Solution - The instances and their relationship with concepts are stored in the database, while the reasoner deals purely with T-Box functionality. Retrieving individuals is then a combination of query against that database and subsumption and classification requests to the reasoner. 4

Instance Store - Architecture 5

Instance Store - Architecture 5

Instance Store – retrieval algorithm 6

Instance Store – retrieval algorithm 6

Instance Store - Implementation • Basic Strategy – stores minimal information in the database,

Instance Store - Implementation • Basic Strategy – stores minimal information in the database, and – relies on the reasoner to provide subsumption testing and hierarchy information. • Classified Descriptions – Cache the classification of each asserted and retrieved description. 7

Comparisons & Evaluation 8

Comparisons & Evaluation 8