6533b85bfe1ef96bd12bbe39

RESEARCH PRODUCT

Handling Context-Sensitive Temporal Knowledge from Multiple Differently Ranked Sources

subject

description

In this paper we develop one way to represent and reason with temporal relations in the context of multiple experts. Every relation between temporal intervals consists of four endpoints’ relations. It is supposed that the context we know is the value of every expert competence concerning every endpoint relation. Thus the context for an interval temporal relation is one kind of compound expert’s rank, which has four components appropriate to every interval endpoints’ relation. Context is being updated after every new opinion is being added to the previous opinions about certain temporal relation. The context of a temporal relation collects all support given by different experts to all components of this relation. The main goal of this paper is to develop tools, which provide formal support for the following manipulations with temporal context: how to derive temporal relation interpreted in context of multiple knowledge sources and how to derive decontextualized value for every temporal relation. Decontextualization of a temporal relation in this paper means obtaining the most appropriate value from the set of Allen’s interval relations, which in the best possible way describes the most supported opinion of all involved experts concerning this relation. We discuss two techniques to obtain the decontextualized value. First one (the minimax technique) takes into account only the worst divergences when calculates the distances between temporal relations. The second one (the weighted mean technique) takes into account all divergences. The modified technique is also considered that takes into account the probability distribution of point relations within the Allen’s set and recalculates the appropriate expert support in accordance with “meta-weights” before making decontextualization.