Akinori Abe
Context (situation, viewpoint etc.) plays a significant role in the (deductive) inference process found in, for instance, context based reasoning and situation theory. In these theories, if the context changes the result of an inference sometimes changes. These theories focus on concept drift according to context change while reasoning. Tracking concept drift is also a hot topic in incremental knowledge acquisition and learning [Helmbold 1994, Widmer 1996]. We are now researching methods for making inferences with incomplete knowledge, and have recently proposed Abductive Analogical Reasoning (AAR). AAR aims to solve the problem that inference with hypothetical reasoning systems like Theorist fails when the knowledge base does not have the necessary hypotheses. It also aims to solve the problem of how to make criterion in selecting the best hypotheses.
AAR employs an analogous clause searching engine using a concept base. This system considers context while searching analogous clauses. Context plays a significant role in guaranteeing the explanatory coherence of hypotheses. However, in AAR, context is not explicitly shown, so it must be retrieved when an inference is made.
Helmbold [Helmbold 1994] and Widmer [Widmer 1996] have examined the retrieval of hidden context and tracking changing concepts. Their methods involve the learning in which target concept is allowed to change over time and tracking concept drift by minimizing disagreement between successive concepts.
On the other hand, since AAR does not always consider the flow of time, it can not retrieve context from the result of previous inferences or through some measure like the Markov process and probabilistic process reported in [Helmbold 1994] and [Widmer 1996].
This page briefly a method for retrieving context by referring to clauses that can be used during inference and for generating hypotheses by considering context. It also discusses the role of context in selecting plausible abductive hypotheses.
References: