Summarizing conceptual descriptions using knowledge representations

Summarizing a body of information is a complex task which mainly depends on the ability to distinguish important information and to condense notions through abstraction. Considering a knowledge representation partially ordering concepts into a directed acyclic graph, this study focuses on the problem of summarizing several human descriptions expressed through sets of concepts. We formally define the problem of summarization in this context and we propose a model mimicking a Human-like Intelligence for scoring alternatives with regard to a specific objective. Several interesting theoretical results related to this problem (e.g. for optimization) are also given. Finally, the evaluation of the proposed approach performed in the domain of odor analysis highlights the benefits of our proposal and shows how it could be used to automatize time-consuming expert summarizing processes. Source code implementing the proposed approach as well as datasets are made available to the community.1