A model for building probabilistic knowledge-based systems using divergence distances. (15th July 2021)

• Record Type: Journal Article
• Authors: Nguyen, Van Tham; Tran, Trong Hieu; Nguyen, Ngoc Thanh
• Is Part Of: Expert systems with applications. Volume 174(2021)
• Publication Date: 2021-07-15
• Subjects: Probabilistic knowledge base -- Divergence distance function -- Merging algorithm -- Probabilistic knowledge-based system; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2020.114494
• Languages: English
• ISSNs: 0957-4174

Abstract

Graphical abstract: Highlights: Propose an architecture of a probabilistic KBS that meets the need of the consistency assurance and the compliance with probability rules. Build a mathematical model for the merging problem that takes into account inconsistency levels, the structural diversity of probabilistic knowledge bases, and ensure the consistency for the joint probabilistic knowledge base. Prove the reliability of the proposed merging model in both the theoretical and experimental aspect. Investigate a family of merging operators with a large range of divergence distance functions between probability distributions. Abstract: The knowledge-based systems (KBSs) in general and solving the knowledge merging problem in particular have seen a great surge of research activity in recent years. However, there still exist two main shortcomings that need to be addressed in the probabilistic framework. Firstly, the current methods only deal with the problems in which original probabilistic knowledge bases (PKBs) are required to be consistent and formed in the same structure. It is a very strong requirement and difficult to apply in practice. Secondly, only a few measures of distance between probability distributions have been studied to apply in existing models. To overcome these disadvantages, in this paper, we introduce a novel framework for merging PKBs. To this end, a theoretical model is introduced and several experiments are implemented. In theoretical model, some theorems are pointed out and proved to provide mathematical background to construct the merging model. Moreover, a deep survey on how to employ divergence distance functions (DDFs) between probability distributions to carry out the merging process are performed. In experimental aspect, a consistency recovery algorithm and some merging algorithms based on DDFs are proposed. Through the results of conducted experiments, issues about the time cost of merging process, the number of iterations, and CPU Time Elapsed parameter to solve the class of optimization problems in the merging process are analyzed, compared, and evaluated.