Synergy-incorporated Bayesian Petri Net: A method for mining “AND/OR” relation and synergy effect with application in probabilistic reasoning

Bayesian networks (BNs) are widely used for knowledge representation and reasoning. However, they suffer from the following limitations: 1) They are unable to explicitly learn “AND” relations and synergy effects from data; 2) They do not depict “AND” relations among causes directly, which sometimes leads to the high complexity of constructing conditional probability tables (CPTs); and 3) During their knowledge representation and reasoning, they fail to express any synergy (catalytic or inhibitory) effect that a non-causal variable may have on a reasoning rule. To address the mentioned issues, this study proposes a method for mining and modelling “AND/OR” relations and synergy effects called Synergy-incorporated Bayesian Petri Net (SBPN). The method integrates a BN with Petri net concepts. It can directly learn “AND/OR” relations and synergy effects from data, harnessing the power of Petri nets to model “AND/OR” relations in BNs, resulting in a compact CPT. Synergy effects are easily depicted via the SBPN, and their influence on the reasoning procedure can be formally expressed using the SBPN. Experimental results on open datasets demonstrate the effectiveness and advantages of the proposed method for knowledge mining and probabilistic reasoning.