New multi-relational naive Bayesian classifier
Yiqing Qin · Systems engineering and electronics · 2008
To avoid the statistical bias existing in the present multi-relational naive Bayesian classifiers,a new multi-relational naive Bayesian classifier named nMRNBC is proposed.First,the definition of semantic relationship graph is extended.Then,a new counting method towards relational individual is presented.Finally,the corresponding naive Bayesian formula is constructed.To achieve high efficiency,the tuple ID propagation method is adopted.To get better accuracy,the attribute filter criterion based on mutual information is used.Experiments show that the new classifier can get good performance.