AN INNOVATIVE SRL ALGORITHM: JOINT-BAYESIAN-NETWORK
Xiaoyuan Xu · 2010
Markov Logic Networks (MLNs) are prominent statistical relational models, but most of the current state-of-the-art algorithms for learning MLNs have focused on relatively small datasets with few descriptive attributes, where predicates are mostly binary and the main task is usually prediction of links between entities. Descriptive attributes are usually nonbinary and can be very informative, but they increase the search space of possible candidate clauses. Our project is to address the problem of learning the structure of an MLN with descriptive attributes on medium to large datasets. We propose and implement an innovative Joint-Bayesian-Network (JBN) algorithm for learning a directed relational model. My main task in this project is to produce an MLN structure via a standard moralization procedure and evaluate the performance through comparsion with a standard MLN algorithm implemented by Alchemy package. Based on the evaluation result, learning MLN structure through JBN algorithm is 200-1000 times faster and its predictive accuracy scores substantially higher than benchmark algorithms on three relational databases.