Accommodating New Relations for e-Business Text Mining Applications

We investigate an approach of accommodating new logic relations, represented in Markov Logic Networks (MLN), from a source domain to a target domain automatically. One characteristic of this problem setting is that logic relations, which are expressed as logic formulae, previously prepared for the source domain dealing with a text mining application are not sufficient for the target domain. Therefore, new logic formulae for the target domain are automatically discovered by capturing the core logic relations for both domains and the candidate relations in the target domain using the unlabeled data in the target domain. Experimental results demonstrate that the new formulae discovered improve the performance on the target domain.