The Research Of Cross-lingual Entity Matching Based On Markov Logic Networks

Cross-lingual entity matching is defined as a task of finding linkagebetween entities of different languages referring to a same entity or object.Cross-lingual entity matching not only extends the sharing of knowledge indifferent languages, but also produces direct contribution to many criticalregions, such as cross language information retrieval and machine learning. Ithas been the core of the research of linked data.Two major challenges in entity matching problem is handling complexityand uncertainty, especially in cross language circumstances. Not only for theentity matching problem, but also for the entire data mining region, how tobetter integrate these two problems in a unified model has become a core spot.Fortunately, the Markov logic networks model proposed by Richardson andDomingos in2006caters to our requirements.Markov logic networks combine first order predicate logic andprobabilistic graphic models, in order to get the likelihood model from linkeddata. MLNs is generally accepted by the modern academic society as a simple logic structural representation which almost perfectly combines first order logicand probabilistic graphic models. It has utmost research value and wide rangeapplications. And also it has been a researching spot in regions of artificialintelligence, machine learning and data mining.This paper analyze the cross-lingual entity matching model based onMarkov logic networks. In the original first order predicate logic system,equality predicate is introduced to formally express semantic equality, whichmakes possible that character sequences in different languages may refer to asame object. We applied Markov logic networks in two empirical experiments,and successfully solved the strong reliance of cross language knowledge incross-lingual entity matching and the ambiguity problem in name translation.We, in certain extent, increase the precision by using traditional methods.