Semantic Relatedness Algorithm Design Between Named Entities Based On Linked Open Data

Measuring semantic relatedness plays an important role in information retrievalandNaturalLanguageProcessing. However, little attentionhasbeenpaidtomeasuringsemantic relatedness between named entities, which is also very signifcant. With theadvance of Semantic Web, more and more documents are annotated with real worldentities. Hence, measuring semantic relatedness between these named entities can beregarded as an efective mean to capture semantic associations between documents,which can be further used for semantic search.The existing work can be divided into two types: knowledge based approachesand statistical based approaches. The former ones basically leverage a high-qualityknowledge source like Wordnet [2] or Wikipedia [1]. The main limitation of this kindofworkisthecoverageissue. WhileWikipediaistheworldlargestdomainindependentknowledge base, it misses a number of entities in some specifc domain. On the otherhand, statistical based approaches mainly exploit the Web for this task. However, theyfail to provide reliable semantic relatedness between words of low frequencies.To solve these problems, we propose a more comprehensive approach by lever-aging Linked Open Data (LOD) to solve these problems. LOD consists of lots of datasources from diferent domains and provides rich a priori knowledge about the entitiesintheworld. Asthedatasourcescovermanydomains, givenanamedentity, itishighlypossible that there is some description about it in LOD. Thus entity coverage problemcan be eased by using LOD. On the other hand, while the statistical based approachesregard named entities which have the same name in all documents as the same entity,LODrepresentsthemasdiferententities. Asaresult, eachentityinLOD,eventhelowfrequency entity, has its own description and it is distinguished from other entities of the same name. By exploiting the semantic associations in LOD, we can solve the pre-vious problems and get a better result on measuring the semantic relatedness betweennamed entities. The experimental results show the high performance and robustness ofour approach.