Research On Entity Linking Discovery Based On Linked Data

With the rapid development of information technology and network technology,there are many data sets being published on the web.The data sets contain analogous data which represent the same resources with so many entity objects in the real world.If those same resources in data sets are linked by correctly building links,then users can conveniently query all needed information from a uniform interface,while query a dataset just once a time.But it is a hard task to build correcting entity linking between datasets,firstly dataset has a huge number of data items and the data's construction is also complex,secondly it will suffer from a lot of comparing computation work to discovery the same resources,so it also makes difficult to link data.Now with more and more linked data applications being used,a lot data choose publish on the web in linked data.Those information published on the web have been transformed to linked data in automatic or semi-automatic ways,so if we can solve the entity linking discovery between linked datasets,then maybe we can give a new method to solve the problems above.But even there are many linked datasets have been published,there are only a few link between entities.It will be inconvenient to share resources for us.So basing on the work of entity linking discovery,we can discovery the real relation between entities and build the entity linking based on linked data,realize the goal of discovering potential entity linking,enhancing the interlinking between datasets and then increasing the accuracy of published linked data!In this thesis,a statistical learning mothed is proposed to recognize entities and build linking across different linked data sets.Before the entities comparing computation,our method first finds class correspondences across datasets and discovers potential attributes correspondences via a statistical learning method,gives a matching relation description for the high correlation attributes,reduces the computing time when compares the similarly of entities;then our method builds entities' linking base on the similarly of attributes to complete the goal of linking discovery across different dataset.When to cluster the attributes correspondences,our method uses K-medoids clustering algorithm to discover the potential attributes correspondences.Then EDOAL language is used to describe the matching relation between those attributes correspondences,and computing the similarly of entities base on the matching attributes decides whether the linking between entities should be building.Our method work out the linking under the SILK framework.In the last Experiments confirm that our method can full fill the linking discovery task and already reaches a high F-measure.So our method can reduce the computing times of comparing entities across different datasets and build correct linking for those entities representing same resources,our method has high feasibility and practicability to solve this problem!...