Citation Context

... Our prototype extends techniques we developed for DogmatiX [14] and allows efficient and effective duplicate detection. Our prototype is extensible and may use alternative methods for classification =-=[7, 8, 12]-=- and comparison strategy [4, 6, 15]. Our choice for extending DogmatiX was motivated by previous research of ours that showed that DogmatiX is best suited in terms of effectiveness in scenarios simila...

Citation Context

... is often desirable to be able to (aggregate and then) use external data/knowledge sources to disambiguate a dataset, as illustrated in Figure 5. This is indeed possible with the approach proposed in =-=[21, 22]-=- and with an adaptive CS model, as illustrated in Figure 6. For that, the dataset and external data need to be first represented as the standard entity-relationship (ER) graph. Then the nodes that are...

Citation Context

... dependencies among coreference decisions [32], [38]. Our proposed disambiguation algorithm is based on analyzing two types of information: object features and the ER graph for the data set. In [17], =-=[27]-=-, and [28], it has been shown that complementing the traditional methodology of analyzing object features with analysis of the ER graph can lead to the improved quality of disambiguation. The idea is ...

Citation Context

...ional approaches exploit textual similarity between reference attributes [18]. More recent approaches consider relational similarities derived from the context of references as additional information =-=[5, 12, 19, 22, 30, 31]-=-. In addition, the similarity measures are no longer static: evidence can be propagated in an iterative way and one resolution decision can affect the others [10, 12]. There are some other approaches ...

Citation Context

.... So if the combined evidence, stored in all the u-v paths, is sufficiently large, the objects are likely to be the same. An in-depth insight into the motivation for this methodology is elaborated in =-=[5]-=-. Formally, the attraction between two nodes u and v via paths is measured using the connection strength measure c(u,v) which is defined as the sum of attractions contributed by each path: c(u,v) = � ...

Citation Context

...s, such as entity resolution and record linkage [1, 4, 7, 9, 10, 19, 20, 23, 24, 27, 28, 31]. Recently, some domain-independent data cleaning approaches for reference disambiguation has been proposed =-=[16,25]-=-, that systematically exploits features and relationships among entities for the purpose of disambiguation. The approach in [16], which we employ to test our adaptive solution, views the dataset as a ...

Citation Context

...reference resolution between instances (also called record linkage [10]) also has been studied for a long time in the database domain. In the classical model proposed in [10] by Fellegi and Sunter the decision regarding whether two individuals (records) describe the same entity was taken based on calculating an aggregated similarity between their field values. This model served as the base for the majority of algorithms developed in the domain (see [11] for a survey). Other proposed approaches exploit additional kinds of available information. For example, the algorithms described in [12] and [13] analyse links between instances and consider graphs instead of atomic data individuals, in [14] ontological schema restrictions are used as evidence and in [15] provenance information about data sources is exploited. While originally in the Semantic Web domain the research primarily focused on the schema-level ontology matching, with the emergence of the Linked Data the problem of instance-level integration also gained importance. Popular existing tools were created and applied to discovery of the owl:sameAs coreference links for Linked Data. These follow the classical Fellegi-Sunter model: e...

Citation Context

...rging objects is done by calculating the distance between data that possibly describe the same real-world entity [5,6], or computing the interconnection strength of their alternative connection paths =-=[15,14]-=-. The most relevant algorithms related to our approach are the ones that identify entities through interconnections between objects found in a given dataset. Ananthakrishna et al. [2] exploit dimensio...