Citation Context

...em [7]. This alleviates the need for user input to establish the decision rule, but requires the availability of a substantial set of robust training data (although some methods, like active learning =-=[10]-=- can reduce the required amount of data). Here we investigate a third category of approaches that relies on the characteristics of the datasets and of the similarity distributions resulting from compa...

Citation Context

... towards automatic discovery of linkage rules (or linkage specifications). To the best of our knowledge, this problem has only recently been studied for Web data, and as a part of SILK [13] and LIMES =-=[17]-=- RDF link discovery frameworks. These approaches use learning algorithms and rely on a number of manually-labeled link candidates. The SILK framework [13] uses a genetic programming paradigm to learn ...

Citation Context

...nstance matching is to determine the matching status of two interested instances. Current interlinking methods can be categorized into two main groups: schemadependent [2,7,10] and schema-independent =-=[1,3,4,9]-=-. The former requires the knowledge about meaning of RDF predicates (e.g. predicate #preLabel declares the label of object) and the predicate alignments (e.g. predicate #preLabel matched with predicat...

Citation Context

...role so far: (1) the determination of time-efficient algorithms [3, 7, 6] for LD and (2) the development of approaches for the efficient computation of link specifications (also called linkage rules) =-=[8, 4, 10, 9]-=-. In most cases, supervised machine learning approaches were used to tackle the second challenge of LD. Approaches developed so far include batch learning using genetic programming [3], the combinatio...

Citation Context

..., SAIM can support domain experts and lay users during the creation of link specifications. Moreover, it implements the timeefficient algorithms for class and property matching algorithms proposed in =-=[5]-=-. SAIM goes beyond existing interfaces for link discovery (e.g., SILK Workbench 6) by supporting several self-configuration algorithms that allow the automatic creation of link specifications. In the ...

Citation Context

...s2 and is very time-efficient. A graphical user interface for the approach is available with SAIM3. This article is an extension of the corresponding workshop article presenting the RAVEN approach in =-=[31]-=-. Changes include an extended discussion of the evaluation results, e.g. the inclusion of property and class matching. Furthermore, the related work part was extended as well as further illustrations ...

Citation Context

...for applications of semantic technologies, e.g. in the areas of data integration and federated queries. While there has been significant recent progress on the automation of this linking process [1], =-=[2]-=-, fully automatic approaches are unlikely to ever achieve extremely high precision and recall over a set of complex scenarios. This means that ensuring a high quality of links requires a manual proces...

Citation Context

...oposed ActiveGenLink [12], which aims to reduce the number of labelled examples using active learning. ActiveGenLink selects link candidates to be labelled by the user from a pool of unlabelled instances through a query strategy. Then, once the user labels a given example, it adds the example to the training data and evolves the population using GenLink. Another semi-supervised technique is EAGLE [17], which is based in a genetic programming technique with active learning, and it aims to generate link specifications as trees. It starts detecting similar classes and data properties using RAVEN [16]. Then, EAGLE evolves an initial random population of link specifications according to genetic operators. After that, the technique computes the most informative links and asks the user to label them. This process is repeated until the stop condition is fulfilled, i.e., a maximal number of iterations is reached, or the fitness value of a link specification is over a given threshold. An unsupervised learning technique was proposed by Nikolov et al. [19], which starts with a random population and keeps iterating over it, applying genetic operators, until a maximal number of iterations is reached...

Citation Context

...que to compare whether two values are similar or not. Machine learning is also used to find out the most efficient comparison pattern, as shown in (Hu, Chen, and Qu 2011; Isele and Bizer 2011; Ngonga =-=Ngomo et al. 2011-=-). That is, which property values should be compared and what comparison functions should be used. There are two machine learning branches: supervised method and unsupervised method. Supervised method...