Abstract
The enormous quantity of digital data necessitates automation, which among other things can help link unstructured to structured data. Such a task requires a systematic approach of mapping entity mentions (e.g., person, location) to corresponding entries in a Knowledge Base. This area of research is rapidly evolving at a breathtaking pace, which has led to the popularization of the Named Entity Disambiguation (NED). NED, also known as Entity Linking, described as the task of removing any ambiguities occurring when processing unstructured data packed with Named Entities. The goal of this paper is to investigate ensemble learning using Support Vector Machines (SVM) for tackling the NED problem. Multiple ensemble learning algorithms were studied, including bagging, boosting and voting using different SVM kernel functions, including Linear, RBF, and Polynomial kernels. Our results on three benchmark corpora show that ensemble learning using SVM produces competitive performance levels compared to well-known entity annotation systems and ensemble models. Specifically, the proposed method was best at the disambiguation of AIDA/CONLL-TestB and AQUAINT with F-measure equals to 78.5 and 71.5%, respectively.