BEL: Bagging for Entity Linking

With recent advances in the areas of knowledge engineering and information extraction, the task of linking textual mentions of named entities to corresponding ones in a knowledge base has received much attention. The rich, structured information in state-of-the-art knowledge bases can be leveraged to facilitate this task. Although recent approaches achieve satisfactory accuracy results, they typically suffer from at least one of the following issues: (1) the linking quality is highly sensitive to the amount of textual information; typically, long textual fragments are needed to capture the context of a mention, (2) the disambiguation uncertainty is not explicitly addressed and often only implicitly represented by the ranking of entities to which a mention could be linked, (3) complex, joint reasoning negatively affects the efficiency. We propose an entity linking technique that addresses the above issues by (1) operating on a textual range of relevant terms, (2) aggregating decisions from an ensemble of simple classifiers, each of which operates on a randomly sampled subset from the above range, (3) following lo-cal reasoning by exploiting previous decisions whenever possible. In extensive experiments on hand-labeled and benchmark datasets, our approach outperformed state-of-the-art entity linking techniques, both in terms of quality and efficiency.