Entity Annotation with Wikipedia Using Neural Networks

The paper describes the application of deep learning methods for the task of building references from the phrases detected in a raw text to Wikipedia. As a result, we develop a system that implements the so-called Wikification task which allows one to extend the representation of the text. The proposed approach uses a three-layer neural system. The first, embedding layer creates word representations for detected phrases, next recursive layer forms a context that allows the third layer through classification perform disambiguation of the detected phrase to its proper meaning represented by the Wikipedia article. Our solution was evaluated using the publicly available AIDA-CoNLL benchmark, achieving high accuracy comparable to the best-published systems.