Disambiguation to Wikipedia: A Language and Domain Independent Approach

Disambiguation to Wikipedia (D2W) is the task of linking mentions of concepts in text to their corresponding Wikipedia articles. Traditional approaches to D2W has focused either in only one language (e.g. English) or in formal texts (e.g. news articles). In this paper, we present a multilingual framework with a set of new features that can be obtained purely from the online encyclopedia, without the need of any natural language specific tool. We analyze these features with different languages and different domains. The approach shows as fully language-independent and has been applied successfully to English, Italian, Polish, with a consistent improvement. We show that only a sufficient number of Wikipedia articles is needed for training. When trained on real-world data sets for English, our new features yield substantial improvement compared to current local and global disambiguation algorithms. Finally, the adaption to the Bridgeman query logs in digital libraries shows the robustness of our approach even in the lack of disambiguation context. Also, as no natural language specific tool is needed, the method can be applied to other languages in a similar manner with little adaptation.