Language and Domain Independent Entity Linking with Quantified Collective Validation

Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 695–704, Lisbon, Portugal, 17-21 September 2015. c⃝2015 Association for Computational Linguistics. Language and Domain Independent Entity Linking with Quantified Collective Validation Han Wang∗,1, Jin Guang Zheng∗,2, Xiaogang Ma1, Peter Fox1, and Heng Ji2 {Tetherless World Constellation1, Computer Science Department2} Rensselaer Polytechnic Institute Troy, NY, USA {wangh17, zhengj6, max7, pfox, jih}@rpi.edu Abstract Linking named mentions detected in a source document to an existing knowl- edge base provides disambiguated entity referents for the mentions. This al- lows better document analysis, knowl- edge extraction and knowledge base pop- ulation. Most of the previous research extensively exploited the linguistic fea- tures of the source documents in a su- pervised or semi-supervised way. These systems therefore cannot be easily ap- plied to a new language or domain. In this paper, we present a novel unsuper- vised algorithm named Quantified Col- lective Validation that avoids excessive linguistic analysis on the source docu- ments and fully leverages the knowledge base structure for the entity linking task. We show our approach achieves state- of-the-art English entity linking perfor- mance and demonstrate successful de- ployment in a new language (Chinese) and two new domains (Biomedical and Earth Science). Experiment datasets and system demonstration are available at http://tw.rpi.edu/web/doc/ hanwang_emnlp_2015 for research purpose. 1 Introduction and Motivation The entity linking (EL) task aims at analyzing each named entity mention in a source document and linking it to its referent in a knowledge base (KB). Consider the following example: “One day after released by the Patriots, Florida born Cald- well visited the Jets. ...... The New York Jets have six receivers on the roster: Cotchery, Coles, ...”. Here “Caldwell” is an ambiguous mention ∗These authors contributed equally to this work. because not only are there thousands of people with different professions named “Caldwell”, but even if as an American football player, as most people would recognize it from the context, there are several “Caldwell”s who are/were associated with either “the Patriots” or “the Jets”. An EL system should be able to disambiguate the men- tion by carefully examining the context and then identify the correct KB referent, which is Reche Caldwell in this case. Although EL has attracted a lot of community attention in the recent years, most research efforts have been focused on developing systems only ef- fective for generic English corpora. When these systems are migrated to a new language or do- main, their performance will usually suffer from a noticeable decline due to the following reasons: 1) State-of-the-art EL systems have developed comprehensive linguistic features from the source documents to generate advanced representations of the mentions and their context. While this methodology has been proved rewarding for a resource-rich language such as English, it prevents the systems from being adopted to a new language, especially to one with limited linguistic resources. One can imagine that it would be very difficult, if not impossible, for an English EL system that benefits from the part-of-speech tagging, depen- dency …

Content truncated to 500 words