Collective Entity Disambiguation with Structured Gradient Tree Boosting

Proceedings of NAACL-HLT 2018, pages 777–786 New Orleans, Louisiana, June 1 - 6, 2018. c⃝2018 Association for Computational Linguistics Collective Entity Disambiguation with Structured Gradient Tree Boosting Yi Yang Ozan Irsoy Kazi Shefaet Rahman Bloomberg LP New York, NY 10022 {yyang464+oirsoy+krahman7}@bloomberg.net Abstract We present a gradient-tree-boosting-based structured learning model for jointly disam- biguating named entities in a document. Gra- dient tree boosting is a widely used machine learning algorithm that underlies many top- performing natural language processing sys- tems. Surprisingly, most works limit the use of gradient tree boosting as a tool for regu- lar classification or regression problems, de- spite the structured nature of language. To the best of our knowledge, our work is the first one that employs the structured gradient tree boosting (SGTB) algorithm for collective en- tity disambiguation. By defining global fea- tures over previous disambiguation decisions and jointly modeling them with local features, our system is able to produce globally opti- mized entity assignments for mentions in a document. Exact inference is prohibitively ex- pensive for our globally normalized model. To solve this problem, we propose Bidirectional Beam Search with Gold path (BiBSG), an ap- proximate inference algorithm that is a variant of the standard beam search algorithm. BiBSG makes use of global information from both past and future to perform better local search. Experiments on standard benchmark datasets show that SGTB significantly improves upon published results. Specifically, SGTB outper- forms the previous state-of-the-art neural sys- tem by near 1% absolute accuracy on the pop- ular AIDA-CoNLL dataset.1 1 Introduction Entity disambiguation (ED) refers to the process of linking an entity mention in a document to its corresponding entity record in a reference knowl- edge base (e.g., Wikipedia or Freebase). As a core information extraction task, ED plays an impor- tant role in the language understanding pipeline, underlying a variety of downstream applications 1When ready, the code will be published at https:// github.com/bloomberg/sgtb. such as relation extraction (Mintz et al., 2009; Riedel et al., 2010), knowledge base population (Ji and Grishman, 2011; Dredze et al., 2010), and question answering (Berant et al., 2013; Yih et al., 2015). This task is challenging because of the inherent ambiguity between mentions and the re- ferred entities. Consider, for example, the men- tion ‘Washington’, which can be linked to a city, a state, a person, an university, or a lake (Fig. 1). Fortunately, simple and effective features have been proposed to capture the ambiguity that are designed to model the similarity between a men- tion (and its local context) and a candidate en- tity, as well as the relatedness between entities that co-occur in a single document. These are typically statistical features estimated from entity- linked corpora, and similarity features that are pre-computed using distance metrics such as co- sine. For example, a key feature for ED is the prior probability of an entity given a specific men- tion, which is estimated from mention-entity co- occurrence statistics. This simple feature alone can yield 70% to 80% accuracy on both …

Content truncated to 500 words