SemLinker, a Modular and Open Source Framework for Named Entity Discovery and Linking

SemLinker, a Modular and Open Source Framework for Named Entity Discovery and Linking Marie-Jean Meurs∗1, Hayda Almeida2, Ludovic Jean-Louis3, Eric Charton4 1Universit´e du Qu´ebec `a Montr´eal, 2Concordia University, 3Netmail, 4Yellow Pages Group Montreal, QC, Canada meurs.marie-jean@uqam.ca, hayda.almeida@concordia.ca ludovic.jean-louis@netmail.com, eric.charton@ypg.com ∗ corresponding author Abstract This paper presents SemLinker, an open source system that discovers named entities, connects them to a reference knowledge base, and clusters them semantically. SemLinker relies on several modules that perform surface form generation, mutual disambiguation, entity clustering, and make use of two annotation engines. SemLinker was evaluated in the English Entity Discovery and Linking track of the Text Analysis Conference on Knowledge Base Population, organized by the US National Institute of Standards and Technology. Along with the SemLinker source code, we release our annotation files containing the discovered named entities, their types, and position across processed documents. Keywords: Entity Discovery, Entity Linking, SemLinker, TAC KBP 1. Introduction Named Entity Recognition (NER) is the task of extract- ing and classifying Named Entities (NEs) in textual doc- uments, while Entity Linking (EL) consists in connecting NEs to their corresponding entries in a Knowledge Base (KB) such as Wikipedia. The English Entity Discovery and Linking (EDL) task (Ji et al., 2014) at the Text Analysis Conference on Knowledge Base Population (TAC KBP1) consists in performing full NE discovery, as well as link- ing and clustering discovered NEs. To handle this task, automated systems identify NEs in a given document col- lection, and link them to their corresponding entities in a reference KB if they exist. Discovered NEs are of type person (PER), organization (ORG), and geo-political en- tity (GPE). For example, in a document with the follow- ing content: “Obama visited Cambridge”, NEs are anno- tated as “⌞Obama⌟[P ER] visited ⌞Cambridge⌟[ORG]”, and linked to relevant Wikipedia entries [Barack Obama] and [University of Cambridge]. NEs linked to KB entries are naturally clustered together, while entities that are not present in the KB, called NIL entities, have to be seman- tically clustered. Extracting NEs in a document collection and associating them to a KB is a complex task. Relevant entities are often polysemous, and can thus be associated with several nodes in a KB. For instance, a linking sys- tem would find multiple candidate nodes in a Wikipedia- based KB to link “Cambridge” to, since the Wikipedia Cambridge (disambiguation) page contains no less than 54 entries. EDL systems have hence to be able to dis- ambiguate NEs. The EDL task is currently investigated in the natural lan- guage processing community with various goals such as populating a new KB from a raw document collection, or improving the retrieval performance of search engines. The 1http://www.nist.gov/tac/2014/index.html SemLinker EDL system was first presented in the TAC KBP Entity Linking 2013 task (Charton et al., 2013; Char- ton et al., 2014a). In 2014, the scope of the task was en- larged to also cover NE discovery, and SemLinker evolved according to this new requirement. The current version of the system relies on generic annotation engines to …

Content truncated to 500 words