Combining Mention Context and Hyperlinks from Wikipedia for Named Entity Disambiguation

Combining Mention Context and Hyperlinks from Wikipedia for Named Entity Disambiguation Ander Barrena IXA NLP Group UPV/EHU Donostia, Basque Country ander.barrena@ehu.eus Aitor Soroa IXA NLP Group UPV/EHU Donostia, Basque Country a.soroa@ehu.eus Eneko Agirre IXA NLP Group UPV/EHU Donostia, Basque Country e.agirre@ehu.eus Abstract Named entity disambiguation is the task of linking entity mentions to their intended refer- ent, as represented in a Knowledge Base, usu- ally derived from Wikipedia. In this paper, we combine local mention context and global hy- perlink structure from Wikipedia in a proba- bilistic framework. Our results show that the two models of context, namely, words in the context and hyperlink pathways to other en- tities in the context, are complementary. We test our method in eight datasets, improving the state-of-the-art results in five, without any tuning, showing that it is robust to out-of- domain scenarios. When tuning combination weights, we match the best reported results on the widely-used AIDA-CoNLL test-b. 1 Introduction Linking mentions occurring in documents to a knowledge base is the main goal of Entity Link- ing or Named Entity Disambiguation (NED). This problem has attracted a great number of papers in the NLP and IR communities, and a large number of techniques, including local context and global in- ference (Ratinov et al., 2011). We propose to use a probabilistic framework that combines entity pop- ularity, name popularity, local mention context and global hyperlink structure, relying on information in Wikipedia alone. Entity and name popularity are useful disambiguation clues in the absence of any context. The local mention context provides di- rect clues (in the form of words in context) to dis- ambiguate each mention separately. The hyperlink structure of Wikipedia provides a global coherence measure for all entities mentioned in the same con- text. The advantages of our method with respect to other alternatives are as follows: (1) It does not in- volve a large number of methods and classifier com- bination. (2) The method learns the parameters di- rectly from Wikipedia so no additional hand-labeled data and training is needed. (3) We combine the global hyperlink structure of Wikipedia with a lo- cal bag-of-words probabilistic model in an intuitive and complementary way. (4) The absence of training allows for robust results in out-of-domain scenarios. The evaluation of NED is fragmented, with sev- eral popular shared tasks, such as TAC-KBP1, ERD2 or NEEL3. Other evaluation datasets include AIDA- CoNLL and KORE504, which are very common in NED evaluation. Note that each dataset poses differ- ent problems. For instance AIDA-CoNLL is com- posed of news, and systems need to disambiguate all occurring mentions. TAC includes news and discus- sion forums, and focuses on a large number of men- tions for a handful of challenging strings. KORE50 includes short sentences with very ambiguous men- tions. Unfortunately, there is no standard dataset, and many contributions in this area report results in just one or two datasets. We report our results on eight datasets, improving the state-of-the-art results on five. 1http://www.nist.gov/tac/2014/KBP/ 2http://web-ngram.research.microsoft. com/erd2014/ 3http://www.scc.lancs.ac.uk/ microposts2015/challenge/index.html 4http://www.mpi-inf.mpg.de/departments/ databases-and-information-systems/ …

Content truncated to 500 words