FBK-IRST: Kernel Methods for Semantic Relation Extraction

Proceedings of the 4th International Workshop on Semantic Evaluations (SemEval-2007), pages 141–144, Prague, June 2007. c⃝2007 Association for Computational Linguistics FBK-IRST: Kernel Methods for Semantic Relation Extraction Claudio Giuliano and Alberto Lavelli and Daniele Pighin and Lorenza Romano FBK-IRST, Istituto per la Ricerca Scientifica e Tecnologica I-38050, Povo (TN), ITALY {giuliano,lavelli,pighin,romano}@itc.it Abstract We present an approach for semantic rela- tion extraction between nominals that com- bines shallow and deep syntactic processing and semantic information using kernel meth- ods. Two information sources are consid- ered: (i) the whole sentence where the re- lation appears, and (ii) WordNet synsets and hypernymy relations of the candidate nom- inals. Each source of information is rep- resented by kernel functions. In particu- lar, five basic kernel functions are linearly combined and weighted under different con- ditions. The experiments were carried out using support vector machines as classifier. The system achieves an overall F1 of 71.8% on the Classification of Semantic Relations between Nominals task at SemEval-2007. 1 Introduction The starting point of our research is an approach for identifying relations between named entities ex- ploiting only shallow linguistic information, such as tokenization, sentence splitting, part-of-speech tag- ging and lemmatization (Giuliano et al., 2006). A combination of kernel functions is used to represent two distinct information sources: (i) the global con- text where entities appear and (ii) their local con- texts. The whole sentence where the entities appear (global context) is used to discover the presence of a relation between two entities. Windows of limited size around the entities (local contexts) provide use- ful clues to identify the roles played by the entities within a relation (e.g., agent and target of a gene in- teraction). In the task of detecting protein-protein interactions, we obtained state-of-the-art results on two biomedical data sets. In addition, promising re- sults have been recently obtained for relations such as work for and org based in in the news domain1. In this paper, we investigate the use of the above approach to discover semantic relations between nominals. In addition to the original feature rep- resentation, we have integrated deep syntactic pro- cessing of the global context and semantic informa- tion for each candidate nominals using WordNet as external knowledge source. Each source of informa- tion is represented by kernel functions. A tree kernel (Moschitti, 2004) is used to exploit the deep syn- tactic processing obtained using the Charniak parser (Charniak, 2000). On the other hand, bag of syn- onyms and hypernyms is used to enhance the repre- sentation of the candidate nominals. The final sys- tem is based on five basic kernel functions (bag-of- words kernel, global context kernel, tree kernel, su- persense kernel, bag of synonyms and hypernyms kernel) linearly combined and weighted under dif- ferent conditions. The experiments were carried out using support vector machines (Vapnik, 1998) as classifier. We present results on the Classification of Seman- tic Relations between Nominals task at SemEval- 2007, in which sentences containing ordered pairs of marked nominals, possibly semantically related, have to be …

Content truncated to 500 words