UNIBA: Combining Distributional Semantic Models and Word Sense Disambiguation for Textual Similarity

Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), pages 748–753, Dublin, Ireland, August 23-24, 2014. UNIBA: Combining Distributional Semantic Models and Word Sense Disambiguation for Textual Similarity Pierpaolo Basile and Annalina Caputo and Giovanni Semeraro Department of Computer Science University of Bari Aldo Moro Via, E. Orabona, 4 - 70125 Bari (Italy) {firstname.surname}@uniba.it Abstract This paper describes the UNIBA team participation in the Cross-Level Semantic Similarity task at SemEval 2014. We pro- pose to combine the output of different se- mantic similarity measures which exploit Word Sense Disambiguation and Distribu- tional Semantic Models, among other lex- ical features. The integration of similar- ity measures is performed by means of two supervised methods based on Gaus- sian Process and Support Vector Machine. Our systems obtained very encouraging results, with the best one ranked 6th out of 38 submitted systems. 1 Introduction Cross-Level Semantic Similarity (CLSS) is the task of computing the similarity between two text fragments of different sizes. The task focuses on the comparison between texts at different lexical levels, i.e. between a larger and a smaller text. The task comprises four different levels: 1) para- graph to sentence; 2) sentence to phrase; 3) phrase to word; 4) word to sense. The task objective is to provide a framework for evaluating general vs. level-specialized methods. Our general approach consists in combining scores coming from different semantic similarity algorithms. The combination is performed by a supervised method using the training data pro- vided by the task organizers. The data set com- prises pairs of text fragments that can be rated with a score between 0 and 4, where 4 indicates the maximum level of similarity. We select algorithms which provide similarities at different levels of semantics: surface (or string- This work is licensed under a Creative Commons At- tribution 4.0 International Licence. Page numbers and pro- ceedings footer are added by the organisers. Licence details: http://creativecommons.org/licenses/by/4.0/ based), lexical (word sense disambiguation), and distributional level. The idea is to combine in a unique system the semantic aspects that pertain text fragments. The following section gives more details about the similarity measures and their combination in a unique score through supervised methods (Section 2). Section 3 describes the system set up for the evaluation and comments on the reported results, while Section 4 concludes the paper. 2 System Description The idea behind our system is to combine the output of several similarity measures/features by means of a supervised algorithm. Those features were grouped in three main categories. The fol- lowing three sub-sections describe in detail each feature exploited by the system. 2.1 Distributional Semantics Level Distributional Semantic Models (DSM) are an easy way for building geometrical spaces of con- cepts, also known as Semantic (or Word) Spaces, by skimming through huge corpora of text in or- der to learn the context of word usage. In the re- sulting space, semantic relatedness/similarity be- tween two words is expressed by the opposite of the distance between points that represent those words. Thus, …

Content truncated to 500 words