Abstract
This paper describes several novel hybrid semantic similarity measures. We study various combinations of 16 baseline measures based on WordNet, Web as a corpus, corpora, dictionaries, and encyclopedia. The hybrid measures rely on 8 combination methods and 3 measure selection techniques and are evaluated on (a) the task of predicting semantic similarity scores and (b) the task of predicting semantic relation between two terms. Our results show that hybrid measures outperform single measures by a wide margin, achieving a correlation up to 0.890 and MAP(20) up to 0.995.
1 Introduction
Semantic similarity measures and relations are proven to be valuable for various NLP and IR applications, such as word sense disambiguation, query expansion, and question answering.
Let R be a set of synonyms, hypernyms, and co-hyponyms of terms C, established by a lexicographer. A semantic relation extraction method aims at discovering a set of relations Rˆ approximating R. The quality of the relations provided by existing extractors is still lower than the quality of the manually constructed relations. This motivates the development of new relation extraction methods.
1 Introduction
Semantic similarity measures and relations are proven to be valuable for various NLP and IR applications, such as word sense disambiguation, query expansion, and question answering.
Let R be a set of synonyms, hypernyms, and co-hyponyms of terms C, established by a lexicographer. A semantic relation extraction method aims at discovering a set of relations Rˆ approximating R. The quality of the relations provided by existing extractors is still lower than the quality of the manually constructed relations. This motivates the development of new relation extraction methods.
No comments:
Post a Comment