Sentence to Sentence Similarity

This paper suggests a novel sentence-to-sentence similarity measure. The proposal makes use of both word embedding and named-entity based semantic similarity. This is motivated by the increasing short text phrases that contain named-entity tags and the importance to detect various levels of hidden semantic similarity even in case of high noise ratio. The proposal is evaluated using a set of publicly available datasets as well as an in-house built dataset, while comparison with some state of art algorithms is performed.