Browsing by Author "Nikola, Mandic"

Word vectors have become corner stone of modern NLP. Researchers are taking embedding ever further by learning to craft embedding vectors with task specific semantics to power wide array of applications. In this thesis we apply simple feed forward network and stacked LSTM on triplets dataset converted to sentence embeddings to evaluate paragraph semantic text similarity. We explore how to leverage existing state of the art sentence embeddings for paragraph semantic text similarity and examine information sentence embeddings used hold.