SeqVectorizer: Sequence Representation in Vector Space

Abstract

The latest strategies for learning vector space portrayals of words have prevailed with regard to catching fine-grained semantic and syntactic consistencies utilizing vector arithmetic. However, the sequence representation is not present in these methods. As a result, to consider the sequence, we are utilizing the sequence neural networks like RNN or statistical techniques like HMM. To represent the sequence through every state vector, we propose a new term or word representation technique called SeqVectorizer, which stands for sequence vectorizer. In SeqVectorizer every state represents a combined vector of two separate joined states, and these are the previous sequence state and the current state probability. Comparing with other representation systems, it shows a state-of-the-art performance on some testing data-sets.