Identifying synonyms in bigram phrases using Compositional DSM and Artificial Neural Network

This paper focuses on identifying whether a bi-gram is a synonym to a given single word, using compositional distributional semantic and Artificial Neural Networks. Understanding the semantics of the natural language text has been a challenging task in natural language processing. In DSM(distributional semantic model) of meaning representation, the words are represented as vectors in multi-dimensional vector space such that words that are semantically similar appear closer to each other whereas words that are dissimilar appear farther apart. This work tries to represent the meaning of bigram in this vector space. Many research works in Distributional semantic try to represent word-pair of the form Adjective-Noun, Verb-Noun, Noun-Noun in multi-dimensional space along with the individual elements of phrases. And these built representations are used for identifying the synonyms of them. Researchers have been successful in representing semantic of words but efficient ways to represent that of phrases are still being investigated. Our work aims to achieve a more efficient method in identifying semantically similar word for the bigram phrase. In this paper, we propose a Neural Network based model which predicts a compositional vector for two-word phrases from its constituent word vectors. The compositional vectors have been evaluated using the classifier to identify synonym word-pairs. PPDB (Paraphrase Database) data set has been used for the experiment. Our model performed better than other non-neural network based models for identifying bigram-unigram synonym pairs.