Utility of Neural Embeddings in Semantic Similarity of Text Data

Abstract

Semantic similarity plays an important role in understanding the context of text data. In this paper, semantic similarity between large text data is computed using different neural embeddings. we review the utility of different deep neural embeddings for text data representation. Most of the earlier papers have studied the semantic similarity of text by using individual word embeddings. In this paper, we have evaluated the neural embedding techniques on large text data with the help of Essay Dataset. We have used recent neural embedding methods such as Google Sentence Encoder, ELMo, and GloVe along with traditional similarity metrics including TF-IDF and Jaccard Index for experimental investigation. Experimental evaluation in this research paper shows that Google Sentence Encoder and ELMo embeddings perform best on semantic similarity task.