Pretrained Sentence Embedding and Semantic Sentence Similarity Language Model for Text Classification in NLP

A key task in the field of Natural Language Processing (NLP) is determining semantic similarity between text sentences. Sentence pair modeling, Textual similarity and language modeling are few important tasks in NLP. Traditional machine learning algorithms require an enormous quantity of training data, but it is a timeconsuming process. Pre-trained models can be modified for a variety of downstream applications since they use methods for generically learning the characteristics of neural network topologies and language representations. Bidirectional Encoder Representations from Transformers- BERT & GPT are the popular architectures in NLP which enable to use minimal fine-tuning effort to produce effective results. In this work a fine-tuned BERT model that is suitable for semantic sentence similarity which predicts the entailment, neutral and contradictory categories of sentence pairs is presented. The fine-tuning feature promotes the training phase of model whichis widely effective across different types of semantic similarity models. The performance analysis of our system shows that the fine-tuned model reduces the number of neurons in the neural network there by reducing storage and time spent in expensive training task to create deep learning model.