"GenSum Adaptive Generalized Text Summarization System using Optimized Transformers"

Abstract

"Abstractive text summarization systems have been integrated with various application in the world to perform text summarization and its nothing new to the field. However, with the prior research it found that in the domain of movies the need for performance improvement is required using latest approaches than the current traditional ML & DL methods, movie review summarization plays a major role in helping users to make better decisions by matching their interest with the reviews of the movie, this saves a lot of time and also improves businesses in their sales. In 2017 researches from Google Brain introduced NLP Transformers, which is a latest approach to solve NLP problems and its increasingly been known and used nowadays over traditional ML & DL approaches like using basic LSTM, RNN approaches. The author explored ways in which to get an optimal solution using Transformer for abstractive text summarization and yet making a generalized solution which can be adapted with respect to any domain (be it hotels, movies, restaurants) and increase its performance as the system gets used over with time. The author was able to experiment with few of the top tier transformer architectures to filter out the optimal model and integrated an automated hyperparameter searching mechanism which will find the best set of hyperparameters to train & customize the model with respect to any domain. ROUGE1 of 80.8, ROUGE2 of 79.42, ROUGEL of 80.8, ROUGELSUM of 80.8 was the optimal evaluation metric result achieved from the BART model giving the best result."