Text Summarization of Customer Reviews Using Natural Language Processing

Humans have a strong capability to summarize complex and lengthy documents in a simple and concise format. However, in processing and summarizing large volumes of documents within a fraction of seconds, machines outperform the humans. In this particular work, a novel text summarization model was developed by combining extractive and abstractive summarization methods to summarize the large volumes of customer reviews extracted from Amazon data set. The extractive method was used to capture a summary that selects the top-ranking sentences in the corpus using a graph-based TextRank algorithm while these summaries are further fed into a neural network of Long short-term memory (LSTM) to produce the final abstractive summary. The effectiveness of this approach has been measured using the most popularly adopted ROUGE metrics for Natural Language Processing Task. Among multiple models tested, Bi-LSTM is shown to effectively capture the salient information present in the reviews achieving high accuracy and resulted in a concise summary without losing the factual meaning of the reviews