Keyphrase Extraction from Bengali Document using LSTM Recurrent Neural Network

Abstract

Keyphrases are single or multiple word phrases of a document which portrays the principal points of that document. These keyphrases help readers to get an overview of the document. In this paper, we proposed a system that uses Long Short-Term Memory (LSTM) Recurrent Neural Network (RNN) to automatically detect keyphrases from a document. We also implemented Multilayer Perceptron (MLP) network to compare the performance of our proposed LSTM approach. We applied several pre-processing steps on a document to generate the candidate keyphrases. Finally, we found better performance from our proposed approach with compared to the MLP network.