Automatic keyphrase extraction using recurrent neural networks

Abstract

Automatic Keyphrase Extraction describes the process of extracting keywords or keyphrases from the body of a document. To our knowledge until now all algorithms rely on a set of manually crafted statistical features to model word importance. In this paper we propose an end-to-end neural keyphrase extraction algorithm using a siamese LSTM network, eliminating the need for manual feature engineering. We train and evaluate our model on the Inspec [6] dataset for keyphrase extraction and achieve comparable results to state-of-the-art algorithms.