Amazing technological breakthrough possible @S-Logix pro@slogix.in

Office Address

  • #5, First Floor, 4th Street Dr. Subbarayan Nagar Kodambakkam, Chennai-600 024 Landmark : Samiyar Madam
  • pro@slogix.in
  • +91- 81240 01111

Social List

Deep keyphrase generation with a convolutional sequence to sequence model - 2017

Deep Keyphrase Generation With A Convolutional Sequence To Sequence Model

Research Area:  Machine Learning

Abstract:

Keyphrases can provide highly condensed and valuable information that allows users to quickly acquire the main ideas. Most previous studies realize the automatic keyphrase extraction through dividing the source text into multiple chunks and then rank and select the most suitable ones. These approaches ignore the deep semantics behind the text and could not predict the keyphrases not appearing in the source text. A sequence to sequence model to generate keyphrases from vocabulary could solve the issues above. However, traditional sequence to sequence model based on recurrent neural network(RNN) suffers from low efficiency problem. We propose an architecture based entirely on convolutional neural networks. Compared to recurrent models, computations can be completely parallelized over all elements so as to better exploit the GPU hardware. Our use of gated linear units alleviates gradient propagation and we equip each decoder layer with a separate attention model. Moreover, we incorporate a copying mechanism to handle out-of-vocabulary phrases. In experiments, we evaluate our model on six datasets, and our proposed model is demonstrated to outperform state-of-the-art baseline models consistently and significantly, both on extracting the keyphrases existing in the source text and generating the absent keyphrases based on the sematic meaning of the text.

Keywords:  

Author(s) Name:  Yong Zhang; Yang Fang; Xiao Weidong

Journal name:  

Conferrence name:  4th International Conference on Systems and Informatics (ICSAI)

Publisher name:  IEEE

DOI:  10.1109/ICSAI.2017.8248519

Volume Information: