We present a deep learning approach for the core digital libraries task of parsing bibliographic reference strings. We deploy the state-of-the-art long short-term memory (LSTM) neural network architecture, a variant of a recurrent neural network to capture long-range dependencies in reference strings. We explore word embeddings and character-based word embeddings as an alternative to handcrafted features. We incrementally experiment with features, architectural configurations, and the diversity of the dataset. Our final model is an LSTM-based architecture, which layers a linear chain conditional random field (CRF) over the LSTM output. In extensive experiments in both English in-domain (computer science) and out-of-domain (humanities) test cases, as well as multilingual data, our results show a significant gain (\$\$p<0.01\$\$) over the reported state-of-the-art CRF-only-based parser.
Description
Neural ParsCit: a deep learning-based reference string parser | SpringerLink
%0 Journal Article
%1 prasad2018neural
%A Prasad, Animesh
%A Kaur, Manpreet
%A Kan, Min-Yen
%D 2018
%J International Journal on Digital Libraries
%K citation deep deeplearning extraction learning lstm network neural
%N 4
%P 323--337
%R 10.1007/s00799-018-0242-1
%T Neural ParsCit: a deep learning-based reference string parser
%U https://doi.org/10.1007/s00799-018-0242-1
%V 19
%X We present a deep learning approach for the core digital libraries task of parsing bibliographic reference strings. We deploy the state-of-the-art long short-term memory (LSTM) neural network architecture, a variant of a recurrent neural network to capture long-range dependencies in reference strings. We explore word embeddings and character-based word embeddings as an alternative to handcrafted features. We incrementally experiment with features, architectural configurations, and the diversity of the dataset. Our final model is an LSTM-based architecture, which layers a linear chain conditional random field (CRF) over the LSTM output. In extensive experiments in both English in-domain (computer science) and out-of-domain (humanities) test cases, as well as multilingual data, our results show a significant gain (\$\$p<0.01\$\$) over the reported state-of-the-art CRF-only-based parser.
@article{prasad2018neural,
abstract = {We present a deep learning approach for the core digital libraries task of parsing bibliographic reference strings. We deploy the state-of-the-art long short-term memory (LSTM) neural network architecture, a variant of a recurrent neural network to capture long-range dependencies in reference strings. We explore word embeddings and character-based word embeddings as an alternative to handcrafted features. We incrementally experiment with features, architectural configurations, and the diversity of the dataset. Our final model is an LSTM-based architecture, which layers a linear chain conditional random field (CRF) over the LSTM output. In extensive experiments in both English in-domain (computer science) and out-of-domain (humanities) test cases, as well as multilingual data, our results show a significant gain ({\$}{\$}p<0.01{\$}{\$}) over the reported state-of-the-art CRF-only-based parser.},
added-at = {2021-03-11T10:40:31.000+0100},
author = {Prasad, Animesh and Kaur, Manpreet and Kan, Min-Yen},
biburl = {https://www.bibsonomy.org/bibtex/212216ebb520d67e99deeb8c8176c46f5/jaeschke},
day = 01,
description = {Neural ParsCit: a deep learning-based reference string parser | SpringerLink},
doi = {10.1007/s00799-018-0242-1},
interhash = {92518150147cfece10d625d292697166},
intrahash = {12216ebb520d67e99deeb8c8176c46f5},
issn = {1432-1300},
journal = {International Journal on Digital Libraries},
keywords = {citation deep deeplearning extraction learning lstm network neural},
month = nov,
number = 4,
pages = {323--337},
timestamp = {2021-05-19T08:35:34.000+0200},
title = {Neural ParsCit: a deep learning-based reference string parser},
url = {https://doi.org/10.1007/s00799-018-0242-1},
volume = 19,
year = 2018
}