SIFRank

The code of our paper SIFRank: A New Baseline for Unsupervised Keyphrase Extraction Based on Pre-trained Language Model

Versions Notes

• 2020/02/21——Initial version Provided the most basic functions.
• 2020/02/28——Second version Added new algorithms DS(document segmentation) and EA(embeddings alignment) to speed up SIFRank and SIFRank+.
• 2020/03/02——Third version A little change of SIFRank+ in ./model/method.py about making a simple normalization of position_score.

Environment

Python 3.6
nltk 3.4.3
StanfordCoreNLP 3.9.1.1
torch 1.1.0
allennlp 0.8.4

Download

• ELMo elmo_2x4096_512_2048cnn_2xhighway_options.json and elmo_2x4096_512_2048cnn_2xhighway_weights.hdf5 from here , and save it to the auxiliary_data/ directory
• StanfordCoreNLP stanford-corenlp-full-2018-02-27 from here, and save it to anywhere

Usage

import nltk
from embeddings import sent_emb_sif, word_emb_elmo
from model.method import SIFRank, SIFRank_plus
from stanfordcorenlp import StanfordCoreNLP
import time

#download from https://allennlp.org/elmo
options_file = "../auxiliary_data/elmo_2x4096_512_2048cnn_2xhighway_options.json"
weight_file = "../auxiliary_data/elmo_2x4096_512_2048cnn_2xhighway_weights.hdf5"

porter = nltk.PorterStemmer()
ELMO = word_emb_elmo.WordEmbeddings(options_file, weight_file, cuda_device=0)
SIF = sent_emb_sif.SentEmbeddings(ELMO, lamda=1.0)
en_model = StanfordCoreNLP(r'E:\Python_Files\stanford-corenlp-full-2018-02-27',quiet=True)#download from https://stanfordnlp.github.io/CoreNLP/
elmo_layers_weight = [0.0, 1.0, 0.0]

text = "Discrete output feedback sliding mode control of second order systems - a moving switching line approach The sliding mode control systems (SMCS) for which the switching variable is designed independent of the initial conditions are known to be sensitive to parameter variations and extraneous disturbances during the reaching phase. For second order systems this drawback is eliminated by using the moving switching line technique where the switching line is initially designed to pass the initial conditions and is subsequently moved towards a predetermined switching line. In this paper, we make use of the above idea of moving switching line together with the reaching law approach to design a discrete output feedback sliding mode control. The main contributions of this work are such that we do not require to use system states as it makes use of only the output samples for designing the controller. and by using the moving switching line a low sensitivity system is obtained through shortening the reaching phase. Simulation results show that the fast output sampling feedback guarantees sliding motion similar to that obtained using state feedback"
keyphrases = SIFRank(text, SIF, en_model, N=15,elmo_layers_weight=elmo_layers_weight)
keyphrases_ = SIFRank_plus(text, SIF, en_model, N=15, elmo_layers_weight=elmo_layers_weight)
print(keyphrases)
print(keyphrases_)

Evaluate the model

Use this eval/sifrank_eval.py to evaluate SIFRank on Inspec, SemEval2017 and DUC2001 datasets We also have evaluation codes for other baseline models. We will organize and upload them later, so stay tuned. F1 score when the number of keyphrases extracted N is set to 5.

Cite

If you use this code, please cite this paper

@article{DBLP:journals/access/SunQZWZ20,
  author    = {Yi Sun and
               Hangping Qiu and
               Yu Zheng and
               Zhongwei Wang and
               Chaoran Zhang},
  title     = {SIFRank: {A} New Baseline for Unsupervised Keyphrase Extraction Based
               on Pre-Trained Language Model},
  journal   = {{IEEE} Access},
  volume    = {8},
  pages     = {10896--10906},
  year      = {2020},
  url       = {https://doi.org/10.1109/ACCESS.2020.2965087},
  doi       = {10.1109/ACCESS.2020.2965087},
  timestamp = {Fri, 07 Feb 2020 12:04:22 +0100},
  biburl    = {https://dblp.org/rec/journals/access/SunQZWZ20.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}