Dataset and Source code of paper 'Enhancing Keyphrase Extraction from Academic Articles with their Reference Information'.

Last update: Nov 24, 2022

Overview

Enhancing Keyphrase Extraction from Academic Articles with their Reference Information

Overview

Dataset and code for paper "Enhancing Keyphrase Extraction from Academic Articles with their Reference Information".

The research content of this project is to analyze the impact of the introduction of reference title in scientific literature on the effect of keyword extraction. This project uses three datasets: SemEval-2010, PubMed and LIS-2000, which are located in the dataset folder. At the same time, we use two unsupervised methods: TF-IDF and TextRank, and three supervised learning methods: NaiveBayes, CRF and BiLSTM-CRF. The first four are traditional keywords extraction methods, located in the folder ML, and the last one is deep learning method, located in the folder DL.

Directory structure

Keyphrase_Extraction:                 Root directory
│  dl.bat:                            Batch commands to run deep learning model
│  ml.bat:                            Batch commands to run traditional models
│ 
├─Dataset:                            Store experimental datasets
│      SemEval-2010:                  Contains 244 scientific papers 
│      PubMed:                        Contains 1316 scientific papers
│      LIS-2000:                      Contains 2000 scientific papers
│ 
├─DL:                                 Store the source code of the deep learning model
│  │  build_path.py:                  Create file paths for saving preprocessed data
│  │  crf.py:                         Source code of CRF algorithm implementation(Use pytorch framework)
│  │  main.py:                        The main function of running the program
│  │  model.py:                       Source code of BiLSTM-CRF model
│  │  preprocess.py:                  Source code of preprocessing function
│  │  textrank.py:                    Source code of TextRank algorithm implementation.
│  │  tf_idf.py:                      Source code of TF-IDF algorithm implementation.
│  │  utils.py:                       Some auxiliary functions
│  ├─models:                          Parameter configuration of deep learning models
│  └─datas
│        tags:                        Label settings for sequence labeling
│ 
└─ML:                                 Store the source code of the traditional models
    │  build_path.py:                 Create file paths for saving preprocessed data
    │  configs.py:                    Path configuration file
    │  crf.py:                        Source code of CRF algorithm implementation(Use CRF++ Toolkit)
    │  evaluate.py:                   Source code for result evaluation
    │  naivebayes.py:                 Source code of naivebayes algorithm implementation(Use KEA-3.0 Toolkit)
    │  preprocessing.py:              Source code of preprocessing function
    │  textrank.py:                   Source code of TextRank algorithm implementation
    │  tf_idf.py:                     Source code of TF-IDF algorithm implementation
    │  utils.py:                      Some auxiliary functions
    ├─CRF++:                          CRF++ Toolkit
    └─KEA-3.0:                        KEA-3.0 Toolkit

Dataset Description

The dataset includes the following three json files:

SemEval-2010: SemEval-2010 Task 5 dataset, it contains 244 scientific papers and can be visited at: https://semeval2.fbk.eu/semeval2.php?location=data.

PubMed: Contains 1316 scientific papers from PubMed (https://github.com/boudinfl/ake-datasets/tree/master/datasets/PubMed).

LIS-2000: Contains 2000 scientific papers from journals in Library and Information Science (LIS).

Each line of the json file includes:

title (T): The title of the paper.

abstract (A): The abstract of the paper.

introduction (I): The introduction of the paper.

conclusion (C): The conclusion of the paper.

body1 (Fp): The first sentence of each paragraph.

body2 (Lp): The last sentence of each paragraph.

full_text (F): The full text of the paper.

references (R): references list and only the title of each reference is provided.

keywords (K): the keywords of the paper and these keywords were annotated manually.

Quick Start

In order to facilitate the reproduction of the experimental results, the project uses bat batch command to run the program uniformly (only in Windows Environment). The dl.bat file is the batch command to run the deep learning model, and the ml.bat file is the batch command to run the traditional algorithm.

How does it work?

In the Windows environment, use the key combination Win + R and enter cmd to open the DOS command box, and switch to the project's root directory (Keyphrase_Extraction). Then input dl.bat, that is, run deep learning model to get the result of keyword extraction; Enter ml.bat to run traditional algorithm to get keywords Extract the results.

Experimental results

The following figures show that the influence of reference information on keyphrase extraction results of TF*IDF, TextRank, NB, CRF and BiLSTM-CRF.

Table 1: Keyphrase extraction performance of multiple corpora constructed using different logical structure texts on the dataset of SemEval-2010

Table 2: Keyphrase extraction performance of multiple corpora constructed using different logical structure texts on the dataset of PubMed

Table 3: Keyphrase extraction performance of multiple corpora constructed using different logical structure texts on the dataset of LIS-2000

Note: The yellow, green and blue bold fonts in the table represent the largest of the P, R and F₁ value obtained from different corpora using the same model, respectively.

Dependency packages

Before running this project, check that the following Python packages are included in your runtime environment.

pytorch 1.7.1

nltk 3.5

numpy 1.19.2

pandas 1.1.3

tqdm 4.50.2

Citation

Please cite the following paper if you use this codes and dataset in your work.

Chengzhi Zhang, Lei Zhao, Mengyuan Zhao, Yingyi Zhang. Enhancing Keyphrase Extraction from Academic Articles with their Reference Information. Scientometrics, 2021. (in press) [arXiv]

Dataset and Source code of paper 'Enhancing Keyphrase Extraction from Academic Articles with their Reference Information'.

Related tags

Overview

Enhancing Keyphrase Extraction from Academic Articles with their Reference Information

Overview

Directory structure

Dataset Description

Quick Start

How does it work?

Experimental results

Dependency packages

Citation

Owner

AntroPy: entropy and complexity of (EEG) time-series in Python

Deeper DCGAN with AE stabilization

[NeurIPS'21] Shape As Points: A Differentiable Poisson Solver

Accurate identification of bacteriophages from metagenomic data using Transformer

Release of SPLASH: Dataset for semantic parse correction with natural language feedback in the context of text-to-SQL parsing

9th place solution in "Santa 2020 - The Candy Cane Contest"

Deep Learning Algorithms for Hedging with Frictions

Semantic code search implementation using Tensorflow framework and the source code data from the CodeSearchNet project

Implementation of "Fast and Flexible Temporal Point Processes with Triangular Maps" (Oral @ NeurIPS 2020)

This repository implements variational graph auto encoder by Thomas Kipf.

Sketch-Based 3D Exploration with Stacked Generative Adversarial Networks

Official repository for "PAIR: Planning and Iterative Refinement in Pre-trained Transformers for Long Text Generation"

Identifying Stroke Indicators Using Rough Sets

pyspark🍒🥭 is delicious，just eat it!😋😋

Code for ICLR 2021 Paper, "Anytime Sampling for Autoregressive Models via Ordered Autoencoding"

Code for CVPR2019 Towards Natural and Accurate Future Motion Prediction of Humans and Animals

A Python package for time series augmentation

LightSeq is a high performance training and inference library for sequence processing and generation implemented in CUDA

Fine-tune pretrained Convolutional Neural Networks with PyTorch

COPA-SSE contains crowdsourced explanations for the Balanced COPA dataset