FNG-IE: an improved graph-based method for keyword extraction from scholarly big-data.

Area of Science:

• Information Science
• Computer Science
• Natural Language Processing

Background:

• Research repositories are growing, creating a need for efficient keyword extraction.
• Existing methods include statistical, machine learning, and graph-based approaches.
• Machine learning methods often require extensive manual training data.

Purpose of the Study:

• To enhance graph-based keyword extraction methods.
• To address the challenge of keyword extraction without large training datasets.
• To improve the identification of influential keywords in massive document collections.

Main Methods:

• Converted a handcrafted dataset into n-gram combinations (unigram to pentagram).
• Enhanced traditional graph-based keyword extraction techniques.
• Developed and tested the proposed FNG-IE method.

Main Results:

• The FNG-IE method demonstrated strong performance in keyword extraction.
• Evaluation using precision, recall, and f-measure showed competitive results.
• The proposed method achieved scores close to those of machine learning approaches.

Conclusions:

• Enhanced graph-based methods can effectively extract keywords without large training datasets.
• The FNG-IE method offers a viable alternative to traditional approaches.
• This research contributes to managing information overload in large-scale data.