Empirical analysis on a keyword-based semantic system

TL;DR

This paper empirically analyzes the statistical and evolutionary properties of keywords in scientific articles, revealing patterns like Zipf's law and decay trends, which enhance understanding of semantic evolution and aid recommender system design.

Contribution

It provides a comprehensive empirical investigation of keyword dynamics across journals, highlighting universal and subject-specific patterns in semantic evolution.

Findings

Keyword frequency follows Zipf's law with exponent 0.86.

Power-law correlation exists between distinct keywords and total occurrences.

Similar decay trends observed across high-impact journals, differing in low-impact ones.

Abstract

Keywords in scientific articles have found their significance in information filtering and classification. In this article, we empirically investigated statistical characteristics and evolutionary properties of keywords in a very famous journal, namely Proceedings of the National Academy of Science of the United States of America (PNAS), including frequency distribution, temporal scaling behavior, and decay factor. The empirical results indicate that the keyword frequency in PNAS approximately follows a Zipf's law with exponent 0.86. In addition, there is a power-low correlation between the cumulative number of distinct keywords and the cumulative number of keyword occurrences. Extensive empirical analysis on some other journals' data is also presented, with decaying trends of most popular keywords being monitored. Interestingly, top journals from various subjects share very similar decaying tendency, while the journals of low impact factors exhibit completely different behavior. Those empirical characters may shed some light on the in-depth understanding of semantic evolutionary behaviors. In addition, the analysis of keyword-based system is helpful for the design of corresponding recommender systems.