An Algorithm to Self-Extract Secondary Keywords and Their Combinations Based on Abstracts Collected using Primary Keywords from Online Digital Libraries

TL;DR

This paper presents an algorithm that automatically extracts secondary keywords and their combinations from abstracts collected via primary keywords, reducing user input over time and enabling efficient keyword analysis across large datasets.

Contribution

The paper introduces a novel algorithm that self-extracts secondary keywords and their combinations from abstracts, minimizing user intervention as dataset size increases.

Findings

User queries decrease as dataset size grows.

Effective extraction of secondary keywords and combinations.

Applicable to large digital library collections.

Abstract

The high-level contribution of this paper is the development and implementation of an algorithm to selfextract secondary keywords and their combinations (combo words) based on abstracts collected using standard primary keywords for research areas from reputed online digital libraries like IEEE Explore, PubMed Central and etc. Given a collection of N abstracts, we arbitrarily select M abstracts (M<< N; M/N as low as 0.15) and parse each of the M abstracts, word by word. Upon the first-time appearance of a word, we query the user for classifying the word into an Accept-List or non-Accept-List. The effectiveness of the training approach is evaluated by measuring the percentage of words for which the user is queried for classification when the algorithm parses through the words of each of the M abstracts. We observed that as M grows larger, the percentage of words for which the user is queried for classification reduces drastically. After the list of acceptable words is built by parsing the M abstracts, we now parse all the N abstracts, word by word, and count the frequency of appearance of each of the words in Accept-List in these N abstracts. We also construct a Combo-Accept-List comprising of all possible combinations of the single keywords in Accept-List and parse all the N abstracts, two successive words (combo word) at a time, and count the frequency of appearance of each of the combo words in the Combo-Accept-List in these N abstracts.