.png){kind=icon}
.png){kind=icon}
Return
Swarm optimized cluster based framework for information retrieval
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2020.113441.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This work explores the integrated power of swarm intelligence and advances in data mining techniques to solve the information retrieval (IR) problem of rapidly growing digital content on the World Wide Web. We propose a swarm optimized cluster based framework with frequent pattern mining techniques to retrieve user-specific knowledge from extensive document collections. In the pre-processing phase, we split the task into two sub-tasks. The first is to decompose the document collection into groups using a bio-inspired K-Flock clustering algorithm, while the second extracts frequent patterns from each cluster using a memory-efficient Recursive Elimination (RElim) algorithm. In the next phase, we implement a cosine similarity based probabilistic model to retrieve query-specific documents from clusters based on the matching scores between the closed frequent patterns of queries and clusters. The performance of a system is evaluated by conducting several experiments which are carried out on five well-known, diverse and variable size datasets viz- TREC 2014-15 CDS (Clinical Decision Support) datasets containing 733,138 records, OHSUMED dataset with 348,566 records from Medline database, NPL dataset with 11,429 records, LISA document collection of 6004 records, CACM (Collection of ACM) dataset of 3204 records. The results show that the proposed IR framework significantly outperforms the traditional sequential IR approach and other state-of-the-art IR approaches, both in terms of the quality of the returned documents and the time of execution. (C) 2020 Elsevier Ltd. All rights reserved.
Keywords:
Information retrieval
Swarm intelligence
Big data clustering
Frequent pattern mining
Unsupervised learning
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W