Semantic Similarity based Web Document Clustering Using Hybrid Swarm Intelligence and FuzzyC-Means

Author Name(s): *J.Avanija, G. Sunitha,K. Reddy Madhavi
Author Email: avans75@yahoo.co.in

Abstract

Information retrieval technology has been central to the success of the web. The volume of information stored and accessed on web is increasing continuously. This enlargement leads to the difficulties such as seeking and managing the existing information. The use of keyword based method in information retrieval processing is the reason behind this limitation and can be overcome by performing semantic search instead of using keywords. Optimization problem occurs even through the use of semantic technologies. Hybrid clustering algorithm combining FuzzyC-Means and PSO has been proposed to generate optimum number of clusters and to get better accuracy in the recovery of documents. Experimental results reveal that the proposed method performs better than hybrid approaches combining PSO and KMeans.

Keywords

Fuzzy C-Means, Clustering, Semantic Similarity, Ontology, K-Means,Particle Swarm Optimization

Introduction

Clustering algorithms are applied on a variety of fields. Appropriate document clustering can be provided through clustering methods [1]. The use of Ontologies improves the keyword search mechanism. Grouping of documents performed based on similarity score using PSO based clustering to improve the document relevancy [2].Documents are categorized using ontology concept weights which improves the accuracy of documents. Similarity measure based clustering of documents along with ontology is proposed [3]. Fuzzy clustering based on semantic analysis using ontology is performed [4].FuzzyC-Means algorithm is the popular clustering method since it is efficient and the implementation is also simple. Even then FCM has drawbacks such as getting trapped to local optima and sensitive to initialization.The K-Means is a hard clustering process in which data is divided into distinct clusters, where each data element belongs to exactly one cluster. FuzzyC-Means also known as soft clustering is well suited for real data set clustering since there may be sharp boundaries between clusters. The membership grades of every document represents the degree by which a document belongs to a particular cluster [5]. To overcome the limitation faced by K-Means Algorithm, PSO is combined with FCM and to improve the speed, fuzziness parameter is included in PSO and combined with FCM. Performance analysis shows that FPSOFCM is better than PSOK-Means (PSOK).

Conclusion

Information retrieval relies on keywords for indexing and retrieving documents. Mostly  keyword based retrieval returns inappropriate  results since same concept can be described using different keywords in documents as well as queries. The background work related to semantic information retrieval indicates that semantic clustering has drawn attention in recent years since it provides accuracy, but efficient clustering techniques are still needed to improve accuracy of document recovery. Further, optimization problem also occurs during clustering process. To overcome this limitations evolutionary approaches like PSO and fuzzy are used along with semantic based clustering process to cluster documents.

Semantic document-clustering using swarm intelligence gives better accuracy, but the computational time is more and is not suitable for large dataset. To handle large dataset and to reduce the computational time, a hybrid approach combining clustering algorithms with evolutionary approaches such as PSO and fuzzy is proposed for clustering web documents. The hybrid approach of clustering combining PSO and K-Means provided optimal solution, but computational time is still higher. Considering this limitation FuzzyC-Means is combined along with PSO to cluster the documents. Document-clustering based on “FPSO+FCM” method shows improvement over algorithms like FuzzyC-Means K-Means and hybrid approaches like KPSO and PSOK. In the “FPSO+FCM” the ability of globalized searching of the PSO algorithm and quick convergence of FCM algorithm are combined. The result from FPSO is used as the initial seed for FCM algorithm, which is applied for refining and generating the final result. Future scope will be to successfully approach real world problems in diverse domains.

1,276 total views, no views today

Download PDF File

About the author: tej