An effective approach for semantic-based clustering and topic-based ranking of web documents

In this large, dynamic and expandable web, extracting desired information of any user query is a significant problem for the search engine. Clustering and Ranking are two important resources which can shed light in this direction. To achieve this potential clustering-ranking mechanism, this study proposes a combined approach of semantic-based clustering and topic-based ranking of web documents. The proposed clustering approach combines the latent semantic indexing (LSI) with min-cut algorithm. To make the clustering technique more effective, a new feature selection method called clustering-based feature selection has been developed that focuses on finding the feature set which gathers the crux of documents in the corpus without deteriorating the outcome of the construction process. While LSI completely overcomes the constraint of synonymy, the min-cut algorithm helps to generate efficient clusters at each stage of the clustering process. For deciding the number of clusters to be formed, silhouette coefficient is used, which is a parameter incorporating both cohesion and separation of clusters. To rank the documents in each semantic cluster, the proposed approach transforms the text into topics using latent Dirichlet allocation and then runs the inverted indexing technique on those topics. 20-Newsgroups and DMOZ datasets are used for experimental work, and the results obtained from the experiment show that the performance of the clustering approach is better than the traditional clustering approaches and the ranking approach is promising.