Citation Context

... that is used incorporates comprehensivesinformation regarding text, hyperlink structure, co-citation,sand the unsupervised clustering method based on spectralsgraph partitioning using normalized cut =-=[7]-=-. Our method issonly concerned with the text but allows overlappingsclusters. A news agent called News Dude, developed bysBillsus and Pazzani [2], learns which stories in the news asuser is interested...

Citation Context

... range that most users browse; and these non-dominant but interesting topics are often missed by users. One way to overcome this difficulty is to use webpage clustering on the retrieved webpages (see =-=[14]-=- for example) so that non-dominant topics are separated into different clusters and then use ranking method to pick up the top webpages on each topic. One may also uses web trawling [18] or network-fl...

Citation Context

...e link structure. Recently, there is a surge of interest in tackling the problem of relational data. Although they obtained performance improvement by leveraging relational links for clustering [2][6]=-=[7]-=-[14][19], most of the previous works only addressed part of the two aforementioned problems. For example, they either assume the inter-type links to be accurate and abundant, and only model the intra-...

Citation Context

...tric which incorporates textual information, hyperlink structure and co-citation relations, He et al. proposed an unsupervised clustering method that was shown to effectively identify relevant topics =-=[15]-=-. The clustering method employs a graph-partitioning method based on normalized cut criterion first developed for image segmentation [16]. Bharat and Henzinger augmented a previous connectivity analys...

Citation Context

...he projection of each document onto the first principal components of D. Web page clustering based on a fusion of textual and hyperlink-based analysis has been explored to good effect by He et al. in =-=[9]-=-. However, He et al. pursue the matter in service to categorization of search results on the Web at large. Although our task—automatic concept extraction from a focused domain—is different than theirs...

Citation Context

...noticed in practical applications [7]. A number of web community discovery algorithms are being developed, e.g., trawling to find bipartite cores [23], network maximum flow [15], and graph-clustering =-=[18]-=-. One advantage of these methods is that weak communities (topics) can be separated from dominant communities and thus identified. Without explicit community discovery, webpages of weak topics are typ...

Citation Context

... The method can be viewed in terms of a random walk on a weighted graph in which the nodes with the greatest activation are selected from the steady state distribution of the random process. He et al =-=[18]-=- describe a spectral graph partitioning method based on the Normalized Cut criterion [19]. Modha and Spangler [20] describe a method of clustering results returned by a search engine using a geometric...