Site-granularity topic distillation on the Web by combining content and hyperlink analysis

Topic distillation on the Web, namely, given a user query to find quality information sources related to the query topic by using hyperlink analysis, has been shown to be useful in Web IR. Based on the analysis of three deficiencies of classical topic distillation algorithm HITS (i.e., failing to meet users' site-granularity information needs; tending to produce unreasonable results; topic drift), this paper presents an improved model and algorithm named s-HITSc (site-granularity HITS enhanced by content analysis). Given a query topic, the new algorithm can model a neighborhood graph at site granularity, compute the relevance weights of the nodes to the topic with content analysis, and apply weighted I/O operations in its iterative hyperlink analysis. Theoretical analysis and experimental results show that the new algorithm can control topic drift and identify more reasonable and meaningful authority and hub sites on a given query topic.