Effect of different network analysis strategies on search engine re-ranking

The research described in this paper examined two different approaches to building the co-citation network that the authors have used in re-ranking the set of results returned by a search engine [22, 23]. The more computationally demanding (in terms of query load) Inter- or Web-wide co-citation approach used in-links from throughout the Web to build the network. In contrast, the Intra co-citation approach only used inlinks inferred from search engine output. Results of this study confirmed the authors' previous findings [23] that reordering based on a network-analytic relevance prediction model significantly improves the precision of top 20 results as compared to the Google search engine. The results also showed (for the queries used) that the Intra co-citation approach is significantly better than the Web-wide co-citation approach, in addition to placing fewer querying demands on the search engine.