An Early DiscoWeb Prototype at TREC8
Brian D. Davison, Apostolos Gerasoulis, Konstantinos Kleisouris, Yingfang Lu, Hyun-ju Seo, Junyu Tian, Song Wang, Wei Wang, Baohua Wu · Text REtrieval Conference · 1999
Recently the notion of popularity and its generalizations have been investigated as a possible alternative approach to text only analysis to rank web pages in search engines (e.g. [Kle98, BP98, CDR + 98, CDDG + 98, BH98, HHMN99] among others). We have built a research prototype that incorporates many link analysis algorithms from the literature and also new algorithms to investigate the impact of the popularity on the ranking of the search engines [DGK + 99]. Our goal in the TREC8 competition was to investigate the quality of the results using the TREC data and in particular the large web track. Unfortunately we did not have the needed hardware in time to generate results for the large web track. We only participated in the Small Web Track (Text Only and Text and Link Analysis). However, our system was designed for large datasets and the quality of the TREC8 results are not representative of the system. More recently we have experimented with larger datasets and we have come to the conclusion that link analysis can signicantly increase the quality of the ranking of search engines, a conclusion that is shared by many others in the literature [BP98, PBMW98, Kle98, CDR + 98, CDDG + 98, CDG + 99]. We will report these new results in a future publication.