Integrating the document object model with hyperlinks for enhanced topic distillation and information extraction

Abstract

Topic distillation is the process of finding authoritative Web pages a comprehensive "hubs" which reciprocally endorse each other and are relevant to a given query. Hyperlink-based topic distillation has been traditionally applied to a macroscopic Web model where documents are nodes in a directed graph and hyperlinks are edges.Mas.M::[KP models miss va lua44 clues such aba4'::M na viga::M paa els,as templa]M2'0]K inclusions, whicha: embedded in HTML paLM using ma0KP taKP Consequently, results of ma:]6:1M2' distillaKP] atillaKP have been deterioraKP] inqua:1 ya s Webpa0: a becoming more complex. We propose a uniformfine-gra'K] model for the Web in which pa:] a represented by theirta trees (aes caesM their Document Object Models or DOMs)aM these DOM trees ar interconnected by ordinaM hyperlinks. Surprisingly, ma]6:[M2K' distillaKKP atillaKK do not work in the finegra -M: scena:]6 We present a new awM0PK1P suitaK1 for the fine-gra2K0 model. It can dis-aggregate hubs into coherent regions by segmenting their DO trees.utua endorsement between hubs as aM0[1['M2K involve these regions, rans, tha single nodes representing complete hubs. Anecdotesae meatesMP' ts using a 28-query, 366000-document benchmark suite, used in ea0]K4 topic distilla[M2 reseai h, reveal two benefits from the new aM:0KK6M2 distillastion quati y improves, a,a by-product of distillation is the aeM14 y to extra0 relevat snippets from hubs which a: nonly payM40[K relevant to the query.