Probabilistic classification techniques to perform geographical labeling of web objects

Abstract

Web search engines provide relevant documents to answer user query. These result-set documents might also contain redundant information which the user might not require. The user has to invest effort in navigating each document to identify relevant information. To overcome such cumbersome overheads, Web object search engines were proposed. These systems provide powerful vertical search facility, so that, the result set of a query will only contain the relevant Web object information. Many techniques have been proposed to geographically label documents for Web search engines, however, geographical labeling of Web objects has got limited attention. The presence of noise in the Web objects due to inaccurate object extraction process complicates the task of assigning geographical labels. Recently in the literature. Gaussian mixture model oriented classification technique was proposed to achieve geographical labeling of Web objects, even then, there is an ample scope to improve labeling accuracy. In this work, two probabilistic classier namely-Bayesian Classifier and Variational Inference Classifier are utilized to achieve geographical labeling of Web objects. The proposed technique provides at least 30% better labeling accuracy and twice better computational efficiency when compared with the contemporary technique.