Approximating Google´s rankings with Latent Semantic Analysis

This study proposed a Latent Semantic Analysis based method to approximate Google´s ranking. We conduct Latent Semantic Analysis on Google´s search results for a given query to find terms with highest LSA weights as features. Then the correlation coefficients between the features and the given query are obtained for use as the feature values. Each result is scored and re-ranked based on a linear combination of weighted sum of feature values that appear in its title, snippet and URL. Experimental results on a small number of popular keywords show that this method is promising to achieve R-Precision up to 0.8 for some combination of search results and features used.