Multivariate Aggregated Ranking Score: A Lightweight and Greedy Ranking Model for Web Search

Quality service mining requires a strong characterization of content, including a measure of the specificity of search results; specificity is a measure of how specific or broad a document is. The attempt to measure this concept, i.e., accepting it as a quantity, should be ruled by clear steps. In this paper, we address these issues by proposing Mars Rank, a multivariate aggregate ranker whose standard deviation calculation takes into consideration the variation with respect to the average topic-sensitive score. Mars Rank has demonstrated monotonicity in experiments. We prove that the score is monotone and introduce the necessary conditions for this. It is intended that the monotonicity property is a quality guarantee. Additionally, a variant is presented that retains the properties of the power method Topic Sensitive page rank and is biased towards pages having a high standard deviation of the score.