| A rapidly growing amount of research focuses on processing uncertain data such as market analysis,WWW,and large sensor networks.Ranking queries on uncertain data have also been studied intensively.The probabilistic Top-k query among them is quite useful to find the k objects that a customer is most interested in.However,the customer's expected objects may not in the query result.Thus,he may pose a Why-not question on probabilistic Top-k queries.On the other hand,the probabilistic reverse Top-k query is applied to find out customers who treat the query object as one of their Top-k favorite with a probability of at least ? a user specified probability threshold.When manufacture's target customers are missing in the query result,he/she should have confusion:Why these customers are not interested in his/her product.Although there are some efforts on Why-not questions,no proposal exists for how to explain the reason Why the expected objects are missing from probabilistic Top-k/reverse Top-k query results,and hence,the existing solutions can not handle our problem directly.Base on the above discussion,in the paper,we study following two prolems.(i)Why-not questions on probabilistic Top-k queries:We solve the Why-not questions on probabilistic Top-k queries by modifying Why-not weighting vectors Wm and k(MWK).MWK refines the original query to make Wm return to the query result.(ii)Why-not questions on probabilistic reverse Top-k queries:We propose a framework called WNPTR to tackle this problem.Given an orginal probabilistic reverse Top-k query and a set of missing tuples,our algorithm returns refined queries from three different aspects,namely,1)modifying the query object;2)modifying the missing weighting vectors Wm and parameter k,3)modifying query object,missing weighting vectors Wm,and parameter k simultaneously.Finally,extensive experimental evaluation using both real and synthetic data sets demonstrates the effectiveness and efficiency of the proposed algorithms. |
PDF Full Text Request