Optimal Trust Mining and Computing on Keyed MapReduce

Abstract

This paper studies trust mining in the framework of keyed MapReduce and trust computing in the context of the Bayesian inferences and makes the following two-fold contributions:In the first fold, a general method for trust mining is introduced and formalized in the context of keyed MapReduce functions. A keyed MapReduce function is a classic MapReduce function associated with a common reference keyword set so that a document is projected on the specified common reference set rather the whole dictionary as that defined in the classic MapReduce function. As a result, keyed MapReduce functions allow one to define flexible trust mining procedures: a look-up table which records the comments of neighbors can be constructed from the inverted index of the keyed MapReduce function;In the second fold, a new method for trust computing is introduced and formalized in the context of maximum likelihood distribution. A look-up table generated in the trust mining stage is now viewed as the current state of the target server and then the maximum likelihood distribution over the look-up table is deduced. We show that the proposed trust computing mechanism is optimal (an upper bound of trust values).