Better Weather Forecasting through truth discovery Analysis

Abstract

In many real world applications, the same object or event may be described by multiple sources. As a result, conflicts among these sources are inevitable and these conflicts cause confusion as we have more than one value or outcome for each object. One significant problem is to resolve the confusion and to identify a piece of information which is trustworthy. This process of finding the truth from conflicting values of an object provided by multiple sources is called truth discovery or fact-finding. The main purpose of the truth discovery is to find more and more trustworthy information and reliable sources. Because the major assumption of truth discovery is on this intuitive principle, the source that provides trustworthy information is considered more reliable, and moreover, if the piece of information is from a reliable source, then it is more trustworthy. However, previously proposed truth discovery methods either do not conduct source reliability estimation at all (Voting Method), or even if they do, they do not model multiple properties of the object separately. This is the motivation for researchers to develop new techniques to tackle the problem of truth discovery in data with multiple properties. We present a method using an optimization framework which minimizes the overall weighted deviation between the truths and the multi-source observations. In this framework, different types of distance functions can be plugged in to capture the characteristics of different data types. We use weather datasets collected by four different platforms for extensive experiments and the results verify both the efficiency and precision of our methods for truth discovery.