Hot Item Detection in Uncertain Data

An object o of a database D is called a hot item, if there is a sufficiently large population of other objects inD that are similar to o. In other words, hot items are objects within a dense region of other objects and provide a basis for many density-based data mining techniques. Intuitively, objects that share their attribute values with a lot of other objects could be potentially interesting as they show a typical occurrence of objects in the database. Also, there are a lot of application domains, e.g. sensor databases, traffic management or recognition systems, where objects have vague and uncertain attributes. We propose an approach for the detection of potentially interesting objects (hot items) of an uncertain database in a probabilistic way. An efficient algorithm is presented which detects hot items, where to each object o a confidence value is assigned that reflects the likelihood that o is a hot item. In an experimental evaluation we show that our method can compute the results very efficiently compared to its competitors.