The recombination step is a very simple one. As a result of the event attribution stage, we have a set of attributes, each being a continuous value indicating confidence in the presence or absence of a particular event, and we also have a set of global attributes that we extracted as well. Combining them is simply an issue of combining the different attributes extracted by each process and ensuring they are associated with the correct class label.
Table 5.3: Global and event attributes for the Blues and Reds domain.
The output of the recombination stage is a training set of attribute-value instances that a conventional attribute-value learner can use. Some of the attributes are global attributes, and the remainder are ``event-present'' attributes which are the confidence that the training instance has a particular event present in it.