3.2 Time Intervals
Global Interval: We compute a single correlation value for the entire input time interval.
Fixed Interval: The input time interval is divided into fixed length sub-intervals with in which correlations are computed.
Sliding Interval: The input time interval is divided into variable length sub-intervals in a way that maximizes correlations. This goal can be achieved by optimizing sizes of correlation intervals, or by using a greedy algorithm that identifies intervals on-the-fly.
We observe, that for very long time intervals, global average sentiment is likely to be close to zero, so global average and zero average effectively become the same. The same is true for local average, which becomes closer to the global one with increased interval sizes. We will use the above variations of the sentiment correlation in our algorithms, which we describe in the following sections.