Finding events that occur together I'm interested in computing hashtag correlation; that is, given one hashtag, which other hashtags are likely? I know how to do correlation when the data is in $(x, y)$ format, but this is in $(timestamp, hashtag)$ format. 
I'm looking for a way to find "clumpiness" of any two given hashtags, measured in some unit of time--a "clumpiness" with a small time value means they hang together, whereas a "clumpiness" with a large time value means there isn't as much of a relationship.
EDIT: Sample data--for instance, looking for a link between exercise and mood. Looking at the table there appears to be a link between #exercised and #goodMood, how would I quantify that? Also, there appears to be a negative correlation between #exercised and #okMood.
September 1, 2013 7:00 AM     #exercised
September 1, 2013 9:00 AM     #goodMood
September 2, 2013 10:00 AM    #okMood
September 3, 2013 7:00 AM     #exercised
September 3, 2013 9:00 AM     #goodMood
September 4, 2013 9:00 AM     #okMood

 A: The problem with what you intend to quantify, is that you would need to record hashtags in pre-defined and recurring time-points, otherwise you will not be able to draw valid inferences.
For example, if the second #goodMood was recorded at September 3, 2013 11:00 AM, but no record of any hashtag was made at September 3, 2013 9:00 AM, then you would not be able to say, "hmm, not much clumpiness here" (the scientifically recognizable word is "clustering") - because you could not say whether #goodMood has already set in at 9:00 AM, or not...  
...except if there is some (actually observed) rule that says that "I record a hashtag the moment I feel it sets in". In such a case, you could easily create a data sample with time intervals (#goodmod set in 2 hours after exercise, 3 hours after exercise, 1 hour after exercise etc)- and then go on and calculate their mean, variance, and other statistics of interest. This would give you for example the average time interval between exercise and #goodMood -an absolute figure, which could then be compared to other central tendency statistics from other pairs, and create a ranking.
A: I'd build a N×N table of N hashtags. Each row corresponds to a hashtag, and each column corresponds to a hashtag too. Each cell is the sum of distances (in hours) between a hashtag row and column. For your example this would be the table:
$$\begin{array}{ccc}\text{#exercised}&\text{#goodMood}&\text{#okMood}\\\hline0&2&4\\2&0&2\\1&2&0\end{array}$$
This tells me that #ex and #go are often close to each other.
