I wanted to figure out how long a good title length, respective to the discipline (dewey class), is.
I was hoping to see results such as: People prefer books on religion with a title length of more than x words, while people prefer books on science with more than y words.

To do so, I aggregated the data first by their dewey class (only the first 3 digits) and then sub-aggregated by the number of words in the title. For each of these sub-groups, I summed up the number of checkouts and the number of books that exist in this sub-group. I then defined the popularity of the subgroup as the ratio of checkouts to number of books (of this dewey class and group of title lenght).

The three dimensions I consequently extracted from the database are 
a) the dewey class
b) the title length
c) the popularity

For each dewey category, I then plotted the popularity vs. the title length.

I encountered several issues with my initial idea:
1) No pattern whatsover was visible. It basically was just a clotter of seemingly random lines.
2) There was too much data. This makes problem 1) worse, but also is a performance issue.
Considering that we have 1000 dewey classes (only first 3 digits) and that the title lengths may span from anywhere between 1 to 40 (in average, there were 14.25 distinct title lengths per dewey), we end up in the ten-thousands of rows.
3) Some dewey categories are obviously more demanded than others. Therfore the lines are quite far out scattered.

To overcome the issues, I implemented following changes:
1) I aggregated the data by the first 2, instead of the first 3 digits of the dewey class. This reduced the data to approx 3000 rows and consequently results in 100 distinct lines. I was hoping that due to the aggregation and therefore more data per group being available, we would also be able to see trends (instead of randomness) more clearly.
2) I extracted the main dewey class from the database (rather than computing it during runtime in processing) in the hope to use it for a better color-coding.
3) I implemented interactive filters.

What I still want to do:
1) Move the filtered main class to foreground when filter is activated.
2) Add a legend for the sub-classes.