task3.2

Assume we have: a set of documents indexed, a search engine with relevance
feedback.  

The protocol from the user perspective is the following:
- A user enters a query
- on the results list, he picks the relevant documents 
- the relevance feedback is performed and new documents are returned by the
search engine
- the user select again relevant documents among the new returned list

After both the first and the second relevance evaluation by the user, we
compute several indicators to measure the efficiency of our search engine:

- The precision at 10 / 20 / 30 / 40 / 50
- The recall at 10 / 20 / 30 / 40 / 50
- The precision-recall curve from 1 to 50

We hope that the precision for the first results (10 / 20 at least) increased
on the relevance feedback results, meaning less unrelevant documents are
listed on top.
In the same way, we would like to increase the recall.

We can also plot the precision-recall curves, and check which curve is above
the other. The higher the precision-recall curve is, the better, especially
for low recall, as the top 10 results is often the only results users look at.