4.5.2 Scoring ¶
Scoring in p-search is simple:
- Each prior will assign a probability score between zero and one to every candidate document.
- The final score is the product of all prior’s probabilities.
- A normalized score is displayed to the user so that all probabilities add to one.
Other search systems employ a scoring system that incorporates additive elements as well as multiplicative. The problem with such an approach for p-search is that it ruins composability. p-search is meant to be a tool to generate search setups on the fly. Having a formula like SCORE = FILE-SCORE * 1.2 + QUERY-SCORE * 1.3 + GIT-AUTHOR * 2
would require the user to evaluate each piece of the equation when any change is made.
A prior should assign a score to a document according to the following criteria:
- Evidence for a document should be indicated with a probability greater than 0.5. The stronger evidence, the closer to 1.0.
- Evidence against a document should be indicated with a probability lower than 0.5. The stronger evidence against, the closer to 0.0.
- Scores of zero or one should not be given in general.