Reciprocal Ranking Fusion

This is an idea where the ranking from multiple systems are fused together to get a better ranking.

Why use reciprocal?

When we've multiple ranking systems, just adding the ranks from multiple systems directly is misleading.

Also note that, reciprocal is just one of the ideas. If we minus the ranks with 100 will also give the same kind of results.

Why the parameter RRF_K?

Without RRF_K, the reciprocal values doesn't have smooth values. The values have stark difference. By using a higher value for RRF_K, we get more smoother rankings.