0

I have a directed graph expressed as (follower,followed) and would like to try to use Mahout to get recommendations for new edges in the graph (aka predict new users that are likely to be followed). This is the code I started from:

mahout recommenditembased --input input.txt --output output.txt -s SIMILARITY_TANIMOTO_COEFFICIENT --booleanData -n 2

This seems to produce an output where all recommendations get an equal score:

2 [4:1.0,3:1.0]
3 [2:1.0]
...

Is there a way to get recommendation scores on a continuous scale, so that I could rank a follower's recommendations based on the score? Should I try a different similarity metric?

ADJ
  • 4,892
  • 10
  • 50
  • 83

1 Answers1

1

That should give weighted results between 0 and 1 for sorting so your results are all very highly ranked. Try SIMILARITY_LOGLIKELIHOOD, which is almost universally better and gives a different type of weighting, still normalized to fit between 0 and 1.

pferrel
  • 5,673
  • 5
  • 30
  • 41