How can the results of affinity propagation be evaluated for my problem?
You can use the resulting net similarity (objective function) to evaluate the quality of the clustering solution; it is the sum of all similarities between non-exemplar data points and their exemplars plus the sum of all exemplar preferences. This is an indicator only for how well the objective function has been maximized given the similarities and may or may not correspond closely to your evaluation criteria. If you measure or compute the input similarities so that maximizing the net similarity corresponds closely to your true objective, then affinity propagation should do a good job. (You can compare it to other algorithms, such as k-centers clustering, which try to maximize the same objective function.
You can use the resulting net similarity (objective function) to evaluate the quality of the clustering solution; it is the sum of all similarities between non-exemplar data points and their exemplars plus the sum of all exemplar preferences. This is an indicator only for how well the objective function has been maximized given the similarities and may or may not correspond closely to your evaluation criteria. If you measure or compute the input similarities so that maximizing the net similarity corresponds closely to your true objective, then affinity propagation should do a good job.
