It really depends on your type of data. I think all classification algorithms can (and are mostly intended to) do online classification if you have enough training data available, it is just a matter of which one has best performance in your data. Also, whether you want to build a model with training data, save the model, and apply the model to new data; OR alternatively have the classification model trained and applied online... What data are you dealing with? Did you try any algorithms "off line" first?

Most of our data are coming from ad server, e-commerce, or CRM.

Currently, we use offline classification algorithms (click or not, answer or not....) built on a datamart and deployed on the stream (Selective Naïve Bayes, random forest, Neural Networks...).

However, for some applications such as ad targeting, or marketing campaigns:

1/ the contextual information is more important than the past behavior of the users,

2/ the data stream cannot be considered stationary.

Hence, we have to deal with contextual information and with concept drift.

I have tested multi-armed bandits which can solve several optimization problems. I'd like to try online classification algorithms.

for labeling, you can use structural similarity. this methods is underlyng in matching for la form and evolution. the curvature and slope are descriptor.

Roberto,

Thanks for your answer. I will read your paper.