Supervised learning algorithm that can be easily retrained with new data I have a web crawler and i want to be able to differentiate a specific class of website (social networks), from others.
My problem is that my starting classified data is really small. What I wan't to do is: 


*

*initially train my algorithm on this data;

*receive feedback from user (correct, wrong) on classified websites and use this to increase my data set and retrain my algorithm.


Is this feasible, or is my new data biased by having been previously already selected as a social network website by my algorithm?
Which supervised learning algorithm would be ideal for being retrained after acquiring new data?
Consider that the features of the website I was considering to give the algorithm for training was something like, number of specific html tags (like a [anchor tag], or img) and number of occurrence of some specific words. (If you have suggestions of better features I would be glad to hear also)
 A: I'm going to answer your question with a qualified yes, you can do that.
Here is the problem, you really aren't doing anything except for collecting more data, and hopefully automating some of your coding of that data. I think what you really need to do is just invest the time and energy into collecting more data, and then training your model with that data.
In terms of collecting your data through the way that you described, you have really only one option, which is to go through the classifications that your algorithm produced, and correct misclassifications by hand. Then retrain your data. Probably, the easiest is to just re-run the training of the algorithm on the larger dataset, but you could set windows task scheduler or cron jobs to run it for you while you sleep.
Remember, YOU MUST correct the data before re-training the algorithm. If you don't the algorithm is just going to get worse and worse after each iteration because it is going to think that the incorrect classifications that it gave before are correct. So make sure that you go through and CORRECT MISCLASSIFICATIONS beforehand.
