Infer information from Tweets. Useful for human-centered computing tasks, such as sentiment analysis, location prediction, authorship profiling and more!
We could allow the user to report an erroneous classifications and to give the correct label on the result page.
We could provide a list of the other possible classes as links, which when clicked would tell the server to log the query and the label that the user assigned. For logging, the serve should be responsible for storing the currently assigned label, rather than providing that as a client side form field (to try to prevent places for abuse).
Assuming we have implemented self-training and we immediately train on the instance, if we get a correction from the user then we need to unlearn that instance. Therefore, we need the classifier to provide something like an untrain or forget method that will undo the results of training on that document.
We could allow the user to report an erroneous classifications and to give the correct label on the result page.
We could provide a list of the other possible classes as links, which when clicked would tell the server to log the query and the label that the user assigned. For logging, the serve should be responsible for storing the currently assigned label, rather than providing that as a client side form field (to try to prevent places for abuse).
Assuming we have implemented self-training and we immediately train on the instance, if we get a correction from the user then we need to unlearn that instance. Therefore, we need the classifier to provide something like an untrain or forget method that will undo the results of training on that document.