Given a sequence of n items associated with a user: (item1, item2, ..., itemn) such that each item is represented by k features. This sequence could represent films watched by a user over time. To detect the most significant changes in a sequence of items, many approaches have been developed (e.g. the Cumulative Sum Control Chart (CUSUM)). However, I do not find a study (or a recent survey) comparing the proposed approaches. What's the state of the art ?
Thanks in advance for your replies !
I hope this paper helps you:
http://dl.acm.org/citation.cfm?id=2674031
Hello,
You are looking for surveys on Concept drift, in predictive analytics and machine learning, the concept drift means that the statistical properties of the target variable, which the model is trying to predict, change over time in unforeseen ways. I would suggest one of João Gama 's papers. Dr Gama is one of the well known researchers in the change detection field; you can find what you'r searching for here :
http://dl.acm.org/citation.cfm?id=2523813
http://arxiv.org/pdf/1010.4784.pdf
http://www.iosrjournals.org/iosr-jce/papers/Vol17-issue1/Version-2/D017122026.pdf
best,
Thank you @mehdi for the references. However, the data I am dealing with is unsupervised: there is no target variable. I would be glad if you have some recent references on the problem of change detection in sequential unlabelled data.
- Badges
- Science topic