This research work explores how noise in web data is currently addressed. We establish that current research works eliminate noise in web data mainly based on the structure and layout of web pages i.e. they consider noise as any data that does not form part of the main web page. However, not all data that form part of the main web page is of a user interest and not every data considered noise is actually noise to a given user. The ability to determine what is useful from noise data taking into account dynamic change of user interests has not been fully addressed by current research works. We aim to justify a claim that it is important to learn noise prior to elimination, not only to decrease levels of noise but also reduce the loss of useful information otherwise eliminated as noise. This is because if the process of eliminating noise in web data is not user-driven, the interestingness of web data available to a user will not reflect their interests given the time of the request.