You can gather statistics for users, like "bans", "flame", "neg attitude". And evaluate devariation of it's grades from others using deviance/expectation statistics method. And even forecast what they will say in future applied cutted SSA-algorithm with zeroed singular values. There is tons of method, every you can apply on. More actions they did then more you can expect for them, even their political relivance.

Thank you  Mr.Ivan N. Filippov .

Mostly the user evaluation is fully based on their previous interactions. I want to know how the provider identifies the malicious user and the normal user when he login's for the first time?Is there any technique?

mechanism of trust in clouds is vague. I think so far trust is not addressed at all. To have proper trust we should have proper monitoring of both sides (providers as well as users). Infact, trust should always be on both sides.