It is absolutely a client issue. For a while (before their API limits became an issue), I dealt with Twitter with a Bayesian network, and it took very little training to massively improve the experience.