With more and more users watch live and on-demand content through set-top boxes, internet TV has become one of China Mobile's main home internet services. Meanwhile, the problems of video freezing and blurred screen have increasingly dominated the user complaints. Because of the large number of network elements involved in Internet TV service, traditional quality inspection methods which mainly consist of manual operation and semiautomatic procedures failed to find abnormalities quickly and accurately. In this paper, we collect set-top box active users and its alarm events through China Mobile's soft probe monitoring system and accurately detect the quality degradation abnormalities in TV service by an Informec-based model, the precision rate of anomaly detection reaches over 80 %, and the accuracy rate reaches over 90 %.