In service-oriented systems, how much a service can be trusted is increasingly crucial for service consumers to make the best decision. Because the methods for deriving trust based on manually assigned feedback cost much time and suffer several drawbacks, automatic trust calculation is the only feasible method for large-scale service-oriented applications. Therefore an automatic trust calculation using other non-trust quality criterion values is proposed. To make the calculation accurate, the Kalman Filter is employed to filter out malicious non-trust values instead of directly filtering out malicious trust values. Furthermore, an improved algorithm is proposed by taking the relationship between the non-trust criterion value and its variance into account to offer higher detection accuracy. Although malicious data can be filtered out, dishonest or inaccurate values can still influence trust values. Hence similarity between consumers is used to weight the values from others. Finally, experiments are carried out to access the validation and robustness of our model. The results show that our algorithm can offer higher detection accuracy under more strategic malicious situations.