we build online models for the auction fraud moderation and detection system designed for a major Asian online auction website. By empirical experiments on a realword online auction fraud detection data, we show that our proposed online probit model framework, which combines online feature selection, bounding coefficients from expert knowledge and multiple instance learning, can significantlyimprove over baselines and the human-tuned model. Note that this online modeling framework can be easily extended to many other applications, such as web spam detection, content optimization and so forth. Regarding to future work, one direction is to include the adjustment of the selection bias in the online model training process. It has been proven to be very effective for offline models .