Feature engineering is one of the key technologies in the research of search advertising recognition. Most of the existing search advertising methods are selected according to the prior knowledge, which is too subjective to be popularized. Taking the advertisement of Ali search advertising as the research object, a feature processing method based on the pre-analysis of store and user data is put forward, and then the conversion rate is predicted with XGBoost. Experiments show that compared with other priori Feature Engineering, the proposed method can significantly improve the prediction results.