In this paper, the knowledge based on decision tree and user context is combined to give a financial product portfolio recommendation method based on matrix decomposition. This project aims to improve the recommendation accuracy of personal finance products. Then, the matrix decomposition method of scenario preference is introduced to obtain a preliminary list of financial product suggestions. The initial recommendation list is filtered twice according to the preferences of users in specific situations obtained by the classification model to get the final recommendation result. For each person's different choices in a specific situation, the initial recommendation list is filtered in a second round to obtain the final recommendation result. Compared with the conventional collaborative filtering algorithm, matrix decomposition algorithm and Baseline algorithm, the situational cognition improves the recommendation accuracy. This algorithm improves the performance of the recommendation system.