Abstract: Clothing has become one of the important commodities in online shopping, and the realization of accurate clothing recommendation meeting user’s personalized aesthetic has become a popular research topic. Aiming at the problem that the fine-grained interest features extracted from users are not comprehensive, which leads to the low accuracy of the recommendation system, this paper proposes a clothing recommendation algorithm integrating long-term and short-term preferences. To solve the problem of sparse and single data leading to low personalization and low diversity of recommendation results, this paper used cross-modal data and attention mechanism to make the model learn more accurate and differential user characteristics. On the real dataset of Clothing Shoes and Jewelry, performance of the designed model (PCR) was compared with the classical recurrent neural network, the MF-BPR model and the TARMF model. The key performance evaluation indexes of the PCR model were improved in NDCG, Precision@K and Recall@K. The experimental results show that the model is feasible and effective in the clothing recommendation.