Recent advances in deep learning have sparked considerable interest in privacy computing. However, obtaining the features of private data is a significant challenge, and it is still difficult to select appropriate privacy computing strategies even if private data features and usage scenarios are given. To tackle these challenges, we introduce a PCSR model that employs mutual information maximization to assess the importance of data features for each privacy computation strategy and determine the relevance of usage scenarios. First, we introduce the MIM model to improve the representation of the privacy computation strategy. Second, we introduce a value extractor module to enhance the representation of data features. Finally, we propose a classification model based on the encoder-decoder architecture to classify privacy computation strategies. The experimental results indicate that our model outperforms other deep learning models, like DistilBERT and XLNet, by an average of 5%.