[Objective] Filter out the information the user wants from the data of the offshore platform through technical methods, and can recommend it to users in a timely and accurate manner.[Methods] Screen the candidate set through content-based recommendation algorithm and item-based collaborative filtering algorithm, and use parallel MapReduce to improve the system's parallel data mining ability; use machine learning algorithms to improve the accuracy of recommendation candidates and truly achieve precision for users Matching, personalized recommendations.[Results] A recommendation list can be effectively generated based on the article clicked by the user. The model evaluation accuracy is 78.5%, and the root mean square error is 0.22.[Limitations] The user features and text features need to be deeply mined; the word segmentation tool is used many times during the experiment, and its accuracy depends on it. The model training algorithm needs to be optimized.[Conclusions] According to the accuracy of the experimental results, the accuracy of the model basically meets the personalized recommendations for users, and can provide good support for platform construction.