With the development of society and technological progress, people pay more and more attention to the inheritance, protection and promotion of cultural heritage. Under the conditions of new technology and new media, communication media and channels are changing rapidly. Improving the speed, breadth and accuracy of the dissemination of information on Guangdong culture in the new era is a new requirement of the times for the dissemination of Guangdong culture. In order to effectively improve the accuracy of Guandong culture recommendation, the study first built a human-machine interactive visual communication platform. Among them, the random forest algorithm was improved by incorporating word embedding technology to facilitate the acquisition of semantic similarity. In addition, it was combined with long-short feature embedding technology and a lightweight gradient boosting tree model, which enabled the acquisition of user interest preferences. The results showed that the model classification accuracy after combining long short-term feature embedding techniques reached 95%, which could better capture the time series features of user behavior. The platform built reduced the risk of overfitting and improved the generalization ability of the model through the integration of multiple decision trees in random forest. Light gradient boosting machine performed excellently and had high computational efficiency when processing large-scale data through an efficient gradient boosting framework. The platform thus built had a recommendation accuracy rate of 97%. At the same time, the platform obtained personalized and similar content recommendation test sections, and the test results were all passed. This proves that the platform can not only accurately recommend Guandong cultural content, but also provide a good communication experience through visualization, providing new ideas and methods for research and practice in the field of Guandong cultural communication.