The goal of this paper is to detect depression based on the posts on social media. The dataset combined both tweet dataset and Weibo dataset scraped from the social media. To classify emotions, researchers have been using traditional models such as Bagging, Support Vector Machines, Decision tree, Multinomial Naïve Bayesian and K-nearest neighbor. In this paper, K- nearest neighbor is chosen based on the better precision in result. The main challenge is Chinese context translation and complexity of the context of the post. Finally, an UI page is designed to complete the mission to input a Weibo ID and output of the depression classification. Our approach achieves relatively higher quality results compared to the previous models in literature, while combining depression detection with the real-time social media posts. With this system, we demonstrate the practicability of our project by predicting the depression situation of Internet users through the model, to bring help to the depressed people or people with potential depression tendencies in society.